TESIS DOCTORAL

El vapor de agua atmosferico sobre laPenınsula Iberica: validacion y efecto

radiativo

Javier Vaquero Martınez

Programa de Doctorado en Modelizacion yExperimentacion en Ciencia y Tecnologıa (R007)

2021

TESIS DOCTORAL

El vapor de agua atmosferico sobre laPenınsula Iberica: validacion y efecto

radiativo

Javier Vaquero Martınez

Programa de Doctorado en Modelizacion yExperimentacion en Ciencia y Tecnologıa (R007)

Conformidad del Director:

La conformidad del director de la tesisconsta en el original en papel de esta

Tesis Doctoral.

Fdo: Manuel Anton Martınez

2021

Agradecimientos

Es peligroso, Frodo, cruzar tu puerta.Pones tu pie en el camino y si no cuidas tus pasos

nunca sabes hacia donde arrastraran.Bilbo a Frodo en El Senor de los Anillos de J.R.R. Tolkien.

Al igual que el hobbit Frodo, hace unos anos cruce mi puerta y comenceeste camino que es el doctorado. Y como a Frodo, mis pasos me han llevadopor caminos que nunca hubiera imaginado. Tambien, como el, he contado conla ayuda de muchas personas que me han ayudado a completar mi tarea, lacual, a pesar no ser mas que una gota en el oceano del conocimiento cientıfico,no hubiera podido llevar a cabo por mı mismo.

Tengo que empezar agradeciendo al Grupo AIRE que me acogiera y ani-mara para comenzar esta andadura por la investigacion cientıfica. Gracias ala financiacion de la Junta de Extremadura y los fondos FEDER (ayuda agrupos GR15137) me fue posible dar mis primeros pasos en el mundo de lainvestigacion cientıfica como tecnico de apoyo a la investigacion en este Grupo.La Universidad de Extremadura fue la siguiente que financio mi investigacion,a traves del Programa Propio de Iniciacion a la Investigacion (Accion II). LaFundacion Tatiana Perez de Guzman el Bueno financio mi contrato predocto-ral EPL03636 durante unos meses. La Junta de Extremadura y el Fondo SocialEuropeo financiaron mi contrato predoctoral PD18029, durante un ano y me-dio aproximadamente. Agradezco toda esta financiacion, que ha hecho posibleque la investigacion no sea mi hobby sino mi profesion. Tambien agradezco alGrupo AIRE, la Junta de Extremadura y los fondos FEDER (ayuda a gruposGR18097) mi corta estancia en el Centro Aeroespacial Aleman (DLR), que setrunco por esta pandemia que tanto dolor esta causando en el mundo. Graciasa todo el grupo del Instituto de Teledeteccion (IMF) de allı que me acogio,

y en especialmente a Diego Loyola, mi supervisor, y a Vıctor Molina, que setomo muchas mas molestias de las necesarias para ayudarme a acomodarmeallı. Una vez mas quiero agradecer al Grupo AIRE toda la financiacion inver-tida en que pueda formarme y difundir mi investigacion en cursos, talleres ycongresos.

Una parte fundamental de esta tesis son aquellas personas con las que hecompartido coautorıa en los distintos artıculos que hemos publicado. No citareaquı a todos ya que estan nombrados en el Apendice A, pero quiero hacer espe-cial mencion al Grupo de Optica Atmosferica, de la Universidad de Valladolid.Sin duda los resultados de esta tesis no podrıan haberse obtenido sin vuestracolaboracion. Ademas, los datos que he utilizado en esta tesis estan disponiblespor las distintas instituciones que los generan o los gestionan, y por tanto lesagradezco que pongan a disposicion de la comunidad cientıfica de forma senci-lla y sin coste. Igualmente estan nombrados en los artıculos correspondientes.Tambien quiero agradecer la disponibilidad de todo el software libre que heutilizado, que ha sido mucho, variado y de gran calidad. En su mayorıa hansido paquetes de R, pero tambien el modelo SBDART, el sistema operativoGNU y el nucleo Linux en varias de sus distribuciones, LATEX, el programaGNU Parallel y otros.

No puedo seguir estos agradecimientos sin mencionar a las personas queforman el Grupo AIRE. El ambiente del grupo y la cercanıa de los profesoresmultiplican, segun mi experiencia, la productividad del doctorando. A Manuel,mi director de tesis, tengo que agradecerle su apoyo constante y su trato tancercano. Es alguien de quien es muy facil aprender. A Agustın jamas podreagradecerle lo suficiente todas las cosas que me ha ensenado y todos los con-sejos que me ha dado, ası como todas las cosas que hace por los que estamosempezando sin que nosotros siquiera sepamos que las necesitamos. Mis com-paneros “juniors” tambien han sido una fuente inagotable de consejos y ayuda.Las noches de Pikando, las Cenas de Empresa y las Jalas Domingueras recon-fortan de cualquiera de los sinsabores de la Ciencia: gracias a Ale, Vıctor ylos demas por ellas. Ademas, tengo mucho que agradecer a mi hermano Josey a Maricruz, que siempre tienen una mano tendida y un consejo cuando losnecesitas.

El apoyo de la familia es fundamental en esta etapa, y yo tengo la suertede tener cerca a la mayorıa de la mıa. Gracias a mi padre, a Yolanda, a Jose yMaricruz (¡otra vez!), a Marıa y Luis, y a Marta por vuestro apoyo y carino.A mi madre tambien le agradezco toda la educacion y el carino que me dio,los cuales, aunque ya no este, los llevo siempre dentro. Y no me olvido de missobrinos Ana, Alejandro, Maricruz y Clara, que son una fuente inagotable dealegrıa.

Mis amigos de siempre Pilar y Javi, y Espe y Flori han estado conmigo enlos momentos mas duros de esta etapa. Gracias por haber estado ahı y por

haberme escuchado cuando lo necesitaba.Tambien quiero dar las gracias a Carmelo y al resto del equipo de ajedrez

Cırculo Pacense por los buenos momentos compitiendo a pesar del poco tiempoque habitualmente tengo para entrenar. Y a Teresa y al grupo de encuentro deClaves por todo el aprendizaje a nivel personal, que permiten a uno alcanzaruna perspectiva de la vida que da mayor calma y consciencia.

No puedo terminar los agradecimientos sin mencionar a la Asociacion deDoctorandos de la Universidad de Extremadura (ADUEx). Ha sido un catali-zador de muchas cosas buenas que han pasado en mi vida y me han permitidoconocer a personas maravillosas, a algunos de cuales puedo llamar hoy amigos.Gracias al grupo de Badajoz, que son personas con las que se enriquece unosolamente de tener una conversacion, dispuestas a ayudarte en aquello quenecesites. Con ellos puedo compartir cualquier dramita de los que ocurren eneste mundo de la investigacion y que solamente los que estamos en el entende-mos. Me aportais mucho cada dıa. Y gracias especialmente a Guada, que meaguanta casi a diario y encima no se cansa.

A mi familia.

Indice general

Capıtulo 1: Introduccion . . . . . . . . . . . . . . . . . . . . . . . . . 1

Capıtulo 2: Datos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.1. GNSS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.2. Radiosondas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.3. Productos satelitales . . . . . . . . . . . . . . . . . . . . . . . . 12

2.3.1. GOME-2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.3.2. MODIS . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.3.3. OMI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.3.4. SEVIRI . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.3.5. SCIAMACHY . . . . . . . . . . . . . . . . . . . . . . . . 142.3.6. AIRS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.4. Otros datos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

Capıtulo 3: Metodologıa . . . . . . . . . . . . . . . . . . . . . . . . . 173.1. Validacion de productos de vapor de agua . . . . . . . . . . . . 173.2. Efecto radiativo del vapor de agua . . . . . . . . . . . . . . . . . 18

Capıtulo 4: Resultados y discusion . . . . . . . . . . . . . . . . . . 214.1. Validacion de productos de vapor de agua . . . . . . . . . . . . 214.2. Efecto radiativo del vapor de agua . . . . . . . . . . . . . . . . . 26

Capıtulo 5: Conclusiones . . . . . . . . . . . . . . . . . . . . . . . . . 29

Bibliografıa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

Apendice A: Artıculos . . . . . . . . . . . . . . . . . . . . . . . . . . 41A.1. Artıculo 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43A.2. Artıculo 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55A.3. Artıculo 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69A.4. Artıculo 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79A.5. Artıculo 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

A.6. Artıculo 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

Apendice B: Abreviaturas y acronimos . . . . . . . . . . . . . . . .113

Apendice C: Erratas de Artıculos . . . . . . . . . . . . . . . . . . .115

Resumen

El vapor de agua es un compuesto atmosferico de gran importancia en elsistema climatico. Es el mayor absorbente de luz infrarroja y, como conse-cuencia, supone una retroalimentacion positiva para el calentamiento global.A pesar de ello, su alta variabilidad, tanto espacial como temporal, lo hacendifıcil de estudiar. Existen muchos tipos de instrumentacion capaces de medirel vapor de agua, cada uno con sus particularidades, ventajas e inconvenien-tes. Por ello, es muy importante hacer validaciones y comparaciones de unosinstrumentos con respecto a otros, para mejorar los productos de vapor deagua, asegurar su calidad y elegir en cada momento las medidas mas adecua-das. Entre las medidas en tierra, destacamos el radiosondeo, que permite unamedicion directa y tradicionalmente utilizada como referencia. No obstante,en las ultimas decadas se ha comenzado a utilizar receptores de los sistemasglobales de navegacion por satelite (GNSS, de los cuales el mas conocido es elsistema de posicionamiento global, GPS) para la medicion del vapor de agua,a traves de la obtencion del retraso troposferico, permitiendo una altısimaresolucion temporal y con una calidad excelente. Por otro lado, las medidassatelitales resuelven un problema habitual de los receptores GNSS y las radio-sondas: la resolucion espacial. Los instrumentos satelitales permiten observartodo el globo cada dıa, llegando a lugares donde no existen redes de GNSS oradiosondeo. Los instrumentos satelitales utilizan tecnicas de teledeteccion decierta complejidad y, por tanto, merece la pena cuantificar la calidad de susproductos de vapor de agua. Esta tesis valida datos de GNSS con respecto aradiosondas para determinar su calidad. Una vez completado este estudio, seutilizan como referencia para compararlos con medidas de distintos satelitescuyos productos de vapor de agua son habitualmente utilizados. Ademas, seplantea el estudio del efecto radiativo del vapor de agua en la Penınsula Iberi-ca mediante datos de receptores GNSS y un modelo de transferencia radiativatanto en onda corta como en onda larga, analizando ademas las tendencias delas series temporales obtenidas.

Capıtulo 1

Introduccion

Nada podrıa ser mas simple que una molecula de agua,pero nada es tan complejo en su comportamiento

En Nature’s Building Blocks: An A-Z Guide to the Elements, de John Emsley.

El vapor de agua es un constituyente gaseoso de la atmosfera que, a pesarde encontrarse en una proporcion relativamente pequena, tiene un papel muyimportante en el sistema climatico. En particular, un papel muy relevante enel transporte de energıa, el ciclo hidrologico o efecto invernadero (Myhre et al.2013).

La molecula de agua esta formada por dos atomos de hidrogeno unidosa uno de oxıgeno, formando los dos enlaces H-O un angulo de 104.45◦ entresı, como se muestra en la Figura 1.1. Dichos enlaces confieren al agua unaserie de caracterısticas que explican su gran importancia como componenteatmosferico.

Figura 1.1. Esquema de la molecula de vapor de agua.

2 Capıtulo 1. Introduccion

Por un lado, estos enlaces son muy polares y dan lugar a puentes dehidrogeno entre el oxıgeno de una molecula y el hidrogeno de otra. Esto haceque, a las temperaturas que se dan en la Tierra, exista agua en estado vapory lıquido, y en estado solido en los lugares mas frıos. Por tanto, la cantidadde vapor de agua en la atmosfera esta modulada en gran medida por las tem-peraturas: las bajas hacen que el vapor de agua condense, mientras que lasaltas provocan una mayor evaporacion de las fuentes de agua lıquida dispo-nibles. Por esta razon, el vapor de agua se encuentra mayoritariamente en latroposfera, con una mayor concentracion en sus capas mas bajas. Ademas, alser el puente de hidrogeno un enlace muy energetico, el calor latente del aguaes muy elevado, ocasionando el transporte de energıa que hemos citado ante-riormente: el agua se evapora a latitudes medias y se transporta en forma devapor a latitudes mas altas y frıas, condensandose y liberando el calor latenteasociado a este cambio de fase.

Figura 1.2. Esquema del efecto invernadero. Fuente: Robert A. Rohde, Wiki-media (https://commons.wikimedia.org/w/index.php?curid=38969784).

Ademas, el enlace H-O tambien es responsable de la estructura del espectrode absorcion y emision del vapor de agua. La Figura 1.3 muestra la absorciondel vapor de agua en el espectro de onda corta (SW, 0.2−4.0 µm, arriba) y enun rango mas amplio (1.0− 30.0 µm, abajo). Dicho espectro hace que el vaporde agua sea el componente atmosferico que mas energıa infrarroja absorbe yre-emite hacia la Tierra. El fenomeno mediante el cual la radiacion solar entraen la Tierra, calentandola, con mayor facilidad de la que el planeta es capaz deliberar energıa en forma de radiacion infrarroja, enfriandose, se conoce comoefecto invernadero. Dicho fenomeno permite que la Tierra mantenga tempera-turas que la hacen habitable para los seres vivos. La diferencia entre la energıaradiativa que entra en el sistema tierra+atmosfera y la que sale de este es

3

llamada balance radiativo y es de gran interes conocer el papel de cada com-ponente atmosferico en este balance. Se conoce como forzamiento al papel quejuegan los distintos componentes de la atmosfera en el balance radiativo, es de-cir, esta definicion indica que el forzamiento de un componente atmosferico esla diferencia entre el balance radiativo real y el que tendrıamos si dicho compo-nente desapareciese instantaneamente (sin dar tiempo a que hubiera ningunaotra modificacion del estado del sistema). Esto permite evaluar los efectos quepuede tener la emision del componente considerado. Sin embargo, el efecto delvapor de agua en el balance radiativo no suele considerarse un forzamiento, yaque las emisiones de vapor de agua que pueda producir la actividad humanano se mantienen en la atmosfera. Como hemos comentado anteriormente, deacuerdo a la temperatura la atmosfera podra aceptar una cantidad mayor omenor de vapor de agua, condensandose el exceso. Una excepcion es el vapor deagua estratosferico, ya que en esta capa sı que se mantiene durante un tiempolargo (Forster y Shine 2002; Smith et al. 2001; Zhong y Haigh 2003). Debidoa esta capacidad para calentar la Tierra y a la posibilidad de cambiar de es-tado, el vapor de agua provoca una retroalimentacion positiva en el sistemaclimatico (Colman 2003, 2015).

0

500

1000

1500

2000

1 2 3 4

Longitud de onda (µm)

Irra

dian

cia

(W m

−2)

Tope de la Atmósfera

Atmósfera seca 'midlatitude summer'

Atmósfera 'midlatitude summer'

Figura 1.3. Arriba: Irradiancia espectral descendente que llega a la superfi-cie terrestre en una atmosfera seca y otra con vapor de agua. La diferen-cia (region gris) representa la absorcion del vapor de agua. Como referen-cia, en negro se muestra la que llega al tope de la atmosfera (unicamen-te debida al sol). Realizacion propia a partir del modelo SBDART. Aba-jo: absorcion en terminos porcentuales del vapor de agua y el dioxido decarbono en un rango de longitudes de onda mas amplio. Fuente: NASA(https://earthobservatory.nasa.gov/features/EnergyBalance/).

4 Capıtulo 1. Introduccion

A pesar del importante papel que desempena el vapor de agua en laatmosfera, existe una gran incertidumbre en la cuantificacion de su efectoradiativo. Aunque se han hecho algunos esfuerzos en estudiarlo para laSW (Di Biagio et al. 2012; Golovko 1999; Haywood et al. 2011; Kawamotoy Hayasaka 2008; Mateos et al. 2013; Obregon et al. 2018; Roman et al.2014), las referencias en la literatura a los efectos en la onda larga (LW) sonescasas (Firsov et al. 2015; Garcıa et al. 2018; Huang et al. 2007; Soden et al.2002). Estos trabajos fundamentalmente se centran en la retroalimentacion ysensibilidad del sistema climatico al vapor de agua y la radiacion de LW, perosin considerar el balance radiativo.

Aunque son habituales distintas medidas del vapor de agua (humedad ab-soluta o relativa en superficie, perfil del vapor de agua, . . . ), esta tesis se centraen el vapor de agua integrado (IWV), que equivale a integrar el perfil de con-centracion de vapor de agua en toda la columna vertical (unidades de densidadsuperficial, habitualmente kg mm−2 o g cm−2). Tambien equivale a tomar todala columna atmosferica con una cierta seccion, hacer precipitar todo el vaporde agua sobre un recipiente con la misma seccion y medir la altura que alcanzael agua en dicho recipiente. Por esta razon, a menudo se conoce esta cantidadcomo vapor de agua precipitable (PWV) y suele medirse en unidades de lon-gitud (mm o cm). Notese que las medidas de densidad superficial y longitudestan relacionadas por la densidad del agua ∼ 1 g cm−3.

Ademas, el vapor de agua presenta una gran variabilidad y, en particular,en la Penınsula Iberica se ha estudiado tanto su ciclo diurno (Ortiz de Galisteoet al. 2011) como anual (Bennouna et al. 2013). El mınimo diario se encuentraentre las 04:30 y las 05:30 h en tiempo universal coordinado (UTC), mientrasque el maximo se encuentra mas extendido segun la localizacion, pero siempreen la segunda mitad del dıa, siendo la variabilidad durante el dıa entre 0.41y 1.35 mm. Estacionalmente, el ciclo diurno durante el invierno es similar entoda la Penınsula, mientras que en verano las diferencias locales son mayores.Tambien en verano el ciclo es mas marcado (amplitud de 1.34 mm), y masdebil en primavera (amplitud de 0.66 mm). Ademas, en general, la zona de lacosta mediterranea tiene valores de IWV mayores que la costa atlantica o elinterior. Respecto al ciclo anual, los valores de IWV mas bajos se encuentranen invierno (∼ 10 mm) y los mas altos en verano (∼ 30 mm). Hay un clarogradiente norte-sur y tambien se observa un patron singular en las estacionesdel sur: muestran un mınimo local en el mes de julio. Una de las dificultadesdel estudio del vapor de agua es esta gran variabilidad tanto espacial comotemporal.

Entre las numerosas tecnicas que permiten su medicion, no existe ningunaque permita captar con suficiente resolucion los cambios del vapor de agua

5

atmosferico de forma global. Una primera clasificacion de dichas tecnicas puedeconsiderar si los instrumentos se encuentran en tierra o se basan en medicionesde satelites que orbitan alrededor del planeta. Entre los equipos en tierra seencuentran los radiometros de microondas (Turner et al. 2007), fotometrossolares (Ichoku et al. 2002), lunares (Barreto et al. 2013), y estelares (Perez-Ramırez et al. 2012), lidar (Turner et al. 2002), sistemas de navegacion globalpor satelite (GNSS, Ortiz de Galisteo et al. 2011) y radiosondeo (Torres et al.2010). Entre ellos destacamos las radiosondas y los GNSS. En particular, lasradiosondas han sido tradicionalmente utilizadas como medida de referenciapor su alta calidad, en parte gracias a que es una observacion muy directa.Consiste en el lanzamiento de un globo-sonda que lleva acoplada una serie desensores (temperatura, presion, humedad, . . . ), los cuales miden los perfiles delas distintas variables conforme el globo asciende. No obstante, el radiosondeotiene un elevado coste economico (ya que las sondas no se recuperan) y portanto solo se lleva a cabo en un numero muy limitado de estaciones y conmuy poca variabilidad temporal (no se suelen realizar mas de cuatro medidasdiarias y es muy habitual llevar a cabo solamente una o dos).

El GNSS mas conocido es el Global Positioning System (GPS) desarrolla-do por Estados Unidos. En la literatura, a menudo se intercambian ambosacronimos. Sin embargo, dado que ya hay operativos otros GNSS, en adelan-te utilizaremos de manera general el termino GNSS. Las medidas de GNSSconsisten en la medicion del retraso que produce la troposfera en la senal queenvıan los satelites a los receptores para su posicionamiento. En el posiciona-miento GNSS, habitualmente se pretende obtener el tiempo de transmision dela senal para obtener la distancia a cada satelite, lo que permite encontrar, portrilateracion, la posicion del receptor. Sin embargo, la senal viene retardadapor diversos efectos (relativista, ionosferico, . . . ) de forma que el tiempo devuelo viene sobrestimado y debe rectificarse. Entre estas correcciones, el retra-so troposferico constituye un ruido para la geodesia y, a la vez, una senal parala meteorologıa. La obtencion precisa del retraso troposferico permite inferirel IWV sobre el receptor GNSS ya que es posible separar el vapor de agua delresto de componentes atmosfericos porque la molecula de agua es la unica conmomento dipolar permanente.

Las medidas de GNSS estan comenzando a utilizarse con asiduidad debidoa sus grandes ventajas: bajo coste, redes cada vez mas densas, medidas decalidad incluso en condiciones adversas para otros instrumentos (nubosidad,precipitacion, . . . ), alta resolucion temporal (tıpicamente entre 5 minutos y 2horas). Tienen un especial potencial para la validacion de instrumentos debidoa estas ventajas: es facil tener coincidencia espacial gracias a su resoluciontemporal y permiten testar otros instrumentos en condiciones de nubosidad

6 Capıtulo 1. Introduccion

o precipitacion (Rohm et al. 2014). Ademas, han sido validados en distintostrabajos (Bokoye et al. 2003; de Haan et al. 2002; Heise et al. 2009; Ohtaniy Naito 2000; Pany et al. 2001; Wang et al. 2007) con buenos resultados.

Por otro lado, las observaciones satelitales se realizan a traves de radiome-tros que se encuentran a bordo de satelites midiendo la radiacion emitida o re-flejada por la superficie terrestre y la atmosfera. Los satelites utilizan metodosde inversion con el fin de obtener informacion sobre el estado de la atmosfe-ra a partir de la medida de radiacion emitida o reflejada. Dependiendo de lainformacion concreta que se quiera obtener y de las longitudes de ondas enlas que mida el radiometro, existen multitud de tecnicas de teledeteccion. Lasmedidas satelitales suelen tener una buena resolucion espacial y habitualmentecubren todo el globo terrestre, pero suelen tener una mala resolucion verticaly temporal (Diedrich et al. 2016). Los productos satelitales de vapor de aguahabitualmente se obtienen a partir de medidas en el infrarrojo (IR, Bennounaet al. 2013), el infrarrojo cercano (NIR, Gao y Li 2008), el visible (Anton et al.2015; Grossi et al. 2015; Roman et al. 2015) y, con menor frecuencia, el mi-croondas (Jones et al. 2009). Todos estos rangos, a excepcion del microondas,tienen problemas cuando se mide en presencia de nubosidad. Sin embargo, laintensidad de radiacion de microondas que puede captar un satelite de formapasiva es muy pequena y, por tanto, tiene otras dificultades.

El primer objetivo de esta tesis es validar los productos de GNSS de vaporde agua tomando el radiosondeo como referencia. Este objetivo se cumplio conla publicacion del Artıculo A.1, en el que se validaron medidas de GNSS conradiosondeos de la red GCOS Reference Upper-Air Network (GRUAN). Unavez se ha asegurado la calidad de las medidas GNSS, se procede a utilizarlascomo referencia en la validacion e inter-comparacion de productos satelitalesen la Penınsula Iberica, lo cual responde al segundo objetivo de la tesis. Suconsecucion llevo a la publicacion de los Artıculos A.2, A.3 y A.4. El prime-ro aborda una intercomparacion de medidas de varios instrumentos satelitalesrespecto a una referencia comun (GNSS). El segundo trata de una validacionen detalle del instrumento satelital Ozone Monitoring Instrument (OMI) conrespecto a GNSS, ya que este es un satelite poco validado hasta el momento.En el Artıculo A.4 se lleva a cabo una validacion en mayor profundidad delinstrumento satelital Moderate Resolution Imaging Spectroradiometer (MO-DIS) con medidas de GNSS como referencia, ya que el producto de vapor deagua de MODIS es ampliamente utilizado. El tercer objetivo se centro encuantificar el efecto radiativo del vapor de agua tanto en SW como LW, ana-lizando ademas su papel en el balance radiativo terrestre. Este tercer objetivodio lugar a la publicacion de dos trabajos, A.5 y A.6. En el Artıculo A.5 seobtiene y analiza el efecto radiativo del vapor de agua en SW en la Penınsula

7

Iberica, mientras que en el Artıculo A.6 se calcula tambien el efecto en LW y seanaliza la serie temporal en las estaciones de la Penınsula Iberica, obteniendoy analizando las tendencias y el papel de los efectos en el balance radiativo.

La estructura de esta tesis doctoral es la siguiente: tras este primer capıtulode Introduccion, el Capıtulo 2 describe los datos utilizados en los distintostrabajos publicados para el desarrollo de la tesis. El Capıtulo 3 explica lametodologıa utilizada para llevar a cabo los objetivos que se plantean en estatesis, mientras que el Capıtulo 4 resume los principales resultados obtenidos.Finalmente, el Capıtulo 5 detalla las conclusiones que se pueden extraer de estatesis. La copia de los trabajos publicados puede encontrarse en el Apendice A.Se ha anadido un ındice de abreviaturas en el Apendice B y una fe de erratasde los artıculos en el Apendicce C.

Capıtulo 2

Datos

¡Datos, datos, datos!¡No puedo hacer ladrillos sin arcilla!

Sherlock Holmes en El misterio de Copper Beeches, de Arthur Conan Doyle.

2.1. GNSSLa metodologıa de obtencion del vapor de agua a partir de datos de GNSS

se describe en Bevis et al. (1992). El posicionamiento GNSS se basa en laobtencion de las distancias entre un receptor (en tierra en nuestro caso) yuna serie de satelites GNSS. Cada satelite GNSS envıa una senal que capta elreceptor. El tiempo de propagacion de la senal permite obtener la distancia realentre satelite y receptor. No obstante, este paso no es trivial, ya que es necesarioevaluar una serie de perturbaciones o retrasos que sufre esta senal, a saber:errores de sincronizacion entre los relojes, correcciones relativistas, retrasosproducidos en la electronica de los instrumentos y los retrasos asociados alpaso de la senal por la atmosfera. En particular, se modela un retraso asociadoa la ionosfera y otro asociado a la troposfera.

El retraso ionosferico se puede eliminar en gran medida mediante el usode receptores de doble frecuencia, los cuales captan (al menos) dos senalesde distinta frecuencia que emiten los satelites GNSS. Las cargas libres de laionosfera hacen que el retraso asociado a la misma sea dispersivo, es decir,

10 Capıtulo 2. Datos

dependiente de la frecuencia. Ası, combinando linealmente estas dos senalesde distinta frecuencia podemos eliminar el retraso ionosferico (dispersivo) ymodelar solamente retrasos no dispersivos.

Por otro lado, se modela el retraso troposferico, que denotaremos comoinclinado (STD) cuando nos refiramos al retraso a lo largo del camino quesigue la senal entre el satelite y el receptor. Sin embargo, lo denotaremos comocenital (ZTD) cuando nos refiramos al retraso en la direccion vertical. Portanto, ZTD se corresponde con el valor que tendrıa el STD si el satelite seencontrase en el cenit del receptor. Ambos valores se relacionan a traves de lasllamadas funciones de mapeo (Niell 2000).

El procesado de los datos de GNSS permite obtener los valores de ZTD paracada estacion y cada instante. De hecho, es habitual que esten disponiblespara su descarga en las paginas web de las redes a las que pertenecen lasestaciones. El ZTD se puede descomponer en la suma de dos retrasos, unorelacionado con la parte no-dipolar de las especies quımicas que componenla atmosfera, conocido como retraso hidrostatico (ZHD) y otro debido a lacomponente dipolar, conocido como retraso humedo (ZWD). La razon de estenombre es que el agua es la unica especie quımica de la atmosfera que tienemomento dipolar permanente.

El ZHD se puede obtener con una precision adecuada mediante la aplica-cion del sencillo modelo de Saastamoinen (1972) que solamente depende de lapresion atmosferica al nivel de la estacion. El modelo se presenta en la Ecuacion(2.1),

ZHD = c1 · P0

1− c2 · cos(2 · λ)− c3 ·H, (2.1)

donde c1 = 2.2779 ± 0.0024 mm, c2 = 0.00266, c3 = 0.00028 km−1, λ es lalatitud y H la altura sobre el geoide a la que se encuentra la estacion. Eldenominador representa una correccion debido al cambio de la aceleracion dela gravedad con la altura y la latitud.

Una vez obtenido el ZHD a traves de la medida de la presion a la altura dela estacion, se puede obtener el ZWD substrayendo al ZTD el ZHD. El ZWDse relaciona directamente con el IWV mediante la Ecuacion (2.2),

IWV = κ · ZWD, (2.2)

siendo κ un parametro que depende de la temperatura media de la atmosferapromediada por el vapor de agua, tambien conocida como la temperatura deDavis (Davis et al. 1985), la cual se define segun la Ecuacion (2.3),

Tm =∫

(Pv/T ) · dz∫(Pv/T 2)dz , (2.3)

2.2. Radiosondas 11

donde Pv es la presion parcial del vapor de agua y T la temperatura, ambasfunciones de la altura z. El parametro κ queda como se muestra en la Ecuacion(2.4),

κ = 106

ρRv [(k3/Tm) + k′2] , (2.4)

donde ρ es la densidad del agua lıquida, Rv es la constante especıfica de losgases ideales para el vapor de agua. La constante k′2 se define como k′2 =k2 −m · k1, siendo m el cociente entre la masa molar del vapor de agua y ladel aire seco, y k1, k2 y k3 las constantes de la refractividad atmosferica.

Los datos para la comparativa de GNSS con radiosondeo de la red GRUANse obtuvieron del procesado que lleva a cabo el Centro de Investigacion Alemande Geociencias (GFZ) de Potsdam, para las estaciones en las que coincidenradiosondeo y GNSS. El procesado GNSS a menudo carecıa de los datos detemperatura y presion (y por tanto de IWV), de modo que para incrementarel numero de datos se calculo la Tm por integracion numerica de los datos deradiosondeo, ası como la presion a la altura de la estacion GNSS.

Los datos de GNSS para la Penınsula Iberica se obtuvieron de una veintenade estaciones de la red Regional Reference Frame Sub-Commission for Europe(EUREF). Puede encontrarse mas informacion de las estaciones en la Figura 1y la Tabla 2 del Artıculo A.2. En las estaciones cercanas a la costa se puedeninducir errores en los algoritmos de obtencion del vapor de agua de los satelites,ya que el pıxel tıpicamente tendra una parte de agua y otra de tierra, lo quehace difıcil modelar su albedo. Por ello, en los Artıculos A.2 y A.3 se consideransolamente nueve estaciones del interior de la Penınsula Iberica. Para el estudiode series temporales (A.6) se eliminaron aquellas estaciones que tenıan huecosimportantes o bien eran series muy cortas.

2.2. RadiosondasLos datos de radiosondas se obtuvieron de la red GRUAN, que proporcio-

na datos para 28 estaciones. Se seleccionaron las cuatro estaciones (LIN, LAU,SOD y NYA, ver Tablas 1 y 2 del Artıculo A.1) que tenıan ademas un receptorGNSS en el mismo lugar. Los lanzamientos de radiosondeo se realizan habi-tualmente a horas especıficas. La estacion LIN suele tener cuatro lanzamientosdiarios (00, 06, 12, 18 h UTC), mientras que NYA normalmente tiene solamen-te datos a las 12 h UTC. Las radiosondas en la estacion SOD se lanzaban alas 00 y 12 h UTC (a veces alguna mas a horas diferentes). Por ultimo, LAUtenıa lanzamientos irregulares (normalemnte un lanzamiento a la semana).

El modelo de las radiosondas usado por la red GRUAN es el Vaisala RS92.Este modelo esta equipado con sensores de temperatura, humedad, presion

12 Capıtulo 2. Datos

y un receptor GNSS. Las fuentes de error principales de este tipo de sondasque pueden afectar al sensor de humedad son el calentamiento del sensor de-bido a la radiacion solar y las correcciones de calibracion que dependen de latemperatura.

Los productos que proporciona la red GRUAN son los siguientes: perfilesde presion, temperatura, humedad, humedad relativa, fraccion de mezcla delvapor de agua, informacion de vientos, radiacion de onda corta y las incerti-dumbres asociadas. El IWV se puede calcular por integracion de la razon demezcla del vapor de agua, como se muestra en la Ecuacion (2.5),

IWV =∫ ps

0WVMR · dp, (2.5)

donde ps es la presion en superficie, WVMR es la razon de mezcla del vaporde agua y p es la presion atmosferica. A este calculo se le impusieron unasrestricciones para asegurar la calidad de los datos:

1. El numero de niveles de altura/presion debıa ser mayor de 15.2. El primer nivel debe tener una altura menor a 1 km.3. El ultimo nivel debe teenr una altura mayor a 9 km.4. El IWV resultante debe tener sentido (entre 0 y 100 mm).

2.3. Productos satelitales

2.3.1. GOME-2Global Ozone Monitoring Instrument - 2 (GOME-2) es un instrumento a

bordo de los satelites MetOp (A, B y C; en esta tesis se utilizaron solamentedatos del MetOp-A). Su principal producto es el ozono, pero tambien aportainformacion sobre gases traza, incluyendo el vapor de agua.

La tecnica para obtener el IWV se conoce como Espectrografıa de Absor-cion Diferencial Optica (DOAS). El algoritmo se encuentra descrito en Wagneret al. (2006, 2003) y tiene tres pasos. El primero es la aplicacion del algoritmoDOAS propiamente dicho en el rango de longitudes de onda 614 − 683 nm,seguido de una correccion por la falta de linealidad de la absorcion. El ultimopaso consiste en la obtencion de la densidad en la columna vertical a partir delfactor de masa optica (AMF) obtenido de la absorcion del oxıgeno molecular.Aunque el oxıgeno molecular tiene un AMF similar al del vapor de agua, noson exactamente iguales y esto puede dar lugar a errores sistematicos. Por ello,en este paso se utiliza una tabla de busqueda para corregir este posible errorsistematico.

2.3. Productos satelitales 13

Una de las grandes ventajas del algoritmo es que no necesita calibracionesexternas y, por tanto, es independiente de otras medidas, ası como de infor-macion a priori.

2.3.2. MODISEl instrumento MODIS se encuentra a bordo de dos plataformas, Aqua y

Terra. Aqua pasa de sur a norte por el ecuador por la tarde, mientras queTerra lo hace de norte a sur por la manana. El producto de vapor de agua seobtiene a partir de pıxeles con resolucion de 1 km × 1 km pero se agrupa en5× 5 pıxeles.

Este instrumento provee datos de vapor de agua con dos algoritmos. Elprimero se basa en radiacion del NIR y por tanto solamente puede utilizarsedurante el dıa. Se utilizan tecnicas de division con dos y tres canales, generandotablas de busqueda para las ratios obtenidas, mediante modelos de transferen-cia radiativa. La cantidad de vapor de agua medida puede ser convertida aIWV teniendo en cuenta la geometrıa solar y observacional. Aunque utilizalos canales 2, 5, 17, 18 y 19, se pueden utilizar otros adicionales cuando haypresencia de nubes. El algoritmo esta explicado en detalle por Gao y Kaufman(1992) y Gao y Li (2008).

Para la noche se utiliza un producto basado en el IR. Este algoritmo ob-tiene a la vez los perfiles de temperatura, humedad, columna total de ozono ytemperatura en superficie. En primer lugar se asocian estadısticamente las ra-diancias en los distintos canales con diferentes perfiles atmosfericos, obtenidosde radiosondas. Esto sirve como una primera aproximacion para comenzar unalgoritmo iterativo basado en la linealizacion de las ecuaciones de transferenciaradiativa. Pueden encontrarse mas detalles sobre este algoritmo en Seemannet al. (2003) y Seemann et al. (2006).

2.3.3. OMIOMI es un instrumento a bordo del satelite Aura, el cual pasa a las 13:30 h

en tiempo local por cada localizacion, de forma sıncrona con el sol. El algo-ritmo usa una ventana en el intervalo de longitudes de onda 430 − 480 nm ytiene tres pasos. En el primero se obtiene la columna inclinada de vapor deagua utilizando un modelo semiempırico que considera varios gases, ası comola correccion de algunos efectos. En el segundo paso, se convierte la columnainclinada a columna vertical, utilizando un AMF obtenido de calculos de trans-ferencia radiativa con tablas de busqueda y, en el ultimo paso, se convierte lasunidades de densidad de columna vertical a unidades de IWV (Wang et al.2014).

14 Capıtulo 2. Datos

Para asegurar la calidad de las medidas es necesario aplicar ciertas restric-ciones. La fraccion de cubierta nubosa (CF) debe ser inferior a 0.1; la presionen el tope de la nube, mayor que 500 hPa; el AMF, por encima de 0.75; elerror cuadratico medio (RMSE) del ajuste, menor que 0.005 y la banderillamaindataqualityflag igual a 0. Los pıxeles afectados por la anomalıa de fila(ver Wang et al. 2014) se han rechazado tambien.

2.3.4. SEVIRISpinning Enhanced Visible and Infrared Imager (SEVIRI) es un instrumen-

to a bordo del satelite Meteosat. SEVIRI cuenta con siete bandas en el IR, enel rango 6.2− 13.4 µm. Entre las cinco bandas utilizadas por el algoritmo, dosde ellas son de absorcion fuerte para el vapor de agua. El algoritmo estima elperfil de temperatura y humedad a partir de las observaciones de la tempera-tura de brillo, utilizando una tecnica de inversion. La solucion generalmente noes unica, de modo que se utiliza un perfil de fondo para constrenir la solucion.Dicho perfil se obtiene a partir de un modelo de prediccion a corto plazo.

Una de las limitaciones del algoritmo es que los productos solamente estandisponibles en condiciones de cielo despejado. En algunos casos, como los cirroso en el borde de las nubes, el algoritmo podrıa no detectar la nube y tratar deestimar el IWV. Sin embargo, en esos casos normalmente el algoritmo falla onecesita un numero inusualmente alto de iteraciones, lo que se marca con unabanderilla. Ademas, las regiones montanosas pueden tener grandes errores sihay diferencias importantes de altura entre la orografıa del modelo y la alturareal.

Siendo el Meteosat un satelite geoestacionario, su resolucion temporal esmuy alta (30 min). Ademas tambien cuenta con una buena resolucion espacial(3 km× 3 km). La resolucion en IWV es de 0.58 mm.

2.3.5. SCIAMACHYSCanning Imaging Absorption SpectroMeter for Atmospheric CHarto-

graphY (SCIAMACHY) es un instrumento montado en el satelite Envisat.Estuvo operativo entre marzo de 2002 y abril de 2012. Notese que los datosde GNSS que tenemos comienzan en 2007, de modo que solo se ha trabajadocon datos de SCIAMACHY desde esa fecha. El satelite pasa por el ecuador alas 10:00 h en tiempo local de cada dıa, en una orbita sıncrona con el sol. Eltamano de sus pıxeles es de unos 60 km× 30 km.

El algoritmo de obtencion del vapor de agua se basa en la tecnica DOAScon masa optica corregida (AMC-DOAS, Noel et al. 2004), utilizando la regionespectral alrededor de los 700 nm. El uso de luz visible hace que solamente

2.3. Productos satelitales 15

se pueda aplicar en momentos con luz solar y en ausencia de nubes. Al igualque con GOME-2, una de las ventajas de este algoritmo es que no depende deinformacion externa y, por tanto, es completamente independiente.

En esta modificacion de la tecnica DOAS se tienen en cuenta los efectosde saturacion de caracterısticas espectrales altamente estructuradas que el ins-trumento no puede resolver. Ademas se utilizan caracterısticas espectrales deloxıgeno molecular en combinacion con el vapor de agua que se ajustan paraobtener una correccion al AMF. Esta correccion trata de representar como deparecidas son las condiciones atmosfericas reales y las modeladas. Por tanto,esta correccion tambien ofrece informacion sobre la calidad del producto.

Por ello las medidas se han filtrado con los siguientes criterios: el angulosolar cenital (SZA) debe ser menor de 88◦ y la correccion del AMF mayor de0.8. Aunque no hay un filtro especıfico para nubes, el criterio de la correcciondel AMF deberıa filtrar la mayorıa de los casos nubosos.

2.3.6. AIRS

El instrumento Atmospheric Infrared Sounder (AIRS) se encuentra a bordodel satelite Aqua. Pasa una o dos veces por la Penınsula Iberica cada dıa,con una resolucion espacial de unos 13 km. El producto concreto utilizado esel “AIRS/Aqua L2 Standard Physical Retrieval (AIRS-only)”, version 6. Elalgoritmo (Barnet et al. 2007) utilizado trata de obtener todos los productosa la vez de forma que satisfagan las observaciones en el sentido de los mınimoscuadrados.

En el algoritmo, las radiancias observadas pasan por una red neuronalpara obtener el estado del sistema, obteniendo los parametros de las nubesy tratando de obtener las radiancias sin nubes. Este proceso se lleva a caboiterativamente y, tras ello, se lleva a cabo un algoritmo fısico de obtencion, conlas radiancias sin nubes y el estado atmosferico como entradas. Tras ello seobtienen parametros de nubes de nuevo y vuelven a eliminarse, obteniendosenuevas radiancias sin nubes en el proceso. El algoritmo tambien elige el tipode superficie y se obtiene el estado final de las variables atmosfericas.

Este producto proporciona una banderilla de calidad (0, 1 y 2). Aquellosdatos con el valor 2 en dicha banderilla se eliminaron, pues no se recomienda suuso. El numero de datos con la banderilla 0, la de mejor calidad y recomendadapara estudios de validaciones in situ, era muy escaso y, por tanto, se decidioutilizar tambien los datos con la banderilla en 1.

16 Capıtulo 2. Datos

2.4. Otros datosPara la validacion de los datos de GNSS respecto a radiosondeo (Artıcu-

lo A.1) se descargaron datos de CF del reanalisis del Centro Europeo de Previ-siones Meteorologicas a Plazo Medio - Interim (ERA-Interim, Dee et al. 2011).

Para la obtencion del efecto radiativo (Artıculos A.5 y A.6) es necesarionutrir al modelo de datos que reflejen el estado de la atmosfera. Para ello seutilizaron datos de ERA-Interim. En particular, se trabajo con medias diariasde columna total de ozono y medias mensuales de albedo de la superficie. Tam-bien se utilizaron datos de insolacion para la seleccion de dıas despejados dela Agencia Estatal de Meteorologıa (AEMet). Para el A.6 se utilizo, ademas,una serie de cubierta nubosa de AEMet. Para modelar la onda larga, se to-maron datos de ERA-Interim de la temperatura en superficie y los perfiles detemperatura, presion, vapor de agua y ozono.

Capıtulo 3

Metodologıa

En esencia, todos los modelos estan equivocados,pero algunos son utiles

George Box en Superficies de respuesta, mezclas y analisis de crestas.

3.1. Validacion de productos de vapor de aguaEn primer lugar, se analizo como se comportan los productos de GNSS

tomando el radiosondeo como referencia. Para ello, se tomaron datos a nivelglobal de estaciones de la red GRUAN en las que ademas de lanzamientos deradiosondas hubiera estaciones de GNSS. Al encontrarse en la misma posicion,solamente se necesita asociar a los datos de radiosondeo (maximo cuatro medi-das al dıa) los datos de GNSS mas cercanos en el tiempo. Se establecio el lımitede 30 minutos de diferencia entre el dato de GNSS y el dato de radiosondeo. Encaso de no haber ningun dato de GNSS en esa ventana de ±30 min centradaen la hora de lanzamiento del radiosondeo, ese dato quedaba descartado.

Para la comparacion de datos satelitales con respecto a GNSS, se aplico elmismo criterio temporal. Sin embargo, tambien era necesario aplicar un criterioespacial: aquellos pıxeles cuyo centro se encontrase mas alla de un radio de100 km de la estacion GNSS eran descartados. En el caso del producto satelitalOMI, se tomo la distancia de un cuadrado de 0.25◦× 0.25◦. En cualquier caso,un estudio anterior (Roman et al. 2015) demostro que la distancia no era un

18 Capıtulo 3. Metodologıa

factor determinante en la validacion de satelites respecto a GNSS, aunquea distancias mayores (del orden de 100 km) sı que se podıa observar unadisminucion en la precision.

La metodologıa seguida para caracterizar las diferencias entre instrumentosconsistio en, una vez casados los datos temporal y espacialmente, calcularla distribucion de diferencias y diferencias relativas como se muestra en lasEcuaciones (3.1) y (3.2):

δ = IWVval − IWVref, (3.1)

δ( %) = 100 · δ/IWVref, (3.2)donde δ es la diferencia en terminos absolutos, δ( %) es la diferencia en terminosrelativos y los subındices val y ref se refieren al instrumento a validar y alutilizado como referencia, respectivamente.

El analisis de la distribucion de dichas diferencias puede proporcionar mu-cha informacion sobre las posibles deficiencias y fortalezas del producto a vali-dar. Tıpicamente se han utilizado el sesgo medio (MBE) para valorar la exac-titud, la desviacion estandar (SD) o el RMSE para valorar la precision. Estosparametros son los mas habituales en la literatura y por ello se han utilizado.Sin embargo, dado que la distribucion de diferencias no suele tener una distri-bucion normal, se ha utilizado de forma mas habitual ındices no parametricos,como la pseudo-mediana (Wilcoxon 1945) o el rango intercuartılico (IQR).Para analizar las dependencias que pueden tener estos ındices con distintasvariables (IWV, SZA, CF, . . . ), se han dividido los datos en bins de dichasvariables y se han obtenido los ındices, lo que permite representar la evoluciondel ındice con dicha variable.

3.2. Efecto radiativo del vapor de aguaDefinimos el efecto radiativo del vapor de agua (WVRE) como la diferencia

entre la la irradiancia neta (descendente menos ascendente) en un determinadonivel (la superficie terrestre o el tope de la atmosfera) bajo unas condicionesatmosfericas reales y la irradiancia neta obtenida en el mismo nivel para unascondiciones ideales de atmosfera seca. La Ecuacion (3.3) refleja esta definicion,

WVRE = (I↓IWV − I↑IWV)− (I↓0 − I↑0 ), (3.3)

en donde I es la irradiancia y el sentido de la flecha vertical en el superındiceindica si la irradiancia es ascendente o descendente, mientras que el subındiceindica si se realiza en una atmosfera real (con vapor de agua, IWV) o en unaatmosfera seca (sin vapor de agua, 0).

3.2. Efecto radiativo del vapor de agua 19

Tambien se calcularon otras variables como la eficiencia radiativa del vaporde agua (WVEFF), definida segun la Ecuacion (3.4), y la tasa de calentamiento(HR), definida en la Ecuacion (3.5).

WVEFF = ∂WVRE/∂IWV (3.4)

HR = ∂T/∂t = g

Cp

∆I∆p (3.5)

En la Ecuacion (3.5), T denota la temperatura, t el tiempo, g = 9.81 ms−2 laaceleracion de la gravedad, Cp ' 1004 J kg−1 K−1 el calor especıfico del aireseco, p la presion y ∆I La irradiancia neta.

El efecto radiativo del vapor de agua se determino para dıas despejadoscon baja carga de aerosol. Para ello, el valor de insolacion diario se dividiopor el valor teorico (tope de la atmosfera) para cada dıa. Se seleccionaronaquellos dıas con un valor de este cociente superior a 0.70. La OrganizacionMeteorologica Mundial (WMO 2008) establece un lımite de 0.75 para filtrarcasos nubosos. Junto a este criterio de umbral de insolacion, en el Artıculo A.6se eliminaron dıas que tuvieran registrados una cubierta nubosa mayor a unaocta.

Todas las irradiancias se calcularon a traves del modelo de transferenciaradiativa Santa Barbara’s Disort Radiative Transfer (SBDART, Ricchiazzi etal. 1998). La irradiancia en SW se calculo en un rango de longitudes de ondaentre las 0.2 µm y las 4.0 µm, con un tamano de paso del 0.50 % y 4 corrientes.Para el modelo de atmosfera se utilizaron las atmosferas que vienen predefi-nidas en SBDART: midlatitude summer para los meses entre marzo y agosto,y midlatitude winter para el resto, modificando los parametros de ozono totaly vapor de agua total con los valores comentados anteriormente. Tambien sedaba al modelo el albedo y para valores de SZA mayores que 90◦ se asigna unvalor nulo de WVRE sin ejecutar el modelo.

El rango de longitudes de onda para el calculo de la irradiancia en LW seestablecio entre 4.0 µm y 100 µm, con un paso de 1 % y 16 corrientes. En vez deusar un perfil de atmosfera predefinido de SBDART, se daba el perfil obtenidode ERA-Interim. A pesar de ello, se seguıa dando el valor de columna totalde ozono y de IWV, de modo que el perfil se re-escalaba de acuerdo a estosvalores. El numero de capas se fijo en 65, con una resolucion de 1 m para lascapas bajas y 900 m para las altas. Se activo la opcion de radiacion termica yse desactivo la radiacion solar.

En cada situacion (estacion e instante determinados) se lanza el modelocuatro veces: dos con vapor de agua (una para SW y otra para LW) y otrasdos con el parametro de vapor de agua puesto a cero, es decir, una atmosferaseca.

20 Capıtulo 3. Metodologıa

Para el estudio de tendencias, se promediaron los datos diariamente y laserie diaria se desestacionalizo. Para ello, se obtuvo en ano tipo (promedio decada dıa del ano para todos los anos) en cada estacion y se resto a cada valordiario el valor del ano tipo de ese dıa. Con los valores diarios desestacionalizadosse calcularon las medias mensuales. Los meses con pocos datos (menos de 5dıas de datos) se calcularon mediante interpolacion lineal. Sobre estas seriesmensuales se utilizo el test de Mann-Kendall (Kendall et al. 1994; Mann 1945)para determinar si la tendencia era significativa o no, mientras que la pendientese obtuvo utilizando el estimador de Sen (1968).

Capıtulo 4

Resultados y discusion

Las ideas no duran mucho.Hay que hacer algo con ellas.

Santiago Ramon y Cajal, recogida en Diccionario de citas, de Luis Senor.

4.1. Validacion de productos de vapor de aguaEl analisis de productos de vapor de agua comenzo con la comparativa del

producto que se considera de mayor calidad, las radiosondas, con el productoque pretendemos utilizar como referencia en otras comparativas, los receptoresGNSS (Artıculo A.1). En primer lugar, se creo un producto de vapor de aguautilizando el ZTD de los receptores (ya procesado por el GFZ de Potsdam), latemperatura media Davis (Ecuacion (2.3)) y la presion de las propias radio-sondas, lo que permitıa ampliar el numero de datos disponibles. Este productose comparo con el procesado por el GFZ de Potsdam y se concluyo que tenıasuficiente calidad como para usarlo en la comparacion GNSS-radiosonda.

Una comparativa general mostro que las diferencias GNSS-radiosonda pre-sentaban errores sistematicos por debajo de 1 mm tanto de media como demediana. Todos estos errores eran negativos, lo cual sugiere que los GNSSpueden tener un pequeno sesgo seco con respecto a las radiosondas. Por otrolado, la precision, evaluada a traves del SD y el IQR, solo superaba 1 mm enla estacion LIN (1.150 mm el IQR y 1.099 mm la SD). Tambien se llevaron a

22 Capıtulo 4. Resultados y discusion

cabo regresiones lineales entre el IWV de ambos instrumentos, mostrando unalto coeficiente de determinacion (R2 ' 0.98) y valores de la pendiente muycercanos la a unidad, ası como ordenadas en el origen del orden de los erroressistematicos mencionados antes.

Tambien se estudio la dependencia de varias variables, en particular: IWV,SZA, presion, CF, movimiento horizontal de la radiosonda. El IWV incrementael MBE y el SD, pero debilmente, de modo que en terminos relativos MBE ySD disminuyen al aumentar IWV. Para distintos valores de SZA, se apreciaun cierto incremento del SD relativo con una bajada brusca en torno a 90◦.El MBE relativo tambien se aleja del cero a medida que se incrementa elSZA, pero se acerca de nuevo al cero tras pasar de esos 90◦. Parte de estecomportamiento puede explicarse como un acoplamiento entre los valores deIWV y los de SZA, es decir, para valores bajos de SZA normalmente el IWVes mayor (debido a la mayor temperatura que supone un radiacion solar masdirecta) y, por tanto, el SD y el MBE relativos tendrıan un valor mas bajo (sinsigno). Tambien era esperable que para valores bajos de SZA las radiosondaspudieran tener un cierto sesgo debido a la radiacion incidente sobre el sensorde humedad (Dirksen et al. 2014; Wang y Zhang 2008). Sin embargo, no seobservo que hubiera grandes diferencias entre el dıa y la noche. Las medidasde la red GRUAN cuentan con correcciones para este tipo de problemas y, portanto, concluimos que estan correctamente aplicadas.

Ademas, se observo que la presion atmosferica tambien afectaba a las dife-rencias entre radiosondas y GNSS. En este caso, se observo un empeoramientode las medidas para presiones altas, lo cual podrıa tener que ver con el modelode Saastamoinen (1972) utilizado para obtener el ZHD a partir de la presionatmosferica. Por otro lado, las altas presiones estan asociadas a IWV menores,de modo que esto podrıa influir adicionalmente en los valores de MBE y SD.

El efecto de la nubosidad tambien se analizo, usando el CF de ERA-Interim.El MBE no mostro ningun patron con esta variable. Por otro lado, aunque elSD tampoco mostro ningun patron relevante, sı se observo una alta variabilidaden las estaciones NYA y SOD (ambas en el Artico).

Por ultimo, se analizo el desplazamiento total de las radiosondas, ya queesto podrıa ser una causa de error con respecto a los GNSS (Seidel et al.2011). En el unico lugar en el que se observa un cambio en el MBE con eldesplazamiento de las radiosondas es NYA, estacion que se encuentra en lacosta, de modo que puede ocurrir que si los vientos mueven la radiosondahacia el mar, los campos de humedad pueden variar bruscamente, induciendoun aumento del MBE. No obstante, el SD sı que aumenta con el desplazamientode las sondas en todas las estaciones.

De este trabajo concluimos que GNSS y radiosondas muestran un acuer-

4.1. Validacion de productos de vapor de agua 23

do excelente y que las dependencias observadas son irrelevantes. Sin embargo,es muy importante contar con medidas redundantes que permitan mejorar lacalidad de los instrumentos e incrementar la resolucion temporal. Las medi-das de GNSS tienen dos principales ventajas: la primera es la alta resoluciontemporal y la segunda es la estabilidad frente a distintas condiciones meteo-rologicas (viento, nubes, etc.). Estas ventajas hacen a los GNSS excelentes paraejercicios de intercomparacion con otras medidas.

Esto nos lleva a comparar medidas satelitales tomando los GNSS comoreferencia en nuestra region principal de estudio, la Penınsula Iberica. Porello, en el Artıculo A.2 se comparan distintos productos satelitales de IWVtomando nueve estaciones del interior de la Penınsula Iberica como referencia,bajo condiciones de cielo sin nubosidad. Los productos inter-comparados son delos siguientes instruentos satelitales: GOME-2, MODIS-Terra, MODIS-Aqua,OMI, SEVIRI, SCIAMACHY y AIRS. En el Artıculo A.3 se trata en mayorprofundidad el producto basado en el instrumento OMI y en el Artıculo A.4,el del instrumento MODIS.

La medida de la exactitud, a traves de la pseudomediana de las diferenciasrelativas satelite-GNSS, mostro valores dispares entre los distintos satelites.GOME-2 tuvo un 16.7 ± 0.8 %, mientras que SEVIRI y SCIAMCHY tuvie-ron valores moderados pero de signo opuesto (−5.2 ± 0.1 % y 6.6 ± 1.2 %).Los demas productos mostraron valores mas cercanos a cero, entre −3.4 %y 2.0 %. El IQR, que indica la dispersion de las diferencias relativas, se en-cuentra entre el 30 y el 50 % en todos los casos. En todos los productos, laspendientes de las regresiones lineales satelite-GNSS fueron menores que la uni-dad, lo que indica que el pıxel del satelite tiende a promediar el IWV en todasu area, de modo que muestran medidas mas suavizadas que las de los GNSS.Los coeficientes de determinacion R2 mostraron valores entre 0.56 (AIRS) y0.83 (GOME-2). En los Artıculos A.3 y A.4, se estudiaron estos ındices porestacion y se observo que habıa algunas diferencias entre estaciones, especial-mente en el sesgo (pseudomediana). Esto podrıa deberse a la mayor o menorhumedad de cada estacion (menores valores de humedad estan asociados a ma-yores sobreestimaciones). Tambien la altura podrıa jugar un papel importante,ya que la correccion AMF esta afectada por la altura (Palmer et al. 2001). Porotro lado, en el Artıculo A.4 se incluyeron estaciones de costa para comprobarsi esto empeora el desempeno de MODIS. Los resultados mostraron que lapresencia de agua en parte del pıxel de MODIS hacıa que la dispersion (IQR)aumentase.

Tambien se analizo la dependencia con distintas variables. La dependenciacon el propio IWV mostro un comportamiento similar en todos los productos.Los valores bajos de IWV tienden a producir una sobrestimacion mientras que

24 Capıtulo 4. Resultados y discusion

los altos se subestiman ligeramente. En cuanto a la precision, el IQR relativodisminuye conforme aumenta IWV. Llama la atencion el alto IQR de OMIpara valores bajos de IWV (0 − 5 mm). En el Artıculo A.4 se observo quelas estaciones de costa tenıan una ligera tendencia a mayor pseudomediana,solamente subestimando los valores de IWV que se encontraban por encima delos 25 mm. Ademas, los datos de dıa (que usan el algoritmo basado en radiacionNIR) siempre tenıan mayores pseudomedianas que los datos de noche (basadosen radiacion IR). No obstante, el algoritmo NIR mostraba menores IQR que elIR para todos los valores de IWV. Por otro lado, en el Artıculo A.3 se computola pseudomediana e IQR para distintos valores de IWV y para valores de SZAaltos y bajos, sin encontrar diferencias importantes entre los dos grupos deSZA.

Ademas, la dependencia con el SZA se estudio en los distintos instrumentossatelitales. Aquellos satelites que utilizan la radiacion solar para las medicionesmuestran una dependencia importante con el SZA, especialmente GOME-2,cuyo sesgo (pseudomediana) aumenta en gran medida conforme aumenta elSZA. Otros productos, como MODIS (especialmente Aqua) y tambien OMI,muestran un comportamiento similar pero no tan marcado. SCIAMACHY,a pesar de utilizar un algoritmo muy similar a GOME-2, no muestra esteaumento. En cuanto al IQR, estos productos tambien muestran valor mayorde IQR relativo a medida que aumenta el SZA, destacando OMI en valoresaltos (de dıa). El Artıculo A.3 destaco que la dependencia de OMI con SZA sedebıa en gran medida a la distribucion de valores de vapor de agua (alto IWVpara SZA bajos, cuando las temperaturas son mayores).

La dependencia con el SZA y el IWV inducen una dependencia estacional enla pseudomediana y el IQR relativos. Se observa que en los meses de verano lapseudomediana tiene valores mas bajos para GOME-2, OMI y SEVIRI, inclusonegativos en algunos casos, mientras que en los meses de invierno tiene un valormas alto para estos satelites. Cabe destacar la gran variacion de GOME-2, quemuestra una cierta sobrestimacion para todos los meses, pero es especialmentealta en los meses de invierno. Esto podrıa deberse a la dependencia con el SZAobservada para este instrumento. AIRS muestra un comportamiento similar alo largo de todo el ano, mientras que los dos instrumentos MODIS tienen unacierta sobreestimacion en verano y la subestimacion se da de forma mas mar-cada en los meses de noviembre y abril. En cuanto al IQR, todos los satelitesmuestran un comportamiento similar, con menor IQR relativo en verano queen invierno. Este comportamiento estacional es menos notable en AIRS (convalores relativamente altos durante todo el ano), mientras que en los demasinstrumentos es mas marcado. OMI es el instrumento con mayor IQR durantegran parte del ano, mientras que GOME-2 es a menudo el que menos IQR

4.1. Validacion de productos de vapor de agua 25

presenta. El bajo IQR en verano puede deberse a un alto IWV en estos mesesen la Penınsula Iberica. Esta relacion se confirmo en los Artıculos A.3 y A.4cuando se separaron, para cada mes, los valores altos y bajos de IWV, y seobservo que en invierno dominan los valores bajos y en verano los altos, mien-tras que en los valores altos y bajos por separado el ciclo estacional era muchomenos marcado.

Por ultimo, aunque todos los calculos anteriores se llevaron a cabo con da-tos de casos sin nubes, se tomaron el resto de casos, con mayor o menor CF,para estudiar la influencia de las nubes en la medicion. No obstante, solamen-te se disponıa de datos de nubosidad de los instrumentos AIRS, GOME-2 yMODIS (Aqua y Terra). En general, se observo que a medida que se incre-menta el CF, la pseudomediana se aleja del cero, aunque en unos casos esnegativo (GOME-2, AIRS, MODIS-Terra) y en otros positivo (MODIS-Aqua).La subestimacion se debe al llamado efecto de apantallamiento, es decir, lasnubes ocultan el vapor de agua que queda por debajo de ellas, de modo que elsatelite solamente mide el vapor que hay por encima y no la columna completa(Kokhanovsky y Rozanov 2008; Roman et al. 2015). La sobrestimacion no tie-ne una explicacion clara, aunque podrıa estar relacionada con el hecho de usarradiacion NIR para el dıa e IR para la noche. Durante la noche, el algoritmopodrıa no ser capaz de distinguir la nube y el vapor de agua, provocando lasobrestimacion. En el Artıculo A.4, este asunto se estudio en mayor detalle,dividiendo los bins en alto IWV, bajo IWV, dıa (NIR) y noche (IR), mostran-do que los valores nocturnos dan una subestimacion que se hace mas intensacon la CF, mientras que los valores de dıa incrementan la sobrestimacion conla CF. En cuanto a la precision, el IQR aumenta a medida que se incrementael CF para los productos de MODIS. AIRS tiene un IQR bastante estable,mientras que GOME-2 muestra una reduccion del IQR conforme se aumentala CF. Una posible explicacion serıa que a medida que aumenta la nubosi-dad se introduce ruido en los algoritmos, aumentando el IQR, pero cuando lacubierta nubosa es mayor, el efecto de apantallamiento podrıa hacer que lasensibilidad se redujese, disminuyendo la variabilidad y el IQR. En la divisionhecha en el Artıculo A.4, el IQR aumenta en todos los casos (alto IWV, bajoIWV, dıa y noche) con la CF, aunque el IQR de los valores nocturnos tienemenor dependencia con la CF. Notese que el Artıculo A.2 tiene una errata enla Figura 10 que se encuentra corregida en la Figura C.1 del Apendice C.

Estos estudios demuestran que el vapor de agua medido a traves de satelitestiene una buena correlacion con las medidas obtenidas por GNSS. Sin embargo,es claro que existen ciertas dependencias a tener en cuenta y que los algoritmose instrumentos futuros deben tratar de mitigarlas para obtener medidas quesean mas representativas de la situacion real.

26 Capıtulo 4. Resultados y discusion

4.2. Efecto radiativo del vapor de aguaEn el Artıculo A.5, el WVRE en onda corta (WVRESW) mostro valores

similares en las estaciones analizadas, con coeficiente de variacion del 7 %,aunque en las estaciones costeras sı se observa un ligero aumento. Esto indicaque la distribucion espacial es menos importante que otras variables, comopuede ser el SZA o la propia variabilidad del IWV a lo largo del ano o deldıa. Los valores de WVRESW obtenidos para un SZA de (60.0 ± 0.5)◦ varıanentre −100.0 Wm−2 y −38.7 Wm−2. Estos valores son comparativamente masintensos que los obtenidos por Mateos et al. (2013) para aerosoles y nubes enGranada. Ademas, el citado estudio considera el papel del vapor de agua comoalgo menor, estimandolo con una variacion entre 25 mm y 5 mm, cuando en elestudio del Artıculo A.5 el 20 % de los datos se encontraban por encima dellımite superior, mientras que el 3 % estaban por debajo del inferior. Aunquehay situaciones en las que no hay nubes, no existen en la realidad situacionesen las que la atmosfera no tenga vapor de agua. Por tanto, es difıcil compararlos efectos radiativos de ambos componentes.

Por otro lado, en el Artıculo A.6 se estudio la variabilidad espacial basa-da en tres zonas, a saber: el interior (I) de la Penınsula Iberica, la region delAtlantico Norte (NA) y la del Mar Mediterraneo (MS). En estas zonas se apre-ciaron ciertas caracterısticas diferenciales, si bien estas eran mas notables parael WVRE en onda larga (WVRELW) y el total, mientras que la variabilidadestaba dominada por la estacionalidad para el WVRESW. El WVRE total essimilar en las zonas I y NA, pero en I la variabilidad es algo mayor, teniendocolas mas largas, mientras que la zona MS muestra valores ligeramente ma-yores a las otras dos zonas. El WVRELW muestra un comportamiento similaral regimen total. Sin embargo, WVRESW muestra valores similares entre lasdistintas zonas y mayor variabilidad dentro de cada zona, lo que tiene que vercon la variabilidad debida al SZA.

El efecto del vapor de agua en la HR de onda corta tiene un marcadocaracter estacional. En invierno, los valores nunca superan 1.0 K d−1 mien-tras que en verano pueden alcanzar los 1.5 K d−1. No obstante, los valoresinferiores alcanzaban el mınimo de 0.3 K d−1 durante todo el ano. Como com-paracion, Valenzuela et al. (2012) mostro para aerosoles una HR por debajo de0.3 K d−1, muy inferior a la HR del vapor de agua. De nuevo, hay que tomarestas comparaciones con precaucion.

Por otro lado, se obtuvo una relacion empırica entre el IWV, el SZA y elWVRE, de acuerdo a la Ecuacion (4.1),

WVRESW = −a · IWVbµc, (4.1)

donde µ = cos SZA y a, b y c son parametros a ajustar. Mediante un ajuste

4.2. Efecto radiativo del vapor de agua 27

por mınimos cuadrados (tras linealizar la ecuacion), se obtuvieron los valoreslog a = 4.144 ± 0.001, b = 0.2661 ± 0.0003 y c = 0.7679 ± 0.0003, con uncoeficiente de determinacion R2 = 0.997. La interpretacion fısica de b es talque un cambio de un 1 % en IWV producirıa un cambio en el WVRE de unb %, en este caso, un 0.2661 %. El mismo ajuste puede hacerse para el WVREnormalizado

(100 % ·WVRE/

(SW↓

noIWV − SW↑noIWV

)).

Una vez obtenida esta relacion empırica, se obtuvo por derivacion de lamisma el WVEFF, como muestra la Ecuacion (4.2):

WVEFFSW = ∂WVRE∂IWV = −a · b · IWVb−1µc = b

WVREIWV , (4.2)

Si se calcula WVEFFSW para cada distintos bins de SZA y de IWV (ver Figu-ra 7 del Artıculo A.5) se obtiene que esta variable decrece conforme aumentael IWV muy rapido para valores bajos de IWV, pero parece saturar para va-lores altos de IWV. Para un bin concreto de IWV, la WVEFFSW disminuyeconforme se incrementa el SZA. Esta relacion puede explicarse de la siguien-te manera: la irradiancia disminuye con el SZA, disminuyendo la cantidad deradiacion que puede absorber el vapor de agua, y por tanto disminuye su efi-ciencia. Ahora bien, podrıa haber un efecto de segundo orden, ya que la masaoptica del vapor de agua se incrementa con el SZA, aumentando el poder deextincion del vapor de agua. Pero esto no se aprecia en la Figura 7 (b) delArtıculo A.5. Sin embargo, si se realizan los mismos calculos con el WVRESWnormalizado, obteniendo el WVEFFSW normalizado, sı se observa este efectode segundo orden. La Figura 7 (b) del Artıculo A.5 muestra que la dependenciacon el SZA se hace mas debil a medida que aumenta el IWV.

El WVEFFSW tambien se calculo para el tope de la atmosfera. Se encontroque podıa establecerse una relacion similar a la de la Ecuacion (4.1), pero eranecesario incluir tambien el albedo de la superficie, como se muestra en laEcuacion (4.3),

WVRESW = −aTOP · IWVbTOPµcTOPαdTOP , (4.3)

donde α es el albedo de la superficie y aTOP, bTOP, cTOP, dTOP son parametrosa ajustar.

Por otro lado, el analisis de la evolucion temporal de IWV y WVRE en losdistintos regımenes (onda corta, larga y total) permiten calcular si existe algu-na tendencia. Estas tendencias deben tomarse con cierta precaucion debido aque tenemos series relativamente cortas (periodo 2007-2015). Se han encontra-do tendencias significativas en la mayorıa de las estaciones. Son positivas parael IWV, el WVRE total y el WVRELW, mientras que el WVRESW muestratendencias negativas. La Tabla 4 del Artıculo A.6 muestra los valores obteni-dos, marcando con un asterisco aquellos que son significativos. Los valores de

28 Capıtulo 4. Resultados y discusion

tendencia obtenidos en este estudio para el vapor de agua son superiores a losmostrados en otros estudios (Chen y Liu 2016; Ning y Elgered 2012). Tantoel WVRE total como el WVRELW muestran tendencias significativas en lasregiones I y NA, pero no significativas en MS.

El balance de las tendencias significativas es, en promedio, el siguien-te: WVRESW tiene una tendencia de −0.09 W m−2 a−1, y el WVRELW,+0.50 W m−2 a−1, mientras que el total coincide aproximadamente conla suma de las dos anteriores, +0.42 W m−2 a−1. Esta tendencia positivapuede explicar parcialmente el aumento de la temperatura en superficieobservada en la Penınsula Iberica durante las dos ultimas decadas (Moratielet al. 2017), la cual, a su vez, incrementa la evaporacion, produciendose unaretroalimentacion positiva (Colman 2015). Ademas, la tendencia observadaen el WVRESW es inferior al promedio global obtenido por Kvalevag y Myhre(2007) (−0.29 W m−2 a−1). Por otro lado, Mateos et al. (2014) mostraron unvalor de +0.36 W m−2 a−1 para el efecto radiativo de aerosoles en SW bajocondiciones de cielo descubierto en la Penınsula Iberica (periodo 2004-2012). Apartir de este resultado, podemos senalar que la tendencia negativa observadapara el WVRESW es un cuarto de la obtenida para el efecto de aerosoles. Portanto, la tendencia del vapor de agua podrıa estar enmascarando parcialmentela magnitud total del papel de los aerosoles en la modulacion de la radiacionde SW en superficie en la Penınsula Iberica.

Capıtulo 5

Conclusiones

Es una vieja maxima mıa que cuando hayas descartado lo imposible,lo que quede, aunque sea improbable, debe ser la verdad

Sherlock Holmes en El signo de los cuatro, de Arthur Conan Doyle.

Las conclusiones principales que se pueden extraer de esta tesis doctoralson las siguientes:

1. Los productos de vapor de agua derivados de receptores GNSS en tierrason de alta calidad mostrando un excelente acuerdo con las medidasde radiosondas de la red GRUAN. Por tanto, pueden utilizarse comoreferencia para la validacion de las medidas proporcionadas por otrosequipos.

2. En general, los productos de vapor de agua inferidos por instrumentossatelitales tienen un buen acuerdo con las medidas GNSS en la PenınsulaIberica. No obstante, dichas estimaciones satelitales deben tomarse concierta precaucion, ya que se han observado dependencias que puedencomprometer la calidad de estos datos, especialmente con el propio vaporde agua, con el SZA (aquellos dependientes de radiacion solar) y con lanubosidad.

3. Se ha propuesto una expresion empırica que relaciona el efecto radiativodel vapor de agua en onda corta con las variables IWV y SZA en laPenınsula Iberica. Las estimaciones del efecto radiativo del vapor de agua

30 Capıtulo 5. Conclusiones

mediante la formula propuesta han mostrado un excelente acuerdo conlas medidas experimentales. A partir de esta expresion empırica ha sidoposible determinar la eficiencia radiativa del vapor de agua.

4. La tendencia positiva observada en la evolucion temporal del vapor deagua atmosferico en la Penınsula Iberica ha provocado un aumento sig-nificativo de su efecto radiativo, el cual se ha cuantificado tanto en ondacorta como en onda larga. El signo positivo obtenido en los valores pro-medios del efecto radiativo en onda larga podrıa explicar parcialmenteel aumento de la temperatura en superficie en la region de estudio. Encambio, se ha obtenido un evidente signo negativo en el efecto radiati-vo en onda corta, lo cual podrıa mitigar parcialmente la influencia deldescenso de la carga de aerosoles en los incrementos de radiacion solarobservados en superficie (brightening).

Bibliografıa

Anton, M., D. Loyola, R. Roman y H. Vomel (2015). ((Validation of GOME-2/MetOp-A Total Water Vapour Column Using Reference Radiosonde Da-ta from the GRUAN Network)). En: Atmospheric Measurement Techniques8.3, pags. 1135-1145. issn: 1867-8548. doi: 10.5194/amt-8-1135-2015.url: https://www.atmos-meas-tech.net/8/1135/2015/.

Barnet, C., E. Manning, P. W. Rosenkranz, L. Strow y J. Susskind (2007).Airs-Team Retrieval for Core Products and Geophysical Parameters. Re-port March. Jet Propulsion Laboratory.

Barreto, A., E. Cuevas, B. Damiri, P. M. Romero y F. Almansa (2013).((Column Water Vapor Determination in Night Period with a LunarPhotometer Prototype)). En: Atmospheric Measurement Techniques 6.8,pags. 2159-2167. issn: 18671381. doi: 10.5194/amt-6-2159-2013. url:http://www.atmos-meas-tech.net/6/2159/2013/.

Bennouna, Y. S., B. Torres, V. E. Cachorro, J. P. O. de Galisteo y C. Toledano(2013). ((The Evaluation of the Integrated Water Vapour Annual Cycle overthe Iberian Peninsula from EOS-MODIS against Different Ground-BasedTechniques)). En: Quarterly Journal of the Royal Meteorological Society139.October, pags. 1935-1956. issn: 00359009. doi: 10.1002/qj.2080.

Bevis, M., S. Businger, T. A. Herring, C. Rocken, R. A. Anthes y R. H. Ware(1992). ((GPS Meteorology: Remote Sensing of Atmospheric Water VaporUsing the Global Positioning System)). En: Journal of Geophysical Research97.D14, pags. 15787-15801. issn: 0148-0227. doi: 10.1029/92JD01517.

Bokoye, A. I., A. Royer, N. O’Niell, P. Cliche, L. McArthur, P. Teillet, G. Fe-dosejevs y J.-M. Theriault (2003). ((Multisensor Analysis of Integrated At-mospheric Water Vapor over Canada and Alaska)). En: Journal of Geophysi-cal Research 108.D15, pags. 21–1 —- 21-16. issn: 0148-0227. doi: 10.1029/2002JD002721. url: http://doi.wiley.com/10.1029/2002JD002721.

Chen, B. y Z. Liu (2016). ((Global Water Vapor Variability and Trend from theLatest 36 Year (1979 to 2014) Data of ECMWF and NCEP Reanalyses,Radiosonde, GPS, and Microwave Satellite: GLOBAL WATER VAPOR

32 Bibliografıa

VARIABILITY AND TREND)). En: Journal of Geophysical Research: At-mospheres 121.19, pags. 11, 442-11, 462. issn: 2169897X. doi: 10.1002/2016JD024917. url: http://doi.wiley.com/10.1002/2016JD024917.

Colman, R. A. (2003). ((A Comparison of Climate Feedbacks in General Circu-lation Models)). En: Climate Dynamics 20, pags. 865-873. issn: 0930-75751432-0894 SU - Earth and Environmental Science. doi: 10.1007/s00382-003-0310-z. pmid: 183550600013.

— (2015). ((Climate Radiative Feedbacks and Adjustments at the Earth’sSurface)). En: Journal of Geophysical Research: Atmospheres 120.8,pags. 3173-3182. issn: 2169897X. doi: 10.1002/2014JD022896.

Davis, J. L., T. A. Herring, I. I. Shapiro, A. E. E. Rogers y G. Elgered (1985).((Geodesy by Radio Interferometry: Effects of Atmospheric Modeling Errorson Estimates of Baseline Length)). En: Radio Science 20.6, pags. 1593-1607.issn: 00486604. doi: 10.1029/RS020i006p01593.

de Haan, S., H. van der Marel y S. Barlag (2002). ((Comparison of GPSSlant Delay Measurements to a Numerical Model: Case Study of a ColdFront Passage)). En: Physics and Chemistry of the Earth, Parts A/B/C27.4-5, pags. 317-322. issn: 14747065. doi: 10.1016/S1474- 7065(02)00006 - 2. url: http : / / linkinghub . elsevier . com / retrieve / pii /S1474706502000062.

Dee, D. P. et al. (2011). ((The ERA-Interim Reanalysis: Configuration andPerformance of the Data Assimilation System)). En: Quarterly Journal ofthe Royal Meteorological Society 137.656, pags. 553-597. issn: 00359009.doi: 10.1002/qj.828.

Di Biagio, C., A. di Sarra, P. Eriksen, S. E. Ascanius, G. Muscari y B. Holben(2012). ((Effect of Surface Albedo, Water Vapour, and Atmospheric Aerosolson the Cloud-Free Shortwave Radiative Budget in the Arctic)). En: ClimateDynamics 39.3-4, pags. 953-969. issn: 0930-7575. doi: 10.1007/s00382-011-1280-1. url: http://link.springer.com/10.1007/s00382-011-1280-1.

Diedrich, H., F. Wittchen, R. Preusker y J. Fischer (2016). ((Representative-ness of Total Column Water Vapour Retrievals from Instruments on Po-lar Orbiting Satellites)). En: Atmospheric Chemistry and Physics 16.13,pags. 8331-8339. issn: 1680-7324. doi: 10.5194/acp-16-8331-2016. url:https://acp.copernicus.org/articles/16/8331/2016/.

Dirksen, R. J., M. Sommer, F. J. Immler, D. F. Hurst, R. Kivi y H. Vomel(2014). ((Reference Quality Upper-Air Measurements: GRUAN Data Pro-cessing for the Vaisala RS92 Radiosonde)). En: Atmospheric MeasurementTechniques 7.12, pags. 4463-4490. issn: 1867-8548. doi: 10.5194/amt-7-4463-2014. url: http://www.atmos-meas-tech.net/7/4463/2014/.

Bibliografıa 33

Firsov, K. M., T. Y. Chesnokova y E. V. Bobrov (2015). ((The Role of theWater Vapor Continuum Absorption in near Ground Long-Wave Radia-tion Processes of the Lower Volga Region)). En: Atmospheric and Ocea-nic Optics 28.1, pags. 1-8. issn: 1024-8560, 2070-0393. doi: 10 . 1134 /S1024856015010030. url: http : / / link . springer . com / 10 . 1134 /S1024856015010030.

Forster, P. M. D. F. y K. P. Shine (2002). ((Assessing the Climate Impact ofTrends in Stratospheric Water Vapor)). En: Geophysical Research Letters29.6, pags. 10–1–10-4. issn: 00948276. doi: 10.1029/2001GL013909. url:http://doi.wiley.com/10.1029/2001GL013909.

Gao, B.-C. y R.-R. Li (2008). ((The Time Series of Terra and Aqua MO-DIS Near-IR Water Vapor Products)). En: IGARSS 2008 - 2008 IEEEInternational Geoscience and Remote Sensing Symposium. Vol. 3. IEEE,pags. 186-189. isbn: 978-1-4244-2807-6. doi: 10 . 1109 / IGARSS . 2008 .4779314.

Gao, B.-C. e Y. J. Kaufman (1992). The MODIS Near-IR Water Vapor Algo-rithm Product ID : MOD05 - Total Precipitable Water. Algorithm TechnicalBackground Document, pags. 1-25.

Garcıa, R. D., A. Barreto, E. Cuevas, J. Grobner, O. E. Garcıa, A. Gomez-Pelaez, P. M. Romero-Campos, A. Redondas, V. E. Cachorro y R. Ra-mos (2018). ((Comparison of Observed and Modeled Cloud-Free LongwaveDownward Radiation (2010–2016) at the High Mountain BSRN Izana Sta-tion)). En: Geoscientific Model Development 11.6, pags. 2139-2152. issn:1991-9603. doi: 10 . 5194 / gmd - 11 - 2139 - 2018. url: https : / / www .geosci-model-dev.net/11/2139/2018/.

Golovko, V. F. (1999). ((Modeling Radiation Absorption by Water Vapor in theAtmosphere within the 0- to 20,000-Cm -1 Spectral Range)). En: Sixth Inter-national Symposium on Atmospheric and Ocean Optics. Ed. por G. G. Mat-vienko y V. P. Lukin. Tomsk, Russia, pag. 632. doi: 10.1117/12.370556.url: http : / / proceedings . spiedigitallibrary . org / proceeding .aspx?doi=10.1117/12.370556.

Grossi, M., P. Valks, D. Loyola, B. Aberle, S. Slijkhuis, T. Wagner, S. Beir-le y R. Lang (2015). ((Total Column Water Vapour Measurements fromGOME-2 MetOp-A and MetOp-B)). En: Atmospheric Measurement Tech-niques 8.3, pags. 1111-1133. issn: 1867-8548. doi: 10.5194/amt-8-1111-2015.

Haywood, J. M., N. Bellouin, A. Jones, O. Boucher, M. Wild y K. P. Shine(2011). ((The Roles of Aerosol, Water Vapor and Cloud in Future Glo-bal Dimming/Brightening)). En: Journal of Geophysical Research 116.D20.

34 Bibliografıa

issn: 0148-0227. doi: 10.1029/2011JD016000. url: http://doi.wiley.com/10.1029/2011JD016000.

Heise, S., G. Dick, G. Gendt, T. Schmidt y J. Wickert (2009). ((Integrated Wa-ter Vapor from IGS Ground-Based GPS Observations: Initial Results froma Global 5-Min Data Set)). En: Annales Geophysicae 27.7, pags. 2851-2859.issn: 1432-0576. doi: 10.5194/angeo-27-2851-2009. url: http://www.ann-geophys.net/27/2851/2009/.

Huang, Y., V. Ramaswamy y B. Soden (2007). ((An Investigation of the Sensi-tivity of the Clear-Sky Outgoing Longwave Radiation to Atmospheric Tem-perature and Water Vapor)). En: Journal of Geophysical Research 112.D5.issn: 0148-0227. doi: 10.1029/2005JD006906. url: http://doi.wiley.com/10.1029/2005JD006906.

Ichoku, C. et al. (2002). ((Analysis of the Performance Characteristics of theFive-Channel Microtops II Sun Photometer for Measuring Aerosol Opti-cal Thickness and Precipitable Water Vapor)). En: Journal of Geophysi-cal Research 107.D13, pags. 5–1—- 5-17. issn: 0148-0227. doi: 10.1029/2001JD001302. url: http://doi.wiley.com/10.1029/2001JD001302.

Jones, A. et al. (2009). ((Evolution of Stratospheric Ozone and Water VapourTime Series Studied with Satellite Measurements)). En: Atmospheric Che-mistry and Physics 9.1, pags. 1157-1209. issn: 1680-7375. doi: 10.5194/acpd-9-1157-2009. url: www.atmos-chem-phys-discuss.net/9/1157/2009/%20http://www.atmos-chem-phys-discuss.net/9/1157/2009/.

Kawamoto, K. y T. Hayasaka (2008). ((Relative Contributions to Surface Short-wave Irradiance over China: A New Index of Potential Radiative Forcing)).En: Geophysical Research Letters 35.17. issn: 0094-8276. doi: 10.1029/2008GL035083. url: http://doi.wiley.com/10.1029/2008GL035083.

Kendall, M. G., A. Stuart y J. K. Ord (1994). The Advanced Theory of Sta-tistics in 3 Volumes. 1 1. London: Griffin. isbn: 978-0-340-61430-3 978-0-470-23380-1 978-0-470-66530-5.

Kokhanovsky, A. y V. Rozanov (2008). ((The Uncertainties of Satellite DOASTotal Ozone Retrieval for a Cloudy Sky)). En: Atmospheric Research 87.1,pags. 27-36. issn: 01698095. doi: 10.1016/j.atmosres.2007.04.006.

Kvalevag, M. M. y G. Myhre (2007). ((Human Impact on Direct and DiffuseSolar Radiation during the Industrial Era)). En: Journal of Climate 20.19,pags. 4874-4883. issn: 0894-8755, 1520-0442. doi: 10.1175/JCLI4277.1.url: http://journals.ametsoc.org/doi/abs/10.1175/JCLI4277.1.

Mann, H. B. (1945). ((Nonparametric Tests Against Trend)). En: Econometrica13.3, pag. 245. issn: 00129682. doi: 10.2307/1907187. JSTOR: 1907187.

Mateos, D., M. Anton, C. Toledano, V. E. Cachorro, L. Alados-Arboledas, M.Sorribas, M. J. Costa y J. M. Baldasano (2014). ((Aerosol Radiative Effects

Bibliografıa 35

in the Ultraviolet, Visible, and near-Infrared Spectral Ranges Using Long-Term Aerosol Data Series over the Iberian Peninsula)). En: AtmosphericChemistry and Physics 14.24, pags. 13497-13514. issn: 1680-7324. doi: 10.5194/acp-14-13497-2014. url: http://www.atmos-chem-phys.net/14/13497/2014/.

Mateos, D., M. Anton, A. Valenzuela, A. Cazorla, F. J. Olmo y L. Alados-Arboledas (2013). ((Short-Wave Radiative Forcing at the Surface for CloudySystems at a Midlatitude Site)). En: Tellus B: Chemical and Physical Meteo-rology 65.1, pag. 21069. issn: 1600-0889. doi: 10.3402/tellusb.v65i0.21069. url: https : / / www . tandfonline . com / doi / full / 10 . 3402 /tellusb.v65i0.21069.

Moratiel, R., B. Soriano, A. Centeno, D. Spano y R. Snyder (2017). ((Wet-Bulb, Dew Point, and Air Temperature Trends in Spain)). En: Theoreticaland Applied Climatology 130.1-2, pags. 419-434. issn: 0177-798X, 1434-4483. doi: 10.1007/s00704-016-1891-x. url: http://link.springer.com/10.1007/s00704-016-1891-x.

Myhre, G. et al. (2013). ((Anthropogenic and Natural Radiative Forcing.)) En:Climate Change 2013: The Physical Science Basis. Contribution of WorkingGroup I to the Fifth Assessment Report of the Intergovernmental Panel onClimate Change. IPCC. IPCC, pags. 659-740. isbn: 978-1-107-05799-1.

Niell, A. E. (2000). ((Improved Atmospheric Mapping Functions for VLBI andGPS)). En: Earth, Planets and Space 52.10, pags. 699-702. issn: 1880-5981.doi: 10.1186/BF03352267.

Ning, T. y G. Elgered (2012). ((Trends in the Atmospheric Water Vapor Con-tent From Ground-Based GPS: The Impact of the Elevation Cutoff An-gle)). En: IEEE Journal of Selected Topics in Applied Earth Observationsand Remote Sensing 5.3, pags. 744-751. issn: 1939-1404, 2151-1535. doi:10.1109/JSTARS.2012.2191392. url: http://ieeexplore.ieee.org/document/6185695/.

Noel, S., M. Buchwitz y J. P. Burrows (2004). ((First Retrieval of Global Wa-ter Vapour Column Amounts from SCIAMACHY Measurements)). En: At-mospheric Chemistry and Physics 4.1, pags. 111-125. issn: 1680-7324. doi:10.5194/acp-4-111-2004.

Obregon, M., M. Costa, A. Silva y A. Serrano (2018). ((Impact of Aerosol andWater Vapour on SW Radiation at the Surface: Sensitivity Study and Ap-plications)). En: Atmospheric Research 213, pags. 252-263. issn: 01698095.doi: 10.1016/j.atmosres.2018.06.001. url: https://linkinghub.elsevier.com/retrieve/pii/S0169809518301388.

Ohtani, R. e I. Naito (2000). ((Comparisons of GPS-Derived Precipitable WaterVapors with Radiosonde Observations in Japan)). En: Journal of Geophy-

36 Bibliografıa

sical Research: Atmospheres 105.D22, pags. 26917-26929. issn: 01480227.doi: 10.1029/2000JD900362.

Ortiz de Galisteo, J. P., V. Cachorro, C. Toledano, B. Torres, N. Laulainen,Y. Bennouna y A. de Frutos (2011). ((Diurnal Cycle of Precipitable Wa-ter Vapor over Spain)). En: Quarterly Journal of the Royal MeteorologicalSociety 137.657, pags. 948-958. issn: 00359009. doi: 10.1002/qj.811.

Palmer, P. I., D. J. Jacob, K. Chance, R. V. Martin, R. J. D. Spurr, T. P.Kurosu, I. Bey, R. Yantosca, A. Fiore y Q. Li (2001). ((Air Mass Fac-tor Formulation for Spectroscopic Measurements from Satellites: Applica-tion to Formaldehyde Retrievals from the Global Ozone Monitoring Ex-periment)). En: Journal of Geophysical Research: Atmospheres 106.D13,pags. 14539-14550. issn: 01480227. doi: 10 . 1029 / 2000JD900772. url:http://doi.wiley.com/10.1029/2000JD900772.

Pany, T., P. Pesec y G. Stangl (2001). ((Atmospheric GPS Slant Path De-lays and Ray Tracing through Numerical Weather Models, a Comparison)).En: Physics and Chemistry of the Earth, Part A: Solid Earth and Geo-desy 26.3, pags. 183-188. issn: 14641895. doi: 10.1016/S1464-1895(01)00044 - 8. url: http : / / linkinghub . elsevier . com / retrieve / pii /S1464189501000448.

Perez-Ramırez, D., F. Navas-Guzman, H. Lyamani, J. Fernandez-Galvez, F. J.Olmo y L. Alados-Arboledas (2012). ((Retrievals of Precipitable Water Va-por Using Star Photometry: Assessment with Raman Lidar and Link to SunPhotometry)). En: Journal of Geophysical Research: Atmospheres 117.D5,n/a-n/a. issn: 01480227. doi: 10.1029/2011JD016450. url: http://doi.wiley.com/10.1029/2011JD016450.

Ricchiazzi, P., S. Yang, C. Gautier y D. Sowle (1998). ((SBDART: A Researchand Teaching Software Tool for Plane-Parallel Radiative Transfer in theEarth’s Atmosphere)). En: Bulletin of the American Meteorological Society79.10, pags. 2101-2114. issn: 0003-0007. doi: 10.1175/1520-0477(1998)079<2101:SARATS>2.0.CO;2. url: http://journals.ametsoc.org/doi/abs/10.1175/1520-0477%281998%29079%3C2101%3ASARATS%3E2.0.CO%3B2.

Rohm, W., Y. Yuan, B. Biadeglgne, K. Zhang y J. L. Marshall (2014).((Ground-Based GNSS ZTD/IWV Estimation System for NumericalWeather Prediction in Challenging Weather Conditions)). En: AtmosphericResearch 138, pags. 414-426. issn: 01698095. doi: 10.1016/j.atmosres.2013.11.026. url: http://linkinghub.elsevier.com/retrieve/pii/S0169809513003499.

Roman, R., M. Anton, V. Cachorro, D. Loyola, J. Ortiz de Galisteo, A. deFrutos y P. Romero-Campos (2015). ((Comparison of Total Water Vapor

Bibliografıa 37

Column from GOME-2 on MetOp-A against Ground-Based GPS Measu-rements at the Iberian Peninsula)). En: Science of The Total Environment533, pags. 317-328. issn: 00489697. doi: 10.1016/j.scitotenv.2015.06 . 124. url: https : / / linkinghub . elsevier . com / retrieve / pii /S0048969715303260.

Roman, R., J. Bilbao y A. de Miguel (2014). ((Uncertainty and Variability inSatellite-Based Water Vapor Column, Aerosol Optical Depth and AngstromExponent, and Its Effect on Radiative Transfer Simulations in the Ibe-rian Peninsula)). En: Atmospheric Environment 89, pags. 556-569. issn:13522310. doi: 10 . 1016 / j . atmosenv . 2014 . 02 . 027. url: http : / /linkinghub.elsevier.com/retrieve/pii/S135223101400123X.

Saastamoinen, J. (1972). ((Atmospheric Correction for the Troposphere andStratosphere in Radio Ranging Satellites)). En: Geophysical Monograph Se-ries. Ed. por S. W. Henriksen, A. Mancini y B. H. Chovitz. Washington, D.C.: American Geophysical Union, pags. 247-251. isbn: 978-1-118-66364-6978-0-87590-015-5. doi: 10.1029/GM015p0247.

Seemann, S. W., W. P. M. J. Li y 2. L. E. Gumley (2003). ((Operational Retrie-val of Atmospheric Temperature, Moisture, and Ozone from MODIS Infra-red Radiances.)) En: Journal of Applied Meteorology 42.1, pags. 1072-1091.

Seemann, S. W., E. E. Borbas, J. Li, W. P. Menzel y L. E. Gumley (2006).Modis Atmospheric Profile Retrieval: Algorithm Theoretical Basis Docu-ment. October. Cooperative Institute for Meteorological Satellite Studies,University of Wisconsin-Madison, pag. 40.

Seidel, D. J., B. Sun, M. Pettey y A. Reale (2011). ((Global Radiosonde BalloonDrift Statistics)). En: Journal of Geophysical Research 116.D7. issn: 0148-0227. doi: 10.1029/2010JD014891. url: http://doi.wiley.com/10.1029/2010JD014891.

Sen, P. K. (1968). ((Estimates of the Regression Coefficient Based on Ken-dall’s Tau)). En: Journal of the American Statistical Association 63.324,pag. 1379. issn: 01621459. doi: 10.2307/2285891. JSTOR: 2285891.

Smith, C. A., J. D. Haigh y R. Toumi (2001). ((Radiative Forcing Due toTrends in Stratospheric Water Vapour)). En: Geophysical Research Letters28.1, pags. 179-182. issn: 00948276. doi: 10.1029/2000GL011846. url:http://doi.wiley.com/10.1029/2000GL011846.

Soden, B. J., R. T. Wetherald, G. L. Stenchikov y A. Robock (2002). ((GlobalCooling After the Eruption of Mount Pinatubo: A Test of Climate Feedbackby Water Vapor)). En: Science 296.5568, pags. 727-730. issn: 00368075.doi: 10.1126/science.296.5568.727. pmid: 17803838. url: http://www.sciencemag.org/cgi/doi/10.1126/science.296.5568.727.

38 Bibliografıa

Torres, B., V. E. Cachorro, C. Toledano, J. P. Ortiz de Galisteo, A. Berjon,A. M. de Frutos, Y. Bennouna y N. Laulainen (2010). ((Precipitable Wa-ter Vapor Characterization in the Gulf of Cadiz Region (SouthwesternSpain) Based on Sun Photometer, GPS, and Radiosonde Data)). En: Jour-nal of Geophysical Research 115.D18103, pags. 1-11. issn: 0148-0227. doi:10 . 1029 / 2009JD012724. url: http : / / doi . wiley . com / 10 . 1029 /2009JD012724.

Turner, D. D., R. A. Ferrare, L. A. H. Brasseur, W. F. Feltz y T. P. Too-man (2002). ((Automated Retrievals of Water Vapor and Aerosol Profilesfrom an Operational Raman Lidar)). En: Journal of Atmospheric and Ocea-nic Technology 19.1, pags. 37-50. issn: 0739-0572. doi: 10.1175/1520-0426(2002 ) 019<0037 : AROWVA > 2 . 0 . CO ; 2. url: http : / / journals .ametsoc.org/doi/abs/10.1175/1520-0426%282002%29019%3C0037%3AAROWVA%3E2.0.CO%3B2.

Turner, D. D., S. A. Clough, J. C. Liljegren, E. E. Clothiaux, K. E.Cady-Pereira y K. L. Gaustad (2007). ((Retrieving Liquid Wat0erPath and Precipitable Water Vapor From the Atmospheric RadiationMeasurement (ARM) Microwave Radiometers)). En: IEEE Transac-tions on Geoscience and Remote Sensing 45.11, pags. 3680-3690.issn: 0196-2892. doi: 10 . 1109 / TGRS . 2007 . 903703. url: http ://ieeexplore.ieee.org/document/4373386/.

Valenzuela, A., F. J. Olmo, H. Lyamani, M. Anton, A. Quirantes y L. Alados-Arboledas (2012). ((Aerosol Radiative Forcing during African Desert DustEvents (2005–2010) over Southeastern Spain)). En: Atmospheric Chemistryand Physics 12.21, pags. 10331-10351. issn: 1680-7324. doi: 10.5194/acp-12-10331-2012. url: http://www.atmos-chem-phys.net/12/10331/2012/.

Wagner, T., S. Beirle, M. Grzegorski y U. Platt (2006). ((Global Trends (1996-2003) of Total Column Precipitable Water Observed by Global Ozone Mo-nitoring Experiment (GOME) on ERS-2 and Their Relation to near-SurfaceTemperature)). En: Journal of Geophysical Research 111.D12, pag. D12102.issn: 0148-0227. doi: 10.1029/2005JD006523.

Wagner, T., J. Heland, M. Zoger y U. Platt (2003). ((A Fast H2O TotalColumn Density Product from GOME: Validation with in-Situ AircraftMeasurements)). En: Atmospheric Chemistry and Physics Discussions 3.1,pags. 323-353. issn: 1680-7375. doi: 10.5194/acpd-3-323-2003.

Wang, H., X. Liu, K. Chance, G. Gonzalez Abad y C. Chan Miller (2014). ((Wa-ter Vapor Retrieval from OMI Visible Spectra)). En: Atmospheric Measure-ment Techniques 7.6, pags. 1901-1913. issn: 1867-8548. doi: 10.5194/amt-7-1901-2014. url: http://www.atmos-meas-tech.net/7/1901/2014/.

Bibliografıa 39

Wang, J. y L. Zhang (2008). ((Systematic Errors in Global Radiosonde Pre-cipitable Water Data from Comparisons with Ground-Based GPS Measu-rements)). En: Journal of Climate 21.10, pags. 2218-2238. issn: 0894-8755,1520-0442. doi: 10 . 1175 / 2007JCLI1944 . 1. url: http : / / journals .ametsoc.org/doi/abs/10.1175/2007JCLI1944.1.

Wang, J., L. Zhang, A. Dai, T. Van Hove y J. Van Baelen (2007). ((ANear-Global, 2-Hourly Data Set of Atmospheric Precipitable Water fromGround-Based GPS Measurements)). En: Journal of Geophysical Research112.D11, pag. D11107. issn: 0148-0227. doi: 10 . 1029 / 2006JD007529.url: http://doi.wiley.com/10.1029/2006JD007529.

Wilcoxon, F. (1945). ((Individual Comparisons by Ranking Methods)). En: Bio-metrics Bulletin 1.6, pag. 80. issn: 00994987. doi: 10 . 2307 / 3001968.pmid: 20983181.

WMO (2008). Guide to Meteorological Instruments and Methods of Observa-tion. Vol. I. Geneva: World Meteorological Organization. isbn: 978-92-63-10008-5.

Zhong, W. y J. D. Haigh (2003). ((Shortwave Radiative Forcing by Stratosphe-ric Water Vapor)). En: Geophysical Research Letters 30.3, pag. 1113. issn:0094-8276. doi: 10.1029/2002GL016042. url: http://doi.wiley.com/10.1029/2002GL016042.

Apendice A

Artıculos

A.1. Artıculo 1 43

A.1. Artıculo 1

Tıtulo: Comparison of Integrated Water Vapor from GNSS andRadiosounding at Four GRUAN Stations.Autores

• Vaquero-Martınez, Javier (U. de Extremadura).• Anton, Manuel (U. de Extremadura).• Ortiz de Galisteo, Jose P. (AEMet).• Roman, Roberto (U. de Valladolid).• Cachorro, Victoria E. (U. de Valladolid).• Mateos, David (U. de Valladolid).

Ano: 2019.Revista: Science of The Total Environment.Paginas: 1639–1648.DOI: 10.1016/j.scitotenv.2018.08.192.Volumen: 648.

Science of the Total Environment 648 (2019) 1639–1648

Contents lists available at ScienceDirect

Science of the Total Environment

j ourna l homepage: www.e lsev ie r .com/ locate /sc i totenv

Comparison of integrated water vapor from GNSSand radiosounding at four GRUAN stations

Javier Vaquero-Martíneza, b,*, Manuel Antóna, b, José Pablo Ortiz de Galisteoc, d,Roberto Románd, Victoria E. Cachorrod, David Mateosd

aDepartamento de Física, Universidad de Extremadura, Badajoz, SpainbInstituto Universitario de Investigación del Agua, Cambio Climático y Sostenibilidad (IACYS), Universidad de Extremadura, Badajoz, SpaincAgencia Estatal de Meteorología (AEMET), Valladolid, SpaindGrupo de Óptica Atmosférica, Universidad de Valladolid, Valladolid, Spain

H I G H L I G H T S

• GRUAN’sGNSSandradiosounding(RS)IWV product show good agreement(R2 0.98).

• Main influence in GNSS-RS differencesis due to IWV values (worse as IWVincreases).

• SZA and seasonality also influence dif-ferences, partly related to IWV values.

• Differences’ influence on pressurecould be partly related to GNSS’ ZHDmodeling.

• Clouds do not show a significativeinfluence in GNSS-RS differences.

G R A P H I C A L A B S T R A C T

A R T I C L E I N F O

Article history:Received 25 May 2018Received in revised form 23 July 2018Accepted 14 August 2018Available online 18 August 2018

Editor: Ashantha Goonetilleke

Keywords:Integrated water vaporGNSSRadiosoundingGRUANValidation

A B S T R A C T

Integrated water vapor (IWV) data from Global Navigation Satellite Systems (GNSS) and radiosounding (RS)are compared over four sites (Lindenberg, Ny-Ålesund, Lauder and Sodankylä), which are part of the GlobalClimate Observing System (GCOS) Reference Upper Air Network (GRUAN). Both datasets show an excellentagreement, with a high degree of correlation (R2 over 0.98). Dependences of GNSS-RS differences on sev-eral variables are studied in detail. Mean bias error (MBE) and standard deviation (SD) increase with IWV,but in relative term, these variables decrease as IWV increases. The dependence on solar zenith angle (SZA)is partially related to the distribution of IWV with SZA, but the increase of SD for low SZA could be associ-ated with errors in the humidity sensor. Large surface pressures worsen performance, which could be dueto the fact that low IWV is typically present in high pressure situations. Cloud cover shows a weak influ-ence on the mentioned MBE and SD. The horizontal displacement of radiosondes generally causes SD toincrease and MBE to decrease (increase without sign), as it could be expected. The results point out thatGNSS measurements are useful to analyze performance to other instruments measuring IWV.

© 2018 Elsevier B.V. All rights reserved.

* Corresponding author at: Departamento de Física, Universidad de Extremadura,Badajoz, Spain.

E-mail address: javier_vm@unex.es (J. Vaquero-Martínez).

1. Introduction

Water vapor has a paramount relevance in the climate system,since it is acknowledged as the most important atmosphericgreenhouse gas, and despite of not being directly involved in global

https://doi.org/10.1016/j.scitotenv.2018.08.1920048-9697/© 2018 Elsevier B.V. All rights reserved.

A.1. Artıculo 1 45

1640 J. Vaquero-Martínez et al. / Science of the Total Environment 648 (2019) 1639–1648

Table 1Location of the GNSS stations and days with IWV and ZTD data available.

Site Corresponding RS site Latitude (◦N) Longitude (◦E) Altitude (m) Days with IWV data Days with ZTD data

ldb0 LIN 52.124 14.070 0.002 2143 2164ldb2 LIN 52.123 14.072 0.160 138 148ldrz LAU −45.022 169.410 0.380 41 98nya1 NYA 78.555 11.515 0.084 1873 1898nya2 NYA 78.555 11.513 0.082 0 27nyal NYA 78.555 11.521 0.082 0 0soda SOD 67.251 26.232 0.300 36 1402sodf SOD 67.216 26.375 0.213 0 1

Table 2Location of RS stations, distance to GNSS sites, and coincident period for both instruments.

Site Latitude (◦N) Longitude (◦E) Altitude (m) Distance (km) Coincident period

LIN 52.210 14.120 112 10.2 12/11/2012 to 04/15/2015LAU −45.050 169.680 370 21.5 06/08/2005 to 01/22/2018SOD 67.370 26.630 179 21.6 05/21/2006 to 05/02/2017NYA 78.923 11.923 16 42.1 05/15/2007 to 01/10/2018

warming, it causes a positive radiative feedback on climate system(Colman, 2003, 2015). It also plays a fundamental role in energytransport, evaporating at low latitudes, and being transported tohigherlatitudeswhereitcondensates,releasinghighamountsof latentheat (Myhre et al., 2013).

Integrated water vapor (IWV) is the variable commonly used tostudy the atmospheric water vapor. IWV is a magnitude equivalentto condensing all the water vapor in the atmospheric vertical column

and measuring the height that it would reach if contained in a ves-sel of unit cross section; being its units those of superficial density(g mm−2) or length (mm).

However, understanding of water vapor effects on climate stillneeds improving because of the high variability of this gas, bothspatially and temporally. It is therefore necessary to retrieve qualitywater vapor data. Radiosounding (RS) is one of the more precise anddirect ways to measure water vapor profiles, and from them IWV

Fig. 1. Scatterplots for GNSS-derived IWV from meteorological data provided by GRUAN (x-axis) and meteorological data provided by radiosounding (y-axis) for the four GRUANstations. Color, continuous lines are regression lines and black, dashed lines are the identity line. (For interpretation of the references to color in this figure legend, the reader isreferred to the web version of this article.)

46 Apendice A. Artıculos

J. Vaquero-Martínez et al. / Science of the Total Environment 648 (2019) 1639–1648 1641

Table 3Statistics of the differences GNSS IWV − RS IWV (all in mm, except slope and R2,which are unitless). MABE is mean absolute bias error, MEDIAN is the median of thedifferences, IQR is the inter-quartile range of the difference and N the number ofdata-points.

Site MBE SD MABE MEDIAN IQR N

LAU −0.767 0.672 0.855 −0.753 0.658 109LIN −0.874 1.099 1.094 −0.833 1.150 7837NYA −0.492 0.614 0.600 −0.449 0.712 2164SOD −0.516 0.830 0.726 −0.435 0.957 2118

data, despite its limitation of temporal resolution (typically one ortwo launches per day). RS is therefore established as a referenceto validate other instruments (du Piesanie et al., 2013; Ohtani andNaito, 2000; Antón et al., 2015). However, it still has some sourcesof errors as explained in Wang and Zhang (2008) and Dirksen et al.(2014), most of them due to the problem of changes in the radiosondemodels and errors in the humidity sensor related to heating by solarradiation.

Moreover, Global Navigation Satellite Systems (GNSS) meteorol-ogy is a relatively recent technique that can be used to derive IWVdata (Bevis et al., 1992). GNSS measurements have some advantages:all-weather availability, high temporal resolution (5 min to 2 h), high

accuracy (less than 3 mm in IWV) and long-term stability. Hence,GNSS data are also used as reference to validate other instruments(Köpken, 2001; Prasad and Singh, 2009; Rama Varma Raja et al.,2008; Román et al., 2015; Vaquero-Martínez et al., 2017a,b, 2018),but as the recent technique that it is, GNSS meteorology still needsvalidation and assessment of quality in different parts of the Globe.

The Global Climate Observing System (GCOS) Reference Upper-Air Network (GRUAN) has recognized the need of having redun-dant water vapor measurements in order to improve their quality(GRUAN, 2007). Hence, GRUAN stations that already measure watervapor with RS are being equipped with GNSS receivers and a GRUANGNSS water vapor product is being developed (WMO, 2008).

The main goal of this study is to analyze the possible errors of thenew GNSS IWV products in order to assess their use for other pur-poses, allowing an improvement in temporal resolution as comparedwith traditional RS. This way, in this article compare the IWV fromGNSS against IWV from RS at the four GRUAN stations with bothRS and GNSS water vapor data currently available, and analyze thecauses of the differences.

This article is organized as follows: Section 2 describes the dif-ferent datasets used and their characteristics, and the methodologyused in this work. Section 3 includes the results and its discussion,validating the GNSS retrieval performed by the authors for com-parison purposes, and analyzing the comparison results. Section 4summarizes the main conclusions.

Fig. 2. Scatterplots for GNSS IWV data (y-axis) and RS IWV data (x-axis) for the four GRUAN stations. Color, dashed lines are regression lines and black, continuous lines are theidentity line. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

A.1. Artıculo 1 47

1642 J. Vaquero-Martínez et al. / Science of the Total Environment 648 (2019) 1639–1648

2. Material and methods

2.1. IWV from GRUAN GNSS

GNSS consists of a series of satellites that communicate throughL-band microwave radiation with receivers, mainly in order to esti-mate these receivers’ locations. The method to obtain IWV fromGNSS measurements is detailed in Bevis et al. (1992), and brieflyexplained in the following lines.

The time spent by the signal in reaching the receiver can be usedto calculate the distance between the satellite and receiver, and tak-ing into account the position of the satellites, to obtain the receiver’sposition. However, several corrections need to be applied, since thesignal suffers a series of delays in its travel to the receiver. There isa particular contribution, the Slant Tropospheric Delay (STD), thatallows IWV calculation. This contribution refers to the delay thatthe troposphere causes in the signal, and is referred to the paththat the signal follows. Mapping functions (Niell, 2000; Boehm etal., 2006a,b) can be applied to obtain the zenithal equivalent of thisamount, the Zenith Tropospheric Delay (ZTD). ZTD is the sum of twocontributions, one related to the non-dipolar contribution of all gasesin the troposphere (Zenith Hydrostatic Delay, ZHD), and anotherrelated to the dipolar contribution of water vapor (Zenith Wet Delay,ZWD) since it is the only compound with dipolar momentum in theatmosphere. A simple model can estimate accurately ZHD (Saasta-moinen, 1972), based on surface pressure. This model is accurate tothe submilimeter region except if that the hydrostatic equilibriumcondition does not hold; in that case errors can reach 1 mm in ZHD.

The performance of other models are similar (Opaluwa et al., 2013).Once ZHD is obtained, ZWD can be estimated as ZWD = ZTD−ZHD.

Additionally, another variable is necessary to convert ZWD toIWV, the water vapor weighted mean temperature in the verticalcolumn (Tm). Tm is defined as Eq. (1):

Tm =

∫ PvT dz∫ Pv

T2 dz, (1)

where Pv is water vapor partial pressure and T is the temperature,both at altitude z. Tm is often estimated from surface temperaturefrom meteorological stations, using empirical fits, or obtained fromre-analysis or radiosondes.

The product used in this work is developed by GRUAN GNSS (GG)Precipitable Water Vapour Task Team. Ground-based GNSS IWV hasbeen identified as a Priority 1 measurement for GRUAN. Therefore,a lot of efforts are being done in the last few years to implementthis kind of measurements in GRUAN sites. The sites are Lindenberg(LIN), Sodankylä (SOD), Lauder (LAU) and Ny-Ålesund (NYA). Despitethe voluntary nature of GG sites, the GG sites must follow a seriesof guidelines in order to ensure the quality of GG IWV data. Thus,these sites must be equipped with automatic meteorological stationsor there must be a nearby station. The GG locations involved in thiswork are detailed in Table 1.

GRUAN network provides both ZTD and IWV products for thosestations equipped with GNSS. However, sometimes meteorological

Fig. 3. MBE (top) and SD (bottom) of GNSS-RS differences (%) with respect to IWV from RS for the four GRUAN stations.

48 Apendice A. Artıculos

J. Vaquero-Martínez et al. / Science of the Total Environment 648 (2019) 1639–1648 1643

data (pressure and temperature) are not available and GRUAN pro-vides only ZTD product. The number of days with GG IWV dataat every station available for this study is also shown in Table 1.It can be observed that LAU and SOD stations exhibit a reducednumber of days with original GG IWV data. To solve this issue andincrease the data number, in this work, GRUAN radiosonde meteoro-logical data (Tm and surface pressure) are used to obtain a new IWVproduct from GG ZTD data (obtained by authors for comparison pur-poses only). This new product, developed for comparison purposes,is named in this work as “re-calculated GG IWV product”, while theGNSS IWV product retrieved directly from GRUAN have been namedas “original GG IWV product”. Table 1 shows the number of avail-able days with this re-calculated GG IWV product. It must be notedthe notable increase of available days, particularly for LAU and SODsites. Some restrictions have been applied to ensure data quality:

• Resulting values of IWV must make sense (0 mm < IWV <100 mm).

• Mean weighted temperature must be lower than 500 K andpositive.

2.2. Radiosoundings from GRUAN network

GRUAN network provides radiosonde data for 28 sites. We haveconsidered those sites that also have a nearby GNSS product from

GRUAN. Table 2 shows the locations of the four sites considered inthis work.

Typically the radiosonde launches are at specific hours. LIN typ-ically has 4 launches a day (00, 06, 12, 18 h), while NYA’s sondesare typically launched at 12 h, and some launches at other hours,specially at 00, 06, and 18 h. Sondes at SOD are launched at 00 and12 h (some others at different hours), and at LAU at different hours(approximately one launch per week).

The radiosondes that provide the data in this work are VaisalaRS92. The RS92 model is equipped with a wire-like capacitivetemperature sensor (“thermocap”); two polymer capacitive mois-ture sensor (“humicap”), a silicon-based pressure sensor and a GPSreceiver. More detailed information about the processing of thedata retrieved can be found at https://www.gruan.org/instruments/radiosondes/sonde-models/vaisala-rs92/ or Dirksen et al. (2014).The main error sources that affect the humidity sensor are daytimesolar heating of the Humicaps (introduces a dry bias), sensor time-lag at temperatures below about −40◦ (this is not a problem in thiswork) and temperature dependent calibration correction.

The GRUAN RS92 product includes data on profiles of pressure,temperature, humidity, relative humidity, water vapor mixing ratio,wind information, frostpoint, short-wave radiation, and associateduncertainties. IWV can be calculated by integration of water vapormixing ratio (WVMR) in pressures as Eq. (2):

IWV =∫ ps

0WVMR • dp, (2)

Fig. 4. MBE (top) and SD (bottom) of GNSS-RS differences (%) with respect to SZA for three GRUAN stations.

A.1. Artıculo 1 49

1644 J. Vaquero-Martínez et al. / Science of the Total Environment 648 (2019) 1639–1648

where WVMR is the water vapor mixing ratio, p is the pressureand ps the surface pressure. In addition, some restrictions have beenconsidered in order to ensure GRUAN data quality:

• Number of levels must be more than 15.• First level must be at height lower than 1 km.• Last level must be at height larger than 9 km.• Resulting values of IWV must make sense 0 mm < IWV <

100 mm.

2.3. Methodology

The followed criterion to match the GNSS and RS data require thattime differences between RS launch and GNSS measurement must bebelow 30 min. For the analysis of differences, RS measurements havebeen considered as reference and two variables have been analyzed,physical difference (GNSS minus RS) and relative difference (differ-ence divided by RS value). The mean of the differences (also known asmean bias error, MBE) and the standard deviation of the differences(SD) have been calculated. The SD have been used as a measurementof precision and the MBE as measurement of accuracy. The MBE iscalculated as Eq. (3):

MBE =1N

N∑i

di, (3)

where di are the physical differences (absolute MBE) or the relativedifferences (relative MBE). Moreover the SD is obtained as Eq. (4):

SD =

√√√√ 1N − 1

N∑i

(di − di

). (4)

In order to study whether these differences depend on other vari-ables or not, the data have been divided into several bins of similarvalues of these variables for the study of the precision and accuracyof IWV in each bin. It must be noticed that data bins with less than15 data have been rejected, as not representative.

3. Results and discussion

3.1. Original GG IWV data vs re-calculated GG IWV data

Fig. 1 shows the correlation between the original and re-calculatedGG IWV data. In all stations both data-sets exhibit an excellent agree-ment (R2 ∼ 0.99). All stations show negative offsets (except NYA,which is positive), but all are quite small, less than 0.4 mm in allcases. Outliers, like the ones in NYA and LIN (differences of more than1.5 mm in IWV), are mainly caused by the differences in pressuremeasurements. However, around 90% of the data pairs differ by lessthan 0.7 mm.

Fig. 5. MBE (top) and SD (bottom) of GNSS-RS differences (%) with respect to pressure for the four GRUAN stations.

50 Apendice A. Artıculos

J. Vaquero-Martínez et al. / Science of the Total Environment 648 (2019) 1639–1648 1645

Therefore, the data-set of GNSS-derived IWV using meteorolog-ical data from radiosonde (GNSSRS) represents very well GRUAN’sIWV product. In order to have a data-set with the same features, allthe data used in this work will come from the GNSS-derived IWVusing meteorological data from radiosonde. The advantages of usingthis data-set are as follows:

1. More data is available (particularly at SOD and LAU stations).2. Davis “mean” temperature can be obtained directly from

radiosonde.3. Temporal interpolation is not necessary.

Needless to say, this is only for comparison purposes, since theradiosonde meteorological data is typically available for at most fourtimes a day, and the GNSS products are available every 15 min.

3.2. Comparison between GNSS IWV and RS IWV

3.2.1. Overall statistics and regressionsTable 3 shows a summary of the statistics of the differences

between IWV from GNSS and RS. MBE values are over −0.9 mm forall stations, being closer to zero for NYA and SOD (around 0.5 mm).SD values are around 0.6–1 mm. Median and MBE values are similar,which indicates that the differences distributions are most likelynormal. Fig. 2 shows the regression lines. Both data-sets are inagreement with R2 around 0.98.

The differences GNSS-RS and relative differences are analyzed inthis section in order to find dependence on different variables. Thedifferences are distributed into bins of similar values of the variable

analyzed, and the evolution of MBE and SD over the different bins isanalyzed. It must be noticed that the data bins with less than 15 dataare not shown, as they are not considered representative.

3.2.2. Dependence of GNSS-RS differences on IWVThe available data-set have been divided into bins of 5 mm. All

stations have a very similar behaviour with respect to IWV. Therelative MBE in Fig. 3 (top) shows that there is a dry bias (around 5%)that decreases in absolute value with IWV. However, for SOD firstbin is closer to zero (∼2.5%) than the rest of the bins (∼5%) of SOD.Absolute MBE (not shown) typically increases in absolute value withIWV, ranging from less than −1 mm up to −2 or −2.5 mm. Such smallrange explains the behaviour of relative MBE: absolute differencesdo not change much, but the reference IWV does, thus the relativevalue decrease (in absolute value) as IWV increases.

Regarding precision (see Fig. 3, bottom), relative SD, decrease asIWV increases, reaching a minimum of around 5% in all cases forIWV above 15 mm. Despite the different ranges of IWV and numberof data of each station, the relative SD is very similar in the lowestbin, between 15 and 17%). A similar interpretation to that of the MBEis appropriate here: SD in absolute terms increases with IWV, butin a range (0.5–2 mm) that is quite smaller than the range of IWVitself (0–40 mm), and therefore relative SD tends to decrease withincreasing IWV. Unfortunately, LAU available data does not show awide range of IWV, so it is difficult to interpret the results, but theyare compatible with those observed in the rest of sites, with val-ues around 5–7% in the range of 5–20 mm. A similar behaviour wasobserved in other comparisons between GNSS and satellite prod-ucts (Román et al., 2015; Vaquero-Martínez et al., 2017a,b, 2018) and

Fig. 6. MBE (top) and SD (bottom) of GNSS-RS differences (%) with respect to cloud fraction (CF) for the four GRUAN stations.

A.1. Artıculo 1 51

1646 J. Vaquero-Martínez et al. / Science of the Total Environment 648 (2019) 1639–1648

between RS and satellite products (Antón et al., 2015). Correlationcoefficient R decreases as IWV increases (not shown), from valuesover 0.8 for low IWV to values below 0.7 for IWV above 30 mm.

3.2.3. Dependence of GNSS-RS differences on SZADifferences related to SZA could be due to errors in radiosonde

sensors (especially humidity sensor, which is affected by solar radia-tion), as stated in Wang and Zhang (2008) and Dirksen et al. (2014).Fig. 4 (top) shows relative MBE of every 5◦ bins. It must be noticedthat LAU does not have bins with enough (>15) data, so its resultsare not considered.

Although there are some differences between stations, relativeMBE generally worsens as SZA increases. LIN shows a sharp increaseat SZA = 90◦ (sunrise and sunset), while worsening of MBE withSZA is more monotonous at SOD and NYA, with some increase from110◦. These behaviours are quite related to typical values of IWV forthose SZA bins, especially at LIN: low SZA causes higher tempera-tures, which causes the atmosphere to accept more water vapor andtherefore causes IWV to increase. The distribution of IWV with SZAwas checked, confirming this hypothesis. Also, an interesting fea-ture at LIN IWV was found: SZA increases rapidly around 90◦ anddecreases for SZA above that value. As NYA and SOD are Arctic sta-tions, the influence of SZA is not so marked. Values are typicallybetween 5 and 10%. GOME-2 water vapor product exhibits a simi-lar behaviour, as shown in Antón et al. (2015), but the sign of MBEis positive in that case. Differences between day and night are not

important, although in Wang and Zhang (2008) Vaisala RS92 showeda worse performance at day than at night.

In relative terms, as Fig. 4 (bottom) reveals, SD increases with SZA.At nighttime, relative SD is higher and more stable, and at daytime,it is lower and has a increasing tendency with SZA. Minimum rela-tive SD for all stations is around 5%, but the maximum differs (10%for LIN, 15% for NYA, and 20% for SOD). This behaviour can be par-tially due to the observed increase in relative SD for low IWV, with asimilar argument to the one provided for relative MBE in this section.In absolute terms (not shown), SD decreases with SZA, which is con-sistent with this argument, but it could also be related to the factthat at low SZA the radiosondes humidity sensor can be affected bysolar radiation (Dirksen et al., 2014; Wang and Zhang, 2008) andpartly because of the typically higher IWV values at low SZA. Severalsatellite product showed similar behaviour (but with less precision)(Vaquero-Martínez et al., 2018).

In this subsection, it is also analyzed the seasonal dependence ofGNSS-RS differences. SZA and IWV both have annual cycles, whichcause the MBE and SD of the differences between IWV from GNSSand RS to have a seasonal dependence as well. LIN and NYA exhibit(not shown) slightly worse relative MBE in winter (low IWV) thanin summer, while SOD (not shown) has worse relative MBE at sum-mer (higher IWV). Relative SD in LIN, NYA and SOD are smallerat summer (low SZA) than in winter. The hypothesis that seasonaldependence on water vapor products performance is mainly affectedby dependences on IWV and SZA is also proposed in other works

Fig. 7. MBE (top) and SD (bottom) of GNSS-RS differences (%) with respect to RS displacement for the four GRUAN stations.

52 Apendice A. Artıculos

J. Vaquero-Martínez et al. / Science of the Total Environment 648 (2019) 1639–1648 1647

where satellite products are compared with GNSS ground-basedmeasurements (Vaquero-Martínez et al., 2017a,b, 2018).

3.2.4. Dependence on pressureSurface pressure also affects to the GNSS-RS differences. Fig. 5

(top) shows the MBE each 5 hPa bins. Relative MBE increases with-out sign as pressure increases. Values are between −15% and 0%approximately. At high pressures, MBE worsens at a sharper rate.This could be caused by the distribution of IWV with surface pres-sure: at high pressure, IWV is smaller, being the relative MBE higher.Another explanation that could contribute partially to this behaviouris related to the way that GNSS IWV is retrieved, since the surfacepressure is needed in Saastamoinen’s model (Saastamoinen, 1972).

Relative SD, shown in Fig. 5) (top), increases with pressure. Valuesare between 5–10% (LIN), around 10% (NYA) and 5–20% (SOD). LAUshows slight lower values, around 5% but these values are only forlow IWV pressure values. As it also happens with MBE, this behaviourcould be partially due to the distribution of IWV with pressure: lowervalues of IWV are generally registered at higher values of pressure.SD in absolute terms (not shown) exhibits a maximum (1000 hPa forLIN, 980 hPa for SOD) that is coincident with a maximum in typicalIWV values.

3.2.5. Dependence of GNSS-RS differences on cloudinessTotal cloud cover data have been obtained from Era-Interim

Reanalysis (Dee et al., 2011), and co-located to the sites and times ofIWV measurements. These data are in the form of cloud fraction (CF),that is to say, a number between 0 (no clouds) and 1 (totally covered)indicating the pixel cloud cover.

Relative MBE, as shown in Fig. 6 (top), is above −4% for LIN andSOD, and between −4 and 12% for NYA. LAU only counts with 1 point,positive relative MBE (less than 2%). However, the results do notshow any dependence of MBE on CF. MBE in absolute terms does notshow any dependence on CF either.

Regarding relative SD, no tendency is observed (see Fig. 6 (bot-tom)). LIN has very stable values around 8%. NYA however, havehighly variable values of SD, some around 7%, other more than 12%,with high uncertainties. Nevertheless, SOD exhibits a slight tendencyto decrease SD as CF increases, although still with high variability(between 7% and 15%) and uncertainties.

3.2.6. Dependence on radiosonde horizontal movementRadiosondes usually move horizontally due to winds. This could

be a source of error (Seidel et al., 2011), so it must be taken intoaccount. The distance is obtained as the horizontal distance betweenthe first (closest to the ground) and last (furthest from the ground)radiosonde positions. 20 km bins have been used to study the evolu-tion of MBE and SD throughout the distances.

Fig. 7 (top) clearly shows that relative MBE is farther from zero ashorizontal displacement increases at NYA, but there is no importanttrend for the other sites. A reason for this could be that NYA site islocated in the Island of Spitsbergen, meaning that a displacement canput the radiosonde over the sea, where differences with the genuinewater vapor vertical profile can be more important. SOD shows a veryhigh variability, which could be due to inhomogeneous terrain (andthus, humidity) in the vicinity of the site. Relative MBE changes from−4% to −9% at LIN, and from −10% to −20% at NYA.

Fig. 7 (bottom) shows the relative SD for several horizontal bins,which clearly increases as the horizontal displacement increases,which is to be expected. LIN goes from 5% to 15%, NYA from 10% to20%, and SOD from 0% to 20%. It must be noted the high variability inSOD relative SD values, which can be caused by the inhomogeneityof the humidity fields in the vicinity around the site.

4. Conclusions

Global Climate Observing System (GCOS) Reference Upper AirNetwork (GRUAN)’s Global Navigation Satellite System (GNSS) andradiosonde (RS) integrated water vapor (IWV) products are in agree-ment at the sites considered. The regression analysis showed a highcorrelation (R2 > 0.98) and certain offset that can be due to the spa-tial separation between GNSS and RS stations. The intercept is pos-itive for all stations except NYA, and the magnitude ranges around0.1–0.2 mm. Values of the standard deviation of the differences (SD)are between 0.6 and 1 mm.

The study on dependences of the GNSS-RS differences showedthat the mean of the differences (MBE) and SD generally increase(omitting the sign of MBE) with IWV, although relative MBE and SDshowed the opposed behaviour. Performance of RS IWV product wasexpected to worsen at low solar zenith angle (SZA) because of errorsin humidity sensor of radiosondes but this was not observed, so cor-rections are being applied correctly. However, SD does increase atlow SZA. Most of the observed dependences on SZA are probablyrelated to the distribution of IWV with SZA (IWV is larger at low SZA,when the temperatures are higher). The dependences on SZA andIWV also cause a seasonal dependence.

MBE (without sign) and SD exhibits an increase with increasingsurface pressure, that can be partially due to the distribution of IWVwith pressure (IWV is smaller at high pressures), and partially toerrors in the modeling of ZHD through Saastamoinen’s model. How-ever, this is an issue that shall be studied closely in future work.Cloud cover did not show a significant influence on MBE and SD.Regarding dependence on horizontal displacement of radiosondes,the relative MBE and SD show that the performance of RS is poorerwhen the horizontal displacement is larger, although this seems tobe very influence by the characteristics of the site’s vicinity.

Insummary, theGNSSandRSvaluesareverysimilarandthedepen-dences on other factors low, but it should be pointed out that it is stillvery necessary to have redundant measurements of water vapor inordertoimproveboththequalityofmeasurementsandthesamplingofthe data. GNSS exhibits two important advantages: first the high tem-poral resolution, and second the stability against the sky conditions(wind, clouds, etc.), which make GNSS IWV measurement particularlywell suited for comparison purposes. However, it must be noticed thatthe low number of stations do not allow to extract conclusions overthe whole range of the variables studied, mainly IWV and SZA.

Acknowledgments

Support from the Junta de Extremadura (Research Group GrantGR15137) is gratefully acknowledged. Work at the Universidad deValladolid is supported by project CMT2015-66742-R. The authorswish to thank the operators at the four observatories (Lindenberg,Lauder, Ny-Ålesund and Sodankylä) for dutifully performing refer-ence radiosoundings and maintenance of GNSS according to theGRUAN standards, as well as GFZ Helmholtz Center Postdam fortheir processing of GNSS data products to obtain ZTD and IWV, andacknowledge ERA-Interim data.

References

Antón, M., Loyola, D., Román, R., Vömel, H., 2015. Validation of GOME-2/MetOp-a totalwater vapour column using reference radiosonde data from the GRUAN network.Atmos. Meas. Tech. 8, 1135–1145. https://doi.org/10.5194/amt-8-1135-2015.

Bevis, M., Businger, S., Herring, T.A., Rocken, C., Anthes, R.A., Ware, R.H., 1992.GPS meteorology: remote sensing of atmospheric water vapor using the globalpositioning system. J. Geophys. Res. 97, 15787–15801. https://doi.org/10.1029/92JD01517. URL: http://doi.wiley.com/10.1029/92JD01517.

Boehm, J., Niell, A., Tregoning, P., Schuh, H., 2006a. Global mapping function (GMF):a new empirical mapping function based on numerical weather model data.Geophys. Res. Lett. 33, 3–6. https://doi.org/10.1029/2005GL025546.

A.1. Artıculo 1 53

1648 J. Vaquero-Martínez et al. / Science of the Total Environment 648 (2019) 1639–1648

Boehm, J., Werl, B., Schuh, H., 2006b. Troposphere mapping functions for GPS and verylong baseline interferometry from European Centre for Medium-Range WeatherForecasts operational analysis data. J. Geophys. Res. Solid Earth 111, 1–9. https://doi.org/10.1029/2005JB003629.

Colman, R., 2003. A comparison of climate feedbacks in general circulation models.Clim. Dyn. 20, 865–873. http://arXiv:183550600013. https://doi.org/10.1007/s00382-003-0310-z. URL: http://link.springer.com/article/10.1007/s00382-003-0310-z.

Colman, R.A., 2015. Climate radiative feedbacks and adjustments at the Earth’s surface.J. Geophys. Res. Atmos. 120, 3173–3182. https://doi.org/10.1002/2014JD022896.URL: http://doi.wiley.com/10.1002/2014JD022896.

Dee, D.P., Uppala, S.M., Simmons, A.J., Berrisford, P., Poli, P., Kobayashi, S., Andrae, U.,Balmaseda, M.A., Balsamo, G., Bauer, P., Bechtold, P., Beljaars, A.C.M., van de Berg, L.,Bidlot, J., Bormann, N., Delsol, C., Dragani, R., Fuentes, M., Geer, A.J., Haimberger, L.,Healy, S.B., Hersbach, H., Hólm, E.V., Isaksen, L., Kållberg, P., Köhler, M., Matricardi,M., McNally, A.P., Monge-Sanz, B.M., Morcrette, J.-J., Park, B.-K., Peubey, C., deRosnay, P., Tavolato, C., Thépaut, J.-N., Vitart, F., 2011. The ERA-Interim reanalysis:configuration and performance of the data assimilation system. Q. J. R. Meteorol.Soc. 137, 553–597. https://doi.org/10.1002/qj.828.

Dirksen, R.J., Sommer, M., Immler, F.J., Hurst, D.F., Kivi, R., Vömel, H., 2014. Referencequality upper-air measurements: GRUAN data processing for the Vaisala RS92radiosonde. Atmos. Meas. Tech. 7, 4463–4490. https://doi.org/10.5194/amt-7-4463-2014. URL: http://www.atmos-meas-tech.net/7/4463/2014/.

du Piesanie, A., Piters, A.J.M., Aben, I., Schrijver, H., Wang, P., Noël, S., 2013. Validationof two independent retrievals of SCIAMACHY water vapour columns usingradiosonde data. Atmos. Meas. Tech. 6, 2925–2940. https://doi.org/10.5194/amt-6-2925-2013. URL: http://www.atmos-meas-tech.net/6/2925/2013/.

GRUAN, 2007. Justification, Requirements, Siting, and Instrumentation Options.URL:https://www.gruan.org/gruan/editor/documents/gcos/gcos-112.pdf.

Köpken, C., 2001. Validation of integrated water vapor from numerical modelsusing ground-based GPS, SSM/I, and water vapor radiometer measurements. J.Appl. Meteorol. 40, 1105–1117. ISSN 0894-8763. https://doi.org/10.1175/1520-0450(2001)040<1105:VOIWVF>2.0.CO;2. URL: http://journals.ametsoc.org/doi/abs/10.1175/1520-0450%282001%29040%3C1105%3AVOIWVF%3E2.0.CO%3B2.

Myhre, G., Shindell, D., Bréon, F.-M., Collins, W., Fuglestvedt, J., Huang, J., Koch,D., Lamarque, J.-F., Lee, D., Mendoza, B., Nakajima, T., Robock, A., Stephens,G., Takemura, T., Zhang, H., 2013. Anthropogenic and natural radiative forcing.Climate Change 2013: The Physical Science Basis. Contribution of Working GroupI to the Fifth Assessment Report of the Intergovernmental Panel on ClimateChange. IPCC., pp. 659–740. (Ipcc ed.).

Niell, A.E., 2000. Improved atmospheric mapping functions for VLBI and GPS. EarthPlanets Space 52, 699–702. https://doi.org/10.1186/BF03352267. URL: http://earth-planets-space.springeropen.com/articles/10.1186/BF03352267.

Ohtani, R., Naito, I., 2000. Comparisons of GPS-derived precipitable water vaporswith radiosonde observations in Japan. J. Geophys. Res. Atmos. 105, 26917–26929. https://doi.org/10.1029/2000JD900362. URL: http://doi.wiley.com/10.1029/2000JD900362.

Opaluwa, Y.D., Adejare, Z.A.T., Abazu, I.C., Adewale, T.O., Odesanmi, A.O., Okorocha,V.C., 2013. Comparative analysis of five standard dry tropospheric delay modelsfor estimation of dry tropospheric delay in GNSS positioning. Am. J. Geogr. Inf.Syst. 11.

Prasad, A.K., Singh, R.P., 2009. Validation of MODIS Terra, AIRS, NCEP/DOE AMIP-II Reanalysis-2, and AERONET Sun photometer derived integrated precipitablewater vapor using ground-based GPS receivers over India. J. Geophys. Res.114, D05107. http://arXiv:16510847. https://doi.org/10.1029/2008JD011230.URL: http://doi.wiley.com/10.1029/2008JD011230.

Rama Varma Raja, M.K., Gutman, S.I., Yoe, J.G., McMillin, L.M., Zhao, J., 2008.The validation of AIRS retrievals of integrated precipitable water vapor usingmeasurements from a network of ground-based GPS receivers over the contiguousUnited States. J. Atmos. Ocean. Technol. 25, 416–428. https://doi.org/10.1175/2007JTECHA889.1.

Román, R., Antón, M., Cachorro, V., Loyola, D., Ortiz de Galisteo, J., de Frutos, A.,Romero-Campos, P., 2015. Comparison of total water vapor column from GOME-2 on MetOp-A against ground-based GPS measurements at the Iberian Peninsula.Sci. Total Environ. 533, 317–328. http://arXiv:26172599. https://doi.org/10.1016/j.scitotenv.2015.06.124. URL: http://linkinghub.elsevier.com/retrieve/pii/S0048969715303260.

Saastamoinen, J., 1972. Atmospheric correction for the troposphere and stratospherein radio ranging satellites. In: Henriksen, S.W., Mancini, A., Chovitz, B.H.(Eds.), Geophysical Monograph Series. American Geophysical Union., pp. 247–251. URL: http://doi.wiley.com/10.1029/GM015p0247. https://doi.org/10.1029/GM015p0247.

Seidel, D.J., Sun, B., Pettey, M., Reale, A., 2011. Global radiosonde balloon drift statis-tics. J. Geophys. Res. 116, https://doi.org/10.1029/2010JD014891. URL: http://doi.wiley.com/10.1029/2010JD014891.

Vaquero-Martínez, J., Antón, M., Ortiz de Galisteo, J.P., Cachorro, V.E., Álvarez-Zapatero, P., Román, R., Loyola, D., Costa, M.J., Wang, H., Abad, G.G., Noël, S., 2018.Inter-comparison of integrated water vapor from satellite instruments usingreference GPS data at the Iberian Peninsula. Remote Sens. Environ. 204, 729–740. https://doi.org/10.1016/j.rse.2017.09.028. URL: http://linkinghub.elsevier.com/retrieve/pii/S0034425717304406.

Vaquero-Martínez, J., Antón, M., Ortiz de Galisteo, J.P., Cachorro, V.E., Costa, M.J.,Román, R., Bennouna, Y.S., 2017a. Validation of MODIS integrated water vaporproduct against reference GPS data at the Iberian Peninsula. Int. J. Appl. Earth Obs.Geoinf. 63, 214–221. http://arXiv:27988187. https://doi.org/10.1016/j.jag.2017.07.008. URL: http://linkinghub.elsevier.com/retrieve/pii/S0048969716327176.

Vaquero-Martínez, J., Antón, M., Ortiz de Galisteo, J.P., Cachorro, V.E., Wang, H., GonzálezAbad, G., Román, R., Costa, M.J., 2017b. Validation of integrated water vapor fromOMI satellite instrument against reference GPS data at the Iberian Peninsula.Sci. Total Environ. 580, 857–864. https://doi.org/10.1016/j.scitotenv.2016.12.032.URL: http://linkinghub.elsevier.com/retrieve/pii/S0048969716327176.

Wang, J., Zhang, L., 2008. Systematic errors in global radiosonde precipitable waterdata from comparisons with ground-based GPS measurements. J. Clim. 21, 2218–2238. https://doi.org/10.1175/2007JCLI1944.1.

WMO, 2008. Report of the GCOS Reference Upper-air Network ImplementationMeeting. https://www.gruan.org/gruan/editor/documents/gcos/gcos-121.pdf.

54 Apendice A. Artıculos

A.2. Artıculo 2 55

A.2. Artıculo 2

Tıtulo: Inter-Comparison of Integrated Water Vapor from Sa-tellite Instruments Using Reference GPS Data at the IberianPeninsula.Autores

• Vaquero-Martınez, Javier (U. de Extremadura).• Anton, Manuel (U. de Extremadura).• Ortiz de Galisteo, Jose P. (AEMet).• Cachorro, Victoria E. (U. de Valladolid).• Alvarez-Zapatero, Pablo (U. de Valladolid).• Roman, Roberto (U. de Valladolid).• Diego Loyola (Centro Aeroespacial Aleman - DLR).• Costa, Maria Joao (U. de Evora).• Wang, Huiqun (Smithsonian Astrophysical Observatory).• Gonzalez Abad, Gonzalo (Smithsonian Astrophysical Observatory).• Noel, Stephan (U. de Bremen).

Ano: 2018.Revista: Remote Sensing of Environment.Paginas: 729–740.DOI: 10.1016/j.rse.2017.09.028.Volumen: 204.

Contents lists available at ScienceDirect

Remote Sensing of Environment

journal homepage: www.elsevier.com/locate/rse

Inter-comparison of integrated water vapor from satellite instruments usingreference GPS data at the Iberian Peninsula

Javier Vaquero-Martíneza,b,*, Manuel Antóna,b, José Pablo Ortiz de Galisteoc,d,Victoria E. Cachorrod, Pablo Álvarez-Zapaterod, Roberto Románe,f, Diego Loyolag,Maria João Costah, Huiquin Wangi, Gonzalo González Abadi, Stefan Noëlj

a Departamento de Física, Universidad de Extremadura, Badajoz, Spainb Instituto Universitario de Investigación del Agua, Cambio Climático y Sostenibilidad (IACYS), Universidad de Extremadura, Badajoz, Spainc Agencia Estatal de Meteorologia (AEMET), Valladolid, Spaind Grupo de Óptica Atmosférica, Universidad de Valladolid, Valladolid, Spaine Department of Applied Physics, University of Granada, Granada, Spainf Andalusian Institute for Earth System Research (IISTA-CEAMA), Granada, Spaing German Aerospace Center (DLR), Oberpfaffenhofen, Germanyh Departamento de Física, Instituto de Ciências da Terra, Escola de Ciências e Tecnología, Universidade de Évora, Évora, Portugali Smithsonian Astrophysical Observatory, Cambridge, Massachusetts, United Statesj Institute of Environmental Physics, University of Bremen, Bremen, Germany

A R T I C L E I N F O

Keywords:Water vaporInter-comparisonIWVGPSSatelliteMODISOMIGOME-2SEVIRISCIAMACHYAIRS

A B S T R A C T

This paper focuses on the inter-comparison of integrated water vapor (IWV) products derived from the followingsatellite instruments: Global Ozone Monitoring Instrument (GOME-2), Moderate-Resolution ImagingSpectroradiometer (MODIS) on the Terra and Aqua satellites, Ozone Monitoring Instrument (OMI), SpiningEnhanced Visible and InfraRed Imager (SEVIRI), Atmospheric Infrared Sounder (AIRS), and Scanning ImagingAbsorption Spectrometer for Atmospheric Chartography (SCIAMACHY). IWV data from GPS in nine ground-based stations located in the Iberian Peninsula are used as reference. The study period extends from 2007 to2012. The results show that, in general, OMI has good accuracy (pseudomedian of the relative differencesbetween OMI and GPS IWV of (−0.7±1.1)%). However, OMI, SCIAMACHY and AIRS show higher inter-quartile range (IQR) (which indicates lower precision) than the rest of satellite instruments. Both MODIS satelliteinstruments and SEVIRI products tend to slightly underestimate reference IWV data while GOME-2 exhibits anotable overestimation (16.7± 0.8%). All satellite instruments showed a tendency to reduce IWV extreme va-lues: low IWV is overestimated while high IWV is underestimated. As for the influence of solar zenith angle(SZA), it can be observed that GOME-2 strongly overestimates the reference for high SZA values (by around 60%for SZA 60−80°). OMI shows, however, a high IQR for high SZA values. Both MODIS instruments show anincrease in the pseudomedian of relative differences and IQR with SZA at daytime, with more stable values atnight. Seasonal dependence is mainly due to the SZA and IWV typical values in each season. In general, insummer the tendency is to underestimate with low IQR (which happens when IWV is high and SZA is low), andin winter the trend is to overestimate with high IQR (which happens when IWV is low and SZA is high).SCIAMACHY shows a high pseudomedian in summer and autumn, and lower in winter and spring. It must benoted that GOME-2 shows a higher overestimation and OMI shows a higher IQR than other satellite instrumentsin winter and autumn. The influence of clouds was also studied, showing an increase of IQR as cloudinessincreases in all satellites. Pseudomedian also worsens as cloudiness increases, generally.

1. Introduction

Water vapor plays a crucial role in Earth's radiative balance, since itis the main absorber of the infrared radiation emitted from Earth's

surface, and therefore responsible for air heating in the low layers.Regarding energy transport, water vapor's latent heat is a very effectivemechanism. Water is evaporated at low latitudes, and water vapor istransported to higher latitudes where condensation releases high

http://dx.doi.org/10.1016/j.rse.2017.09.028Received 20 February 2017; Received in revised form 12 September 2017; Accepted 20 September 2017

* Corresponding author.E-mail address: javier_vm@unex.es (J. Vaquero-Martínez).

Remote Sensing of Environment 204 (2018) 729–740

Available online 28 September 20170034-4257/ © 2017 Elsevier Inc. All rights reserved.

T

A.2. Artıculo 2 57

amounts of heat (Myhre et al., 2013). Water vapor is the most im-portant natural greenhouse gas, indispensable for life on Earth. Its hy-droxyl (H −O) bond allows absorption in the infrared region. More-over, it involves a positive feedback loop in climate change, accordingto general circulation models (Colman, 2003). If the temperature ofatmosphere rises, air can hold more water vapor, as the saturationvapor pressure increases with temperature. This further increases thegreenhouse effect, warming the atmosphere.

Quality data for integrated water vapor (IWV) are critical for im-proving current understanding of the effect of water vapor in the cli-mate system. Nevertheless, monitoring water vapor has some difficul-ties. First, its high variability, both temporally and spatially. Watervapor exhibits both an annual cycle (Ortiz de Galisteo et al., 2014) anda diurnal one (Ortiz de Galisteo et al., 2011). Second, the challenge toobtain data under a wide range of sky conditions. Additionally, ground-based water vapor data are particularly scarce over polar and oceanicregions. As a result, satellite measurements are necessary to improvethe spatial coverage.

There are numerous techniques for measuring IWV, both fromground and from space. Among ground-based measurements there aremicrowave radiometers (Turner et al., 2007), sun-photometers (Ichokuet al., 2002), lunar-photometers (Barreto et al., 2013), star-photometers(Pérez-Ramírez et al., 2012), Lidar (Turner et al., 2002), GPS system(Ortiz de Galisteo et al., 2011), and radio-sounding (Torres et al.,2010). Space measurements are performed using satellites which collectinformation from different parts of the electromagnetic spectrum: mi-crowave (Jones et al., 2009), visible (Román et al., 2015; Wang et al.,2014), near-infra-red (Grossi et al., 2015) and infra-red (Bennounaet al., 2013).

Radiosonde and GPS are the most powerful techniques to measureIWV. However, temporal coverage of radiosonde is very limited (gen-erally one or two measurements a day). Because of this, GPS is used inthis study as reference to validate satellite IWV data. GPS ground-basedretrieval of water vapor has been studied broadly, as in Ortiz deGalisteo et al. (2010), for GPS antenna corrections, and in Pany et al.(2001) and De Haan et al. (2002), where GPS data were compared witha numerical model. One of the key features of GPS IWV retrieval is itsindependence of meteorological events (Rohm et al., 2014), such ascloudiness or precipitation, along with its high temporal resolution, asmentioned above.

Nevertheless, the coverage of GPS stations is currently not sufficientto represent the high spatial variability of water vapor. Some applica-tions, such as weather forecasts and climate studies, need global datawith higher spatial resolution, and therefore satellite observations areuseful in those cases. However, satellite retrievals have two main pro-blems (Diedrich et al., 2016). On the one hand, if they are low Earthorbiting satellites, they do not adequately sample the diurnal cycle(only one or two measurements a day). On the other hand, if visible orNIR spectra are used, the opacity of clouds makes the measurementsunder cloudy-sky condition unreliable (Diedrich et al., 2016).

In this work, a detailed inter-comparison between IWV data fromseven satellite instruments against reference GPS measurements is

performed. The instruments are: Global Ozone Monitoring Instrument(GOME-2), Moderate-Resolution Imaging Spectroradiometer (MODIS)on the Terra and Aqua satellites, Ozone Monitoring Instrument (OMI),Spining Enhanced Visible and InfraRed Imager (SEVIRI), AtmosphericInfrared Sounder (AIRS), and Scanning Imaging AbsorptionSpectrometer for Atmospheric Chartography (SCIAMACHY). GOME-2IWV data have been widely validated (Noël et al., 2008; Antón et al.,2015; Grossi et al., 2015; Román et al., 2015; Kalakoski et al., 2016), aswell as MODIS water vapor products (Li et al., 2003; Gao and Li, 2008;Prasad and Singh, 2009; Bennouna et al., 2013; Chang et al., 2015;Ningombam et al., 2016; Vaquero-Martínez et al., 2017a). However, thevalidation of OMI IWV product has only been found in Wang et al.(2016a) and Vaquero-Martínez et al. (2017b), AIRS IWV products inHagan et al. (2004), Rama Varma Raja et al. (2008), Milstein andBlackwell (2016), SCIAMACHY IWV products in Bovensmann et al.(1999), Noël et al. (2005), Schrijver et al. (2009), du Piesanie et al.(2013), and SEVIRI IWV products in (Hanssen et al., 2001; Schroedter-Homscheidt et al., 2008).

To our knowledge, an intercomparison between seven satellite in-struments against a common reference dataset has not been performedbefore. Therefore, the main goal of this article is to analyze the dif-ferences and similarities in the performance of different satellite IWVproducts in order to improve the understanding of the quality of sa-tellite IWV observations.

2. Instruments and data

2.1. Satellite instruments and their IWV products

Some of the main characteristics of the satellite instruments aresummarized in Table 1. A more detailed description of the satelliteinstruments and their IWV products can be found in the followingsubsections.

2.1.1. GOME-2GOME-2 (Callies et al., 2000) is an improved version of the GOME

instrument, a medium-resolution UV-VIS-NIR spectrometer. The pri-mary product of the GOME-2 satellite is the total atmospheric contentof ozone and the vertical ozone profile. Additionally, it also providesinformation about other trace gases in the atmosphere, such as the totalcolumn amount of water vapor, sulphur dioxide, total and troposphericnitrogen dioxide, tropospheric ozone and bromine oxide. Currently,there are two operational GOME-2 sensors on-board the MetOp-A andMetOp-B satellites. The default scan widths are 960 km and 1920 km,enabling the combined GOME-2 sensors to cover Earth's surface in adaily basis with a ground pixel of 40 km × 40 km (EUMETSAT, 2011).

The IWV data used in this work, obtained from GOME-2 MetOp-A,were derived from the GOME Data Processor (GDP, version 4.6) gen-erated by the German Aerospace Center, Remote Sensing TechnologyInstitute (DLR-IMF) in the framework of the EUMETSAT satelliteApplication Facility on Atmospheric Chemistry Monitoring (O3 M SAF)(Grossi et al., 2015). The period of study extends from 2007 to 2012.

Table 1Summary with main characteristics of the instruments used.

Satellite Algorithm Pixel size λ range Period Passing freq. Cloud filter? Cloud info?

OMI SAO OMH2O v. 1.0 Level 2 13 km× 24 km 430–480 nm Once a day 2007–2009 Yes Not availableSEVIRI SPhR-PGE13 v2.0 3 km× 3 km Around 6.7μm 2008–2012 15–30 min No NoSCIAMACHY AMC-DOAS 30 km × 60 km Around 700 nm 2007–(April)2012 Around once every

6 daysIndirectly No

GOME-2 GDP v. 4.6 80 km× 40 km 614−684 nm 2007–2012 Twice every three days Yes YesMODIS 5 km× 5 km NIR(nighttime) IR

(daytime)2007–2012 1–2 per day Yes Yes

AIRS AIRS/Aqua L2 St. Phys. Ret. (AIRS-only)

13.5 km IR 2007–2012 1–2 times a day Yes Yes

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

730

58 Apendice A. Artıculos

The retrieval method implemented in GDP is based on DifferentialOptical Absorption Spectrography (DOAS). This algorithm, described indetail in Wagner et al. (2006, 2003), consists of three steps:

1. DOAS fitting: water vapor, O2 and O4 absorptions are taken intoaccount. H2O cross section is based on line-by-line computationsusing HITRAN H2O line parameter for a fixed temperature andpressure. The broadband filtering is improved by including threetypes of vegetation spectra, as well as a correction for the ring effect(see Wagner et al., 2009).

2. Non-linearity absorption correction: GOME-2 cannot spectrally re-solve the water vapor (and oxygen) absorption bands, the watervapor slant column density is not linear with IWV, and a correctionmust be applied. The correction factors are obtained by means of themathematical convolution of H2O spectrum with the instrument slitfunction. Such effect is more important for large H2O SCDs.

3. Vertical column density calculation: The corrected SCD must beconverted to vertical column densities (VCDs) to make them geo-metry-independent. This is achieved by dividing SCD by a con-venient air mass factor (AMF), which is derived from oxygen ab-sorption. AMF is obtained dividing O2 SCD by the O2 VCD for astandard atmosphere. AMF for water vapor and oxygen is assumedto be similar, which can cause some systematic errors. O2 AMF isexpected to be larger than water vapor's, since O2 scale height islarger than H2O scale height. In order to correct this, a look-up tablewith correction factors is applied, which depends on SZA, line ofsight angle, relative azimuth and surface albedo. The correctionfactors are calculated through radiative transfer calculations.

The fitting algorithm uses the wavelength region between 614 and683 nm, where the spectral resolution is about 0.54 nm. The main ad-vantages of IWV products from GOME-2 are their independence ofexternal calibration sources and their accuracy both over land and overocean, and the lack of assumptions on atmospheric pressure, tempera-ture, radiative transfer, or other a-priori information.

2.1.2. MODIS-Terra and MODIS-AquaMODIS is on-board Terra and Aqua satellite platforms (King et al.,

1992). Terra's orbit around the Earth is scheduled to overpass theequator from north to south in the morning, while Aqua passes fromsouth to north over the equator in the afternoon. They cover the wholeplanet in 1–2 days. Its swath width is 2330 km.

MODIS has 36 spectral bands, some of which (890–920 nm,931–941 nm and 915–965 nm) are related to atmospheric watervapor. These bands have a spatial resolution of 1 km, but Level 2moisture profiles are binned using 5×5 pixels. Thus, the resolution ofthe IWV product is 5 km × 5 km. The water vapor product is generatedfor both daytime (using NIR bands) and night (using IR bands).

For daytime, NIR bands (channels 2, 5, 17, 18, 19) are used (solarradiation reflected by Earth + atmosphere). The NIR algorithm uses 2-channel and 3-channel rationing techniques. Look-up tables are gen-erated with values of these ratios, calculated from radiative transferprograms. The total amount of water vapor can be transformed intoIWV by taking into account the solar and observational geometries. Ifclouds are present, other channels in the range of 0.8−2.5μm regionare used, since they contain information on absorptions due to watervapor above and within clouds. The algorithm is thoroughly explainedin Gao and Kaufman (1992), Gao and Li (2008).

For nighttime, IR bands are used (radiation emitted by Earth +atmosphere). The algorithm employs a statistical retrieval with an op-tion for a subsequent nonlinear physical retrieval (Seemann et al.,2003). The algorithm calculates MODIS infrared band radiances from adataset of radiosonde observations, in order to associate computed ra-diances with atmospheric profiles. The MODIS atmospheric water-vapor product is then estimated from the total column water vapor,integrating MODIS infrared retrievals of atmospheric moisture profiles

in clear-sky scenes.The data are included in the water vapor product (MOD05_L2 and

MYD05_L2) collection 6. It is, however, obtained from the MODISAtmospheric Profile (MOD07 and MYD07) Collection 6 product, simplyadded to product MOD05 for convenience.

2.1.3. OMIOMI (Levelt et al., 2006) was developed by the Netherland's Agency

for Aerospace Programs (NIVR) and the Finnish Meteorological In-stitute (FMI) to the EOS Aura mission. It is on-board NASA's EarthObserving System (EOS) Aura satellite platform. Aura has a Sun-syn-chronous polar orbit, which allows OMI to sample the whole planetdaily at 1330 local time (LT). The nominal OMI pixel size is 13 km×24 km at nadir.

The OMI IWV data used in this study are the first version of theSmithsonian Astrophysical Observatory (SAO) OMH2O level 2 retrievalwhich uses the algorithm presented in Wang et al. (2014). The algo-rithm uses a window of 430 nm − 480 nm, and it follows three steps:(1) direct fitting of Slant Column Density (SCD), using a semi-empiricalmodel that considers several gases (water vapor, ozone, nitrogen di-oxide, liquid water, and more), as well as some effects (the ring effect,wavelength shift, and more); (2) SCD conversion to Vertical ColumnDensity (VCD) using the Air Mass factor (AMF), which is calculatedusing radiative transfer calculations in look-up tables at 442 nm, and(3) conversion of VCD to IWV by a conversion of units.

Following the guidelines from Wang et al. (2014), some restrictionshave been applied to the OMH2O product to assure its quality. Cloudfraction has to be below 0.1, cloud top pressure over 500 HPa, AMFgreater than 0.75, retrieval root mean square (RMS) value for the fittingSlant Column Density lower than 0.005, maindataqualityflag flag equalto 0. Pixels affected by the row anomaly (see Wang et al., 2014) havebeen rejected as well.

2.1.4. SEVIRIMeteosat are a series of geostationary satellites operated by

EUMETSAT. Meteosat Satellites are equipped with SEVIRI, whichcounts with 7 IR bands in the range 6.2−13.4μm. The retrieval algo-rithm uses the bands WV6.2, WV7.3, IR10.8, IR12.0 and IR13.4, wherethe first two are bands of strong absorption by water vapor. The re-trieval process deals with obtaining the profile of temperature andhumidity from infrared brightness temperature observations, using aninversion technique, i.e. trying to find an atmospheric profile thatwould reproduce the observations. The solution to this problem is notgenerally unique, so a background profile is used as a constraint. Thisbackground profile is obtained from a short range forecast model, and itis slowly varied until its radiative properties fit the observations. Thealgorithm of retrieval is detailed in AEMET and NWC SAF (2013).

One of the limitations of this algorithm is that its products are onlyavailable under clear conditions. In some cases, such as cirrus clouds orin the edge of clouds, NWCSAF/MSG Cloud Mask module might notdetect clouds and the algorithm would try to estimate IWV over thosepixels. However, the retrieval in those cases usually fails or needs a highnumber of iterations, which is detected by a quality flag. Moreover,mountain regions can exhibit large errors if there are differences be-tween NWP topography, and the same can happen with temperatureover very hot or cold pixels, where NWP first guess and the actual skintemperature can be quite different. Additionally, the effect of emissivitytemporal variation is not handled, and fixed values from IREMISmonthly datasets have been used.

As Meteosat is geostationary, data are available with very hightemporal resolution. The product temporal resolution is 30 min. Onlythe temporally closer datum to every GPS datum was selected. Itsspatial resolution is 3 km ×3 km. SEVIRI IWV resolution is around0.58 mm.

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

731

A.2. Artıculo 2 59

2.1.5. SCIAMACHYSCIAMACHY (Bovensmann et al., 1999) is an instrument on-board

the Envisat satellite. It was operational from March 2002 to April 2012.Thus, our study period in this work for SCIAMACHY is from 2007 untilApril 2012. Envisat orbited the Earth in a sun-synchronous orbit, over-passing the equator at 10.00 h LT every day. It sampled the wholeplanet in 6 days in nadir mode. SCIAMACHY's ground pixel size is ty-pically 60 km×30 km.

The retrieval algorithm for SCIAMACHY data is based on the AirMass Corrected Differential Absorption Spectroscopy (AMC-DOAS)method (Noël et al., 2004). This method allows to obtain the IWV frommeasurements in the spectral region around 700 nm. The use of visiblelight makes the method only applicable to daytime and (almost) cloud-free scenes. One of the main advantages of AMC-DOAS is that it pro-vides a completely independent data set, since the IWV products do notdepend on external information.

AMC-DOAS algorithm is based on a modification of DOAS approach.In this modification, the saturation effects from highly structured dif-ferential spectral features that are not resolved by the instruments areaccounted for. Moreover, O2 absorption features are fitted in combi-nation with H2O to derive a correction for the Air Mass Factor (AMF).This correction tries to compensate the lack of information on back-ground and topographic characteristics, and represents how similar theatmospheric conditions and the conditions in the model calculationsare. For example, if the correction were 1 it would indicate a perfectmatch (the correction ranges from 0 to 1). Therefore, the correctionfactor also contains information about the quality of the retrieved IWV.

In order to assess the quality of data, SCIAMACHY data are filteredusing the following criteria: local SZA below 88° and AMF correctiongreater than 0.8. There is no specific cloud filter applied, but the AMFcorrection criterion takes out most of the cloudy scenes.

2.1.6. AIRSAIRS (Aumann et al., 2003) is a high-spectral resolution infrared

sounder aboard NASA's Aqua satellite platform. It surpasses the IberianPeninsula 1–2 times a day. The IR bands used in the retrieval processhave a spatial resolution of 13 km.

The AIRS products used for this work were AIRS/Aqua L2 StandardPhysical Retrieval (AIRS-only) V6. This product has a quality flag forIWV data. The algorithm used in the retrieval (Barnet and Nedis, 2007)has been designed so that all data products simultaneously satisfy themeasurements in a least-squares sense. The Standard Product includesmeasurements of cloud and surface properties, profiles of retrievedtemperature, water vapor, ozone, and a flag for cloud ice or water, aswell as the errors associated with these quantities.

Observed radiances are passed through a neural network to obtainthe atmospheric state, from which cloud parameters are retrieved andthen a cloud clearing is performed to obtain cloud-cleared radiances.This process is done iteratively twice and then a first physical retrievalalgorithm is applied, with the cloud-cleared radiances and the atmo-spheric states as inputs. Then, a new cloud parameter retrieval processis performed and another cloud clearing as well, with new cloud-cleared radiances as output. Then, the type of surface is chosen by thealgorithm, obtaining the final state of the whole set of atmosphericvariables. For more details, see Olsen et al. (2013a,b).

Data with quality flag 2 are rejected in this work, while data withflag 1 or 0 are accepted. Quality flag 2 data are not recommended foruse, while data with quality flag 1 may be used for statistical climatestudies. Data with quality flag 0, recommended for comparison with insitu measurements, would be more suitable, but the number of data-points was scarce for the purpose of this work. The bands for watervapor retrieval are 938cm−1, 1310−1606cm−1 and2607−2657cm−1, respectively.

Data were downloaded from AIRS Science Team/Joao Texeira(2013), AIRS/Aqua L2 Standard Physical Retrieval (AIRS-only) V006,version 006, Greenbelt, MD, USA, Goddard Earth Sciences Data and

Information Services Center (GES DISC), Accessed September 2016,10.5067/AQUA/AIRS/DATA202.

2.2. GPS IWV data

The method to obtain IWV from GPS measurements is briefly de-scribed in this paper. A more detailed explanation can be found in Beviset al. (1992).

The satellites that form the constellation of GPS communicatethrough L-band microwave radiation with ground-based receivers.Usually, the time spent by the signal in reaching the receiver is used tocalculate the distance between the satellites and the receiver. However,several corrections need to be accounted for. In particular, the tropo-sphere produces a delay in the signal, which is usually called SlantTropospheric Delay (STD). It can be converted to the ZenithTropospheric Delay (ZTD) through the so-called mapping functions. Inthis case, Niell's mapping function (Niell, 2000) was used.

=

m EZTD STD

( ) (1)

Once the ZTD is obtained, it can be separated in two differentcontributions: the Zenith Hydrostatic Delay (ZHD) and Zenith WetDelay (ZWD).

= +ZTD ZHD ZWD (2)

The former is due to the tropospheric gases, while water vapor isresponsible for the latter. The ZHD can be modeled and removed ifsurface temperature and atmospheric pressure at the station are known.The quality of these meteorological data is important to minimize errorsin the final product (Wang et al., 2016c). IWV is obtained from theremaining ZWD. The relation between ZWD and IWV is linear,

=IWV Π·ZWD (3)

The constant Π depends on the water vapor - weighted mean tem-perature (Wang et al., 2016b), which can be derived from surfacetemperature.

The GPS IWV data used in this work have been obtained fromground-based GPS measurements of the zenith total delay (ZTD). Thetropospheric products were provided by the Spanish GeographicInstitute “Instituto Geográfico Nacional”, which is a local analysiscenter of the European Reference Frame (EUREF). The analysis is per-formed using Bernese 5.0 software for GNSS data processing. Two stepsare required: in a first step, the coordinates of the stations are obtainedwith high precision, and in the second step, ZTD is obtained. Themethod is based on the resolution of the equation for double differencesof phase (Leick, 1995; Rohm et al., 2014), which uses a network ofground-based receiver stations and differences of time in reaching thesignal between different stations of the network to calculate the stationspositions and other delays and sources of error.

As is described above, once we get the ZTD, two variables areneeded to model ZHD: temperature and pressure at the location of theGPS stations. This information was provided by the SpanishMeteorological State Agency (AEMet). AEMet stations are not ne-cessarily in the same exact location where the GPS receiver is located.However, the stations are as close as possible, usually in the same re-gion. In the case of altitude difference, temperature was corrected byassuming a vertical gradient of temperature of 6.5 K. Data are inter-polated to the time of the GPS measurements. In the case of tempera-ture, data were interpolated linearly. As for pressure, the barometrictide was taken into account to interpolate.

IWV data at the nine GPS stations were available for this work from2007 to 2012. These GPS data, which have a temporal resolution of onehour, have been used to perform other validation exercises on satelliteIWV data (Román et al., 2015; Bennouna et al., 2013; Vaquero-Martínez et al., 2017a; Vaquero-Martínez et al., 2017b).

The stations selected for this research were located at the interior of

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

732

60 Apendice A. Artıculos

the Iberian Peninsula. Coastal stations were rejected in order to avoidpossible influences from error caused by sea or mixed sea-land pixels insatellite observations. Table 2 lists information for the nine stationsselected and the map in Fig. 1 shows their locations in the Iberian Pe-ninsula.

3. Methodology

3.1. Collocation and comparison criteria

Two different criteria were followed for spatial collocation. The firstcriterion was to take the pixel whose center was the closest to the

ground-based GPS station. The second criterion was to average theclosest pixels (those within 0.25°×0.25° distance to the ground-basedGPS station). The first criterion was used for the collocation betweenGOME-2 and GPS, between MODIS-Terra and GPS, and SEVIRI andGPS.

The temporal criterion followed was to match GPS and satellite IWVvalues whose temporal difference was the closest. In all cases suchdifference had to be below 30 min.

Satellite data under cloudy-sky conditions (cloud fraction given byeach satellite algorithm larger than zero) have been rejected for allanalyses, except for the study of cloud dependence (see Section 4.5),where all sky conditions were considered for those satellite datasetsthat provide information on cloudiness (i.e. GOME-2, MODIS-Terra andMODIS-Aqua, and AIRS).

3.2. Statistical analysis

Once the temporal and spatial match between the satellite and theGPS data is achieved, there is a dataset for each satellite, where everyrow has a satellite IWV value, a GPS IWV value, the location (station),and other columns with additional information, such as the date andtime, SZA or cloud fraction (CF). The relative differences (Eq. (4))studied in this work are calculated as:

= ⋅

−

δw w

w100i s

i s i s

i s,

,sat

,GPS

,GPS (4)

where the index s denotes a satellite, the index i represents a fixed lo-cation and time and w is the IWV measurement by the satellite (sat) orGPS.

The distribution of the satellite-GPS differences is analyzed for eachground-based station using several variables. First, two indices arecalculated, the pseudomedian and the interquartile range (IQR). Thepseudomedian is obtained using the Wilcoxon signed rank test withcontinuity correction (Wilcoxon, 1946). The pseudomedian is definedas the median of all the midpoints of pairs of observations, which agreeswith the median if the dataset is symmetric. The pseudomedian of therelative differences provides information about the accuracy of thesatellite data, while IQR reports about their precision. Pseudomedianhas been chosen over median as index because it is a better estimatorwhen the distribution is asymmetric, which is typically the case for δdistribution when applied to binned data.

Furthermore, a linear regression analysis between the GPS and thesatellite data was performed in order to analyze their proportionalityand similarity. Then, in order to study the dependence with certainvariables, the two indices are applied to bins of data. The bin widths are5° for SZA, 5 mm for IWV and 0.10 for CF. Moreover, the seasonaldependence of relative differences was also analyzed in detail. Binswith less than 50 data points have been rejected. The dependence ofdistance satellite pixel - GPS ground-based station was not considered inthis work, since Román et al. (2015) did not show an important impactin the satellite IWV data.

Table 2Characteristics of the GPS stations.

Station Acronym Latitude Longitude

(ºN) (ºE)Córdoba coba 37.92 −4.72León leon 42.59 −5.65Logroño rioj 42.46 −2.5Salamanca sala 40.95 −5.5Sonseca sons 39.68 −3.96Teruel teru 40.35 −1.12Valladolid vala 41.70 −4.71Villafranca vill 40.44 −3.95Cáceres cace 39.48 −6.34

coba

leon rioj

sala

sons

teru

vala

vill

cace

35.0

37.5

40.0

42.5

45.0

−12 −8 −4 0

lon

lat

Fig. 1. Location of the nine stations selected.

Table 3Statistical analysis of sat-GPS relative differences. The pseudomedian (pMedian) and IQR of the δ distribution, the number of data (N) and the coefficients of the regression analysis areshown. y0 column shows the intercept, b stands for the slope and R2 is Pearson's coefficient of determination. The numbers after± are the 95% confidence interval.

Satellite pMedian IQR N y0 b R2

(%) (%) (mm)OMI −0.7± 1.1 40.80 3895 2.65± 0.28 0.78± 0.02 0.63SEVIRI −5.2± 0.1 33.31 187375 2.89± 0.03 0.690± 0.002 0.67SCIAMACHY 6.6± 1.2 45.72 2629 0.92± 0.36 0.96± 0.02 0.70GOME-2 16.7±0.8 32.58 4317 3.40± 0.18 0.88± 0.01 0.83MODIS-Terra −0.9± 0.5 34.58 13651 1.01± 0.14 0.915± 0.009 0.74MODIS-Aqua −3.4± 0.4 33.24 13581 0.99± 0.14 0.89± 0.01 0.71AIRS 2.0± 1.8 47.84 1832 3.05± 0.41 0.73± 0.03 0.56

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

733

A.2. Artıculo 2 61

4. Results and discussion

4.1. Statistical analysis

Table 3 shows the pseudomedian and IQR of the satellite-GPS dif-ferences (Eq. (1)) for the seven satellite instruments. The results in-dicate that GOME-2, SCIAMACHY and AIRS highly overestimate, onaverage, the reference GPS data (positive pseudomedian values), whileMODIS-Aqua, MODIS-Terra and SEVIRI have a small tendency to un-derestimate IWV (negative pseudomedian values). OMI pseudomedian,however, shows that there is no significant bias in OMI IWV with re-spect to reference GPS IWV. IQR is between 30% and 35% for GOME-2,both MODIS, and SEVIRI, while it is higher than 40% for OMI,SCIAMACHY and AIRS. The regression analyses performed for eachsatellite instrument show that the intercept y0 is always positive and theslope b is always lower than 1. This suggests that satellite instrumentstend to overestimate low IWV data, and underestimate high values. Thisresult is in agreement with other studies (Rama Varma Raja et al., 2008;Bennouna et al., 2013; Antón et al., 2015; Román et al., 2015;Scheepmaker et al., 2015; Vaquero-Martínez et al., 2017b; Vaquero-Martínez et al., 2017a). Correlation coefficient R2 shows a fair agree-ment. The agreement is better for GOME-2 and both MODIS instru-ments, and worse for AIRS. The validation of GOME-2 in Antón et al.(2015) against radiosonde showed a slightly better agreement(R2=0.95).

Fig. 2 presents a time series of each instrument (columns) and eachstation (rows). It can be observed that all satellites represent the sea-sonal variation of water vapor correctly. The lack of available data insome periods at some stations can be identified. For instance, teru sta-tion time series starts in 2009, because the GPS receiver in that stationwas not operative until 2009. Moreover, it can be observed that OMIdata are only available in the period 2007–2009, as mentioned inSection 2.1. The different density of data-points is related to the sa-tellite's passing frequency and the quality filters mentioned inSection 2.1. The differences between satellite and GPS IWV are alsorepresented, showing that in all satellites these are approximatelycentered around 0 mm.

4.2. IWV dependence

Fig. 3 shows the pseudomedian of the sat-GPS differences againstreference (GPS) IWV data in bins of 5 mm. The error bars are the 95%confidence interval in the Wilcoxon signed rank test with continuitycorrection. It can be seen that the behavior is similar in all satelliteinstruments: the pseudomedian is positive for the lowest IWV values inall of them, while satellite data tend to underestimate large IWV values.This is in agreement with the behavior observed in other studies (Antónet al., 2015; Vaquero-Martínez et al., 2017a,b). AIRS, GOME-2 andSEVIRI have the largest range of variation. Their pesudomedians reachalmost +40% (AIRS and SEVIRI) and +60% (GOME-2) for low IWVvalues, while they decrease to −30% (AIRS), −25% (SEVIRI) and−10% (GOME-2) in large IWV cases. Both MODIS instruments performsimilarly, with Terra being slightly higher than Aqua. It can be noticedthat SCIAMACHY and GOME-2 (whose retrieval algorithms use DOAStechniques) tend to slightly overestimate IWV for intermediate values(∼ 10–25 mm), while the rest of satellites tend to underestimate IWV inthis range of IWV values. The behavior of GOME-2 was also reported inAntón et al. (2015). In that work, GOME-2 showed discrepancies withreference radiosonde IWV data under 20% when data are grouped bysimilar SZA values. The strongest differences between Antón et al.(2015) and the present work are at low IWV values, suggesting that SZAmight play an important role.

Regarding the precision statistical, IQR, Fig. 4 shows similar valuesfor all satellite instruments except for OMI, which has much higher IQRfor low IWV values (over 100%, the rest being around 50%). IQR de-creases with increasing IWV in all cases, reaching values under 25% forhigh IWV. The satellite instrument with the lowest IQR in the wholerange of IWV is GOME-2. The behavior of SCIAMACHY water vaporproduct is different. It keeps a high IQR for low and medium IWV (up to25 mm approximately), only becoming lower than 20% at high IWV(>30 mm). A similar pattern was reported in Noël et al. (2004) whenECMWF IWV data were used as reference.

4.3. SZA dependence

The influence of SZA on the pseudomedian is different for eachsatellite instrument, as seen in Fig. 5. OMI and GOME-2, which usevisible radiation for IWV retrieval, show an increase of the

Fig. 2. Time series of every collocated dataset of every satellite instrument in every station. Blue line is the satellite IWV and red line is the difference between satellite measurements andGPS data.

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

734

62 Apendice A. Artıculos

pseudomedian with SZA. As SZA increases, the amount of IWV that thesunlight encounters increases. This could affect the correction factorused to calculate the air mass factor (AMF). In the case of OMI thischange is very smooth, and could be explained by the correlation of SZAand IWV values (high IWV values occur when temperature is higher,which happens when SZA is low, and vice versa), as reported inVaquero-Martínez et al. (2017b). The increase of the pseudomedianwith SZA is specially strong in GOME-2, from very small values (under5%) for low SZA to very high values (around 80%) for high SZA, as ithas already been reported in the literature (Kalakoski et al., 2011;Antón et al., 2015; Román et al., 2015). By contrast, SCIAMACHY,which also uses visible radiation, shows the opposite behavior: a de-crease of relative difference with increasing SZA. This can also be

related to the quality of AMF correction being influenced by SZA in theretrieval algorithm used for this satellite instrument.

In the case of satellites that use IR radiation for IWV retrieval, i.e.the MODIS instruments (Terra and Aqua) and SEVIRI, the influence ofSZA at daytime is similar to OMI. This fact suggests that the SZA de-pendence may be related to other variables that change with SZA (i.e.the amount of water vapor). In the case of AIRS, the pseudomedianseems to slightly decrease with SZA. Furthermore, when using IR ra-diation it is possible to make measurements in the nighttime. AIRS has anotably good performance at nighttime, with pseudomedian close to 0for the whole nighttime range. The rest of the instruments have nega-tive pseudomedian of the error at nighttime, above −20%. A strongdiscontinuity is observed between daytime and nighttime

Fig. 3. Pseudomedian of sat-GPS relative differences against reference IWV (GPS). Error bars are the 95% confidence interval in the Wilcoxon signed rank test.

Fig. 4. IQR of sat-GPS relative differences against reference IWV (GPS).

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

735

A.2. Artıculo 2 63

measurements of MODIS. This is probably related to the fact that theIWV retrieval is different for daytime and nighttime. SEVIRI and AIRS,which use the same retrieval algorithm for both day and night, have aquite similar response in the whole SZA range.

The variation of IQR with SZA is plotted in Fig. 6. Again, OMI andGOME-2 behave similarly, but in this case GOME-2 performs better: itsIQR ranges from under 20% for low SZA, to 50% for high SZA. Bycontrast, OMI IQR changes from 30% to more than 70%, increasingwith SZA. SCIAMACHY has a similar behavior as well, with highervalues of IQR than OMI up to SZA=50°, and between OMI and GOME-2 from that SZA on. Both MODIS instruments have similar IQR com-pared to GOME-2. SEVIRI has a more stable IQR with SZA, alwaysbetween 15% and 40%. For nighttime, SEVIRI, MODIS-Aqua and

MODIS-Terra have similar IQR, slightly increasing with SZA. AIRS IQRat nighttime is clearly higher than the rest, while at daytime it is above50%. The increase of IQR with daytime SZA can be explained if we takeinto consideration the increasing corrections to obtain the proper AMFof water vapor. These corrections introduce noise in the measurements,which are stronger as the corrections are larger. Moreover, at high SZAIWV is usually lower, so the relative difference is higher for the sameabsolute difference.

4.4. Seasonal dependence

Satellite performance displays a dependence on the season of theyear, related to the annual cycle of water vapor and SZA values. In

Fig. 5. Pseudomedian of sat-GPS relative differences against SZA.

Fig. 6. IQR of sat-GPS relative differences against SZA.

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

736

64 Apendice A. Artıculos

Fig. 7, the pseudomedian of the relative differences is shown in bins ofone month. GOME-2 shows the strongest seasonal dependence, withpseudomedian values ranging from +5% in summer to over +50% inwinter, which is probably connected to the strong dependence on SZAshown above. This is in agreement with Román et al. (2015), and showshigher pseudomedians than in Antón et al. (2015), where the referenceinstrument was radiosondes. The rest of the satellites have mediansbetween −25 % and +25%. OMI has a similar behavior to GOME-2,with an overestimation (positive pseudomedian) in winter and a slightunderestimation (negative pseudomedian) in summer, in agreementwith Vaquero-Martínez et al. (2017b). However, both MODIS satelliteinstruments show overestimation in summer and underestimation al-most the rest of the year (except for a slight overestimation in De-cember). MODIS-Terra has slightly higher pseudomedian values insummer than MODIS-Aqua. Bennouna et al. (2013) showed that MODISalgorithm performed worse in winter. The reason for the discrepancycould be related to differences in datasets, such as the years used andthe stations selected. If atmospheric conditions change, IWV willchange too, and thus performance of the algorithm can be different.Moreover, SEVIRI underestimates from April to November and over-estimates from December to March. AIRS is the closest to the zero linethroughout the year. SCIAMACHY, however, has a special behavior:summer and autumn months are overestimated (up to 25%), whilewinter and spring are slightly underestimated.

The seasonal dependence on the precision index can be seen inFig. 8. All satellite instruments have a similar behavior: IQR is higher inwinter than in summer. OMI has the higher IQR in winter and autumn,reaching more than 70% in December, while AIRS has IQR over 40%throughout the year, for almost all months. However, the rest of thesatellite instruments have IQR from 20% to 55%. GOME-2 data havethe best performance except in winter, where all satellite instrumentsexcept OMI (higher IQR) perform similarly. This behavior can be re-lated to the fact that in winter, IWV is smaller and thus the relativedifference tends to be higher, as commented in Section 4.2. OMI be-havior is in agreement with Vaquero-Martínez et al. (2017b).

4.5. Cloudiness dependence

The influence of cloudiness on the pseudomedian is represented in

Fig. 9 for those satellite instruments that provide information aboutcloudiness and were not filtered (AIRS, GOME-2, MODIS-Aqua andMODIS-Terra). In general, as CF increases the pseudomedian is furtherfrom the zero line: it can be below 0, underestimating the IWV (AIRS,GOME-2 and MODIS-Terra) or over 0, overestimating (MODIS-Aqua).The underestimation can be due to the so called shielding effect (Románet al., 2015; Kokhanovsky and Rozanov, 2008): clouds can “hide” thewater vapor under them. The differences between MODIS-Aqua andMODIS-Terra could be related to their different passing times and theuse of NIR radiation in daylight and IR during nighttime. At nighttime,the algorithm could confuse the presence of clouds with water vapor,causing the overestimation.

IQR, the precision index, is shown in Fig. 10. IQR computed for bothMODIS data products increases as cloudiness increases, AIRS seems tohave a stable value of IQR and GOME-2 shows a certain decrease of IQRas CF is higher. The reason for this could be that clouds introduce noisein the measurements, but if there are too many clouds, the shieldingeffect reduces the sensitivity to water vapor, decreasing the variability(IQR).

5. Conclusions

The analysis of the relative differences between satellite and GPSmeasurements has found some similarities and differences among thesatellite measurements. In general, AIRS and OMI measurements areaccurate (pesudomedian of the differences close to zero), but they areless precise than the rest of the satellites. Regarding precision the rest ofthe satellites perform similarly, but GOME-2 overestimates IWV whileSEVIRI and both MODIS underestimate the measurements. Regressionanalysis showed that all satellites tend to homogenize water vapor: lowIWV tends to be overestimated, while high IWV tends to be under-estimated. This result was confirmed when studying the dependence ofthe relative differences on IWV data. The reason for this could be thatspatial resolution of satellites is much lower than GPS ground-basedstations, and thus IWV measurement is somehow averaged over thewhole pixel. The precision index (IQR) showed that measurements aremore precise as IWV increases. OMI precision is especially low (highIQR) at low IWV. IQR computed for SCIAMACHY data seems to be highup to 20 mm, when IQR starts to decrease as IWV increases.

Fig. 7. Seasonal evolution of the pseudomedian of sat-GPS relative differences. December has been rearranged as the first month in order to make easier to identify the different seasons.

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

737

A.2. Artıculo 2 65

The study on the influence of SZA on the relative differences showedthat GOME-2 highly overestimates IWV at high SZA. There is a generaltendency to overestimate for SZA between 60° and 80°. OMI performsreasonably well although its precision quickly becomes lower (higherIQR) as SZA increases. SEVIRI has a quite stable IQR over the wholeSZA range. Nighttime measurements are underestimated for all IR sa-tellites (SEVIRI and MODIS-Terra and Aqua) except AIRS, which pre-sents a good accuracy in nighttime.

The annual variations of the two indices are studied as well. Theperformance of all satellites is similar, with the following exceptions.GOME-2 shows a high overestimation during winter and autumn,probably the cause of its high overestimation in the general analysis.SCIAMACHY shows a high pseudomedian in summer and autumn, and

lower in winter and spring. OMI shows very high IQR (low precision) inwinter.

The influence of clouds is studied for those satellites that provideinformation about cloudiness. The presence of clouds increases thedeviation of satellite IWV data with respect to the reference GPS mea-surements, whether overestimating (MODIS-Aqua) or underestimating(MODIS-Terra, GOME-2, AIRS). IQR generally increases or remainsstable, except for GOME-2, which shows a slight decrease of IQR withCF.

Although satellite retrievals can provide good spatial coverage ofIWV values, they still need improvements in order to reduce the notabledifferences and dependences observed when the satellite IWV productsare compared against reference GPS data. This study indicates that

Fig. 8. Seasonal evolution of the IQR of sat-GPS relative differences. December has been rearranged as the first month in order to make easier to identify the different seasons.

Fig. 9. Pseudomedian of sat-GPS relative differences against CF.

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

738

66 Apendice A. Artıculos

more work is needed to increase OMI precision and GOME-2 accuracyfor low IWV, and to improve AIRS precision under all conditions.

Acknowledgments

This work was supported by the Spanish Ministry of Economy andCompetitiveness through project CGL2014-56255-C2. Manuel Antónthanks Ministerio de Ciencia e Innovación and Fondo Social Europeo(RYC-2011-08345) for the award of a postdoctoral grant (Ramón yCajal). Support from the Junta de Extremadura (Research Group GrantsGR15137) is gratefully acknowledged. The GOME-2/MetOp-A productswere generated at DLR under the auspices of the O3MSAF projectfunded by EUMETSAT and national contributions. The generation ofSCIAMACHY data was supported by ESA, DLR Bonn and by theUniversity of Bremen, Germany. Work at Universidad de Valladolid issupported by projects CMT2015-66742-R and MINECO VA100U14.Work at Universidad de Granada was supported by the AndalusiaRegional Government (project P12-RNM-2409) and the SpanishMinistry of Economy and Competitiveness and FEDER funds under theprojects CGL2013-45410-R, CGL2016-81092-R and “Juan de la Cierva-Formación” program. Work at SAO is supported by NASA’sAtmospheric Composition: Aura Science Team program (sponsor con-tract number NNX14AF56G). Work at Universidade de Évora is co-funded by the European Union through the European RegionalDevelopment Fund, included in the COMPETE 2020 (OperationalProgram Competitiveness and Internationalization) through the ICTproject (UID/GEO/04683/2013) with the reference POCI-01-0145-FEDER-007690.

References

AEMET,NWC, S.A.F., 2013. Algorithm Theoretical Basis Document for “SEVIRI PhysicalRetrieval Product” (SPhR-PGE13 v2.0). Technical Report.

Antón, M., Loyola, D., Román, R., Vömel, H., 2015. Validation of GOME-2/MetOp-A totalwater vapour column using reference radiosonde data from the GRUAN network.Atmos. Meas. Tech. 8, 1135–1145. http://dx.doi.org/10.5194/amt-8-1135-2015.

Aumann, H.H., Chahine, M.T., Gautier, C., Goldberg, M.D., Kalnay, E., Mcmillin, L.M.,Revercomb, H., Rosenkranz, P.W., Smith, W.L., Staelin, D.H., Strow, L.L., Susskind,J., 2003. AIRS / AMSU / HSB on the Aqua mission : design, science objectives, dataproducts, and processing systems. Processing 41, 253–264.

Barnet, C., Nedis, N., 2007. Airs-Team Retrieval for Core Products and Geophysical.Technical Report March.

Barreto, A., Cuevas, E., Damiri, B., Romero, P.M., Almansa, F., 2013. Column water vapordetermination in night period with a lunar photometer prototype. Atmos. Meas. Tech.6, 2159–2167. http://dx.doi.org/10.5194/amt-6-2159-2013.

Bennouna, Y.S., Torres, B., Cachorro, V.E., Ortiz de Galisteo, J.P., Toledano, C., 2013. Theevaluation of the integrated water vapour annual cycle over the Iberian Peninsulafrom EOS-MODIS against different ground-based techniques. Q. J. R. Meteorol. Soc.139, 1935–1956. http://dx.doi.org/10.1002/qj.2080.

Bevis, M., Businger, S., Herring, T. a., Rocken, C., Anthes, R. a., Ware, R.H., 1992. GPSmeteorology: remote sensing of atmospheric water vapor using the Global PositioningSystem. J. Geophys. Res. 97, 15787–15801. http://dx.doi.org/10.1029/92JD01517.

Bovensmann, H., Burrows, J.P., Buchwitz, M., Frerick, J., Noël, S., Rozanov, V.V., Chance,K.V., Goede, a.P.H., 1999. SCIAMACHY: mission objectives and measurement modes.J. Atmos. Sci. 56, 127–150. http://dx.doi.org/10.1175/1520-0469(1999)056<0127:SMOAMM>2.0.CO;2.

Callies, J., Corpaccioli, E., Eisinger, M., Hahne, a., Lefebvre, a., 2000. GOME-2 - Metop'ssecond-generation sensor for operational ozone monitoring. ESA Bull. Eur. SpaceAgency 102, 28–36.

Chang, L., Gao, G., Jin, S., He, X., Xiao, R., Guo, L., 2015. Calibration and evaluation ofprecipitable water vapor from Modis infrared observations at night. IEEE Trans.Geosci. Remote Sens. 53, 2612–2620. http://dx.doi.org/10.1109/TGRS.2014.2363089.

Colman, R., 2003. A comparison of climate feedbacks in general circulation models. Clim.Dyn. 20, 865–873. http://dx.doi.org/10.1007/s00382-003-0310-z.

De Haan, S., Van Der Marel, H., Barlag, S., 2002. Comparison of GPS slant delay mea-surements to a numerical model: case study of a cold front passage. Phys. Chem.Earth 27, 317–322. http://dx.doi.org/10.1016/S1474-7065(02)00006-2.

Diedrich, H., Wittchen, F., Preusker, R., Fischer, J., 2016. Representativeness of totalcolumn water vapour retrievals from instruments on polar orbiting satellites. Atmos.Chem. Phys. Discuss. 16, 8331–8339. http://dx.doi.org/10.5194/acp-2016-99.

EUMETSAT, 2011. European Organisation for the Exploitation of MeteorologicalSatellites: GOME-2 Product Guide.

Gao, B.-C., Kaufman, Y.J., 1992. The MODIS near-IR water vapor algorithm product ID:MOD05 - total precipitable water. Algorithm Tech. Backgr. Doc. 1–25 URL:Biblioteca_Digital_SPRGao1992_ATBD.pdf.

Gao, B.C., Li, R.R., 2008. The time series of Terra and Aqua MODIS near-IR water vaporproducts. Int. Geosci. Remote Sens. Symp. (IGARSS) 3, 186–189. http://dx.doi.org/10.1109/IGARSS.2008.4779314.

Grossi, M., Valks, P., Loyola, D., Aberle, B., Slijkhuis, S., Wagner, T., Beirle, S., Lang, R.,2015. Total column water vapour measurements from GOME-2 MetOp-A and MetOp-B. Atmos. Meas. Tech. 8, 1111–1133. http://dx.doi.org/10.5194/amt-8-1111-2015.

Hagan, D.E., Webster, C.R., Farmer, C.B., May, R.D., Herman, R.L., Weinstock, E.M.,Christensen, L.E., Lait, L.R., Newman, P.A., 2004. Validating AIRS upper atmospherewater vapor retrievals using aircraft and balloon in situ measurements. Geophys. Res.Lett. 31, 10–13. http://dx.doi.org/10.1029/2004GL020302.

Hanssen, R.F., Feijt, A.J., Klees, R., 2001. Comparison of precipitable water vapor ob-servations by spaceborne radar interferometry and meteosat 6.7−μmradiometry. J.Atmos. Oceanic Tech. 18, 756–764. http://dx.doi.org/10.1175/1520-0426(2001)058<0756:COPWVO>2.0.CO;2.

Ichoku, C., Levy, R., Kaufman, Y.J., Remer, L.A., Li, R.R., Martins, V.J., Holben, B.N.,Abuhassan, N., Slutsker, I., Eck, T.F., Pietras, C., 2002. Analysis of the performancecharacteristics of the five-channel Microtops II Sun photometer for measuring aerosoloptical thickness and precipitable water vapor. J. Geophys. Res.-Atmos. 107. http://

Fig. 10. IQR of sat-GPS relative differences against CF.

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

739

A.2. Artıculo 2 67

dx.doi.org/10.1029/2001JD001302.Jones, A., Urban, J., Murtagh, D.P., Eriksson, P., Brohede, S., Haley, C., Degenstein, D.,

Bourassa, A., Von Savigny, C., Sonkaew, T., Rozanov, A., Bovensmann, H., Burrows,J., 2009. Evolution of stratospheric ozone and water vapour time series studied withsatellite measurements. Atmos. Chem. Phys. Discuss 9, 1157–1209. http://dx.doi.org/10.5194/acpd-9-1157-2009.

Kalakoski, N., Kujanpää, J., Sofieva, V., Tamminen, J., Grossi, M., Valks, P., 2016.Validation of GOME-2/Metop total column water vapour with ground-based and insitu measurements. Atmos. Meas. Tech. 9, 1533–1544. http://dx.doi.org/10.5194/amt-9-1533-2016.

Kalakoski, N., Wagner, T., Mies, K., Beirle, S., Slijkhuis, S., Loyola, D., 2011. O3M SafValidation Report. Technical Report 2/2011. . http://o3msaf.fmi.fi/docs/vr/Validation_Report_OTO_H2O_Mar_2011.pdf.

King, M.D., Kaufman, Y.J., Menzel, W.P., Tanré, D., 1992. Remote sensing of cloud,aerosol, and water vapor properties from the Moderate Resolution ImagingSpectrometer (MODIS). IEEE Trans. Geosci. Remote Sens. 30, 2–27. http://dx.doi.org/10.1109/36.124212.

Kokhanovsky, A.A., Rozanov, V.V., 2008. The uncertainties of satellite DOAS total ozoneretrieval for a cloudy sky. Atmos. Res. 87, 27–36. http://dx.doi.org/10.1016/j.atmosres.2007.04.006.

Leick, A., 1995. GPS Satellite Surveying. Wiley.Levelt, P.F., van den Oord, G.H.J., Dobber, M.R., Malkki, A., Visser, H., de Vries, J.,

Stammes, P., Lundell, J.O.V., Saari, H., 2006. The ozone monitoring instrument. IEEETrans. Geosci. Remote Sens. 44, 1093–1101.

Li, Z., Muller, J.P., Cross, P., 2003. Comparison of precipitable water vapor derived fromradiosonde, GPS, and Moderate-Resolution Imaging Spectroradiometer measure-ments. J. Geophys. Res. Atmos. 108. http://dx.doi.org/10.1029/2003JD003372.

Milstein, A.B., Blackwell, W.J., 2016. Neural network temperature and moisture retrievalalgorithm validation for AIRS/AMSU and CrIS/ATMS. J. Geophys. Res. Atmos. 121,1414–1430. http://dx.doi.org/10.1002/2015JD024008.

Myhre, G., Shindell, D., Bréon, F.-M., Collins, W., Fuglestvedt, J., Huang, J., Koch, D.,Lamarque, J.-F., Lee, D., Mendoza, B., Nakajima, T., Robock, A., Stephens, G.,Takemura, T., Zhang, H., 2013. Anthropogenic and Natural Radiative Forcing.Technical Report. http://dx.doi.org/10.1017/CBO9781107415324.018.

Niell, A.E., 2000. Improved atmospheric mapping functions for VLBI and GPS. Earth,Planets and Space 52, 699–702. http://dx.doi.org/10.1186/BF03352267.

Ningombam, S.S., Jade, S., Shrungeshwara, T.S., Song, H.J., 2016. Validation of watervapor retrieval from Moderate Resolution Imaging Spectro-radiometer (MODIS) innear infrared channels using GPS data over IAO-Hanle, in the trans-Himalayan re-gion. J. Atmos. Sol. Terr. Phys. 137, 76–85. http://dx.doi.org/10.1016/j.jastp.2015.11.019.

Noël, S., Buchwitz, M., Bovensmann, H., Burrows, J.P., 2005. Validation of SCIAMACHYAMC-DOAS water vapour columns. Atmos. Chem. Phys. 5, 1835–1841. http://dx.doi.org/10.5194/acp-5-1835-2005.

Noël, S., Buchwitz, M., Burrows, J.P., 2004. First retrieval of global water vapour columnamounts from SCIAMACHY measurements. Atmos. Chem. Phys. 4, 111–125. http://dx.doi.org/10.5194/acp-4-111-2004.

Noël, S., Mieruch, S., Bovensmann, H., Burrows, J.P., 2008. Preliminary results of GOME-2 water vapour retrievals and first applications in polar regions. Atmos. Chem. Phys.8, 1519–1529. http://dx.doi.org/10.5194/acp-8-1519-2008.

Olsen, E.T., Blaisdell, J., Iredell, L., Susskind, J., 2013a. AIRS/AMSU/HSB Version 6Retrieval Flow. Technical Report. Jet Propulsion Laboratory.

Olsen, E.T., Fetzer, E., Hulley, G., Manning, E., Blaisdell, J., Iredell, L., Susskind, J.,Warner, J., Wei, Z., Blackwell, W., Maddy, E., 2013b. AIRS / AMSU / HSB Version 6Level 2 Product User Guide. Technical Report.

Ortiz de Galisteo, J.P., Bennouna, Y., Toledano, C., Cachorro, V., Romero, P., Andrés,M.I., Torres, B., 2014. Analysis of the annual cycle of the precipitable water vapourover Spain from 10-year homogenized series of GPS data. Q. J. R. Meteorol. Soc. 140,397–406. http://dx.doi.org/10.1002/qj.2146.

Ortiz de Galisteo, J.P., Cachorro, V., Toledano, C., Torres, B., Laulainen, N., Bennouna, Y.,de Frutos, A., 2011. Diurnal cycle of precipitable water vapor over Spain. Q. J. R.Meteorol. Soc. 137, 948–958. http://dx.doi.org/10.1002/qj.811.

Ortiz de Galisteo, J.P., Toledano, C., Cachorro, V., Torres, B., 2010. Improvement in PWVestimation from GPS due to the absolute calibration of antenna phase center varia-tions. GPS Solutions 14, 389–395. http://dx.doi.org/10.1007/s10291-010-0163-y.

Pany, T., Pesec, P., Stangl, G., 2001. Atmospheric GPS slant path delays and rays tracingthrough numerical weather models, a comparison. Phys. Chem. Earth Solid EarthGeod. 26, 183–188. http://dx.doi.org/10.1016/S1464-1895(01)00044-8.

Pérez-Ramírez, D., Navas-Guzmán, F., Lyamani, H., Fernández-Gálvez, J., Olmo, F.J.,Alados-Arboledas, L., 2012. Retrievals of precipitable water vapor using star photo-metry: assessment with Raman lidar and link to sun photometry. J. Geophys. Res.Atmos. 117, 1–10. http://dx.doi.org/10.1029/2011JD016450.

Prasad, A.K., Singh, R.P., 2009. Validation of MODIS Terra, AIRS, NCEP/DOE AMIP-IIReanalysis-2, and AERONET Sun photometer derived integrated precipitable watervapor using ground-based GPS receivers over India. J. Geophys. Res. Atmos. 114,1–20. http://dx.doi.org/10.1029/2008JD011230.

Rama Varma Raja, M.K., Gutman, S.I., Yoe, J.G., McMillin, L.M., Zhao, J., 2008. Thevalidation of AIRS retrievals of integrated precipitable water vapor using measure-ments from a network of ground-based GPS receivers over contiguous United States.J. Atmos. Ocean. Technol. 25, 416–428. http://dx.doi.org/10.1175/2007JTECHA889.1.

Rohm, W., Yuan, Y., Biadeglgne, B., Zhang, K., Marshall, J.L., 2014. Ground-based GNSSZTD/IWV estimation system for numerical weather prediction in challenging weatherconditions. Atmos. Res. 138, 414–426. http://linkinghub.elsevier.com/retrieve/pii/S0169809513003499http://dx.doi.org/10.1016/j.atmosres.2013.11.026.

Román, R., Antón, M., Cachorro, V.E., Loyola, D., Ortiz de Galisteo, J.P., de Frutos, A.,Romero-Campos, P.M., 2015. Comparison of total water vapor column from GOME-2on MetOp-A against ground-based GPS measurements at the Iberian Peninsula. Sci.Total Environ. 533, 317–328. http://dx.doi.org/10.1016/j.scitotenv.2015.06.124.

Scheepmaker, R.A., Frankenberg, C., Deutscher, N.M., Schneider, M., Barthlott, S.,Blumenstock, T., Garcia, O.E., Hase, F., Jones, N., Mahieu, E., Notholt, J., Velazco, V.,Landgraf, J., Aben, I., 2015. Validation of sciamachy HDO/H2O measurements usingthe TCCON and NDACC-MUSICA networks. Atmos. Meas. Tech. 8, 1799–1818.http://dx.doi.org/10.5194/amt-8-1799-2015.

Schrijver, H., Gloudemans, A.M.S., Aben, I., 2009. Water vapour total columns fromSCIAMACHY spectra in the 2.36 mu m window. Atmos. Meas. Tech. 2, 561–571.http://dx.doi.org/10.5194/amt-2-561-2009.

Schroedter-Homscheidt, M., Drews, A., Heise, S., 2008. Total water vapor column re-trieval from MSG-SEVIRI split window measurements exploiting the daily cycle ofland surface temperatures. Remote Sens. Environ. 112, 249–258. http://dx.doi.org/10.1016/j.rse.2007.05.006.

Seemann, S.W., Li, J., Menzel, W.P., Gumley, L.E., 2003. Operational retrieval of at-mospheric temperature, moisture, and ozone from MODIS infrared radiances. J. Appl.Meteor. 42, 1072–1091.

Torres, B., Cachorro, V.E., Toledano, C., Ortiz De Galisteo, J.P., Berjón, A., De Frutos,A.M., Bennouna, Y., Laulainen, N., 2010. Precipitable water vapor characterization inthe Gulf of cadiz region (southwestern Spain) based on Sun photometer, GPS, andradiosonde data. J. Geophys. Res. Atmos. 115, 1–11. http://dx.doi.org/10.1029/2009JD012724.

Turner, D.D., Clough, S.A., Liljegren, J.C., Clothiaux, E.E., Cady-Pereira, K.E., Gaustad,K.L., 2007. Retrieving liquid water path and precipitable water vapor from the at-mospheric radiation measurement (ARM) microwave radiometers. In: IEEETransactions on Geoscience and Remote Sensing. vol. 45. pp. 3680–3689. http://dx.doi.org/10.1109/TGRS.2007.903703.

Turner, D.D., Ferrare, R.A., Heilman Brasseur, L.A., Feltz, W.F., Tooman, T.P., 2002.Automated retrievals of water vapor and aerosol profiles from an operational Ramanlidar. J. Atmos. Ocean. Technol. 19, 37–50. http://dx.doi.org/10.1175/1520-0426(2002)019<0037:AROWVA>2.0.CO;2.

Vaquero-Martínez, J., Antón, M., Ortiz de Galisteo, J.P., Cachorro, V.E., Costa, M.J.,Román, R., Bennouna, Y.S., 2017a. Validation of MODIS integrated water vaporproduct against reference GPS data at the Iberian Peninsula. Int. J. Appl. Earth Obs.Geoinf. 63, 214–221. http://linkinghub.elsevier.com/retrieve/pii/S0048969716327176http://dx.doi.org/10.1016/j.jag.2017.07.008.

Vaquero-Martínez, J., Antón, M., Ortiz de Galisteo, J.P., Cachorro, V.E., Wang, H.,González Abad, G., Román, R., Costa, M.J., 2017b. Validation of integrated watervapor from OMI satellite instrument against reference GPS data at the IberianPeninsula. Sci. Total Environ. 580, 857–864. http://linkinghub.elsevier.com/retrieve/pii/S0048969716327176http://dx.doi.org/10.1016/j.scitotenv.2016.12.032.

Wagner, T., Beirle, S., Deutschmann, T., 2009. Three-dimensional simulation of the Ringeffect in observations of scattered sun light using Monte Carlo radiative transfermodels. Atmos. Meas. Tech. 2, 113–124. http://www.atmos-_meas-_tech.net/2/113/2009/http://dx.doi.org/10.5194/amt-2-113-2009.

Wagner, T., Beirle, S., Grzegorski, M., Platt, U., 2006. Global trends (1996-2003) of totalcolumn precipitable water observed by Global Ozone Monitoring Experiment(GOME) on ERS-2 and their relation to near-surface temperature. J. Geophys. Res.Atmos. 111, 1–15. http://dx.doi.org/10.1029/2005JD006523.

Wagner, T., Heland, J., Zöger, M., Platt, U., 2003. A fast H< sub>2</sub>O totalcolumn density product from GOME& ndash; validation with in-situ aircraft mea-surements. Atmos. Chem. Phys. Discuss. 3, 323–353. http://dx.doi.org/10.5194/acpd-3-323-2003.

Wang, H., Gonzalez Abad, G., Liu, X., Chance, K., 2016a. Validation of OMI total columnwater vapor product. Atmos. Chem. Phys. Discuss. 181, 1–23. http://dx.doi.org/10.5194/acp-2016-181.

Wang, H., Liu, X., Chance, K., González Abad, G., Chan Miller, C., 2014. Water vaporretrieval from OMI visible spectra. Atmos. Meas. Tech. 7, 1901–1913. http://dx.doi.org/10.5194/amt-7-1901-2014.

Wang, X., Zhang, K., Wu, S., Fan, S., Cheng, Y., 2016b. Water vapor-weighted meantemperature and its impact on the determination of precipitable water vapor and itslinear trend. J. Geophys. Res. Atmos. 121, 833–852. http://doi.wiley.com/10.1002/2015JD024181http://dx.doi.org/10.1002/2015JD024181.

Wang, X., Zhang, K., Wu, S., He, C., Cheng, Y., Li, X., 2016c. Determination of zenithhydrostatic delays and the development of new global long-term GNSS-derived pre-cipitable water vapor. Atmos. Meas. Tech. Discuss. 10, 1–17. http://www.atmos-_meas-_tech-_discuss.net/amt-_2016-_264/http://dx.doi.org/10.5194/amt-2016-264.

Wilcoxon, F., 1946. Individual comparisons of grouped data by ranking methods. J. Econ.Entomol. 39, 269. http://dx.doi.org/10.2307/3001968.

du Piesanie, A., Piters, A.J.M., Aben, I., Schrijver, H., Wang, P., Noël, S., 2013. Validationof two independent retrievals of sciamachy water vapour columns using radiosondedata. Atmos. Meas. Tech. 6, 2925–2940. http://www.atmos-_meas-_tech.net/6/2925/2013/http://dx.doi.org/10.5194/amt-6-2925-2013.

J. Vaquero-Martínez et al. Remote Sensing of Environment 204 (2018) 729–740

740

68 Apendice A. Artıculos

A.3. Artıculo 3 69

A.3. Artıculo 3

Tıtulo: Validation of Integrated Water Vapor from OMI Sate-llite Instrument against Reference GPS Data at the IberianPeninsula.Autores

• Vaquero-Martınez, Javier (U. de Extremadura).• Anton, Manuel (U. de Extremadura).• Ortiz de Galisteo, Jose P. (AEMet).• Cachorro, Victoria E. (U. de Valladolid).• Wang, Huiqun (Smithsonian Astrophysical Observatory).• Gonzalez Abad, Gonzalo (Smithsonian Astrophysical Observatory).• Roman, Roberto (U. de Valladolid).• Costa, Maria Joao (U. de Evora).

Ano: 2017.Revista: Science of The Total Environment.Paginas: 857–864.DOI: 10.1016/j.scitotenv.2016.12.032.Volumen: 580.

Science of the Total Environment 580 (2017) 857–864

Contents lists available at ScienceDirect

Science of the Total Environment

j ourna l homepage: www.e lsev ie r .com/ locate /sc i totenv

Validation of integrated water vapor from OMI satellite instrumentagainst reference GPS data at the Iberian Peninsula

Javier Vaquero-Martíneza, b,*, Manuel Antóna, b, José Pablo Ortiz de Galisteoc, d, Victoria E. Cachorrod,Huiqun Wange, Gonzalo González Abade, Roberto Románf, h, Maria João Costag

aDepartamento de Física, Universidad de Extremadura, Badajoz, SpainbInstituto Universitario de Investigación del Agua, Cambio Climático y Sostenibilidad (IACYS), Universidad de Extremadura, Badajoz, SpaincAgencia Estatal de Meteorología (AEMET), Valladolid, SpaindGrupo de Óptica Atmosférica, Universidad de Valladolid, Valladolid, SpaineSmithsonian Astrophysical Observatory, Cambridge, MA, United StatesfDepartamento de Física Aplicada, Universidad de Granada, Granada, SpaingDepartamento de Física, Instituto de Ciências da Terra, Escola de Ciências e Tecnología, Universidade de Évora, Évora, PortugalhAndalusian Institute for Earth System Research (IISTA-CEAMA), University of Granada, Autonomous Government of Andalusia, 18006, Granada, Spain

H I G H L I G H T S

• Version 1.0 OMI IWV product ispromising, in fairly good agreementwith GPS data.

• OMI data can sometimes be unrep-resentative to possible extreme localIWV values.

• Low IWV data show great vari-ability (∼ 100%) and overestimation(∼ +40%).

• High IWV data show less variabilityand underestimation (∼−20%).

• Seasonal and SZA dependence of OMI-GPS differences is mainly related toIWV.

G R A P H I C A L A B S T R A C T

A R T I C L E I N F O

Article history:Received 14 October 2016Received in revised form 3 December 2016Accepted 4 December 2016Available online 14 December 2016

Editor: D. Barcelo

Keywords:OMIWater vaporValidationIWVGPSSatellite

A B S T R A C T

This paper shows the validation of integrated water vapor (IWV) measurements retrieved from the OzoneMonitoring Instrument (OMI), using as reference nine ground-based GPS stations in the Iberian Peninsula.The study period covers from 2007 to 2009. The influence of two factors, - solar zenith angle (SZA) and IWV -,on OMI-GPS differences was studied in detail, as well as the seasonal dependence. The pseudomedian of therelative differences is −1 ± 1% and the inter-quartile range (IQR) is 41%. Linear regressions calculated overeach station show an acceptable agreement (R2 up to 0.77). The OMI-GPS differences display a clear depen-dence on IWV values. Hence, OMI substantially overestimates the lower IWV data recorded by GPS (∼40%),while underestimates the higher IWV reference values (∼20%). In connection to this IWV dependence, therelative differences also show an evident SZA dependence when the whole range of IWV values are analyzed(OMI overestimates for high SZA values while underestimates for low values). Finally, the seasonal varia-tion of the OMI-GPS differences is also associated with the strong IWV dependence found in this validationexercise.

© 2016 Elsevier B.V. All rights reserved.

* Corresponding author at: Departamento de Física, Universidad de Extremadura, Badajoz, Spain.E-mail address: javier_vm@unex.es (J. Vaquero-Martínez).

http://dx.doi.org/10.1016/j.scitotenv.2016.12.0320048-9697/© 2016 Elsevier B.V. All rights reserved.

A.3. Artıculo 3 71

858 J. Vaquero-Martínez et al. / Science of the Total Environment 580 (2017) 857–864

1. Introduction

Water vapor is the most important greenhouse gas on Earth, andplays a key role in the hydrological cycle (Myhre et al., 2013). Addi-tionally, it provides latent heating when it condenses, and, accordingto general circulation models (Colman, 2003), it represents a positiveclimate feedback.

However, water vapor is also one of the most variable gases inthe troposphere, both spatially and temporally (Myhre et al., 2013;Ortiz de Galisteo et al., 2011, 2014). Therefore, in order to assess cli-mate change, knowledge of the spatio-temporal distribution of watervapor is fundamental. Since ground-based observations cannot pro-vide a uniform global coverage (being specially scarce over polarand oceanic regions), it is necessary to use satellite measurements toimprove spatial cover.

Water vapor is usually quantified using the column-integratedamount of atmospheric water vapor (IWV), equivalent to condensingall the water vapor in the atmospheric column and measuring theheight that it would reach in a vessel of unit cross section; it can bemeasured in superficial density (g/mm2) or in length (height) units(mm).

Over the years, multiple remote-sensing techniques have beendeveloped to measure IWV both from ground sites and from spaceplatforms. Among them, we find microwave radiometers (Jones et al.,2009; Turner et al., 2007), sun-photometers (Ichoku et al., 2002),Lidar (Turner et al., 2002), satellite measurements (Bennouna et al.,2013; Román et al., 2015; Wang et al., 2014), Global PositioningSystem (GPS) (Ortiz de Galisteo et al., 2011) and radiosounding(Jakobson et al., 2005; Torres et al., 2010).

Among ground-based water vapor instruments, GPS receiver sta-tions are one of the most powerful techniques to measure IWV. Ithas been widely studied, as in Pany et al. (2001) and De Haan et al.(2002) (tested against a numerical model), and Ortiz de Galisteo et al.(2010) (for GPS antenna corrections). One of its main advantages isthe independence of meteorological events, such as cloudiness orprecipitation, along with the possibility of high temporal resolution(up to a few minutes) and low cost of the receivers, allowing a densecoverage (Köpken, 2001).

Unfortunately, ground-based measurements cannot resolve thespatialstructuresofglobalwatervaporfields,andcoverageisrestrictedmainly to land areas. Satellite observations are more suitable forweather forecasts and climate studies, due to high accuracy and highspatial resolution of IWV products. There are, however, two majordrawbacks in polar orbiting satellite observations (Diedrich et al.,2016). First, most areas are sampled only once per day (or even less),depending on latitude and swath width of the instruments. Secondly,clouds are opaque in the visible and NIR spectra and therefore satelliteIWV data under cloudy conditions are not reliable. Therefore, to re-assure the quality of the IWV data derived from satellite instruments,validation exercises using reference measurements are required.

Numerous satellite instruments provide IWV data which havebeen widely inter-compared against reference ground-based mea-surements, namely, Global Ozone Monitoring Experiment-2 (GOME-2) (Grossi et al., 2015; Kalakoski et al., 2016; Noël et al., 2008;Román et al., 2015), MODerate-resolution Imaging Spectroradiome-ter (MODIS) (Bennouna et al., 2013; Chang et al., 2015; Gao andLi, 2008; Li et al., 2003; Ningombam et al., 2016; Prasad andSingh, 2009; Román et al., 2014), Meteosat (Hanssen et al., 2001;Schroedter-Homscheidt et al., 2008), MEdium Resolution ImagingSpectrometer (MERIS) (Diedrich et al., 2016; Li et al., 2006) or SCIA-MACHY (Bovensmann et al., 1999; Noël et al., 2005; Schrijver et al.,2009). Additionally, the Ozone Monitoring Instrument (OMI) alsoprovides IWV data using the retrieval algorithm proposed by Wanget al. (2014). However, to our knowledge, only one validation exer-cise using the IWV product from OMI can be found in literature(Wang et al., 2016) and that paper presented comparisons on a global

scale using reference data that are different from those used in thepresent work.

This paper focuses on the validation of the IWV data obtainedfrom the OMI satellite instrument using as reference the GPS IWVdata recorded at nine stations in the Iberian Peninsula, covering theperiod 2007–2009. The main objective of this paper is to quantify thedifferences between IWV obtained from OMI and GPS, considered asreference, in order to improve the understanding of the quality andaccuracy of the OMI IWV data.

The paper is organized as follows. Datasets are described inSection 2. Section 3 shows the methodology to carry out the study.Results are presented in Section 4, and, finally, conclusions are drawnin Section 5.

2. Data

2.1. OMI data

OMI (Levelt et al., 2006) was launched on 15 of July 2004 on-board NASA Earth Observing System (EOS) Aura satellite into aSun-synchronous polar orbit. Developed by the Netherland’s Agencyfor Aerospace Programs (NIVR) and the Finnish Meteorological Insti-tute (FMI), OMI UV/Vis spectrograph samples the whole planet dailyat 1330 local time (LT).

The OMI IWV data used in this study are the first version of Smith-sonian Astrophysical Observatory (SAO) OMH2O level 2 retrievalswhich uses the SAO operational retrieval algorithm presented indetail in González Abad et al. (2015).

The visible channel (349 nm –504 nm) of OMI covers severalwater vapor spectral bands. These bands are weak compared withbands at longer wavelengths. Using the 7m and 6m + d polyadsbetween 430 –480 nm helps to avoid non-linearity due to saturation.Another feature that makes this retrieval valuable and unique amongsatellite retrievals is the more uniform albedo over the globe makingresults over land and water consistent. Despite albedo uniformity,validation analysis carried on by Wang et al. (2016), showed a signif-icant bias (around 5% lower) of SAO OMH2O version 1 compared toin-situ measurements over the oceans.

The retrieval follows these steps: (1) direct fitting of Slant ColumnDensity (SCD) using a semi-empirical model considering several gases(water vapor, ozone, nitrogen dioxide, O2-O2, glyxoal, liquid water),the Ring effect, the water Ring effect, 3rd order closure polynomials,wavelength shift, under-sampling correction, and common mode. (2)SCD conversion to Vertical Column Density (VCD) by dividing SCD bythe Air Mass factor (AMF). AMF are calculated using radiative trans-fer calculations saved in look-up-tables (LUT) at 442 nm. LUTs aredependent on viewing geometry (solar zenith angle (SZA), viewingzenith angle (VZA) and relative azimuth angle (RAA)) and surfaceproperties (pressure and albedo). It is important here to mention thatAMF is notably sensitive to cloudiness (Wang et al., 2016). Finally,VCD can be converted easily to IWV multiplying by a factor (molecu-lar weight of water divided by Avogadro constant). A full descriptionof the retrieval set up can be found in Wang et al. (2014).

Following the guidelines provided by Wang et al. (2014) forOMH2O quality the cloud fraction had to be lower than 0.1, cloudtop pressure greater than 500 HPa, air mass factor greater than0.75 and retrieval root mean square (RMS) value for the fittingSlant Column Density lower than 0.005. Moreover, only pixel whosemaindataqualityflag flag were equal to 0 were chosen, and pixelsaffected by the row anomaly (Wang et al., 2014) have been rejectedas well.

2.2. GPS data

GPS IWV data used in this work were derived from ground-basedGPS measurements of zenith total delay (ZTD) using tropospheric

72 Apendice A. Artıculos

J. Vaquero-Martínez et al. / Science of the Total Environment 580 (2017) 857–864 859

products from Spanish Geographic Institute “Instituto GeográficoNacional”, a local analysis center of the European Reference Frame(EUREF). These GPS data have been successfully used to performexhaustive validation exercises on satellite IWV data derived fromGOME-2 (Román et al., 2015), and MODIS (Bennouna et al., 2013).The IWV retrieval from GPS method is thoroughly explained in Beviset al. (1992). In the present paper, a brief description is provided.

GPS consists of a constellation of satellites that communicatethrough microwave with ground-based receivers. The time that themicrowave signal spends on reaching the receiver is used to calculatethe distance from the receiver to the satellites. Several correctionsmust be made (relativistic corrections, ionospheric delay, and so on).The effect of troposphere can be accounted for in the Slant Tropo-spheric Delay (STD), which can be converted through the so-calledmapping functions Niell (2000) to Zenith Tropospheric Delay (ZTD).After calculating ZTD with GPS measurements, it can be divided intothe sum of two contributions: Zenith Hydrostatic Delay (ZHD), due totropospheric gases, and Zenith Wet Delay (ZWD) due to water vapor.ZHD can be modeled and removed by knowing the surface atmo-spheric pressure and temperature at station level. Thus, ZWD can becalculated and converted to IWV.

In order to estimate the hydrostatic contribution of the delay, twovariables are needed, temperature and pressure at the location ofGPS station, which are provided by the Spanish Meteorological StateAgency (AEMET). It must be noted that these meteorological stationsare not in the same place as GPS stations, but elsewhere in the sametown or very close to each other, as well as not measuring at the sametimes. Thus, interpolation is needed. Temperature measurement dataare interpolated linearly to the time of GPS measurements. For pres-sure interpolation the barometric tide was taken into account. Sucheffect has an amplitude of about 0.5 mb, but GPS IWV observationsare quite sensitive to uncertainties in pressure (Hagemann et al.,2003). Moreover, the diurnal IWV cycle is quite low. Hence, if a linearinterpolation were used for pressure, part of pressure’s diurnal cyclewould be transferred to IWV’s cycle. Furthermore, a correction basedon the difference of altitude between GPS and meteorological sta-tions was made, considering a standard atmosphere with a verticalgradient of temperature of 6.5 ◦C/km (Ortiz de Galisteo et al., 2014).The IWV data at the nine GPS stations are available for the period2007–2009.

The GPS dataset has a temporal resolution of an hour. Nine GPSstations in the interior of the Iberian Peninsula were chosen, in orderto avoid introducing data that might be influenced by sea or mixedsea-land pixels in the satellite observations. A summary of the sta-tions and their IWV characteristics can be found in Table 1, alongwith a map showing their locations in Fig. 1. In Table 1, the medianand IQR of the GPS IWV of every station is shown. coba station hasthe highest median (16.29 mm), while leon has the lowest median(9.27 mm). The rest of stations have medians of 11–13 mm except

Fig. 1. Location of the nine stations selected.

for rioj, which has a median of 14.89 mm. The IQR show higher vari-ability in rioj station (11.90), while leon has the lowest IQR (7.55). TheIQR of the rest of stations are between 8 and 10 mm.

3. Methodology

3.1. Comparison criteria

The spatial co-location between GPS and OMI follows these crite-ria: OMI pixels have to have their center within 0.25◦ longitude and0.25◦ latitude distance to the GPS station. If there is more than onepixel, the average is calculated, weighted by the fitting error.

Regarding temporal co-location, OMI collects data over theIberian Peninsula at 13.30 h (LT) every day. Thus, GPS data selectedto match OMI data is always at 13.30 h, without any delay betweenboth instruments measurements. Data with very high uncertainties(s(ZTD) > 1.5 mm for GPS and DIWVOMI > 10 mm for OMI) havebeen rejected.

Table 1Location and statistics of IWV at the GPS stations considered. The Median and IQR columns show the median and IQR of IWV obtained by GPS. All stations include data from Startdate (last column) to 12/31/09.

Station Acronym Latitude Longitude Altitude N Median IQR Start date

(◦) (◦) (m) (mm) (mm) mm/dd/yy

Córdoba coba 37.92 −4.72 162 605 16.3 9.2 01/01/07León leon 42.59 −5.65 915 390 9.3 7.6 09/09/07Logroño rioj 42.46 −2.5 452 365 14.9 11.9 01/21/07Salamanca sala 40.95 −5.5 800 524 11.4 8.0 01/01/07Sonseca sons 39.68 −3.96 755 429 11.1 9.2 09/09/07Teruel teru 40.35 −1.12 956 188 11.7 9.7 09/28/08Valladolid vala 41.70 −4.71 766 264 12.6 8.3 04/20/08Villafranca vill 40.44 −3.95 596 532 12.8 9.2 01/01/07Cáceres cace 39.48 −6.34 384 598 12.8 8.8 01/01/07All All – – – 3895 12.8 9.2 –

A.3. Artıculo 3 73

860 J. Vaquero-Martínez et al. / Science of the Total Environment 580 (2017) 857–864

3.2. Statistical analysis

The study is carried out by building a dataset where every rowhas an OMI datum, a GPS datum, the station, and additional informa-tion, such as Solar Zenith Angle (SZA), date, and so on. The GPS andOMI data are matched according to the spatial and temporal crite-ria mentioned in the previous section. The relative differences (d) areanalyzed. The relative difference is calculated as:

di = 100 •wOMI

i − wGPSi

wGPSi

where the index i denotes a certain location and date, w is the IWVmeasured by OMI or GPS. In order to study the distribution of d,two indices are applied: the pseudomedian (d) and the interquar-tile range (IQR). The pseudomedian is calculated through Wilcoxonsigned rank test with continuity correction (Wilcoxon, 1946). Thepseudomedian agrees with the median (the datum that has as manydata over it as under it) in a symmetric dataset. It is defined as themedian of all the midpoints of pairs of observations. The pseudome-dian gives a measurement of the accuracy. If it was close to zero, OMIand GPS would be very close, but if it was positive, it would meanthat OMI is overestimating IWV, while a negative value of the pseu-domedian would be a sign of an underestimation tendency of OMIdata. IQR is the difference between the first (the datum that has 25%of data under it and 75% over it) and the third (the datum that has75% of data under it and 25% over it) quartile. This gives the widthof the central half of the data, being therefore a measurement of theprecision of the satellite measurement.

In this context, a statistical analysis per station has been per-formed, in order to detect differences between stations. In the anal-ysis, indices over each station have been calculated, and a linearregression analysis between the GPS and the OMI data is performedfor each ground-based GPS station and for all stations together inorder to analyze their proportionality and similarity. In order tostudy the influence of other parameters (SZA, IWV and season) onthese indices, data have been binned for these variables. The indicescalculated over the binned data have been plotted against the centralvalue of the bin. For IWV, the bin width was 5 mm, for SZA 5◦ andfor seasonal dependence, one month. Bins with less than 30 data areignored in this paper.

4. Results and discussion

4.1. Statistical analysis

A statistical analysis was performed on the nine stations. Table 2shows the results of this analysis, where the pseudomedian andIQR of the d distribution are calculated, and a regression analysis

is performed between OMI and GPS IWV data. The pseudomedianvalues show that OMI both underestimates and overestimates thereference GPS data, with values between −19% for rioj and +23% forteru. The values of pseudomedian and IQR have been also calculatedin absolute value (in mm), reporting values of −2.8 mm (rioj) and+2.5 mm (teru) for the pseudomedian, and around 4 mm for the IQR

This over/under-estimation shown in the signs of the pseudome-dian values in Table 2 can be partially related to the IWV distributionat each ground-based station. In order to clarify this issue, the his-togram of the station with highest overestimation (teru) and under-estimation (rioj) is shown in Fig. 2. In this plot, it can be appreciatedthat teru tends to have lower IWV than rioj, where higher IWV valuesare more common. This suggests that those stations with low valuesof IWV tend to be overestimated by OMI, while those with highervalues of IWV tend to be underestimated. This IWV dependence willbe analyzed in detail in the next subsection. Another reason for thedifferences between stations could be their altitude, since the AMFcorrection is affected by this variable (Palmer et al., 2001). From ourresults, most ground-based stations located at higher altitude showgreater IQR and pseudomedian values than the stations at loweraltitude.

The all-stations row shows that, on average, OMI observationsslightly underestimate the reference GPS data (∼1%), which is ingood agreement with the results reported by Wang et al. (2016).IQR ranges approximately between 30% and 50%, with a value of 41%when all data are analyzed together. Part of the variability is relatedto the stripes in OMI data which can be up to 15–20% (Wang et al.,2014). Nevertheless, the high IQR values in Table 2 indicate a notablevariability in the OMI-GPS relative differences over the study region.

The slopes derived from the linear regression analysis are around0.8 while the intercepts are generally positive, suggesting that datatend to be overestimated for lower values of IWV and underesti-mated for the higher ones. In order to show this more clearly, Fig. 3shows the scatter-plot between OMI and GPS using all data from thenine stations. The regression line is above the 1:1 line when IWV islower than ∼12 mm. The R2 values are between 0.6 and 0.7 in moststations which indicates a fairly good agreement between OMI andGPS data.

In Fig. 4, time series plots are shown for each station. As shownin Table 1, some GPS stations started working after 01/01/2007. GPSand OMI appear to agree reasonably well. Stations coba, rioj andvala show a trend towards negative OMI-GPS differences, while sala,sons and teru are prone to positive OMI-GPS differences. The rest ofstations do not show a clear behavior.

4.2. IWV dependence

The OMI-GPS differences display an evident dependence on IWVvalues as can be observed in Fig. 5 (top). In this plot, the pseudome-dians of relative differences are represented against IWV in bins of

Table 2Statistical analysis of OMI-GPS relative differences. The pseudomedian (pMedian) and IQR of the d distribution, the number of data (N) and the coefficients of the regressionanalysis are shown. y0 column shows the intercept, b stands for the slope and R2 is Pearson’s coefficient of determination. The numbers in parenthesis are the 95% confidenceinterval.

Station pMedian IQR pMedian IQR N y0 b R2

(mm) (mm) (%) (%) (mm)

coba −2.3(0.3) 5.0 −13(2) 27.5 605 1.7(0.8) 0.76(0.04) 0.67leon −0.7(0.4) 4.9 −5(4) 51.6 390 1.8(0.9) 0.77(0.08) 0.51rioj −2.8(0.3) 4.3 −19(2) 26.7 365 −0.3(0.8) 0.84(0.05) 0.77sala 1.4(0.3) 4.9 15(3) 48.4 524 3.1(0.7) 0.86(0.06) 0.64sons 1.6(0.3) 3.9 18(3) 41.3 429 3.4(0.7) 0.85(0.05) 0.71teru 2.5(0.6) 4.6 23(6) 43.1 188 3.2(1.3) 0.96(0.09) 0.69vala −0.8(0.4) 3.6 −5(3) 29.8 264 2.2(0.9) 0.78(0.06) 0.68vill −0.7(0.3) 4.5 −4(3) 35.7 532 2.4(0.7) 0.78(0.05) 0.66cace −0.1(0.3) 4.7 0(3) 37.4 598 2.3(0.8) 0.84(0.05) 0.63All −0.3(0.1) 5.1 −1(1) 40.8 3895 2.7(0.3) 0.78(0.02) 0.63

74 Apendice A. Artıculos

J. Vaquero-Martínez et al. / Science of the Total Environment 580 (2017) 857–864 861

Fig. 2. Histogram of IWV values (GPS) for the stations teru and rioj.

5 mm. The error bars are the 95% confidence interval in the Wilcoxonsigned rank test with continuity correction. A strong overestimation(40%) is observed for low IWV values. By contrast, high IWV valuestend to be underestimated (about 15%). Additionally, the OMI-GPSdifferences corresponding to intermediate IWV values, around 10–15 mm are the closest to 0. This IWV dependence can be probablyrelated to the fact that the IWV is averaged over the closest pixelsat 0.25◦ latitude × 0.25◦ longitude neighborhood pixels, while GPSprovides the local value. As a consequence, if a ground-based stationexhibits an extreme value of IWV (very low or very high) compared

0

10

20

30

40

50

0 10 20 30 40 50

GPS IWV(mm)

OM

I IW

V(m

m)

Lines Fitted Ideal

Fig. 3. Scatter-plot of OMI IWV vs. GPS IWV. The red, dashed line is the linearregression and the black, solid line is the 1:1 line.

with the area nearby the average OMI IWV will tend to overestimatelow IWV and underestimate high IWV. Moreover, two subsets arealso plotted in Fig. 5: high (above 40◦) and low (below 40◦) SZA val-ues. The behavior is not significantly different between them, whichpoints out that the OMI-GPS dependence on IWV does not depend onthe SZA range selected for the analysis.

Fig. 5 (bottom) shows the influence of IWV over the IQR derivedfrom the OMI-GPS relative differences. There is a sharp decrease ofIQR with increasing IWV, with OMI-GPS differences greatly disperse(over 100%) for IWV values below 5 mm and less than 20% for IWVabove 20 mm. Additionally, the difference between high and low SZAis very little, with the values for low SZA slightly lower than those forhigh SZA.

4.3. SZA dependence

The influence of SZA on the satellite instruments performancehas been reported in Antón et al. (2015) and Román et al. (2015) forGOME-2, which uses a window in the visible part of the spectrumlike OMI. Fig. 6 (top) shows a strong dependence of pseudomedianIWV values on SZA when the whole range of IWV values is usedin the analysis. For low SZA (below 35◦), OMI underestimates theGPS IWV data. This pattern is the opposite when dealing with highSZA, being the pseudomedian of differences positive and uncertaintyhigher. The best agreement between both instruments is achievedfor intermediate SZAs (35◦ < SZA < 60◦). However, if a distinctionis made between low (below 12.75 mm) and high IWV (above 12.75mm), there is a clear difference in the behavior of these two subsetswith respect to the whole dataset. Low IWV values are notably over-estimated, while high IWV data are underestimated, in agreementwith the results shown in the previous section. Moreover, the pseu-domedian of both subsets does not significantly change with SZA,suggesting that the SZA dependence observed for the whole datasetcan be associated with the seasonal cycle of IWV data over the studyregion with the smallest IWV values in winter (high SZA ) and thelargest in summer (low SZA). In fact, at low SZA, temperatures arehigher, and more water vapor can be accepted by the atmosphere.However, at higher SZA, temperatures are lower, and not so muchwater vapor can be accepted by the atmosphere.

IQR, however, is clearly lower for low SZA (less than 30%) thanfor high SZA (75%), as it can be observed in Fig. 6 (bottom). There isan increasing tendency of IQR with SZA. Nevertheless, high IWV datahave IQR values between 25% and 30% for all SZA, while low IWVdata have a larger variation with SZA. The values of IQR for the wholedataset seem to be heavily influenced by the value of IWV.

4.4. Seasonal dependence

Fig. 7 (top) shows a clear seasonal dependence of the OMI-GPSdifferences due to the fact that conditions of humidity (IWV) differfrom one season to another. Regarding all IWV conditions, summermonths (June, July and August) are underestimated by OMI betweena 5% and a 10%. Winter months (December, January and February)are, however, overestimated, with greater uncertainty. The monthswhen OMI-GPS values are closer to zero are in the spring (March,April and May) and autumn (September, October and November),except for March (which behaves like winter months) and Septem-ber (which behaves like summer months). This behavior is similar tothe one reported in Wang et al. (2016). In Fig. 7 two subsets havebeen considered: low IWV and high IWV, with the values consideredin Section 4.3. High IWV subsets do not change notably through-out the year, being underestimated. The low IWV subset is mostlyoverestimated.

IQR is shown in Fig. 7 (bottom). For all IWV conditions, Summermonths have the lowest IQR (30%). Winter months have veryhigh IQR, reaching 75% for December. Spring and autumn show

A.3. Artıculo 3 75

862 J. Vaquero-Martínez et al. / Science of the Total Environment 580 (2017) 857–864

cace coba leon

rioj sala sons

teru vala vill

0

20

40

0

20

40

0

20

40

07 08 09 10 07 08 09 10 07 08 09 10

Date

IWV

(m

m)

Instrument GPS OMI

Fig. 4. Time series of IWV from both instruments, OMI and reference GPS, for each station . The green line shows the difference between OMI and GPS IWV measurements. Thedate is in years.

Fig. 5. IWV dependence on the pseudomedian (top) and the IQR (bottom) of OMI-GPSrelative differences. The size of dots represents the number of data used to obtain theindex for that data point (N).

Fig. 6. SZA dependence on the pseudomedian (top) and the IQR (bottom) of OMI-GPSrelative differences. The size of dots represents the number of data used to obtain theindex for that data point (N).

76 Apendice A. Artıculos

J. Vaquero-Martínez et al. / Science of the Total Environment 580 (2017) 857–864 863

Fig. 7. Seasonal dependence on the pseudomedian (top) and IQR (bottom) of OMI-GPS relative differences. The size of dots represents the number of data used to obtainthe index for that data point (N). December has been rearranged as the first month inorder to facilitate the identification of the different seasons.

intermediate performance. Regarding the subsets studied, the lowIWV subset has clearly higher IQR than the high IWV subset. Thispattern is very different from the one reported in Wang et al. (2016),where the index used was the standard deviation of the differencesOMI-GPS (in mm), and summer months have larger differences thanwinter months. One reason for this could be the fact that during sum-mer months IWV is higher, and thus the relative differences are lowin percentage. In winter, the opposite happens: IWV is low, makingsmall absolute differences appear large in terms of percentage. It isworth noticing that the standard deviation reported by Wang et al.(2016) ranges from 5 mm to 7 mm, while IWV ranges from 1 mm to39 mm. It can be observed that low IWV subset presents lower IQRthat high IWV throughout the year, and that the decrease in summermonths of IQR for all IWV conditions is due to larger amount of highIWV cases than low IWV cases.

5. Conclusions

The period 2007 –2009 has been studied to validate the IWVdata retrieved by OMI against the GPS reference data at nine landground-based stations in the Iberian Peninsula. The IWV valuesretrieved from OMI show a strong dependence on IWV values. Satel-lite IWV data tend to overestimate (pseudomedian values around

40%) low IWV data, showing high variability. For high IWV data, thebehavior is the opposite: OMI underestimates the reference GPS data(pseudomedian value around 20%) and the variability is lower. Theover/under-estimation can be partially related to OMI – GPS data co-location: OMI data is averaged using the pixels in the neighborhoodof GPS station. Additionally, satellite IWV data are retrieved usingalbedo, surface pressure and other information. This is in contrast tothe local nature of GPS measurement. Consequently, OMI measure-ments can sometimes be unrepresentative to possible extreme localIWV values.

The effect of SZA on the water vapor column retrieved by OMIsatellite instrument is highly dependent on the seasonal cycle of IWVvalues over the study region, with the highest IWV values in summerand the lowest in winter. Thus, low SZA data has similar accuracy andprecision as high IWV data. Low SZA data tend to be underestimatedand have less variability. High SZA, however, is related to low IWVvalues, which leads to overestimation and high variability. The lit-tle impact of SZA on OMI-GPS IWV differences contrast with Románet al. (2015), where a strong dependence of SZA on the differencesbetween GOME-2 IWV and GPS in the Iberian Peninsula was found.This points out that OMI retrieval appears to be more independent ofSZA values than GOME-2 retrieval.

The seasonal variation found for the OMI–GPS differences ismainly related to IWV. Summer months have high IWV values(higher temperatures), exhibiting underestimation and low variabil-ity, while winter months are associated with low IWV (low temper-atures), with overestimation and high variability. Even in the bestcases of OMI-GPS agreement, the IQR is high (more than 25%) whichsuggests increase variability of the relative differences.

In summary, although the version 1.0 OMI satellite water vaporproduct is very promising, in a fairly good agreement with referenceGPS data, it still needs improvements in order to reduce the vari-ability and IWV dependence. An updated version 2.1 OMI algorithmshows the potential to improve data quality, as reported in Wang etal. (2016). Its performance should be tested once the data is released.

Acknowledgements

This work was supported by the Spanish Ministry of Economyand Competitiveness through project CGL2014-56255-C2. ManuelAntón thanks Ministerio de Ciencia e Innovación and Fondo SocialEuropeo (RYC-2011-08345) for the award of a postdoctoral grant(Ramón y Cajal). Support from the Junta de Extremadura (ResearchGroup Grants GR15137) is gratefully acknowledged. Work at Univer-sidad de Valladolid is supported by project CMT2015-66742-R. Workat Universidad de Granada was supported by the Andalusia RegionalGovernment (project P12-RNM-2409) and the Spanish Ministry ofEconomy and Competitiveness and FEDER funds under the projectsCGL2013-45410-R and “Juan de la Cierva-Formación” program. Workat SAO is supported by NASA’s Atmospheric Composition: AuraScience Team program (sponsor contract number NNX14AF56G).Work at Universidade de Évora is co-funded by the EuropeanUnion through the European Regional Development Fund, includedin the COMPETE 2020 (Operational Program Competitiveness andInternationalization) through the ICT project (UID/GEO/04683/2013)with the reference POCI-01-0145-FEDER-007690.

References

Antón, M., Loyola, D., Román, R., Vömel, H., 2015. Validation of GOME-2/MetOp-A totalwater vapour column using reference radiosonde data from the GRUAN network.Atmos. Meas. Tech. 8, 1135–1145. http://dx.doi.org/10.5194/amt-8-1135-2015.

Bennouna, Y.S., Torres, B., Cachorro, V.E., Ortiz de Galisteo, J.P., Toledano, C., 2013. Theevaluation of the integrated water vapour annual cycle over the iberian peninsulafrom EOS-MODIS against different ground-based techniques. Q. J. Roy. Meteorol.Soc. 139, 1935–1956. http://dx.doi.org/10.1002/qj.2080.

A.3. Artıculo 3 77

864 J. Vaquero-Martínez et al. / Science of the Total Environment 580 (2017) 857–864

Bevis, M., Businger, S., Herring, T. a., Rocken, C., Anthes, R. a., Ware, R.H., 1992. GPSmeteorology: remote sensing of atmospheric water vapor using the global posi-tioning system. J. Geophys. Res. 97, 15787–15801. http://doi.wiley.com/10.1029/92JD01517. http://dx.doi.org/10.1029/92JD01517.

Bovensmann, H., Burrows, J.P., Buchwitz, M., Frerick, J., Noël, S., Rozanov, V.V.,Chance, K.V., Goede, a.P.H., 1999. SCIAMACHY: mission objectives and mea-surement modes. J. Atmos. Sci. 56, 127–150. http://dx.doi.org/10.1175/1520-0469(1999)056&iexcl;0127:SMOAMM&iquest;2.0.CO;2.

Chang, L., Gao, G., Jin, S., He, X., Xiao, R., Guo, L., 2015. Calibration and evalua-tion of precipitable water vapor from MODIS infrared observations at night.IEEE Trans. Geosci. Remote Sens. 53, 2612–2620. http://dx.doi.org/10.1109/TGRS.2014.2363089.

Colman, R., 2003. A comparison of climate feedbacks in general circulation models.Climate Dynam. 20, 865–873. http://link.springer.com/article/10.1007/s00382-003-0310-z. http://dx.doi.org/10.1007/s00382-003-0310-z.

De Haan, S., Van Der Marel, H., Barlag, S., 2002. Comparison of GPS slant delay mea-surements to a numerical model: case study of a cold front passage. Phys. Chem.Earth 27, 317–322. http://dx.doi.org/10.1016/S1474-7065(02)00006-2.

Diedrich, H., Wittchen, F., Preusker, R., Fischer, J., 2016. Representativeness of total col-umn water vapour retrievals from instruments on polar orbiting satellites. Atmos.Chem. Phys. Discuss. 16, 8331–8339. http://dx.doi.org/10.5194/acp-2016-99.

Gao, B.C., Li, R.R., 2008. The time series of terra and aqua MODIS near-IR water vaporproducts. Int. Geosci. Remote Sens. Symp. (IGARSS) 3, 186–189. http://dx.doi.org/10.1109/IGARSS.2008.4779314.

González Abad, G., Liu, X., Chance, K., Wang, H., Kurosu, T.P., Suleiman, R., 2015.Updated smithsonian astrophysical observatory ozone monitoring instrument(SAO OMI) formaldehyde retrieval. Atmos. Meas. Tech. 8, 19–32. http://dx.doi.org/10.5194/amt-8-19-2015.

Grossi, M., Valks, P., Loyola, D., Aberle, B., Slijkhuis, S., Wagner, T., Beirle, S.,Lang, R., 2015. Total column water vapour measurements from GOME-2metop-a and metop-b. Atmospheric Measurement Techniques 8, 1111–1133.http://www.atmos-meas-tech.net/8/1111/2015/.http://dx.doi.org/10.5194/amt-8-1111-2015.

Hagemann, S., Bengtsson, L., Gendt, G., 2003. On the determination of atmosphericwater vapor from GPS measurements - art. no. 4678. J. Geophys. Res.-Atmos. 108,4678.

Hanssen, R.F., Feijt, A.J., Klees, R., 2001. Comparison of precipitable water vaporobservations by spaceborne radar interferometry and meteosat 6.7−lm radiom-etry. J. Atmos. Ocean. Technol. 18, 756–764. http://dx.doi.org/10.1175/1520-0426(2001)058&iexcl;0756:COPWVO&iquest;2.0.CO;2.

Ichoku, C., Levy, R., Kaufman, Y.J., Remer, L.A., Li, R.R., Martins, V.J., Holben, B.N.,Abuhassan, N., Slutsker, I., Eck, T.F., Pietras, C., 2002. Analysis of the performancecharacteristics of the five-channel microtops II sun photometer for measuringaerosol optical thickness and precipitable water vapor. J. Geophys. Res. Atmos.107, http://dx.doi.org/10.1029/2001JD001302.

Jakobson, E., Ohvril, H., Okulov, O., Laulainen, N., 2005, jun. Variability ofradiosonde-observed precipitable water in the Baltic region. Nord. Hydrol. 36,423–433.

Jones, A., Urban, J., Murtagh, D.P., Eriksson, P., Brohede, S., Haley, C., Degenstein, D.,Bourassa, A., Von Savigny, C., Sonkaew, T., Rozanov, A., Bovensmann, H., Burrows,J., 2009. Evolution of stratospheric ozone and water vapour time series studiedwith satellite measurements. Atmos. Chem. Phys. Discuss 9, 1157–1209. www.atmos-chem-phys-discuss.net/9/1157/2009/. http://dx.doi.org/10.5194/acpd-9-1157-2009.

Kalakoski, N., Kujanpää, J., Sofieva, V., Tamminen, J., Grossi, M., Valks, P., 2016. Vali-dation of GOME-2/MetOp total column water vapour with ground-based and insitu measurements. Atmos. Meas. Tech. 9, 1533–1544. http://dx.doi.org/10.5194/amt-9-1533-2016.

Köpken, C., 2001. Validation of integrated water vapor from numerical mod-els using ground-based GPS, SSM/I, and water vapor radiometer measure-ments. J. Appl. Meteorol. 40, 1105–1117. http://dx.doi.org/10.1175/1520-0450(2001)040&iexcl;1105:VOIWVF&iquest;2.0.CO;2.

Levelt, P.F., van den Oord, G.H.J., Dobber, M.R., Malkki, A., Visser, H., de Vries, J.,Stammes, P., Lundell, J.O.V., Saari, H., 2006. The ozone monitoring instrument.IEEE Trans. Geosci. Remote Sens. 44, 1093–1101. http://dx.doi.org/10.1109/TGRS.2006.872333.

Li, Z., Muller, J., Cross, P., Albert, P., Fischer, J., Bennartz, R., 2006. Assessment of thepotential of MERIS near-infrared water vapour products to correct ASAR inter-ferometric measurements. Int. J. Remote Sens. 27, 349–365. http://dx.doi.org/10.1080/01431160500307342.

Li, Z., Muller, J.P., Cross, P., 2003. Comparison of precipitable water vapor derived fromradiosonde, GPS, and moderate-resolution imaging spectroradiometer measure-ments. J. Geophys. Res. Atmos. 108, http://dx.doi.org/10.1029/2003JD003372.

Myhre, G., Shindell, D., Bréon, F.-M., Collins, W., Fuglestvedt, J., Huang, J., Koch, D.,Lamarque, J.-F., Lee, D., Mendoza, B., Nakajima, T., Robock, A., Stephens, G.,Takemura, T., Zhang, H., 2013, . Anthropogenic and natural radiative forcing.Technical Report WG1AR5. http://dx.doi.org/10.1017/CBO9781107415324.018.

Niell, A.E., 2000. Improved atmospheric mapping functions for VLBI and GPS. EarthPlanets Space 52, 699–702. http://dx.doi.org/10.1186/BF03352267.

Ningombam, S.S., Jade, S., Shrungeshwara, T.S., Song, H.J., 2016. Validation of watervapor retrieval from moderate resolution imaging spectro-radiometer (MODIS)in near infrared channels using GPS data over IAO-Hanle, in the trans-himalayanregion. J. Atmos. Sol. Terr. Phys. 137, 76–85. http://dx.doi.org/10.1016/j.jastp.2015.11.019.

Noël, S., Buchwitz, M., Bovensmann, H., Burrows, J.P., 2005. Validation of SCIAMACHYAMC-DOAS water vapour columns. Atmos. Chem. Phys. 5, 1835–1841. http://dx.doi.org/10.5194/acp-5-1835-2005.

Noël, S., Mieruch, S., Bovensmann, H., Burrows, J.P., 2008. Preliminary results of GOME-2 water vapour retrievals and first applications in polar regions. Atmos. Chem.Phys. 8, 1519–1529. http://dx.doi.org/10.5194/acp-8-1519-2008.

Ortiz de Galisteo, J.P., Bennouna, Y., Toledano, C., Cachorro, V., Romero, P., Andrés, M.I.,Torres, B., 2014. Analysis of the annual cycle of the precipitable water vapour overSpain from 10-year homogenized series of GPS data. Q. J. Roy. Meteorol. Soc. 140,397–406. http://dx.doi.org/10.1002/qj.2146.

Ortiz de Galisteo, J.P., Cachorro, V., Toledano, C., Torres, B., Laulainen, N., Bennouna, Y.,de Frutos, A., 2011. Diurnal cycle of precipitable water vapor over Spain. Q. J. Roy.Meteorol. Soc. 137, 948–958. http://dx.doi.org/10.1002/qj.811.

Ortiz de Galisteo, J.P., Toledano, C., Cachorro, V., Torres, B., 2010. Improvement in PWVestimation from GPS due to the absolute calibration of antenna phase centervariations. GPS Solutions 14, 389–395. http://dx.doi.org/10.1007/s10291-010-0163-y.

Palmer, P.I., Jacob, D.J., Chance, K., Martin, R.V., Spurr, R.J.D., Kurosu, P., Bey, I., Yantosca,R., Fiore, A., Li, Q., 2001. Air mass factor formulation for spectroscopic measure-ments from satellites: application to formaldehyde retrievals form the globalozone monitoring experiment. J. Geophys. Res. 106, 14539–14550. http://dx.doi.org/10.1029/2000JD900772.

Pany, T., Pesec, P., Stangl, G., 2001. Atmospheric GPS slant path delays and rays tracingthrough numerical weather models, a comparison. Phys. Chem. Earth Solid EarthGeod. 26, 183–188. http://dx.doi.org/10.1016/S1464-1895(01)00044-8.

Prasad, A.K., Singh, R.P., 2009. Validation of MODIS terra, AIRS, NCEP/DOE AMIP-II reanalysis-2, and AERONET sun photometer derived integrated precipitablewater vapor using ground-based GPS receivers over India. J. Geophys. Res. Atmos.114, 1–20. http://dx.doi.org/10.1029/2008JD011230.

Román, R., Antón, M., Cachorro, V.E., Loyola, D., Ortiz de Galisteo, J.P., de Fru-tos, A., Romero-Campos, P.M., 2015. Comparison of total water vapor columnfrom GOME-2 on MetOp-A against ground-based GPS measurements at theIberian Peninsula. Sci. Total. Environ. 533, 317–328. http://dx.doi.org/10.1016/j.scitotenv.2015.06.124.

Román, R., Bilbao, J., de Miguel, A., 2014. Uncertainty and variability in satellite-basedwater vapor column, aerosol optical depth and Angström exponent, and its effecton radiative transfer simulations in the Iberian Peninsula. Atmos. Environ. 89,556–569. http://dx.doi.org/10.1016/j.atmosenv.2014.02.027.

Schrijver, H., Gloudemans, A.M.S., Aben, I., 2009. Water vapour total columns fromSCIAMACHY spectra in the 2.36 lm window. Atmos. Meas. Tech. 2, 561–571.http://dx.doi.org/10.5194/amt-2-561-2009.

Schroedter-Homscheidt, M., Drews, A., Heise, S., 2008. Total water vapor columnretrieval from MSG-SEVIRI split window measurements exploiting the daily cycleof land surface temperatures. Remote Sens. Environ. 112, 249–258. http://dx.doi.org/10.1016/j.rse.2007.05.006.

Torres, B., Cachorro, V.E., Toledano, C., Ortiz De Galisteo, J.P., Berjón, A., De Frutos, A.M.,Bennouna, Y., Laulainen, N., 2010. Precipitable water vapor characterization inthe Gulf of Cadiz region (southwestern Spain) based on sun photometer, GPS, andradiosonde data. J. Geophys. Res. Atmos. 115, 1–11. http://dx.doi.org/10.1029/2009JD012724.

Turner, D.D., Clough, S.A., Liljegren, J.C., Clothiaux, E.E., Cady-Pereira, K.E., Gaustad, K.L.,2007. Retrieving liquid water path and precipitable water vapor from the atmo-spheric radiation measurement (ARM) microwave radiometers. IEEE Transactionson Geoscience and Remote Sensing. vol. 45. pp. 3680–3689. http://dx.doi.org/10.1109/TGRS.2007.903703.

Turner, D.D., Ferrare, R.A., Heilman Brasseur, L.A., Feltz, W.F., Tooman, T.P., 2002.Automated retrievals of water vapor and aerosol profiles from an operationalraman lidar. J. Atmos. Ocean. Technol. 19, 37–50. http://dx.doi.org/10.1175/1520-0426(2002)019&iexcl;0037:AROWVA&iquest;2.0.CO;2.

Wang, H., Gonzalez Abad, G., Liu, X., Chance, K., 2016. Validation of OMI total columnwater vapor product. Atmos. Chem. Phys. Discuss. 181, 1–23. http://dx.doi.org/10.5194/acp-2016-181.

Wang, H., Liu, X., Chance, K., González Abad, G., Chan Miller, C., 2014. Water vaporretrieval from OMI visible spectra. Atmos. Meas. Tech. 7, 1901–1913. http://dx.doi.org/10.5194/amt-7-1901-2014.

Wilcoxon, F., 1946. Individual comparisons of grouped data by ranking methods. J.Econ. Entomol. 39, 269. http://dx.doi.org/10.2307/3001968.

78 Apendice A. Artıculos

A.4. Artıculo 4 79

A.4. Artıculo 4

Tıtulo: Validation of MODIS Integrated Water Vapor Productagainst Reference GPS Data at the Iberian Peninsula.Autores

• Vaquero-Martınez, Javier (U. de Extremadura).• Anton, Manuel (U. de Extremadura).• Ortiz de Galisteo, Jose P. (AEMet).• Cachorro, Victoria E. (U. de Valladolid).• Costa, Maria Joao (U. de Evora).• Roman, Roberto (U. de Valladolid).• Bennouna, Yasmine S. (U. de Valladolid).

Ano: 2017.Revista: International Journal of Applied Earth Observation and Geoin-formation.Paginas: 214–221.DOI: 10.1016/j.jag.2017.07.008.Volumen: 63.

Contents lists available at ScienceDirect

Int J Appl Earth Obs Geoinformation

journal homepage: www.elsevier.com/locate/jag

Validation of MODIS integrated water vapor product against reference GPSdata at the Iberian Peninsula

Javier Vaquero-Martíneza,b,⁎, Manuel Antóna,b, José Pablo Ortiz de Galisteoc,d,Victoria E. Cachorrod, Maria João Costae, Roberto Románf,g, Yasmine S. Bennounah

a Departamento de Física, Universidad de Extremadura, Badajoz, Spainb Instituto Universitario de Investigación del Agua, Cambio Climático y Sostenibilidad (IACYS), Universidad de Extremadura, Badajoz, Spainc Agencia Estatal de Meteorologia (AEMET), Valladolid, Spaind Grupo de Óptica Atmosférica, Universidad de Valladolid, Valladolid, Spaine Departamento de Física, Instituto de Ciências da Terra, Escola de Ciências e Tecnología, Universidade de Évora, Évora, Portugalf Department of Applied Physics, University of Granada, Granada, Spaing Andalusian Institute for Earth System Research (IISTA-CEAMA), Granada, Spainh Grupo de Óptica Atmosférica, Universidad de Valladolid, Valladolid, Spain

A R T I C L E I N F O

Keywords:MODISWater vaporValidationIWVGPSSatellite

A B S T R A C T

In this work, the water vapor product from MODIS (MODerate-resolution Imaging Spectroradiometer) instru-ment, on-board Aqua and Terra satellites, is compared against GPS water vapor data from 21 stations in theIberian Peninsula as reference. GPS water vapor data is obtained from ground-based receiver stations whichmeasure the delay caused by water vapor in the GPS microwave signals. The study period extends from 2007until 2012. Regression analysis in every GPS station show that MODIS overestimates low integrated water vapor(IWV) data and tends to underestimate high IWV data. R2 shows a fair agreement, between 0.38 and 0.71. Inter-quartile range (IQR) in every station is around 30–45%. The dependence on several parameters was also ana-lyzed. IWV dependence showed that low IWV are highly overestimated by MODIS, with high IQR (low preci-sion), sharply decreasing as IWV increases. Regarding dependence on solar zenith angle (SZA), performance ofMODIS IWV data decreases between 50° and 90°, while night-time MODIS data (infrared) are quite stable. Theseasonal cycles of IWV and SZA cause a seasonal dependence on MODIS performance. In summer and winter,MODIS IWV tends to overestimate the reference IWV value, while in spring and autumn the tendency is tounderestimate. Low IWV from coastal stations is highly overestimated (∼60%) and quite imprecise (IQR around60%). On the contrary, high IWV data show very little dependence along seasons. Cloud-fraction (CF) depen-dence was also studied, showing that clouds display a negligible impact on IWV over/underestimation. However,IQR increases with CF, except in night-time satellite values, which are quite stable.

1. Introduction

Water vapor is the most important atmospheric greenhouse gas, andits phase changes involve exchanges of latent heat energy. Water isevaporated at low latitudes, and its vapor is transported towards higherlatitudes to condensate, releasing high amounts of heat (Myhre et al.,2013). Moreover, it is well known that water vapor involves a positivefeedback in climate change, according to general circulation models(Colman, 2003). Usually, water vapor is quantified using the columnintegrated amount of atmospheric water vapor (IWV), which isequivalent to condensing all the water vapor in the atmospheric columnand measuring the height that it would reach in a vessel of unit cross

section; IWV can be measured in columnar mass density (g/cm2 or kg/m2) or in length (height) units (mm).

Some details about the role of water vapor in the atmosphere arestill to be completely understood. Thus, it is necessary to monitor watervapor, but this is not a trivial issue for two reasons: the first one is itshigh variability, both temporal and spatial. Water vapor exhibits bothan annual (Ortiz de Galisteo et al., 2014) and diurnal (Ortiz de Galisteoet al., 2011) cycle; therefore, good temporal resolution is very im-portant in water vapor monitoring, especially for some applications.The second reason is that the world coverage is not homogeneous.Water vapor data is scarce over polar and oceanic regions, due to thelack of ground-based observations. Over land, there is still some

http://dx.doi.org/10.1016/j.jag.2017.07.008Received 2 May 2017; Received in revised form 13 July 2017; Accepted 14 July 2017

⁎ Corresponding author at: Departamento de Física, Universidad de Extremadura, Badajoz, Spain.E-mail address: javier_vm@unex.es (J. Vaquero-Martínez).

Int J Appl  Earth Obs Geoinformation 63 (2017) 214–221

Available online 23 August 20170303-2434/ © 2017 Published by Elsevier B.V.

MARK

A.4. Artıculo 4 81

scarcity over some parts of Africa, South America, and North Asia (seeWang et al., 2007).

There are several instruments for measuring IWV: micro-waveradiometers (Turner et al., 2007), star-photometers (Pérez-Ramírezet al., 2012), moon-photometers (Barreto et al., 2013), sun-photometers(Ichoku et al., 2002), lidar (Turner et al., 2002), GPS system (Ortiz deGalisteo et al., 2011), radiosounding (Torres et al., 2010). Additionally,numerous instruments on board satellite platforms can also retrieveIWV data using different parts of the electromagnetic spectrum: mi-crowave by MLS (Livesey and Van Snyder, 2004) and SSM/I (Wentz andSpencer, 1998); visible by GOME-2 (Grossi et al., 2015), OMI (Wanget al., 2014) and SCIAMACHY (Noël et al., 2004); near infra-red byMODIS (Gao and Kaufman, 1992; Gao and Li, 2008), and infra-red byMODIS (Seemann et al., 2006), AIRS (Barnet et al., 2007), AMSR-E(Wentz and Meissner, 2007) and SEVIRI (Schroedter-Homscheidt et al.,2008; Martinez, 2013).

Radiosounding and GPS are among the most powerful techniques tostudy IWV. However, temporal resolution of radiosounding is generallylimited to one or two measurements per day. In contrast, GPS providesa high temporal resolution with numerous records throughout the day-time and night-time. Hence, GPS measurements of water vapor havebeen validated, as in Wang et al. (2007) (against radiosonde, micro-wave radiometer and satellite data), Ohtani and Naito (2000) (againstradiosonde), Bokoye et al. (2003) (against radiosonde and radiometer),Heise et al. (2009) (against ECMWF reanalysis), Schneider et al. (2010)(intercomparison with spectrometer, radiometer and sunphotometer,and radiosondes), and Pany et al. (2001) and de Haan et al. (2002)(tested against a numerical model). From all these validation exercises,GPS IWV data have been checked as a reliable reference, with biasaround 2 mm and standard deviation of about 1.22 mm (see Wanget al., 2007).

GPS measurements of water vapor, however, have two relevantdrawbacks. First, ground-based stations are necessary, so coverage islimited to land areas. Second, spatial resolution depends on density ofthe networks available. Some applications, such as weather forecastsand climate studies, need higher spatial resolution to represent properlythe high spatial variability of water vapor. Satellite retrievals have,however, some issues. On the one hand, polar orbiting satellites have alow temporal coverage (one or two measurement a day, usually) de-pending on the latitude of the area and the swath width of the satellite.On the other hand, visible or NIR radiation is usually used, makingcloudy-scene measurements unreliable due to the opacity of clouds.

This work aims to validate data from MODIS satellite radiometeragainst reference GPS network in the Iberian Peninsula. MODIS datahave been validated before (Li et al., 2003; Gao and Li, 2008; Prasadand Singh, 2009; Chang et al., 2015; Ningombam et al., 2016) in otherareas. However, over the Iberian Peninsula, MODIS has only been va-lidated in Bennouna et al. (2013), Román et al. (2014). This paper aimsto study the dependence of several parameters – IWV, solar zenith angle(SZA), seasonality and clouds – on MODIS performance, which hasnever been done before to our knowledge. Therefore, it is expected thatthis paper will contribute to understand the main drawbacks of the IWVproduct derived from MODIS, allowing the comparison with other re-gions and possible improvements for the retrieval algorithm.

2. Instruments and data

2.1. MODIS

MODIS is a radiometer on board Terra (launched in 1999) and Aqua(launched in 2002) satellite platforms (Salomonson et al., 1989; Kinget al., 1992). Both platforms are sun-synchronous, polar-orbiting sa-tellites, covering the whole planet in 1–2 days. Terra's orbit around theEarth is scheduled to overpass the equator from north to south in themorning, while Aqua passes from south to north over the equator in theafternoon. MODIS swath width is 2330 km.

MODIS measures in 36 spectral bands, covering the range0.4–15 μm acquiring data at three spatial resolutions – 250, 500, and1000 m. However, level 2 moisture profiles and IWV are derived for5 × 5 pixels, which have 1 km2 resolution, thus, the resolution of theIWV product is 5 km × 5 km.

Water vapor is generated for both daytime and night. For daytimefive NIR bands (channels 2, 5, 17, 18, 19) are used (solar radiationreflected by Earth + atmosphere), and for nighttime only IR bands areused (radiation emitted by Earth + atmosphere).

NIR algorithm uses 2-channel and 3-channel rationing techniques,generating look-up tables with values of these ratios and total amountof water vapor associated with such values, using radiative transferalgorithms. Once the total amount of water vapor is obtained, it can beconverted to IWV taking into account the solar and observationalgeometries. In the presence of clouds, other channels in the 0.8–2.5 μ mregion are used, since they contain information about absorptions dueto water vapor above and within clouds. More detailed informationabout the algorithm can be found in Gao and Kaufman, 1992.

IR algorithm consists on a statistical synthetic regression with asubsequent nonlinear physical retrieval that iteratively improves theMODIS solution fit. It uses 25-36 bands, covering the spectral regionbetween 3 and 14.5 μm. More details can be found in Seemann et al.(2006).

The data are included in the water vapor product (MOD05_L2 andMYD05_L2) collection 6. It is, however, obtained from the MODISAtmospheric Profile (MOD07 and MYD07) Collection 6 product, andthen appended to MOD05 product for convenience. MODIS cloudfraction (CF) data has been used as well to select clear-sky scenes and tostudy cloudiness dependence.

2.2. GPS IWV data

GPS stations can be used to derive atmospheric water vapor pro-ducts. The method is explained in the following lines, although a morethorough description can be found in Bevis et al. (1992). The strategyused to determine the position of a receiver deals with the measurementof the time spent by the microwave signal on reaching the receiver atGPS station. However, the signal suffers a series of delays along its path.Among them, the delay caused by the tropospheric gases is called SlantTropospheric Delay (STD). Zenith Tropospheric Delay (ZTD), which isthe delay that the signal would have if the GPS satellite was exactly atthe station's zenith, can be computed from the STD using the mappingfunctions (Niell, 2000). Once ZTD is computed, it can be separated intotwo different contributions: Zenith Hydrostatic Delay (ZHD) and ZenithWet Delay (ZWD). The former is due to tropospheric gases while thelatter is caused by water vapor's dipolar momentum. If surface pressureis known, ZHD can be modeled, and thus ZWD obtained. Then, IWV canbe obtained from ZWD if surface temperature and pressure are known.

In this work, ZTD from 21 GPS ground-based stations were used toobtain IWV products (see Fig. 1 and Table 1). Tropospheric products(ZTD) were provided by Spanish Geographic Institute “Instituto Geo-gráfico Nacional”, which is a local analysis center for the EuropeanReference Frame (EUREF). The meteorological variables (surface pres-sure and temperature) needed to retrieve IWV from ZTD products wereprovided by the Spanish Meteorological State Agency (AEMet). Data areinterpolated to the time of GPS measurements (one measurement perhour). In the case of temperature (hourly data), data were interpolatedlinearly, but in the case of pressure (4 data per day) the barometric tideneeded to be accounted for. This resulted in an IWV data-set for thetwenty-one GPS stations in the period 2007–2012. Every row in thisdata-set has several columns: site, date, hour, IWV, CF, SZA, and othercolumns for additional information.

These data have been used to perform other validation exercises onIWV data from OMI (Vaquero-Martínez et al., 2017), GOME-2 (Románet al., 2015), and MODIS (Bennouna et al., 2013).

J. Vaquero-Martínez et al. Int J Appl  Earth Obs Geoinformation 63 (2017) 214–221

215

82 Apendice A. Artıculos

3. Methodology

3.1. Comparison criteria

In order to match GPS data and MODIS data, some criteria wereapplied. First, the distance between the center of the satellite pixel andthe ground-based GPS station had to be the lowest. Second, time dif-ference between both measurements had to be the lowest, and alwayslower than 30 min. Otherwise, the measurement was rejected for thisstudy. Only cloud-free (CF = 0) data have been used, except for theanalysis on CF dependence, where the whole data-set have been con-sidered.

3.2. Statistical analysis

Once data from MODIS and GPS are co-located according to thecriteria mentioned above, a data-set is built, with every row containinga MODIS IWV datum, a GPS IWV datum, SZA, date, CF, and so on. Therelative differences between MODIS and GPS are calculated as:

=

−

δw w

w100i

i i

i

MODIS GPS

GPS

where the index i represents a specific location and date, w is the IWVmeasured by MODIS or GPS. Two indices are applied to the distributionof relative differences: the pseudomedian (δ ) and the interquartilerange (IQR). The pseudomedian is calculated through Wilcoxon signedrank test with continuity correction (Wilcoxon, 1945).

The pseudomedian agrees with the median in a symmetric data-set,and it is defined as the median of all the midpoints of pairs of ob-servations. This has information about the accuracy of the MODIS IWVdata. Pseudomedian values close to zero indicate that MODIS and GPSagree well, but positive values of pseudomedian would show thatMODIS is overestimating IWV. Negative values, therefore, would besignals of underestimation.

IQR is the difference between the first and the third quartile of therelative differences, which gives the width of the central half of thedata. IQR allows measuring the precision of MODIS IWV product.

A statistical analysis per station has been performed in this context,in order to detect differences between stations. In the analysis, bothindices (pseudomedian and IQR) over each station have been calcu-lated, and a linear regression model has been applied to the MODIS(dependent variable) and GPS (independent variable) IWV data, inorder to analyze their proportionality an similarity.

In order to study the influence of other parameters (SZA, IWV,season and CF), data were binned for these variables, and the indicesare calculated over those bins of data. Then, the indices are plottedagainst the variables. The bin widths have been the following: 5 mm(IWV), 5° (SZA), one month (season), and 0.1 (CF). Bins with few data(less than 30) have been ignored in this paper.

4. Results and discussion

4.1. Statistical analysis

A statistical analysis per stations was performed, whose results canbe seen in Table 2. Pseudomedian values show that some stations tendto overestimate reference GPS data, while others tend to underestimatethem. However, there is not a clear pattern. IQR, is quite homogenous,with values between 30 and 45%, which shows that MODIS variabilityis high. This result is similar to that obtained for OMI IWV product inVaquero-Martínez et al. (2017). Regarding regression parameters, theintercept, y0, varies from 0.4 mm to 6.8 mm between stations, but in allcases the intercept is positive, suggesting that low IWV values areprobably overestimated by MODIS product. Coastal stations tend tohave higher intercepts than inland stations. The slope b, on the con-trary, is lower than 1 in all cases (except for mall, which is slightlyhigher than 1). This shows that high IWV values are prone to be un-derestimated by MODIS. Pearson's coefficient, R2, indicates a fairagreement between the data, with values from 0.38 to 0.71. In Li et al.(2003), regression between MODIS daytime (NIR) and GPS IWV inGermany showed slopes greater than 1 and intercepts below 0 mm. As itwill be mentioned in Section 4.3, MODIS daytime algorithm seems tooverestimate IWV with respect to GPS products, which could be thecause for the difference between the results in the present work andthose in Li et al. (2003). However, in Ningombam et al. (2016), theresults of the regression of MODIS NIR product against GPS measure-ments of IWV at the dry (IWV typically between 1 and 16 mm trans-Himalayan region were quite similar to those in Table 2, as well as in

Fig. 1. Location of the twenty-one stations selected. Coastal stations are in red and inlandstations in blue. (For interpretation of the references to color in this figure legend, thereader is referred to the web version of the article.)

Table 1Location of GPS stations considered.

Station Acronym Latitude Longitude Altitude(°N) (°E) km

A_Coruña acor 43.36 −8.40 0.01Santander cant 43.47 −3.80 0.05Vigo vigo 42.18 −8.81 0.03Córdoba coba 37.92 −4.72 0.16León leon 42.59 −5.65 0.92Logroño rioj 42.46 −2.50 0.45Salamanca sala 40.95 −5.50 0.80Sonseca sons 39.68 −3.96 0.76Teruel teru 40.35 −1.12 0.96Valladolid vala 41.70 −4.71 0.77Villafranca vill 40.44 −3.95 0.60Alicante alac 38.34 −0.48 0.01Almería alme 36.85 −2.46 0.08Burriana borr 39.91 −0.08 0.02Ceuta ceu1 35.89 −5.31 0.05Creus creu 42.32 3.32 0.08Mallorca mall 39.55 2.63 0.06Valencia vale 39.48 −0.34 0.03Cáceres cace 39.48 −6.34 0.38Huelva huel 37.20 −6.92 0.03San_Fernando sfer 36.46 −6.21 0.04

J. Vaquero-Martínez et al. Int J Appl  Earth Obs Geoinformation 63 (2017) 214–221

216

A.4. Artıculo 4 83

Raja et al. (2008) in north America. In such work the correlation be-tween MODIS and GPS were very high, around R2 = 0.91, which couldbe due to the smaller range of measurement and the selection of goodquality data.

Fig. 2 shows the scatterplot between MODIS and GPS IWV data. Itcan be observed that both data sets agree well, better for inland stationsand Terra platform. Coastal stations tend to have more dispersion,probably due to the presence of water in the pixels of MODIS, whichmakes the retrieval more challenging. Furthermore, the better

performance of Terra could be due to its typical passing hours (in themorning).

4.2. IWV dependence

As mentioned above, in order to study IWV dependence, data weregrouped in bins of 5 mm, and pseudomedian and IQR were calculatedfor each bin. In Fig. 3 top, it can be observed that low IWV is clearlyoverestimated, but the rest of IWV are quite close to the zero line, beingslightly underestimated. Coastal stations have a slight tendency tooverestimation, except for large IWV (IWV > 25 mm), which is un-derestimated and small IWV (IWV < 10 mm), which is significantlyoverestimated. Daylight subset pseudomedians are always abovenighttime ones. Overestimating daylight (NIR) product has been ob-served in other studies, such as Albert et al., 2005. Moreover, IQR de-pendence on IWV, shown in Fig. 3 bottom, clearly decreases with in-creasing IWV. Differences between daytime and nighttimemeasurements are small (daylight shows a slightly lower IQR thannighttime). Coastal and inland stations behave very similarly, butcoastal stations’ IQR is generally higher.

Table 2MODIS statistical analysis. The pseudomedian and IQR of the δ distribution, the numberof data (N) and the coefficients of the regression analysis are shown. y0 column shows theintercept, b stands for the slope and R2 is Pearson's coefficient of determination. Thenumbers in parenthesis are the 95% confidence interval. The double line separates coastal(top) and inland (bottom) stations.

Station pMedian IQR N y0 b R2

(%) (%) (mm)

acor 1(1) 39.89 6021 3.4(0.37) 0.79(0.02) 0.49cant −9(1) 37.91 5378 3.5(0.36) 0.70(0.02) 0.50vigo 3(1) 40.82 6673 4.0(0.34) 0.78(0.02) 0.50alac 2(1) 35.84 6902 2.7(0.36) 0.89(0.02) 0.60alme −11(1) 31.52 7342 1.3(0.33) 0.84(0.02) 0.58borr −3(1) 30.87 5698 2.1(0.30) 0.86(0.01) 0.71ceu1 4(1) 42.15 5116 6.8(0.46) 0.66(0.02) 0.38creu 5(1) 41.03 6080 4.9(0.28) 0.73(0.02) 0.61mall 12(1) 35.94 6668 1.9(0.39) 1.04(0.02) 0.65vale −7(1) 32.89 6669 2.7(0.30) 0.78(0.01) 0.64huel 14(1) 40.79 6572 4.2(0.39) 0.89(0.02) 0.52sfer 18(1) 45.10 4496 5.1(0.50) 0.87(0.03) 0.46

coba −9(1) 30.39 6876 0.9(0.31) 0.87(0.02) 0.62leon −2(1) 41.09 6042 1.0(0.22) 0.90(0.02) 0.62rioj −15(1) 32.79 5496 0.4(0.25) 0.83(0.02) 0.68sala 3(1) 41.04 6522 1.1(0.24) 0.95(0.02) 0.63sons 17(1) 43.03 6288 1.1(0.26) 1.09(0.02) 0.68teru −5(1) 38.52 4787 0.8(0.26) 0.90(0.02) 0.66vala −4(1) 36.36 5206 0.4(0.28) 0.95(0.02) 0.66vill −3(1) 34.00 6683 0.5(0.24) 0.95(0.02) 0.68cace 12(1) 39.78 6842 1.8(0.29) 1.00(0.02) 0.61

All 0.9(0.7) 39.36 128357 2.3(0.07) 0.870(0.004) 0.61

Fig. 2. Scatterplot between MODIS retrieved IWV data and GPS IWV data. Data is dividedinto coastal and inland stations, and into Terra or Aqua data.

Fig. 3. Pseudomedian (top) and IQR (bottom) of MODIS-GPS relative differences as afunction of different IWV bins. Error-bars in pseudomedian are the 95% confidence in-terval in the Wilcoxon rank test.

J. Vaquero-Martínez et al. Int J Appl  Earth Obs Geoinformation 63 (2017) 214–221

217

84 Apendice A. Artıculos

4.3. SZA dependence

Daylight values can be expected to perform differently for differentSZA values, as solar radiation diminishes with increasing SZA.However, nighttime values are not expected to have any difference inperformance, except maybe the difference due to the diurnal cycle ofIWV and the IWV dependence. Fig. 4 top shows the pseudomedianvalues of MODIS bins against the bin central SZA. It can be observedthat MODIS presents a tendency to overestimate IWV for high SZA(SZA > 50°) at daytime. A similar behavior was observed in Románet al. (2015) for GOME-2 (visible), and in Vaquero-Martínez et al.(2017) for OMI (visible). Similarly to OMI behavior, part of MODIS SZAdependence can be explained by variations in the typical IWV for thatSZA (diurnal cycle of water vapor). Coastal stations show higherpseudomedian (overestimation) for low IWV subset at daytime. Night-time values tend to be underestimated, being this underestimation moresensed in inland stations. High IWV tend to be underestimated in mostof the SZA range, only overestimated at low SZA values (under 40°),while low IWV tend to be overestimated in the whole range, except fornighttime measurements, from 125° on.

Regarding IQR, Fig. 4 shows that, again, daytime values with highSZA (50° > SZA > 90°) present high IQR values for both coastal and

inland stations, similarly to GOME-2 behavior in Román et al. (2015)and OMI behavior in Vaquero-Martínez et al. (2017). Nighttime valuesare quite stable, with some variability when using high/low IWV se-paration. Nighttime IQR in coastal stations seems to increase as SZAincreases, but this is likely related to the IWV dependence. Around100–125°, high IWV values are more numerous, and as mentionedabove, high IWV values are associated with low IQR values. However,over 125°, low IWV values dominate and thus IQR increases. The de-pendencies observed for low and high IWV subsets may be related tothe fact that after the sunset some sunlight still remains. Generallynighttime IQR is greater than daylight IQR, which is in agreement withAlbert et al., 2005; Bennouna et al., 2013

4.4. Seasonal dependence

Regarding the seasonal variation of pseudomedian and IQR indices,Fig. 5 shows the results of grouping data in bins of 1-month and cal-culating the pseudomedian and IQR for each bin. Pseudomedian valuesshow overestimation in summer and winter, and underestimation inspring and autumn. This is in agreement with the results in Prasad andSingh, 2009. The behavior of the different subsets considered (low/highIWV, daylight and nighttime) is similar, but low IWV and daytime

Fig. 4. Pseudomedian and IQR of MODIS-GPS relative differences as a function of dif-ferent IWV bins. Error-bars in pseudomedian are the 95% confidence interval in theWilcoxon rank test. IWV is considered low if below 14 mm, and high if above 14 mm.

Fig. 5. Seasonal dependence on the pseudomedian of MODIS-GPS relative differences.December has been rearranged as the first month in order to make easier to identify thedifferent seasons.

J. Vaquero-Martínez et al. Int J Appl  Earth Obs Geoinformation 63 (2017) 214–221

218

A.4. Artıculo 4 85

subsets are prone to overestimation while nighttime and high IWVsubsets are underestimated. The difference between the two algorithms(daylight and nighttime) agrees with the one observed in Bennounaet al. (2013). Although this behavior is similar to the one observed forOMI in Vaquero-Martínez et al. (2017) and for AIRS in Raja et al.(2008), in the former work OMI was shown to underestimate duringsummer, instead of overestimate reference GPS IWV. In the presentwork, all months are within± 30%, except for coastal stations’ lowIWV subset in summer, which has a exceptionally high overestimation∼60%.

The IQR is generally lower in summer than in winter. This isprobably due to IWV being typically higher in summer than in winter.Still, coastal summer low IWV is very high (more than 60% in July). It isnoticeable that high IWV is very stable. This is similar to the seasonaldependence shown in Vaquero-Martínez et al. (2017) for OMI, althoughin the case of OMI IQR values are more extreme, ranging from less than30% to more than 70%. Nighttime and daytime IQR behave similarly.

4.5. Cloudiness dependence

In order to study cloudiness dependence, cloudy data rejected be-fore are now included in the data set. In Fig. 6, data is grouped in bins

of 0.1 CF width and the pseudomedian and IQR are represented againstCF bins. Pseudomedian is quite stable in the whole CF range. There is aslight tendency to increase overestimation (daylight and low IWV datasets) and underestimation (nighttime and high IWV data sets) as CFincreases. Therefore, the treatment of cloud scenes in the MODIS re-trieval seems to be adequate. Pseudomedian is between± 20% ap-proximately, less than other parameters (IWV or SZA). Cloudy scenes,however, were reported to worsen IWV in Prasad and Singh (2009),where MODIS nighttime measurements are noted to be sensitive only towater vapor above the clouds. Satellite retrievals that do not apply aspecific strategy when dealing with cloudy scenes show higher differ-ences (leading to underestimation) when compared against GPS IWV, asshown, for GOME-2 retrieval, in Román et al. (2015).

IQR, however, clearly increases as CF increases. This is expected asclouds add noise to the measurements. Nighttime values are quitestable, due to the use of Earth + atmosphere radiation instead ofsunlight. Low IWV values have the highest IQR, probably caused by thefact that a small error in low IWV leads to higher relative errors than inhigh IWV. Daytime clearly shows a higher IQR (less precision) thannighttime subset.

5. Conclusions

In this work, a validation of MODIS water vapor Level 2 Collection 6(MOD05_L2 and MYD05_L2) product from the period 2007–2012 in theIberian Peninsula has been made. MODIS agrees well with GPS ground-based station measurements. However, some dependences have beenobserved.

IWV dependence is especially important at very low IWV values,where the agreement between MODIS and GPS is not good. MODISstrongly overestimates (pseudomedian around 40%) IWV under 5 mm,with a high variability (IQR around 60%). However, overestimationand variability quickly decrease as IWV increases.

Performance of MODIS water vapor product also varies with SZA.Measurements generally worsen between 50° and 90°, overestimatinglow IWV and underestimating high IWV, and increasing IQR. Nighttimemeasurements (SZA > 90°) are quite stable, with a slight tendency tounderestimation.

The previous dependences are the cause for a seasonal dependence.Seasonal pseudomedian analysis showed that summer (more daytimehours) and winter (lower IWV) tend to be overestimated, while springand autumn underestimated. Coastal low IWV subset is overestimated,especially in summer (∼60%). IQR is lower in summer and increases inwinter. Again, coastal low IWV subset shows high IQR, which is par-ticularly notable in summer (∼60%). However, high IWV subset showsvery little seasonal dependence.

Finally, all-sky data were considered to study CF dependence. CFhas a small influence in the pseudomedian of the relative differences,since positive and negative errors are compensated. Increasing CFworsens subsets (low IWV and daylight by overestimation and high IWVand nighttime by underestimation). Regarding IQR, it increases as CFincreases, for all subsets, except nighttime measurements, which arequite stable.

The results in this paper show that the quality of MODIS watervapor product in the Iberian Peninsula is similar to that of other areas.Therefore, this study assures the performance of MODIS in the IberianPeninsula in terms of the dependence of such performance on severalvariables. It is expected that this study enables improvements of MODISNIR and IR algorithms.

Acknowledgments

This work was supported by the Spanish Ministry of Economy andCompetitiveness through project CGL2014-56255-C2. Manuel Antónthanks Ministerio de Ciencia e Innovación and Fondo Social (RYC-2011-08345) Europeo for the award of a postdoctoral grant (Ramón y Cajal).

Fig. 6. Pseudomedian and IQR of MODIS-GPS relative differences as a function of dif-ferent CF bins. Errorbars in pseudomedian are the 95% confidence interval in theWilcoxon rank test.

J. Vaquero-Martínez et al. Int J Appl  Earth Obs Geoinformation 63 (2017) 214–221

219

86 Apendice A. Artıculos

Support from the Junta de Extremadura (Research Group GrantsGR15137) is gratefully acknowledged. Work at Universidad deValladolid is supported by project CMT2015-66742-R. Work atUniversidad de Granada was supported by the Andalusia RegionalGovernment (project P12-RNM-2409) and the Spanish Ministry ofEconomy and Competitiveness and FEDER funds under the projectsCGL2016-81092-R and “Juan de la Cierva-Formación” program (FJCI-2014-22052). Work at Universidade de Évora is co-funded by theEuropean Union through the European Regional Development Fund,included in the COMPETE 2020 (Operational Program Competitivenessand Internationalization) through the ICT project (UID / GEO / 04683/2013) with the reference POCI-01-0145-FEDER-007690. The MODISdatasets were acquired from the Level-1 and AtmosphereArchive & Distribution System (LAADS) Distributed Active ArchiveCenter (DAAC), located in the Goddard Space Flight Center inGreenbelt, Maryland (https://ladsweb.nascom.nasa.gov/).

References

Albert, P., Bennartz, R., Preusker, R., Leinweber, R., Fischer, J., 2005. Remote sensing ofatmospheric water vapor using the moderate resolution imaging spectroradiometer.Am. Meteorol. Soc. 22, 309–314.

Barnet, C., Manning, E., Rosenkranz, P.W., Strow, L., Susskind, J., 2007. Airs-TeamRetrieval for Core Products and Geophysical Parameters. Technical Report March JetPropulsion Laboratory.

Barreto, A., Cuevas, E., Damiri, B., Romero, P.M., Almansa, F., 2013. Column water vapordetermination in night period with a lunar photometer prototype. Atmos. Meas. Tech.6, 2159–2167. http://dx.doi.org/10.5194/amt-6-2159-2013. http://www.atmos-meas-tech.net/6/2159/2013/.

Bennouna, Y.S., Torres, B., Cachorro, V.E., Ortiz de Galisteo, J.P., Toledano, C., 2013. Theevaluation of the integrated water vapour annual cycle over the Iberian Peninsulafrom EOS-MODIS against different ground-based techniques. Q. J. R. Meteorol. Soc.139, 1935–1956. http://dx.doi.org/10.1002/qj.2080.

Bevis, M., Businger, S., Herring, T.A., Rocken, C., Anthes, R.A., Ware, R.H., 1992. GPSmeteorology: remote sensing of atmospheric water vapor using the global positioningsystem. J. Geophys. Res. 97, 15787–15801. http://dx.doi.org/10.1029/92JD01517.

Bokoye, A.I., Royer, A., O’Niell, N., Cliche, P., McArthur, L., Teillet, P., Fedosejevs, G.,Thériault, J.-M., 2003. Multisensor analysis of integrated atmospheric water vaporover Canada and Alaska. J. Geophys. Res. 108, 21-1–21-16. http://dx.doi.org/10.1029/2002JD002721.

Chang, L., Gao, G., Jin, S., He, X., Xiao, R., Guo, L., 2015. Calibration and evaluation ofprecipitable water vapor from MODIS infrared observations at night. IEEE Trans.Geosci. Rem. Sens. 53, 2612–2620. http://dx.doi.org/10.1109/TGRS.2014.2363089.http://ieeexplore.ieee.org/document/6945895/.

Colman, R., 2003. A comparison of climate feedbacks in general circulation models. Clim.Dyn. 20, 865–873. http://dx.doi.org/10.1007/s00382-003-0310-z.

Gao, B.-C., Kaufman, Y.J., 1992. The MODIS Near-IR Water Vapor Algorithm Product ID:MOD05 – Total Precipitable Water. Algorithm Technical Background Document. pp.1–25.

Gao, B.-C., Li, R.-R., 2008. The time series of terra and aqua MODIS near-IR water vaporproducts. IGARSS 2008 – 2008 IEEE International Geoscience and Remote SensingSymposium, IEEE, vol. 3. pp. 186–189. http://dx.doi.org/10.1109/IGARSS.2008.4779314. http://ieeexplore.ieee.org/document/4779314/.

Grossi, M., Valks, P., Loyola, D., Aberle, B., Slijkhuis, S., Wagner, T., Beirle, S., Lang, R.,2015. Total column water vapour measurements from GOME-2 MetOp-A and MetOp-B. Atmos. Meas. Tech. 8, 1111–1133. http://dx.doi.org/10.5194/amt-8-1111-2015.http://www.atmos-meas-tech.net/8/1111/2015/.

de Haan, S., van der Marel, H., Barlag, S., 2002. Comparison of GPS slant delay mea-surements to a numerical model: case study of a cold front passage. Phys. Chem.Earth A/B/C 27, 317–322. http://dx.doi.org/10.1016/S1474-7065(02)00006-2.http://linkinghub.elsevier.com/retrieve/pii/S1474706502000062.

Heise, S., Dick, G., Gendt, G., Schmidt, T., Wickert, J., 2009. Integrated water vapor fromIGS ground-based GPS observations: initial results from a global 5-min data set. Ann.Geophys. 27, 2851–2859. http://dx.doi.org/10.5194/angeo-27-2851-2009. http://www.ann-geophys.net/27/2851/2009/.

Ichoku, C., Levy, R., Kaufman, Y.J., Remer, L.A., Li, R.-R., Martins, V.J., Holben, B.N.,Abuhassan, N., Slutsker, I., Eck, T.F., Pietras, C., 2002. Analysis of the performancecharacteristics of the five-channel Microtops II Sun photometer for measuring aerosoloptical thickness and precipitable water vapor. J. Geophys. Res. 107, 5-1–5-17.http://dx.doi.org/10.1029/2001JD001302.

King, M., Kaufman, Y., Menzel, W., Tanre, D., 1992. Remote sensing of cloud, aerosol, andwater vapor properties from the moderate resolution imaging spectrometer (MODIS).IEEE Trans. Geosci. Rem. Sens. 30, 2–27. http://dx.doi.org/10.1109/36.124212.http://ieeexplore.ieee.org/document/124212/.

Li, Z., Muller, J.-P., Cross, P., 2003. Comparison of precipitable water vapor derived fromradiosonde, GPS, and moderate-resolution imaging spectroradiometer measurements.J. Geophys. Res. 108, 4651. http://dx.doi.org/10.1029/2003JD003372.

Livesey, N.J., Van Snyder, W., 2004. EOS MLS Retrieval Processes Algorithm TheoreticalBasis. Technical Report part 3 Jet Propulsion Laboratory.

Martinez, M.A., 2013. Algorithm Theoretical Basis Document for ”SEVIRI Physical

Retrieval Product” (SPhR-PGE13 v2.0). Technical Report Agencia Estatal deMeteorología (AEMET). . http://www.nwcsaf.org/scidocs/Documentation/SAF-NWC-CDOP2-INM-SCI-ATBD-13_v2.0.pdf.

Myhre, G., Shindell, D., Bréon, F.-M., Collins, W., Fuglestvedt, J., Huang, J., Koch, D.,Lamarque, J.-F., Lee, D., Mendoza, B., Nakajima, T., Robock, A., Stephens, G.,Takemura, T., Zhang, H., 2013. Anthropogenic and natural radiative forcing. ClimateChange 2013: The Physical Science Basis. Contribution of Working Group I to theFifth Assessment Report of the Intergovernmental Panel on Climate Change. pp.659–740.

Niell, A.E., 2000. Improved atmospheric mapping functions for VLBI and GPS. EarthPlanet. Space 52, 699–702. http://dx.doi.org/10.1186/BF03352267.

Ningombam, S.S., Jade, S., Shrungeshwara, T., Song, H.-J., 2016. Validation of watervapor retrieval from Moderate Resolution Imaging Spectro-radiometer (MODIS) innear infrared channels using GPS data over IAO-Hanle, in the trans-Himalayan re-gion. J. Atmos. Solar-Terr. Phys. 137, 76–85. http://dx.doi.org/10.1016/j.jastp.2015.11.019. http://linkinghub.elsevier.com/retrieve/pii/S1364682615301000.

Noël, S., Buchwitz, M., Burrows, J.P., 2004. First retrieval of global water vapour columnamounts from SCIAMACHY measurements. Atmos. Chem. Phys. 4, 111–125. http://dx.doi.org/10.5194/acp-4-111-2004. http://www.atmos-chem-phys.net/4/111/2004/.

Ohtani, R., Naito, I., 2000. Comparisons of GPS-derived precipitable water vapors withradiosonde observations in Japan. J. Geophys. Res.: Atmos. 105, 26917–26929.http://dx.doi.org/10.1029/2000JD900362.

Ortiz de Galisteo, J.P., Bennouna, Y., Toledano, C., Cachorro, V., Romero, P., Andrés,M.I., Torres, B., 2014. Analysis of the annual cycle of the precipitable water vapourover Spain from 10-year homogenized series of GPS data. Q. J. R. Meteorol. Soc. 140,397–406. http://dx.doi.org/10.1002/qj.2146.

Ortiz de Galisteo, J.P., Cachorro, V., Toledano, C., Torres, B., Laulainen, N., Bennouna, Y.,de Frutos, A., 2011. Diurnal cycle of precipitable water vapor over Spain. Q. J. R.Meteorol. Soc. 137, 948–958. http://dx.doi.org/10.1002/qj.811.

Pany, T., Pesec, P., Stangl, G., 2001. Atmospheric GPS slant path delays and ray tracingthrough numerical weather models, a comparison. Phys. Chem. Earth A: Solid EarthGeod. 26, 183–188. http://dx.doi.org/10.1016/S1464-1895(01)00044-8. http://linkinghub.elsevier.com/retrieve/pii/S1464189501000448.

Pérez-Ramírez, D., Navas-Guzmán, F., Lyamani, H., Fernández-Gálvez, J., Olmo, F.J.,Alados-Arboledas, L., 2012. Retrievals of precipitable water vapor using star photo-metry: assessment with Raman lidar and link to sun photometry. J. Geophys. Res.:Atmos. 117. http://dx.doi.org/10.1029/2011JD016450.

Prasad, A.K., Singh, R.P., 2009. Validation of MODIS Terra, AIRS, NCEP/DOE AMIP-IIReanalysis-2, and AERONET Sun photometer derived integrated precipitable watervapor using ground-based GPS receivers over India. J. Geophys. Res. 114, D05107.http://dx.doi.org/10.1029/2008JD011230.

Raja, M.K.R.V., Gutman, S.I., Yoe, J.G., McMillin, L.M., Zhao, J., 2008. The validation ofAIRS retrievals of integrated precipitable water vapor using measurements from aNetwork of ground-based GPS receivers over the contiguous United States. J. Atmos.Ocean. Technol. 25, 416–428. http://dx.doi.org/10.1175/2007JTECHA889.1.

Román, R., Antón, M., Cachorro, V., Loyola, D., Ortiz de Galisteo, J., de Frutos, A.,Romero-Campos, P., 2015. Comparison of total water vapor column from GOME-2 onMetOp-A against ground-based GPS measurements at the Iberian Peninsula. Sci. TotalEnviron. 533, 317–328. http://dx.doi.org/10.1016/j.scitotenv.2015.06.124.

Román, R., Bilbao, J., de Miguel, A., 2014. Uncertainty and variability in satellite-basedwater vapor column, aerosol optical depth and Angström exponent, and its effect onradiative transfer simulations in the Iberian Peninsula. Atmos. Environ. 89, 556–569.http://dx.doi.org/10.1016/j.atmosenv.2014.02.027.

Salomonson, V., Barnes, W., Maymon, P., Montgomery, H., Ostrow, H., 1989. MODIS:advanced facility instrument for studies of the Earth as a system. IEEE Trans. Geosci.Rem. Sens. 27, 145–153. http://dx.doi.org/10.1109/36.20292. http://ieeexplore.ieee.org/document/20292/.

Schneider, M., Romero, P.M., Hase, F., Blumenstock, T., Cuevas, E., Ramos, R., 2010.Continuous quality assessment of atmospheric water vapour measurement techni-ques: FTIR, Cimel, MFRSR, GPS, and Vaisala RS92. Atmos. Meas. Tech. 3, 323–338.http://dx.doi.org/10.5194/amt-3-323-2010. http://www.atmos-meas-tech.net/3/323/2010/.

Schroedter-Homscheidt, M., Drews, A., Heise, S., 2008. Total water vapor column re-trieval from MSG-SEVIRI split window measurements exploiting the daily cycle ofland surface temperatures. Rem. Sens. Environ. 112, 249–258. http://dx.doi.org/10.1016/j.rse.2007.05.006. http://linkinghub.elsevier.com/retrieve/pii/S0034425707001952.

Seemann, S.W., Borbas, E.E., Li, J., Menzel, W.P., Gumley, L.E., 2006. Modis AtmosphericProfile Retrieval: Algorithm Theoretical Basis Document. Technical Report OctoberCooperative Institute for Meteorological Satellite Studies. University of Wisconsin-Madison.

Torres, B., Cachorro, V.E., Toledano, C., Ortiz de Galisteo, J.P., Berjón, A., de Frutos,A.M., Bennouna, Y., Laulainen, N., 2010. Precipitable water vapor characterization inthe Gulf of Cadiz region (southwestern Spain) based on Sun photometer, GPS, andradiosonde data. J. Geophys. Res. 115, 1–11. http://dx.doi.org/10.1029/2009JD012724.

Turner, D.D., Clough, S.A., Liljegren, J.C., Clothiaux, E.E., Cady-Pereira, K.E., Gaustad,K.L., 2007. Retrieving liquid water path and precipitable water vapor from theAtmospheric Radiation Measurement (ARM) microwave radiometers. IEEE Trans.Geosci. Rem. Sens. 45, 3680–3690. http://dx.doi.org/10.1109/TGRS.2007.903703.http://ieeexplore.ieee.org/document/4373386/.

Turner, D.D., Ferrare, R.A., Brasseur, L.A.H., Feltz, W.F., Tooman, T.P., 2002. Automatedretrievals of water vapor and aerosol profiles from an operational Raman lidar. J.Atmos. Ocean. Technol. 19, 37–50.. http://dx.doi.org/10.1175/1520-0426(2002)019<0037:AROWVA>2.0.CO;2.

J. Vaquero-Martínez et al. Int J Appl  Earth Obs Geoinformation 63 (2017) 214–221

220

A.4. Artıculo 4 87

Vaquero-Martínez, J., Antón, M., Ortiz de Galisteo, J.P., Cachorro, V.E., Wang, H.,González Abad, G., Román, R., Costa, M.J., 2017. Validation of integrated watervapor from OMI satellite instrument against reference GPS data at the IberianPeninsula. Sci. Total Environ. 580, 857–864. http://dx.doi.org/10.1016/j.scitotenv.2016.12.032. http://linkinghub.elsevier.com/retrieve/pii/S0048969716327176.

Wang, H., Liu, X., Chance, K., González Abad, G., Chan Miller, C., 2014. Water vaporretrieval from OMI visible spectra. Atmos. Meas. Tech. 7, 1901–1913. http://dx.doi.org/10.5194/amt-7-1901-2014. http://www.atmos-meas-tech.net/7/1901/2014/.

Wang, J., Zhang, L., Dai, A., Van Hove, T., Van Baelen, J., 2007. A near-global, 2-hourly

data set of atmospheric precipitable water from ground-based GPS measurements. J.Geophys. Res. 112, D11107. http://dx.doi.org/10.1029/2006JD007529.

Wentz, F.J., Meissner, T., 2007. Supplement 1 Algorithm Theoretical Basis Document forAMSR-E Ocean Algorithms. Technical Report.

Wentz, F.J., Spencer, R.W., 1998. SSM/I Rain retrievals within a unified all-weatherocean algorithm. J. Atmos. Sci. 55, 1613–1627. http://dx.doi.org/10.1175/1520-0469(1998)055<1613:SIRRWA>2.0.CO;2.

Wilcoxon, F., 1945. Individual comparisons by ranking methods. Biom. Bull. 1, 80.http://dx.doi.org/10.2307/3001968.

J. Vaquero-Martínez et al. Int J Appl  Earth Obs Geoinformation 63 (2017) 214–221

221

88 Apendice A. Artıculos

A.5. Artıculo 5 89

A.5. Artıculo 5

Tıtulo: Water Vapor Radiative Effects on Short-Wave Radiationin Spain.Autores

• Vaquero-Martınez, Javier (U. de Extremadura).• Anton, Manuel (U. de Extremadura).• Ortiz de Galisteo, Jose P. (AEMet).• Roman, Roberto (U. de Valladolid).• Cachorro, Victoria E. (U. de Valladolid).

Ano: 2018.Revista: Atmospheric Research.Paginas: 18–25.DOI: 10.1016/j.atmosres.2018.02.001.Volumen: 205.

Contents lists available at ScienceDirect

Atmospheric Research

journal homepage: www.elsevier.com/locate/atmosres

Water vapor radiative effects on short-wave radiation in Spain

Javier Vaquero-Martíneza,b,*, Manuel Antóna,b, José Pablo Ortiz de Galisteoc,d, Roberto Románe,f,Victoria E. Cachorrod

a Departamento de Física, Universidad de Extremadura, Badajoz, Spainb Instituto Universitario de Investigación del Agua, Cambio Climático y Sostenibilidad (IACYS), Universidad de Extremadura, Badajoz, Spainc Agencia Estatal de Meteorologia (AEMET), Valladolid, Spaind Grupo de Óptica Atmosférica, Universidad de Valladolid, Valladolid, Spaine Department of Applied Physics, University of Granada, Granada, Spainf Andalusian Institute for Earth System Research (IISTA-CEAMA), Granada, Spain

A R T I C L E I N F O

Keywords:Short-waveRadiative effectRadiative efficiencyWater vaporIWVIberian Peninsula

A B S T R A C T

In this work, water vapor radiative effect (WVRE) is studied by means of the Santa Barbara's Disort RadiativeTransfer (SBDART) model, fed with integrated water vapor (IWV) data from 20 ground-based GPS stations inSpain. Only IWV data recorded during cloud-free days (selected using daily insolation data) were used in thisstudy. Typically, for SZA=60.0± 0.5° WVRE values are around − 82 and − 66 Wm−2 (first and thirdquartile), although it can reach up − 100 Wm−2 or decrease to − 39 Wm−2. A power dependence of WVRE onIWV and cosine of solar zenith angle (SZA) was found by an empirical fit. This relation is used to determine thewater vapor radiative efficiency (WVEFF= ∂WVRE/∂IWV). Obtained WVEFF values range from − 9 and 0Wm−2 mm−1 (− 2.2 and 0%mm−1 in relative terms). It is observed that WVEFF decreases as IWV increases, butalso as SZA increases. On the other hand, when relative WVEFF is calculated from normalized WVRE, an increaseof SZA results in an increase of relative WVEFF. Heating rates were also calculated, ranging from 0.2 Kday−1 to1.7 Kday−1. WVRE was also calculated at top of atmosphere, where values ranged from 4 Wm−2 to 37 Wm−2.

1. Introduction

The climate system is interactive, and all its elements (atmosphere,Earth's surface and biosphere) are interconnected (Denman andBrasseur, 2007). Water, presented in its three states in the Earth-at-mosphere system, is one of the elements of paramount importance.Water vapor is acknowledged as the most important atmosphericgreenhouse gas, and although it is not directly involved in climatechange since its concentration is regulated by temperature more thananthropogenic emissions, it causes a positive radiative feedback onclimate system (Colman, 2003).

Currently, the radiative effect of water vapor is considered a feed-back rather than a forcing, since the water vapor concentration ismainly dependent on the temperature on a global scale, and the typicalresidence time of water vapor is ten days (Myhre et al., 2013). For thesereasons, anthropogenic emissions of water vapor have a negligibleimpact on global climate. The main anthropogenic impact in watervapor content is due to the emission of other greenhouse gases, whichcause temperature increase and therefore an increase in water vaporcontent (Santer et al., 2007). Emissions in the stratosphere, however,

can be considered as a forcing (Smith et al., 2001; Forster and Shine,2002; Zhong and Haigh, 2003; Solomon et al., 2010), because in thestratosphere water vapor emissions (i.e., caused by stratosphericflights) manage to stay in the long term.

Water vapor in the atmosphere can be quantified using the columnintegrated amount of water vapor (IWV), which is equivalent to con-densing all the water vapor in the atmospheric column and measuringthe height that it would reach in a vessel of unit cross section. It can bemeasured in columnar mass density (g cm−2 or kgm−2) or in length(height) units (mm) (Román et al., 2015). The instantaneous watervapor radiative effect (WVRE) at surface is defined as the net change inshort-wave (SW) solar radiation at surface taking as reference a dryatmosphere (adapted from Mateos et al., 2013a). It can be also calcu-lated at top of atmosphere (TOA) (WVRETOP). Therefore, water vaporefficiency (WVEFF) can be defined as the variation on WVRE that iscaused by an increase of 1 unit of atmospheric water vapor, that is tosay, the first derivative of WVRE with respect to IWV.

In this work, the WVEFF focused on the SW range is analyzed usinga radiative transfer code fed with IWV data recorded from several GPSground-based stations in the Iberian Peninsula. Although other works

https://doi.org/10.1016/j.atmosres.2018.02.001Received 25 September 2017; Received in revised form 24 January 2018; Accepted 3 February 2018

* Corresponding author at: Departamento de Física, Universidad de Extremadura, Badajoz, Spain.E-mail address: javier_vm@unex.es (J. Vaquero-Martínez).

Atmospheric Research 205 (2018) 18–25

Available online 08 February 20180169-8095/ © 2018 Published by Elsevier B.V.

T

A.5. Artıculo 5 91

have studied the change in surface radiation due to water vapor (Sodenet al., 2002; Di Biagio et al., 2012; Román et al., 2014), none quantifiesnor analyzes the WVEFF or WVRE and its dependences on IWV andSZA, as it has already been done for clouds (Mateos et al., 2013b,2014b), aerosols (Mateos et al., 2013a, 2014a) and ozone (Antón andMateos, 2013; Antón et al., 2016). This paper aims to be useful for abetter understanding of the individual contributions of water vapor tothe radiation budget in the Iberian Peninsula, and evaluate the WVEFFunder different conditions of SZA and IWV in this context. Knowledgeabout surface energy balance sensitivity to variations of IWV is im-portant to assess the system's response to future climate changes.

2. Integrated water vapor data

IWV data used in this work were recorded from 20 GPS Spanishstations located mostly in the Iberian Peninsula (see Fig. 1 and Table 1).For a full description of the method to derive IWV data from GPS, referto Bevis et al. (1992). In the process of positioning a GPS ground-basedstation, the fundamental idea is to determine the distance to severalGPS satellites in order to triangulate the receiver position. The distanceis obtained by measuring the time that the microwave signals take fromGPS satellites to GPS receivers. The signals, however, suffer some delaysalong their way. One of those delays is called the slant troposphericdelay (STD), which is caused by the tropospheric gases. STD is due totwo contributions, one related to water molecule's dipolar momentum,slant wet delay (SWD), and a non-dipolar contribution, due to all gases(including water vapor), which is known as slant hydrostatic delay(SHD)

= +STD SWD SHD (1)

Such delay can be converted to zenith tropospheric delay (ZTD) byapplying mapping functions. Mapping functions are different for SHDand SWD, but they are similar, so an approximation can be made

= + =m E m E m ESTD ( )ZWD ( )ZHD ( )ZTDwet hydrostatic (2)

= +ZTD ZHD ZWD (3)

If pressure at surface is known, ZHD can be modeled, and ZWDobtained from subtracting ZTD minus ZHD. ZWD is proportional to IWV

= κZWD IWV (4)

The constant κ can be determined from the mean temperature of theatmosphere weighted by the water vapor content. This mean

Fig. 1. Location of the twenty stations selected. Coastal stations are written in red and inland stations in blue. (For interpretation of the references to color in this figure legend, the readeris referred to the web version of this article.)

Table 1Location of GPS stations considered.

Station Acronym Latitude Longitude Altitude

(°N) (°E) (m)

A Coruña acor 43.36 −8.40 12Alicante alac 38.34 −0.48 10Almería alme 36.85 −2.46 77Burriana borr 39.91 −0.08 22Cáceres cace 39.48 −6.34 384Ceuta ceu1 35.89 −5.31 53Córdoba coba 37.92 −4.72 162Huelva huel 37.20 −6.92 29León leon 42.59 −5.65 915Logroño rioj 42.46 −2.50 452Mallorca mall 39.55 +2.63 62Salamanca sala 40.95 −5.50 800San Fernando sfer 36.46 −6.21 4Santander cant 43.47 −3.80 48Sonseca sons 39.68 −3.96 755Teruel teru 40.35 −1.12 956Valencia vale 39.48 −0.34 28Valladolid vala 41.70 −4.71 766Vigo vigo 42.18 −8.81 33Villafranca vill 40.44 −3.95 596

J. Vaquero-Martínez et al. Atmospheric Research 205 (2018) 18–25

19

92 Apendice A. Artıculos

temperature can be estimated from an empirical relationship if tem-perature at the station level is known.

The dataset used in this work covers from 2007 to 2015. SpanishGeographic Institute “Instituto Geográfico Nacional”, which is a localanalysis center for the European Reference Frame (EUREF), providedthe tropospheric products. Surface pressure and temperature, needed toretrieve IWV from ZTD products, were provided by the SpanishMeteorological State Agency (AEMet). The temperature was inter-polated to the time of measurements linearly, and pressure was inter-polated as well, taking into account the barometric tide. The IWV da-taset obtained has already been used in other works to performvalidation exercises on satellite IWV data such as Román et al. (2015),Bennouna et al. (2013), and Vaquero-Martínez et al. (2017a,b).

Daily insolation data were provided from AEMet as well. Insolationis divided by the theoretical insolation in a cloud-free situation to ob-tain an insolation factor. In order to filter out cloudy cases or cases witha significant load of aerosol days with an insolation factor below 0.75(75%) were not considered. The World Meteorological Organization(2008) recommends using a 0.70 threshold to filter out cloudy scenes,so 0.75 is a proper threshold to remove both cloudy scenes and heavyaerosol load situations.

3. Water vapor radiative effect

SW irradiances at surface were simulated by means of SantaBarbara's DISORT Radiative Transfer model (SBDART), under cloud andaerosol free conditions using a radiative transfer solver with 4 streams.Detailed information about this radiative transfer code can be found inRicchiazzi et al. (1998). This model was fed with hourly IWV data,recorded during cloud-free days. Additionally, total column ozone(daily means from ERA-Interim Reanalysis) and surface albedo(monthly means from ERA-Interim Reanalysis) were used as input inthe simulations. For more information on ERA-Interim Reanalysis, referto Dee et al. (2011). The spectral region considered ranges from 0.2 μmto 4.0 μm. The wavelength step chosen was 0.50%, as a compromisebetween computational economy and precision. This becomes stepsranging from 0.001 μm up to 0.02 μm. The atmosphere models(McClatchey et al., 1972) used were SBDART's mid-latitude summerfrom March to August (both included) and mid-latitude winter, for therest of the year. The water vapor profile and ozone profile are re-scaledto the total IWV and total column ozone that the model is fed with.Thermal radiation is not considered in these computations, since it isnegligible in the wavelength range considered. The model was runtwice for each hourly GPS measurement: once with all data mentionedabove and other with the same data except for water vapor, which is setto 0 cm. This allows to obtain the WVRE as the difference between thenet (downwards minus upwards) irradiance at surface simulated underan atmosphere with water vapor and the net irradiance assuming nowater vapor.

= − − −↓ ↑ ↓ ↑WVRE (SW SW ) (SW SW )IWV IWV noIWV noIWV (5)

At surface, this equation can be written as= − −↓ ↓αWVRE (1 )(SW SW )IWV noIWV . Because SW radiation comes from

the sun, nighttime (SZA>90°) WVRE is automatically set to zero,without running the radiative transfer model. The heating rates can beobtained using the expression from Liou (2002)

∂

∂=

Tt

gC p

ΔSWΔp (6)

where T is the temperature, t is the time, g=9.81m s−2 is the grav-itational acceleration, Cp ≃ 1004 J kg−1 K−1 is the specific heat of dryair, SW is the net flux in the range mentioned above, and p is thepressure. In this study, the water vapor heating rate is calculated for thewhole atmospheric column, which is the difference in heating ratesbetween an atmosphere with water vapor and a dry atmosphere.

Once the WVRE is obtained, it is possible to calculate the watervapor efficiency (WVEFF) as the partial derivative of WVRE with re-spect to IWV,

=∂

∂WVEFF WVRE

IWV (7)

if a functional form for WVRE depending on IWV is suggested. Thisefficiency is a relevant magnitude to analyze the sensitivity of WVREvalues to IWV changes in SW radiation, reporting about the relationshipbetween the absolute variations (in physical units) in WVRE and IWVvalues. Thus, this magnitude can be useful to quantify the impact ofIWV increases (associated with the global warming) on net solar ra-diation at surface and TOA.

4. Results and discussion

4.1. Sensitivity study

In order to account for the effects that the uncertainties of the inputvariables may have on the WVRE computation, a sensitivity analysishave been performed. For these computations, several IWV and SZAvalues were used (see Fig. 2). Both albedo and total column ozone wereconsidered, with extreme values of both. In the simulations with extremealbedo values, an intermediate value of ozone (319 DU) was used, whilefor simulations with extreme ozone values, an intermediate value of al-bedo was used (0.160). Mid-latitude winter atmosphere profile was used.

The differences between extreme values of albedo and ozone areshown in Fig. 2. A change in albedo from its minimum to its maximumvalue produces a small but noticeable change in WVRE (up to around8Wm−2). This represents less than 5% (if using the mean value asreference). However, changes in ozone values are not important, alwaysunder 0.24Wm−2 (less than 1.5%).

Regarding SZA and IWV sensitivity, Fig. 3 shows the relative dif-ference (or error) associated to an increase of 0.5° of SZA (top) or 1mmof IWV (bottom). The differences are below 3.5% for both the change inSZA and IWV. However, in most cases differences are under 1%. Theabsolute differences are under 3.5Wm−2 for SZA errors and 6Wm−2

for IWV errors.

4.2. Spatial variability

In order to study the differences between stations, some statisticshave been calculated for a SZA window of (60.0± 0.5)°, shown in abox-plot in Fig. 4. The SZA window reduces the variability due to SZA,which allows a clearer analysis of the spatial differences. It can be

Fig. 2. Sensitivity analysis of albedo and ozone. Differences between WVRE obtainedusing maximum and minimum values of albedo (0.146 and 0.187) and ozone (228 and493 DU) have been calculated for several SZA and IWV values. Circles are differences ofWVRE with different albedo and triangles are differences of WVRE with different totalcolumn ozone.

J. Vaquero-Martínez et al. Atmospheric Research 205 (2018) 18–25

20

A.5. Artıculo 5 93

observed that all stations present similar values, although coastal sta-tions generally have higher IWV (because of the proximity to watermasses) and stronger WVRE (because of the higher IWV). The first andthird quartiles of WVRE are around − 86.3 and − 71.0Wm−2, al-though some of them can reach up to − 100.0Wm−2 or decrease to −38.7Wm−2. SD is around 10Wm−2, while the coefficient of variation

(CV) is around 7%. The distributions are quite symmetric, since themedian and the mean are quite similar for every station. Mateos et al.(2013b) obtained SW radiative forcing in Granada for clouds andaerosols (SZA of 60°), reporting − 50Wm−2 and − 19Wm−2, re-spectively, and − 69Wm−2 for the combined effect of both clouds andaerosols. In the mentioned work, experimental data was used, and anempirical model was used to estimate cloud free radiation. The modelwas dependent on SZA and aerosol optical depth steps. This resultshows that water vapor could have a greater radiative effect than cloudsand aerosols in the Iberian Peninsula. On the contrary, the role of watervapor is claimed to be minor in the mentioned study. This could berelated to the fact that maximum IWV considered was 25mm, while inthe present study around 20% of the data are beyond that limit.Moreover, the reference was 5mm, instead of a totally dry atmosphere.Around a 3% of the IWV data in the present study are below this value.Additionally, it is important to notice that while there are situationswhere there are no clouds, there are no situations with no water vaporat all, so the radiative effects of both clouds and water vapor are dif-ficult to compare in a real situation.

Di Biagio et al. (2012) obtained values for WVRE in the arctic regionbetween − 100 and − 20Wm−2. This is somewhat below the valuesobtained in the present study, probably due to the fact that in the arcticregion, IWV is smaller (1–16mm), and SZA values are greater as well.

Because the results show that WVRE distribution does not have asignificant spatial dependence, in the following subsections all stationswill be averaged together.

4.3. Water vapor effect on heating rates

Water vapor effects on heating rates, which are the difference be-tween the heating rates (see Eq. (6)) with and without water vapor,show a strong dependence on the hour of the day and the season.Generally, they range between from 0.2 K day−1 to 1.7 K day−1. Theseasonal and hourly dependence can be observed in Fig. 5. DJF valuesare always under 1.0 K day−1, while JJA can reach 1.5 K day−1 in thecentral hours of the day. MAM and SON exhibit intermediate values inthese hours. It must be noticed that the minimum values are in the fourseasons quite similar, around 0.3 K day−1. This is a quite strong value ifcompared with aerosols, as shown in Valenzuela et al. (2012), whereaerosol heating rates are reported to be always below 0.3 K day−1.

Fig. 3. Sensitivity analysis of SZA and IWV. Relative differences between WVRE obtainedusing the value in the legend and the value plus 0.5° of SZA (top) or 1mm (bottom) havebeen calculated for several SZA and IWV values.

Fig. 4. Boxplot of the WVRE in the ground-based stations — SZA= (60.0± 0.5)°.

J. Vaquero-Martínez et al. Atmospheric Research 205 (2018) 18–25

21

94 Apendice A. Artıculos

4.4. Empirical model for WVRE

The results of WVRE for every hour show a high degree of corre-lation with IWV and SZA (more concisely, with μ=cosSZA), as shownin Fig. 6. On the one hand, the amount of water vapor will obviouslyhave an impact on SW radiation at surface: the higher the IWV, thestronger the abortion effects. On the other hand, SZA has a double ef-fect. First, optical mass increases with SZA, increasing radiative effect.Additionally, the larger the SZA is, the smaller the intensity of incomingradiation on a horizontal surface, due to the stronger absorption by

other atmospheric gases and the geometric effect caused by non-verti-cality.

Because of the linear behavior of the log-log plots in Fig. 6, the bestfit appears to be the one shown in Eq. (8)

= − ⋅a μWVRE IWVb c (8)

which can be linearized for a multi-linear regression in the form of Eq.(9)

= + +a b c μlog |WVRE| log( ) log IWV log( ) (9)

In these equations, WVRE is in Wm−2 and IWV in mm. Similarempirical models have been proposed for other atmospheric gases (i.e.ozone, see Madronich, 2007). The result of this multi-linear model givesa Pearson's Coefficient of R2= 0.997. The coefficients are log(a)= 4.144±0.001, b=0.2661±0.0003 and c=0.7679±0.0003.

This model represents an empirical formula for WVRE depending onμ, the cosine of SZA, and IWV. Black, dashed lines in Fig. 6 representthis fit for SZA=30.05° and SZA=50.05° (left) and forIWV=25.05mm and IWV=40.05mm (right), with very good agree-ment.

In this model, the physical meaning of the slope b is the ratio be-tween relative changes in WVRE and relative changes of IWV. For smallchanges of IWV, we can derive b from Eq. (8) as

=b ΔWVRE/WVREΔIWV/IWV

.(10)

This means that a change of 1% in IWV would cause a change of b%in WVRE, that is to say, ∼0.27%. The interpretation of b is similar tothat of the Radiation Amplification Factor (RAF) used as a measure ofsensitivity of ultraviolet solar radiation to changes in total ozonecolumn (McKenzie et al., 1991).

Normalized WVRE ⋅ −↓ ↑(100% WVRE/(SW SW ))noIWV noIWV is between− 6.9 and − 28.1%. The same approach can be followed with thisvariable. The multi-linear model gives good results (R2= 0.9891),with log(aN)= 1.7374±0.0004, bN=0.2826± 0.0001 and cN=−0.3252± 0.0002.

4.5. Water vapor efficiency calculation

The empirical model obtained in the previous section can be used toobtain water vapor radiative efficiency (WVEFF) as the derivative ofWVRE with respect to IWV (see Eq. (7)).

Fig. 5. Boxplot of heating rates according to seasonality and hour of the day.

Fig. 6. WVRE against IWV (up) and μ (down) in a log-log plot. The dashed, black linesshow the fit. The legends show the intervals of SZA and IWV values considered.

J. Vaquero-Martínez et al. Atmospheric Research 205 (2018) 18–25

22

A.5. Artıculo 5 95

= − ⋅ ⋅ ⋅ =−a μ b bWVEFF IWV WVREIWV

c b 1(11)

WVEFF has been calculated for some different SZA bins. The resultis shown in Fig. 7, where WVEFF is plotted against SZA (a) and IWV (b).It can be noticed that in all cases WVEFF decreases as SZA increases,and for a certain value of SZA, WVEFF decreases as IWV increases.Fig. 7 (b) shows that WVEFF decreases as IWV increases, very sharply atsmall IWV, while saturating for greater IWV. For a fixed value of IWV,WVEFF decreases as SZA increases. A similar dependence of cloudsradiative efficiency on SZA was observed in Mateos et al. (2014b), al-though the functional form was different. WVEFF dependence on SZAcan be explained in the following way: vertical solar irradiance de-creases with SZA, decreasing the amount of radiation available forwater vapor to absorb, and therefore decreasing its efficiency. Never-theless, there could be a second order effect, as water vapor opticalmass increases with SZA, increasing the extinguishing power of watervapor, but this is not noticeable in these results. However, using thesame approach for normalized WVRE as for WVRE, as shown in Fig. 7(c) and (d), we can eliminate the first effect and the second is revealed.In Fig. 7 (c), it can be seen that the dependence on SZA is weaker asIWV increases, due to the saturating effect of high IWV. As an increaseof SZA causes an increase in the water vapor optical mass, if IWV isalready large the saturation causes small values of WVEFF.

Fig. 8 shows a time series of WVEFF and IWV. The WVEFF valuesare typically around − 8 and 0Wm−2 mm, and normalized WVEFFbetween − 2 and 0%/mm. It can be observed that for small IWV,WVEFF is stronger. Therefore, the annual and diurnal cycle of WVEFF isrelated to IWV and modulated by SZA.

4.6. Effects on top of atmosphere

The WVRE and WVEFF has also been calculated at TOA (WVRETOPand WVEFFTOP). Fig. 4 shows the boxplots of WVRETOP for every sta-tion. The variability is quite small, varying from 11.3 to 21.3Wm−2. Ifthe relative WVRETOP is calculated, the values are between 1.7 and5.7%. These small values are expected, since the downwards flux is thesame with and without the water vapor, and the upward fluxes are

small in both cases. The small normalized values are explained takinginto account that the neat flux without water vapor is quite similar attop of the atmosphere and at surface (maximum variation are around6%). So the denominator is similar in both cases, while the numerator(WVRE) is smaller at TOA than at surface. The influence of albedo inWVRETOP is more important than at surface. The reason is that thedownwards fluxes are the same with and without water vapor, so theycancel out, so the upwards fluxes (which depend on albedo) are themain contribution to WVRETOP. The values are always positive.Therefore, a empirical expression for WVRETOP as a function of SZA,IWV and albedo can be found:

= ⋅a μ αWVRE IWVb c dTOP TOP TOP TOP TOP (12)

Correlation is R2= 0.9931, and the coefficients are log(aTOP)= 4.567± 0.003, bTOP= 0.2264±0.0003, cTOP= 0.8785±0.0003 and dTOP= 0.933±0.002. The linear relationship and the ef-fect of albedo can be noticed in Fig. 9.

Using the same methodology, = bWVEFFTOP topWVRE

IWVTOP . The de-

pendence of WVEFFTOP on SZA and IWV is quite similar to the observedfor WVEFF in Fig. 7, but the scale is different: WVEFFTOP ranges from 0to 1.6Wm−2 mm−1 and in relative terms, from 0 to 0.3mm−1. Theeffect is weaker at TOA than at surface, both in absolute and relativeterms.

5. Conclusions

In this work, WVRE under cloud-free conditions has been obtainedfrom radiative transfer model SBDART in the context of the IberianPeninsula. Values, for 59.5°< SZA<60.5°, are between − 100.0 (forIWV=39.8 mm) and− 38.7Wm−2 (for IWV=1.4mm), which pointsout the high radiative effect related to the water vapor. All stationsconsidered showed similar values of WVRE, although slight differencescould be noticed between coastal and inland stations. Heating rateswere also calculated, being always between 0.2 and 1.5 K day−1.Moreover, a power relation between WVRE and μ and IWV has beenproposed, with a high degree of correlation. The same approach hasbeen followed for normalized WVRE, with similar results.

Fig. 7. WVEFF (without sign) against SZA (a) for several IWV bins, and against IWV (b) for several SZA bins. (c) and (d) are similar but for normalized WVEFF (without sign).

J. Vaquero-Martínez et al. Atmospheric Research 205 (2018) 18–25

23

96 Apendice A. Artıculos

Then, from the proposed empirical relation, WVEFF has been found,by applying the first derivative of WVRE with respect to IWV. WVEFFshowed a decrease with increasing IWV, sharply at small IWV and sa-turating for greater IWV. This happens because the more water vapor inthe atmosphere, the more radiation is absorbed by it, and thus lessradiation is available for the lower layers of water vapor to absorb. Thisresults in a decrease of efficiency.

With increasing SZA, WVEFF decreases, in a relatively steadymanner. When SZA increases, the incoming radiation is smaller, andthus efficiency is diminished. WVEFF values are around − 8 and0Wm−2 mm−1 (− 1.8 and 0%/mm−1).

Following the same approach as before, relative WVEFF was cal-culated from normalized WVRE. It showed a similar relation with IWV,but an opposed relation with SZA: increasing SZA resulted in higherrelative WVEFF. When SZA increases, water vapor's optical mass in-creases, increasing its efficiency.

The effect of water vapor was also analyzed at TOA, where it is

positive and weaker than in the surface. For 59.5°< SZA<60.5°, itgoes from 11.3 (for IWV=1.44mm) to 20.3Wm−2 (forIWV=39.8 mm). The influence of albedo is higher and was included inthe empirical formula. The behavior of WVEFFTOP is similar to WVEFF,but positive and much weaker, ranging between 0 and1.6Wm−2 mm−1 (0 and 0.3%mm−1).

Acknowledgments

This work was supported by the Spanish Ministry of Economy andCompetitiveness through project CGL2014-56255-C2. Support from theJunta de Extremadura (Research Group Grant GR15137) is gratefullyacknowledged. Work at the Universidad de Valladolid is supported byproject CMT2015-66742-R. Work at the Universidad de Granada wassupported by the Andalusia Regional Government (project P12-RNM-2409) and the Spanish Ministry of Economy and Competitiveness andFEDER funds under the projects CGL2016-81092-R and “Juan de laCierva-Formación” program (FJCI-2014-22052). Some free softwarewas used in this work: GNU Parallel (Tange, 2011), for parallel pro-cessing of SBDART's calls; R (R Core Team, 2017), for data analysis, aswell as some of its packages, such as ggplot2 (Wickham, 2009), ggmap(Kahle and Wickham, 2013), xtable (Dahl, 2016), reshape (Wickham,2007), plyr (Wickham, 2011), chron (James and Hornik, 2014) andmemisc (Elff, 2017).

References

Antón, M., Cazorla, A., Mateos, D., Costa, M.J., Olmo, F.J., Alados-Arboledas, L., 2016.Sensitivity of UV erythemal radiation to total ozone changes under different skyconditions: results for Granada, Spain. Photochem. Photobiol. 92, 215–219. http://dx.doi.org/10.1111/php.12539.

Antón, M., Mateos, D., 2013. Shortwave radiative forcing due to long-term changes oftotal ozone column over the Iberian Peninsula. Atmos. Environ. 81, 532–537. http://dx.doi.org/10.1016/j.atmosenv.2013.09.047.

Bennouna, Y.S., Torres, B., Cachorro, V.E., Ortiz de Galisteo, J.P., Toledano, C., 2013. Theevaluation of the integrated water vapour annual cycle over the Iberian Peninsulafrom EOS-MODIS against different ground-based techniques. Q. J. R. Meteorol. Soc.139, 1935–1956. http://dx.doi.org/10.1002/qj.2080.

Bevis, M., Businger, S., Herring, T.A., Rocken, C., Anthes, R.A., Ware, R.H., 1992. GPSmeteorology: remote sensing of atmospheric water vapor using the global positioningsystem. J. Geophys. Res. 97, 15787–15801. http://dx.doi.org/10.1029/92JD01517.

Colman, R., 2003. A comparison of climate feedbacks in general circulation models. Clim.Dyn. 20, 865–873. http://dx.doi.org/10.1007/s00382-003-0310-z.

Dahl, D.B., 2016. xtable: Export Tables to LaTeX or HTML. https://CRAN.R-project.org/package=xtable R package version 1.8-2.

Dee, D.P., Uppala, S.M., Simmons, A.J., Berrisford, P., Poli, P., Kobayashi, S., Andrae, U.,Balmaseda, M.A., Balsamo, G., Bauer, P., Bechtold, P., Beljaars, A.C.M., van de Berg,L., Bidlot, J., Bormann, N., Delsol, C., Dragani, R., Fuentes, M., Geer, A.J.,Haimberger, L., Healy, S.B., Hersbach, H., Hólm, E.V., Isaksen, L., Kållberg, P.,Köhler, M., Matricardi, M., McNally, A.P., Monge-Sanz, B.M., Morcrette, J.-J., Park,B.-K., Peubey, C., de Rosnay, P., Tavolato, C., Thépaut, J.-N., Vitart, F., 2011. TheERA-Interim Reanalysis: configuration and performance of the data assimilationsystem. Q. J. R. Meteorol. Soc. 137, 553–597. http://dx.doi.org/10.1002/qj.828.

Denman, K.L., Brasseur, G., 2007. Couplings between changes in the climate system andbiogeochemistry. Climate Change 2007: The Physical Science Basis. Contribution ofWorking Group I to the Fourth Assessment Report of the Intergovernmental Panel onClimate Change. . http://www.ipcc.ch/pdf/assessment-report/ar4/wg1/ar4-wg1-

Fig. 8. Time series of WVEFF (a) and IWV (b).

Fig. 9. WVRE against IWV (up) and μ (down) in a log-log plot. The legends show theintervals of SZA and IWV values considered.

J. Vaquero-Martínez et al. Atmospheric Research 205 (2018) 18–25

24

A.5. Artıculo 5 97

chapter7.pdf.Di Biagio, C., di Sarra, A., Eriksen, P., Ascanius, S.E., Muscari, G., Holben, B., 2012. Effect

of surface albedo, water vapour, and atmospheric aerosols on the cloud-free short-wave radiative budget in the Arctic. Clim. Dyn. 39, 953–969. http://dx.doi.org/10.1007/s00382-011-1280-1.

Elff, M., 2017. memisc: Tools for Management of Survey Data and the Presentation ofAnalysis Results. https://CRAN.R-project.org/package=memisc R package version0.99.13.

Forster, P.M.D.F., Shine, K.P., 2002. Assessing the climate impact of trends in strato-spheric water vapor. Geophys. Res. Lett. 29http://dx.doi.org/10.1029/2001GL013909. 10-1–10-4.

James, D., Hornik, K., 2014. chron: Chronological Objects Which Can Handle Dates andTimes. http://CRAN.R-project.org/package=chron R package version 2.3-45. Soriginal by David James, R port by Kurt Hornik.

Kahle, D., Wickham, H., 2013. ggmap: spatial visualization with ggplot2. R J. 5, 144–161.http://journal.r-project.org/archive/2013-1/kahle-wickham.pdf.

Liou, K., 2002. An Introduction to Atmospheric Radiation. Elsevier: Elsevier, New York,USA.

Madronich, S., 2007. Analytic formula for the clear-sky UV index. Photochem. Photobiol.83, 1537–1538. http://dx.doi.org/10.1111/j.1751-1097.2007.00200.x.

Mateos, D., Antón, M., Sanchez-Lorenzo, A., Calbó, J., Wild, M., 2013a. Long-termchanges in the radiative effects of aerosols and clouds in a mid-latitude region(1985–2010). Glob. Planet. Chang. 111, 288–295. http://linkinghub.elsevier.com/retrieve/pii/S0921818113002233http://dx.doi.org/10.1016/j.gloplacha.2013.10.004.

Mateos, D., Antón, M., Valenzuela, A., Cazorla, A., Olmo, F.J., Alados-Arboledas, L.,2013b. Short-wave radiative forcing at the surface for cloudy systems at a midlatitudesite. Tellus B Chem. Phys. Meteorol. 65, 21069. http://dx.doi.org/10.3402/tellusb.v65i0.21069.

Mateos, D., Antón, M., Toledano, C., Cachorro, V.E., Alados-Arboledas, L., Sorribas, M.,Costa, M.J., Baldasano, J.M., 2014a. Aerosol radiative effects in the ultraviolet,visible, and near-infrared spectral ranges using long-term aerosol data series over theIberian Peninsula. Atmos. Chem. Phys. 14, 13497–13514. http://www.atmos-chem-phys.net/14/13497/2014/http://dx.doi.org/10.5194/acp-14-13497-2014.

Mateos, D., Antón, M., Valenzuela, A., Cazorla, A., Olmo, F., Alados-Arboledas, L., 2014b.Efficiency of clouds on shortwave radiation using experimental data. Appl. Energy113, 1216–1219. http://linkinghub.elsevier.com/retrieve/pii/S0306261913007046http://dx.doi.org/10.1016/j.apenergy.2013.08.060.

McClatchey, R.A., Fenn, R.W., Selby, J.E.A., Volz, F.E., Garing, J.S., 1972. OpticalProperties of the Atmosphere, Third edition. . http://oai.dtic.mil/oai/oai?verb=getRecord&metadataPrefix=html&identifier=AD0753075.

McKenzie, R.L., Matthews, W.A., Johnston, P.V., 1991. The relationship between er-ythemal UV and ozone, derived from spectral irradiance measurements. Geophys.Res. Lett. 18, 2269–2272. http://dx.doi.org/10.1029/91GL02786.

Myhre, G., Shindell, D., Bréon, F.-M., Collins, W., Fuglestvedt, J., Huang, J., Koch, D.,Lamarque, J.-F., Lee, D., Mendoza, B., Nakajima, T., Robock, A., Stephens, G.,Takemura, T., Zhang, H., 2013. Anthropogenic and natural radiative forcing. In:Climate Change 2013: The Physical Science Basis. Contribution of Working Group I tothe Fifth Assessment Report of the Intergovernmental Panel on Climate Change, pp.659–740.

R Core Team, 2017. R: A Language and Environment for Statistical Computing. RFoundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/.

Ricchiazzi, P., Yang, S., Gautier, C., Sowle, D., 1998. SBDART: a research and teachingsoftware tool for plane-parallel radiative transfer in the Earth's atmosphere. Bull. Am.Meteorol. Soc. 79, 2101–2114. http://dx.doi.org/10.1175/1520-0477(1998)

079<2101:SARATS>2.0.CO;2.Román, R., Antón, M., Cachorro, V., Loyola, D., Ortiz de Galisteo, J., de Frutos, A.,

Romero-Campos, P., 2015. Comparison of total water vapor column from GOME-2 onMetOp-A against ground-based GPS measurements at the Iberian Peninsula. Sci. TotalEnviron. 533, 317–328. http://linkinghub.elsevier.com/retrieve/pii/S0048969715303260http://dx.doi.org/10.1016/j.scitotenv.2015.06.124.

Román, R., Bilbao, J., de Miguel, A., 2014. Uncertainty and variability in satellite-basedwater vapor column, aerosol optical depth and Angström exponent, and its effect onradiative transfer simulations in the Iberian Peninsula. Atmos. Environ. 89, 556–569.http://linkinghub.elsevier.com/retrieve/pii/S135223101400123Xhttp://dx.doi.org/10.1016/j.atmosenv.2014.02.027.

Santer, B.D., Mears, C., Wentz, F.J., Taylor, K.E., Gleckler, P.J., Wigley, T.M.L., Barnett,T.P., Boyle, J.S., Bruggemann, W., Gillett, N.P., Klein, S.A., Meehl, G.A., Nozawa, T.,Pierce, D.W., Stott, P.A., Washington, W.M., Wehner, M.F., 2007. Identification ofhuman-induced changes in atmospheric moisture content. Proc. Natl. Acad. Sci. 104,15248–15253. http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=1986574&tool=pmcentrez&rendertype=abstracthttp://dx.doi.org/10.1073/pnas.0702872104.

Smith, C.A., Haigh, J.D., Toumi, R., 2001. Radiative forcing due to trends in stratosphericwater vapour. Geophys. Res. Lett. 28, 179–182. http://dx.doi.org/10.1029/2000GL011846.

Soden, B.J., Wetherald, R.T., Stenchikov, G.L., Robock, A., 2002. Global cooling after theeruption of Mount Pinatubo: a test of climate feedback by water vapor. Science 296,727–730. http://dx.doi.org/10.1126/science.296.5568.727.

Solomon, S., Rosenlof, K., Portmann, R., Daniel, J., Davis, S., Sanford, T., Plattner, G.-K.,2010. Contributions of stratospheric water vapor to decadal changes in the rate ofglobal warming. Science 327, 1219–1223. http://www.sciencemag.org/cgi/content/abstract/327/5970/1219http://dx.doi.org/10.1126/science.1182488.

Tange, O., 2011. GNU parallel: the command-line power tool; login. USENIX Mag. 36,42–47. https://www.usenix.org/system/files/login/articles/105438-Tange.pdf.

Valenzuela, A., Olmo, F.J., Lyamani, H., Antón, M., Quirantes, A., Alados-Arboledas, L.,2012. Aerosol radiative forcing during African desert dust events (2005–2010) overSoutheastern Spain. Atmos. Chem. Phys. 12, 10331–10351. http://www.atmos-chem-phys.net/12/10331/2012/http://dx.doi.org/10.5194/acp-12-10331-2012.

Vaquero-Martínez, J., Antón, M., Ortiz de Galisteo, J.P., Cachorro, V.E., Costa, M.J.,Román, R., Bennouna, Y.S., 2017. Validation of MODIS integrated water vapor pro-duct against reference GPS data at the Iberian Peninsula. Int. J. Appl. Earth Obs.Geoinf. 63, 214–221. http://linkinghub.elsevier.com/retrieve/pii/S0048969716327176http://dx.doi.org/10.1016/j.jag.2017.07.008.

Vaquero-Martínez, J., Antón, M., Ortiz de Galisteo, J.P., Cachorro, V.E., Wang, H.,González Abad, G., Román, R., Costa, M.J., 2017, Feb, Feb. Validation of integratedwater vapor from OMI satellite instrument against reference GPS data at the IberianPeninsula. Sci. Total Environ. 580, 857–864. http://linkinghub.elsevier.com/retrieve/pii/S0048969716327176http://dx.doi.org/10.1016/j.scitotenv.2016.12.032.

Wickham, H., 2007. Reshaping data with the reshape package. J. Stat. Softw. 21, 1–20.http://www.jstatsoft.org/v21/i12/.

Wickham, H., 2009. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag, NewYork. http://ggplot2.org.

Wickham, H., 2011. The split-apply-combine strategy for data analysis. J. Stat. Softw. 40,1–29. http://www.jstatsoft.org/v40/i01/.

World Meteorological Organization, 2008. Guide to Meteorological Instruments andMethods of Observation. vol. I WMO, Geneva OCLC: 785715260.

Zhong, W., Haigh, J.D., 2003. Shortwave radiative forcing by stratospheric water vapor.Geophys. Res. Lett. 30, 1113. http://dx.doi.org/10.1029/2002GL016042.

J. Vaquero-Martínez et al. Atmospheric Research 205 (2018) 18–25

25

98 Apendice A. Artıculos

A.6. Artıculo 6 99

A.6. Artıculo 6

Tıtulo: Evaluation of Water Vapor Radiative Effects Using GPSData Series over Southwestern Europe.Autores

• Vaquero-Martınez, Javier (U. de Extremadura).• Anton, Manuel (U. de Extremadura).• Sanchez-Lorenzo, Arturo (U. de Extremadura).• Cachorro, Victoria E. (U. de Valladolid).

Ano: 2020.Revista: Remote Sensing.Paginas: 1307.DOI: 10.3390/rs12081307.Volumen: 12.

remote sensing

Letter

Evaluation of Water Vapor Radiative Effects UsingGPS Data Series over Southwestern Europe

Javier Vaquero-Martínez 1,2,* , Manuel Antón 1,2, Arturo Sanchez-Lorenzo 1,2

and Victoria E. Cachorro 3

1 Departamento de Física, Universidad de Extremadura, CP 06006 Badajoz, Spain; mananton@unex.es (M.A.);arturosl@unex.es (A.S.-L.)

2 Instituto Universitario de Investigación del Agua, Cambio Climático y Sostenibilidad (IACYS),Universidad de Extremadura, CP 06006 Badajoz, Spain

3 Grupo de Óptica Atmosférica, Universidad de Valladolid, Paseo Belén 7, CP 47011 Valladolid, Spain;chiqui@goa.uva.es

* Correspondence: javier_vm@unex.es

Received: 2 April 2020; Accepted: 18 April 2020 ; Published: 21 April 2020�����������������

Abstract: Water vapor radiative effects (WVRE) at surface in the long-wave (LW) and short-wave(SW) spectral ranges under cloud and aerosol free conditions are analyzed for seven stations in Spainover the 2007–2015 period. WVRE is calculated as the difference between the net flux obtained bytwo radiative transfer simulations; one with water vapor from Global Positioning System (GPS)measurements and the other one without any water vapor (dry atmosphere). The WVRE in the LWranges from 107.9 Wm2 to 296.7 Wm−2, while in the SW it goes from −64.9 Wm−2 to −6.0 Wm−2.The results show a clear seasonal cycle, which allows the classification of stations in three sub-regions.In general, for total (SW + LW) and LW WVRE, winter (DJF) and spring (MAM) values are lowerthan summer (JJA) and autumn (SON). However, in the case of SW WVRE, the weaker values are inwinter and autumn, and the stronger ones in summer and spring. Positive trends for LW (and total)WVRE may partially explain the well-known increase of surface air temperatures in the study region.Additionally, negative trends for SW WVRE are especially remarkable, since they represent about aquarter of the contribution of aerosols to the strong brightening effect (increase of the SW radiationflux at surface associated with a reduction of the cloud cover and aerosol load) observed since the2000s in the Iberian Peninsula, but with opposite sign, so it is suggested that water vapor could bepartially masking the full magnitude of this brightening.

Keywords: water vapor; radiative effect; long-wave; Spain; Sourthwestern Europe; Europe;radiative transfer

1. Introduction

Water vapor is acknowledged as a crucial element in the climate system. Its latent heat has animportant role in energy transport and it is obviously a fundamental part of the hydrological cycle [1].Water vapor is also known to be the main absorber of the infrared radiation emited by Earth’s surfaceand atmosphere, which allows heating of the low atmosphere. Its hydroxyl (H–O) bond is the cause ofthe absorption in the infrared region.

The infrared absorption of water vapor involves a positive feedback [2,3]. If the atmosphere’stemperature rises, the air can hold more water vapor from evaporation, since the saturation vaporpressure increases as temperature rises. This further increases the temperature of the climate systembecause of water vapor heating. Therefore, the effect of water vapor is considered a feedback ratherthan a forcing, since, on a global scale, the water vapor concentration is mainly dependent on

Remote Sens. 2020, 12, 1307; doi:10.3390/rs12081307 www.mdpi.com/journal/remotesensing

A.6. Artıculo 6 101

Remote Sens. 2020, 12, 1307 2 of 12

the temperature, and the typical residence time of water vapor is about ten days [1]. Hence, theanthropogenic emissions of water vapor have no significant effect on the global climate, except thosein the stratosphere, where due to the conditions of stability, pressure and temperature, water vaporemissions manage to stay in the long term, and therefore can be considered a forcing [4–7].

Water vapor is not evenly distributed in the atmospheric column. The lower layers of air generallyhold most of the water vapor, sharply decreasing with height. Therefore, water vapor is generallyquantified using the column integrated amount of water vapor or integrated water vapor (IWV). This isequivalent to condensing all the water vapor in the atmospheric column and measuring the height thatit would reach in a vessel of unit cross section. The units are those of columnar mass density (gcm−2

or kgm−2). Since 1 g of liquid water has a volume of 1 cm3, the columnar mass density can be writtenas units of length (1 cm).

However, there is a great uncertainty about the quantification of the radiative effects of watervapor. Although some efforts have been made to study it in the short-wave (SW) region (i.e., [8–16]),the references in the literature related to the LW effects of water vapor are scarce (i.e., [17–21]).These works mainly focus in the feedback and sensitivity of climate system with respect to water vaporand the downwelling long-wave (LW) radiation, but not considering the radiative balance. This workaims to shed some light on the role of water vapor in the radiative balance, not only in SW as previousstudies have done, but also in LW. The implications of this work can help to understand the trends insolar radiation at surface, as well as the increase of temperatures in the Iberian Peninsula in the studyperiod. Similar approaches have been conducted with other atmospheric compounds, like aerosols(i.e., [22]), aerosols and clouds [23,24], ozone [25,26], even stratospheric water vapor [7].

In this work, the water vapor radiative effect (WVRE) is defined as the net change in radiationat surface taking as reference a dry atmosphere (adapted from [9]). Daily values (calculated as theintegration of hourly values) of WVRE in both SW and LW regions are presented for GPS stationsin Spain, for the period 2007–2015. The total WVRE is also analyzed, obtained as the sum of the SWand LW effects. The paper has the following structure: Section 2 describes the data-sets used in thiswork and in Section 3 the methodology is explained. The results are exhibited in Section 4, while thediscussion of results is carried out in Section 5. Finally, conclusions are drawn in Section 6.

2. Data

2.1. IWV Data from GPS stations

Global Positioning System (GPS) ground-based stations can be used to measure IWV as thoroughlydetailed in Bevis et al. [27]. In short, in GPS positioning, the troposphere causes a delay in thesignal between the GPS satellite and the GPS receiver that must be estimated for precise positioning,since such delay is of the order of a few meters. The direct measurement of this delay is known as SlantTropospheric Delay (STD). STD is the result of two contributions, one related to the non-dipolarcontribution that all gases in the atmosphere cause (known as Slant Hydrostatic Delay, SHD),and another contribution due to the dipolar effects in water vapor molecules, which is known as SlantWet Delay (SWD). The sum of both contributions gives the STD, as Equation (1) shows.

STD = SWD + SHD (1)

However, the slant delays change with the geometry. They depend on the angle between thesatellite–receiver line and the zenith. Therefore, a mapping function [28–30] is needed to convert slantdelays to zenith delays, as shown in Equation (2).

STD = mw(E)ZWD + mh(E)ZHD = m(E)(ZWD + ZHD) = m(E)ZTD (2)

where E is the elevation angle, m are the mapping functions, ZTD is the Zenith Tropospheric Delay,ZWD the Zenith Wet Delay and ZHD the Zenith Hydrostatic Delay.

102 Apendice A. Artıculos

Remote Sens. 2020, 12, 1307 3 of 12

ZTD is provided by GPS processing methods. ZHD can be obtained using a simple model [31]based on the surface pressure. The difference between ZTD and ZHD gives ZWD, which can beconverted to IWV with a conversion factor ZWD = κ · IWV. The constant κ can be determined fromthe mean air temperature of the atmospheric column weighted by the water vapor content. This meanair temperature is estimated from an empirical linear relationship with temperature at the station level.

The period covered in this work ranges from 2007 to 2015, since this is the time span of availableGPS data. The tropospheric products are provided by the Spanish Geographic Institute “InstitutoGeográfico Nacional”, which is a local analysis center for the European Reference Frame (EUREF).The stations selected are those that meet the quality standards for EUREF network, have a longuninterrupted time-series, have nearby meteorological automatic stations and have a representativenumber of cloud-free days in all (or most) months of the year. Surface pressure and temperaturewere provided by the Spanish Meteorological State Agency (AEMet). Temperature is provided hourly,while pressure is measured four times a day. Temperature was linearly interpolated to the time ofmeasurements, and pressure was interpolated taking into account the barometric tide. This is donefor the seven stations shown in Figure 1. A summary of the stations positions is presented in Table 1.The temporal resolution of the IWV data-set is one hour.

Table 1. Position of the GPS stations. Latitude and longitude are given in degrees north and east, whilealtitude is given in meters. Zones are: North atlantic (NA), Mediterranean Sea (MS) and interior (I).

Station Acronym Latitude Longitude Altitude Zone

A Coruña acor 43.36 −8.40 12 NACórdoba coba 37.92 −4.72 162 MS

Villafranca vill 40.44 −3.95 596 IAlicante alac 38.34 −0.48 10 MSAlmería alme 36.85 −2.46 77 MSValencia vale 39.48 −0.34 28 MSCáceres cace 39.48 −6.34 384 I

acor

coba

vill

alac

alme

valecace

0 100 200 300 400 km

N

ZoneNAMSI

Figure 1. Map of the Global Positioning System (GPS) stations included in this work. Zones are: Northatlantic (NA), Mediterranean Sea (MS) and interior (I).

A.6. Artıculo 6 103

Remote Sens. 2020, 12, 1307 4 of 12

2.2. Auxiliary Data

Some additional data sets were used in order to characterize the state of the atmosphere andsurface. Profiles of temperature, ozone, and water vapor were obtained from ERA-Interim reanalysis.The latter were re-scaled to the IWV value from GPS stations. The temperature of the surface was alsoobtained from ERA-Interim reanalysis for LW calculations.

The ERA-Interim [32] is the latest global reanalysis model produced by the European Centre forMedium-Range Weather Forecasts (ECMWF) to replace ERA-40. The data is available at 4 times aday (00, 06, 12, 18 h). For every IWV measurement the nearest pixel at the closest time is selectedto represent the state of the atmosphere and surface. The ERA-Interim grid has a resolution of0.75◦ × 0.75◦ with 37 vertical levels.

Additionally, daily sunshine duration records and cloud cover (CC) data provided by AEMet arealso used for the selection of days with cloud free and low aerosol load conditions.

3. Methodology

Both LW and SW irradiances have been simulated using the Santa Barbara’s DIScrete OrdinateRadiative Transfer (DISORT) Radiative Transfer model (SBDART, [33]), only for those days consideredas cloud and aerosol free (Rayleigh atmosphere). For that, firstly, days with CC less than or equal to1 okta are selected. Subsequently, from these cloud-free days, to select cases with low aerosol load,daily sunshine duration is divided by its theoretical value under the assumption of cloud-free sky, anda threshold value of 0.75 is used to filter out heavy aerosol load situations. WMO [34] recommends thevalue of 0.70, being 0.75 a more restrictive threshold to ensure aerosol free days.

The SW wavelength range considered was between 0.2 µm and 4.0 µm (0.5% steps, ranging from0.001 to 0.02 µm). More detailed description on the variable inputs to the SW simulations can befound in Vaquero-Martínez et al. [13]. The atmosphere model were “mid-latitude summer” fromMarch to August (both included) and “mid-latitude winter” for the rest of the year, both includedin SBDART [35]. However, these atmosphere models are modified with re-scaled profiles for IWVand ozone to the total IWV (GPS data) and total column ozone (ERA-Interim data-set). The numberof streams was set to 4. Additionally, thermal radiation is unset. SW WVRE for solar zenith angle(SZA) > 90◦ are set to 0 without performing the calculation, since no SW radiation is available underthis condition.

The LW simulations, however, used a different configuration. The LW wavelength range wasbetween 4.0 µm and 100 µm (steps of 1% width, that is to say, from 0.04 to 1 µm). The number ofstreams was set to 16, which is a value used in other works in the LW range [19,36]. The number ofatmosphere layers was set to 65, with a resolution between 1 m for the lower layers and 900 m for thehigher ones. The atmospheric composition profile from reanalysis is given to SBDART as input, aswell as the temperature of the surface from ERA-Interim reanalysis and IWV from GPS (to re-scale thewater vapor in the layers). LW calculations have sun radiation unset and thermal radiation activated.

The model was run twice, for each hourly GPS measurement: in one the IWV value fromGPS measurement is used, while in the other one, the IWV is set to 0 cm. The irradiances arethen used to calculate the WVRE at surface which is defined [13] as the difference between the netirradiances (downwards minus upwards) with and without water vapor in the atmosphere, as shownin Equation (3).

WVRE =(

I↓IWV − I↑IWV

)−

(I↓dry − I↑dry

)(3)

Total WVRE is defined as the sum of both LW and SW WVRE. The calculated variables areintegrated to daily values, in order to compare LW and SW contributions. Missing values are filled withlinearly interpolated data, but days with more than 50% missing data are filtered out. The integrationis done as an average of the 24 hourly data of each day. It must be noticed that the effects on SW

104 Apendice A. Artıculos

Remote Sens. 2020, 12, 1307 5 of 12

radiation are limited by insolation, while the effects on LW are active for the whole day, and thereforethey must be integrated to daily values before any comparison.

For the study of trends, daily data are deseasonalized, that is to say, we obtained the anomalies.The process is to subtract to every data-point the mean of the data with coincident day of year and site.Then, monthly means of daily deseasonalized data (anomalies) are calculated. Monthly means with5 days or less are replaced by the linear interpolation at that month. The test used to calculate the trendand decide if it is significant are the Mann–Kendall [37,38] test and Sen’s slope [39], since the IWVand WVRE data do not follow a normal distribution. The confidence level considered for significanceis 0.05.

4. Results

4.1. Spatial and Seasonal Variability

In order to study the spatial variability of the WVRE, the sites have been divided into threegroups: North Atlantic (NA), Mediterranean Sea (MS) and Interior (I). Figure 1 shows the distributionof stations from each zone in Spain. These groups have a geographical meaning (NA are stations closeto the North Atlantic Sea, in Nothern Spain; I stations are in the interior of Iberia, and MS stationsare close to the Mediterranean Sea in the East and South of Iberia). This division have been alreadyapplied in other water vapor related studies [40,41]. Also, the WVRE exhibits similar features betweenthe sites that belong to the same group.

Table 2 shows some statistics of WVRE by zone and regime. The mean total WVRE is similarin I and NA, but the I zone shows more variability, with longer low-tail (minimum and first quartileare lower than in NA) and high-tail (larger maximum and first quartile than in NA). MS zone showsgenerally higher values of total WVRE than the other two zones. Standard deviation (SD) values arearound 20 Wm−2, while coefficient of variation (CV) values are around 10% for NA and 12.5%. The LWregime shows a behavior close to the total regime, with MS WVRE values being higher than the othertwo zones, and I and MS zones being more disperse than NA. Regarding the SW regime, values in thethree zones are quite similar. Thus, considering all stations, the mean WVRE is −39.2 Wm−2, with aSD of 15.9 Wm−2 and a CV of 37.9%. The increased variability (approximately 3 times the LW CV) isdue to the seasonality of the solar zenith angle and sunlight hours that heavily affect the SW WVRE.

Table 2. Summary statistics of WVRE. All values are in W m−2, except CV (in %). The table shows theminimum (min), the first quartile (Q1), the median, the mean, the third quartile (Q3), the maximum(max), the standard deviation (sd) and the coefficient of variation (CV). The zones are: North Atlantic(NA), Mediterranean Sea (MS) and Interior (I).

Regime Zone min Q1 Median Mean Q3 max sd CV

Total NA 99.4 150.0 161.5 160.9 171.8 206.9 15.4 9.5Total MS 107.6 161.5 179.9 179.5 198.0 235.8 22.4 12.5Total I 96.5 150.7 166.4 166.3 181.8 222.4 20.8 12.5Total All 96.5 156.9 173.0 174.0 192.4 235.8 22.6 13.0LW NA 132.1 179.5 200.2 198.1 217.3 253.0 23.5 11.9LW MS 120.5 190.6 224.0 220.2 249.8 296.7 33.9 15.4LW I 107.9 176.0 207.6 203.3 228.6 274.9 31.9 15.7LW All 107.9 185.4 214.3 213.3 242.0 296.7 33.8 15.8SW NA −62.6 −50.7 −38.2 −37.1 −25.2 −9.5 14.7 39.6SW MS −64.9 −54.2 −45.2 −40.6 −25.0 −9.8 15.1 37.2SW I −61.3 −49.4 −40.1 −36.8 −23.3 −6.0 14.1 38.2SW All −64.9 −52.4 −42.8 −39.2 −24.4 −6.0 14.9 37.9

Regarding the seasonal behavior, Figure 2 shows a box-plot of the WVRE by months, and Table 3displays the average values by season: autumn (SON), summer (JJA), spring (MAM) and winter (DJF).It is observed that total WVRE values are quite similar in autumn (SON) as in summer (JJA), while

A.6. Artıculo 6 105

Remote Sens. 2020, 12, 1307 6 of 12

spring (MAM) values are similar to winter (DJF) ones. Hence, two seasons, for the purpose of totalWVRE, could be considered: a cold (spring + winter) season and a warm (summer + autumn) season.For MS and I, summer total WVRE is slightly over the autumn total WVRE, while in NA, the oppositeoccurs. In Figure 2, it can be observed that SW WVRE is quite similar in the three regions, with thelargest effect during summer months (∼ −50 Wm−2) and the smallest one during winter months(∼ −20 Wm−2).

Table 3. Seasonal mean values of WVRE by regime and zone. Values are in Wm−2. Seasons are winter(December, January, February), spring (March, April, May), summer (June, July, August), and autumn(September, October, November).

Regime Zone Winter Spring Summer Autumn

Total NA 151.8 147.6 164.3 173.5Total MS 159.0 159.9 195.6 188.1Total I 144.9 151.8 179.9 171.6Total All 154.4 156.4 188.8 181.3SW NA −16.6 −38.3 −53.2 −31.6SW MS −20.0 −41.6 −55.1 −33.2SW I −17.5 −37.0 −50.4 −29.3SW All −19.1 −40.0 −53.5 −31.7LW NA 168.4 186.0 217.5 205.1LW MS 179.0 201.5 250.7 221.3LW I 162.4 188.8 230.3 200.9LW All 173.5 196.4 242.2 213.0

NA MS I

TotalLW

SW

Jan

Feb

Mar Apr

May Jun

Jul

Aug

Sep

Oct

Nov

Dec Jan

Feb

Mar Apr

May Jun

Jul

Aug

Sep

Oct

Nov

Dec Jan

Feb

Mar Apr

May Jun

Jul

Aug

Sep

Oct

Nov

Dec

100

150

200

100

150

200

250

300

−60

−40

−20

Month

WV

RE

(Wm

−2)

Figure 2. Water vapor radiative effects (WVRE) seasonal evolution in the regions considered in thisstudy, in the form of a box-plot.

4.2. Trends

Figure 3 shows the evolution of the mean annual IWV and WVRE for the three regimes (LW, SWand Total) on the seven stations. The data have been deseasonalized and then, the monthly averageshave been used to build the yearly time series. It is remarkable that, despite the small number of years

106 Apendice A. Artıculos

Remote Sens. 2020, 12, 1307 7 of 12

(less than a decade) in the time-series, most of the sites and variables show statistically significantresults, in all cases with the same sign. IWV has positive trend in all sites, as well as total WVREand LW WVRE. By contrast, SW WVRE has negative trend, due to the positive trends shown by IWV.Specifically, Table 4 shows IWV and (total, LW and SW) WVRE by stations and zones. IWV trends areabout 0.01− 0.02 cm year−1, statistically significant in all stations but alme and vill.

Table 4. Trends of IWV and WVRE (total, LW and SW). Values are in cm year−1 for IWV andWm−2 year−1. P-values are shown in parenthesis, and an asterisk (*) is added to statisticallysignificant results.

Zone Site IWV LW WVRE SW WVRE Total WVRE

NA acor 0.011 (1.8 · 10−03)* 0.394 (1.6 · 10−03)* -0.097 (4.2 · 10−04) * 0.348 (2.9 · 10−03) *MS alac 0.013 (3.4 · 10−03) * 0.199 (2.6 · 10−01) −0.079 (6.0 · 10−03) * 0.102 (4.8 · 10−01)MS alme 0.0016 (8.1 · 10−01) −0.0810 (7.3 · 10−01) 0.0314 (3.9 · 10−01) −0.0512 (8.4 · 10−01)MS coba 0.010 (2.5 · 10−02) * 0.377 (7.7 · 10−02) −0.085 (2.9 · 10−02)* 0.346 (5.6 · 10−02)MS vale 0.0178 (1.3 · 10−03) * 0.1104 (5.8 · 10−01) 0.0095 (8.1 · 10−01) 0.1117 (4.8 · 10−01)

I cace 0.014 (2.5 · 10−03) * 0.770 (6.4 · 10−04) * -0.103 (6.7 · 10−03) * 0.707 (4.0 · 10−04) *I vill 0.0059 (1.2 · 10−01) 0.5005 (2.4 · 10−02) * -0.0399 (2.8 · 10−01) 0.4230 (2.6 · 10−02) *

IWV

WV

RE

LW W

VR

ES

W W

VR

E

2008 2010 2012 2014 2016

−0.2−0.1

0.00.10.20.3

−10−5

05

−10−5

05

10

−2−1

01

Date

Ano

mal

ies

aver

aged

ove

r al

l sta

tions

Figure 3. Time-series of station-averaged integrated water vapor (IWV) and WVRE (SW, LW and totalregimes) anomalies. Values are in cm for IWV and in Wm−2 for WVRE. Blue solid lines point out thelinear trends (see text).

5. Discussion

5.1. Spatial and Seasonal Variability

Mateos et al. [23] reported an averaged value of−57.1 Wm−2 (SD of 16.2 Wm−2) for the combinedradiative effect caused by clouds and aerosols over the Iberian Peninsula for the period 1985–2010.Additionally, Mateos et al. [22] derived the aerosol radiative effect under cloud-free conditions at sixstations located in the Iberian Peninsula, reporting annual averages in the range of−8.8 to−5.7 Wm−2.Hence, the comparison of these radiative effects with the results reported in Figure 2 and Table 3 for

A.6. Artıculo 6 107

Remote Sens. 2020, 12, 1307 8 of 12

the WVRE points out the great influence of the water vapor on the SW irradiance in the study region.Moreover, the large differences between seasons shown in Figure 2 can be related to the markedseasonal cycle of water vapor column over the Iberian Peninsula [42], in which IWV is larger insummer and smaller in winter.

The differences among regions are more noticeable in the LW range, as well as in the total WVRE(because of the LW contribution). Region I shows values below the MS region, but a similar seasonalvariability. However, the NA region shows a less marked seasonal variability, being the differencebetween winter and summer more subtle than in the other two cases. Therefore, NA winter LW WVREis similar to MS winter LW WVRE, while NA summer LW WVRE is similar to I summer LW WVRE.

5.2. Trends

The trend values reported in Table 4 for IWV are about three times higher than those obtainedby Ning and Elgered [43] for Sweden and Finland, using GPS data during the period 1997–2016(∼ 0.004 cm year−1). Chen and Liu [44] showed values around 0.002 cm year−1 in temperate latitudesfor the period 2000–2014, one order of magnitude below our results. Vicente-Serrano et al. [45] alsodetected positive trends in specific humidity at surface in Spain in the period 1961–2011.

The positive trends in IWV cause the total WVRE trend to be significantly positive in stationsfrom regions I and NA, while non-significant in MS stations. The LW WVRE trends are positive aswell (except for alme, with a non-significant negative trend), while the SW WVRE trends are negative(except for alme and vale both with positive but non-significant trends). The balance between both LWand SW WVRE gives a positive trend (except for alme, with a non-significant negative trend). The SWWVRE significant trends are around −0.09 Wm−2 year−1, while the LW WVRE significant trends arearound 0.50 Wm−2 year−1. Then the overall significant trends on WVRE are around 0.42 Wm−2 year−1.This positive trend can partially explain the rise in surface air temperatures observed in the IberianPeninsula during the last two decades (e.g., [46]), which in turn increase the evaporation in a positiveclimate feedback [3].

The mean SW WVRE trend values are weaker (−0.09 Wm−2 year−1) than Kvalevåg andMyhre [47], where global SW trends due to water vapor are estimated as −0.29 Wm−2 per year.This result indicates that water vapor could have a role in modulating the widespread increase ofSW surface radiation, also known as brightening, reported in the literature since the 1980s [48,49].Mateos et al. [24] determined that this SW radiation trend is +0.7 Wm−2 year−1 on average for theperiod 2003–2012 in the Iberian Peninsula, using both ground-based and satellite SW data. Theseauthors also showed that three fourths of the trend is explained by clouds, while the other one fourthis related to aerosol change, in line with the observed reductions in total cloud cover and aerosolload over the study region. Additionally, Mateos et al. [22] reported a statistically significant trend of+0.36 Wm−2 year−1 for the aerosol radiative effect under cloud-free conditions in the Iberian Peninsula(period 2004–2012). Hence, it must be pointed out that the negative trend for SW WVRE is about aquarter of this positive trend for the aerosol radiative effect. Therefore, the trends of water vapor couldbe partially masking the full magnitude of the role of aerosol load in the modulation of SW radiationat surface over the study region.

6. Conclusions

This work has provided some insight about the radiative effects of the water vapor in SothwesternEurope in both the long wave and short wave bands under cloud and aerosol load free conditions.The results show that the three regions considered in the Iberian Peninsula have total WVRE around173 Wm−2, with total maximum of 235.8 Wm−2 and minimum of 96.5 Wm−2. The LW WVREis therefore larger in absolute value than the SW WVRE. Specifically, LW WVRE values rangefrom 107.9 Wm2 to 296.7 Wm−2, while SW WVRE exhibits negative values, from −64.9 Wm−2 to−6.0 Wm−2.

108 Apendice A. Artıculos

Remote Sens. 2020, 12, 1307 9 of 12

The distribution of the WVRE has a marked seasonal cycle in all zones considered. In general,LW WVRE is higher in summer and autumn, and lower in winter and spring. However, the SWWVRE has stronger values (more negative) in spring and summer, and weaker in autumn and winter.Overall, the total WVRE follows the LW WVRE pattern, with stronger values in summer and autumn,and weaker WVRE in winter and spring.

The trends have been calculated for IWV and WVRE in the total, LW and SW regimes. IWVtrends are positive in all cases, with a mean value of 0.013 mm year−1, and this causes the LW andtotal WVRE trends to be positive (mean values of 0.56 and 0.49 Wm−2year−1, respectively) and SWWVRE trends to be negative (−0.09 Wm−2year−1).

It must be highlighted that the positive radiative effect in the whole spectral range, associatedwith the increase of the water vapor over the Iberian Peninsula, may partially explain the notableincrease of the surface air temperature reported in the literature in this region. Additionally, thenegative radiative effect in SW, due to the notable increase of IWV values over the Iberian Peninsuladuring the last decade, may play a key role in mitigating the SW radiation increases associated with areduction of the cloud cover and aerosol load over this region. Therefore, this increase of the watervapor could partially offset the strong brightening effect (increase of SW radiation at surface) recordedin the Iberian Peninsula since 2000s.

Author Contributions: J.V.-M. performed the data analysis and wrote the main draft of the paper. M.A. providedthe main ideas and contributed to the data analysis and paper writing. A.S.-L. contributed to the discussionof results and paper writing. V.E.C. provided GPS, temperature, pressure and sunshine duration data andcontributed to paper writing. All authors have read and agreed to the published version of the manuscript.

Funding: This work was partly supported by the Ministerio de Economia y Competitividad of the SpanishGovernment (CGL2017-87917-P) and by Junta de Extremadura and FEDER funds (IB18092). V.E.C. is gratefulto the Spanish Ministry of Science, Innovation and Universities for the support through the ePOLAAR project(RTI2018-097864-B-I00). A.S.-L. was supported by a postdoctoral fellowship RYC-2016–20784 funded by theSpanish Ministry of Science, Innovation and Universities. J.V.-M. was supported by a predoctoral fellowship(PD18029) from Junta de Extremadura and European Social Fund.

Acknowledgments: Authors thank AEMet and ECMWF for providing the data necessary for this work, and toPaul Ricchiazzi for the SBDART radiative transfer code. Authors also acknowledge the R [50] packages tidyverse1.2.1 [51], lubridate 1.7.1 [52], ggpubr 0.2.4 [53], zoo 1.8-7 [54], trend 1.1.2 [55].

Conflicts of Interest: Authors declare no conflict of interest.

Abbreviations

IWV Integrated Water VaporWVRE Water Vapor Radiative EffectsSW short-waveLW long-waveGPS Global Positioning SystemNA North Atlantic zoneI Interior zoneMS Mediterranean Sea zoneZTD Zenith Tropospheric DelayZWD Zenith Wet DelayZHD Zenith Hydrostatic DelayEUREF European Reference FrameAEMet Spanish Meteorological State Agencysd standard deviationCV Coefficient of Variation

A.6. Artıculo 6 109

Remote Sens. 2020, 12, 1307 10 of 12

References

1. Myhre, G.; Shindell, D.; Bréon, F.M.; Collins, W.; Fuglestvedt, J.; Huang, J.; Koch, D.; Lamarque, J.F.; Lee, D.;Mendoza, B.; et al. Anthropogenic and Natural Radiative Forcing. In Climate Change 2013: The PhysicalScience Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel onClimate Change; IPCC., Ed.; IPCC: Geneva, Switzerland, 2013; pp. 659–740.

2. Colman, R. A Comparison of Climate Feedbacks in General Circulation Models. Clim. Dyn. 2003, 20, 865–873.[CrossRef]

3. Colman, R. Climate Radiative Feedbacks and Adjustments at the Earth’s Surface. J. Geophys. Res. Atmos.2015, 120, 3173–3182. [CrossRef]

4. Forster, P.M.D.F.; Shine, K.P. Assessing the Climate Impact of Trends in Stratospheric Water Vapor. Geophys.Res. Lett. 2002, 29, 10. [CrossRef]

5. Smith, C.A.; Haigh, J.D.; Toumi, R. Radiative Forcing Due to Trends in Stratospheric Water Vapour. Geophys.Res. Lett. 2001, 28, 179–182. [CrossRef]

6. Solomon, S.; Rosenlof, K.; Portmann, R.; Daniel, J.; Davis, S.; Sanford, T.; Plattner, G.K. Contributions ofStratospheric Water Vapor to Decadal Changes in the Rate of Global Warming. Science 2010, 327, 1219–1223.[CrossRef]

7. Zhong, W.; Haigh, J.D. Shortwave Radiative Forcing by Stratospheric Water Vapor. Geophys. Res. Lett. 2003,30, 1113. [CrossRef]

8. Di Biagio, C.; di Sarra, A.; Eriksen, P.; Ascanius, S.E.; Muscari, G.; Holben, B. Effect of Surface Albedo, WaterVapour, and Atmospheric Aerosols on the Cloud-Free Shortwave Radiative Budget in the Arctic. Clim. Dyn.2012, 39, 953–969. [CrossRef]

9. Mateos, D.; Antón, M.; Valenzuela, A.; Cazorla, A.; Olmo, F.J.; Alados-Arboledas, L. Short-Wave RadiativeForcing at the Surface for Cloudy Systems at a Midlatitude Site. Tellus B Chem. Phys. Meteorol. 2013, 65, 21069.[CrossRef]

10. Román, R.; Bilbao, J.; de Miguel, A. Uncertainty and Variability in Satellite-Based Water Vapor Column,Aerosol Optical Depth and Angström Exponent, and Its Effect on Radiative Transfer Simulations in theIberian Peninsula. Atmosp. Environ. 2014, 89, 556–569. [CrossRef]

11. Obregón, M.A.; Costa, M.J.; Serrano, A.; Silva, A.M. Effect of Water Vapor in the SW and LW DownwardIrradiance at the Surface during a Day with Low Aerosol Load. IOP Conf. Ser. Earth Environ. Sci. 2015,28, 012009. [CrossRef]

12. Obregón, M.; Costa, M.; Silva, A.; Serrano, A. Impact of Aerosol and Water Vapour on SW Radiation at theSurface: Sensitivity Study and Applications. Atmos. Res. 2018, 213, 252–263. [CrossRef]

13. Vaquero-Martínez, J.; Antón, M.; Ortiz de Galisteo, J.P.; Román, R.; Cachorro, V.E. Water Vapor RadiativeEffects on Short-Wave Radiation in Spain. Atmos. Res. 2018, 205, 18–25. [CrossRef]

14. Golovko, V.F. Modeling Radiation Absorption by Water Vapor in the Atmosphere within the 0- to 20,000-cm−1

Spectral Range. In Proceedings of the Sixth International Symposium on Atmospheric and Ocean Optics,Tomsk, Russia, 19 November 1999; p. 632. [CrossRef]

15. Kawamoto, K.; Hayasaka, T. Relative Contributions to Surface Shortwave Irradiance over China: A NewIndex of Potential Radiative Forcing. Geophys. Res. Lett. 2008, 35. [CrossRef]

16. Haywood, J.M.; Bellouin, N.; Jones, A.; Boucher, O.; Wild, M.; Shine, K.P. The Roles of Aerosol, Water Vaporand Cloud in Future Global Dimming/Brightening. J. Geophys. Res. 2011, 116. [CrossRef]

17. Soden, B.J.; Wetherald, R.T.; Stenchikov, G.L.; Robock, A. Global Cooling After the Eruption of MountPinatubo: A Test of Climate Feedback by Water Vapor. Science 2002, 296, 727–730. [CrossRef]

18. Huang, Y.; Ramaswamy, V.; Soden, B. An Investigation of the Sensitivity of the Clear-Sky OutgoingLongwave Radiation to Atmospheric Temperature and Water Vapor. J. Geophys. Res. 2007, 112. [CrossRef]

19. Viúdez-Mora, A.; Calbó, J.; González, J.A.; Jiménez, M.A. Modeling Atmospheric Longwave Radiation atthe Surface under Cloudless Skies. J. Geophys. Res. 2009, 114. [CrossRef]

20. Firsov, K.M.; Chesnokova, T.Y.; Bobrov, E.V. The Role of the Water Vapor Continuum Absorption in nearGround Long-Wave Radiation Processes of the Lower Volga Region. Atmos. Ocean. Opt. 2015, 28, 1–8.[CrossRef]

110 Apendice A. Artıculos

Remote Sens. 2020, 12, 1307 11 of 12

21. García, R.D.; Barreto, A.; Cuevas, E.; Gröbner, J.; García, O.E.; Gómez-Peláez, A.; Romero-Campos, P.M.;Redondas, A.; Cachorro, V.E.; Ramos, R. Comparison of Observed and Modeled Cloud-Free LongwaveDownward Radiation (2010–2016) at the High Mountain BSRN Izaña Station. Geosci. Model Dev. 2018,11, 2139–2152. [CrossRef]

22. Mateos, D.; Antón, M.; Toledano, C.; Cachorro, V.E.; Alados-Arboledas, L.; Sorribas, M.; Costa, M.J.;Baldasano, J.M. Aerosol Radiative Effects in the Ultraviolet, Visible, and near-Infrared Spectral RangesUsing Long-Term Aerosol Data Series over the Iberian Peninsula. Atmos. Chem. Phys. 2014, 14, 13497–13514.[CrossRef]

23. Mateos, D.; Antón, M.; Sanchez-Lorenzo, A.; Calbó, J.; Wild, M. Long-Term Changes in the Radiative Effectsof Aerosols and Clouds in a Mid-Latitude Region (1985–2010). Glob. Planetary Chang. 2013, 111, 288–295.[CrossRef]

24. Mateos, D.; Sanchez-Lorenzo, A.; Antón, M.; Cachorro, V.E.; Calbó, J.; Costa, M.J.; Torres, B.; Wild, M.Quantifying the Respective Roles of Aerosols and Clouds in the Strong Brightening since the Early 2000sover the Iberian Peninsula. J. Geophys. Res. Atmos. 2014, 119, 10382–10393. [CrossRef]

25. Antón, M.; Mateos, D. Shortwave Radiative Forcing Due to Long-Term Changes of Total Ozone Columnover the Iberian Peninsula. Atmos. Environ. 2013, 81, 532–537. [CrossRef]

26. Antón, M.; Mateos, D.; Román, R.; Valenzuela, A.; Alados-Arboledas, L.; Olmo, F.J. A Method to Determinethe Ozone Radiative Forcing in the Ultraviolet Range from Experimental Data. J. Geophys. Res. Atmos. 2014,119, 1860–1873. [CrossRef]

27. Bevis, M.; Businger, S.; Herring, T.A.; Rocken, C.; Anthes, R.A.; Ware, R.H. GPS Meteorology: Remote Sensingof Atmospheric Water Vapor Using the Global Positioning System. J. Geophys. Res. 1992, 97, 15787–15801.[CrossRef]

28. Boehm, J.; Niell, A.; Tregoning, P.; Schuh, H. Global Mapping Function (GMF): A New Empirical MappingFunction Based on Numerical Weather Model Data. Geophys. Res. Lett. 2006, 33, 3–6. [CrossRef]

29. Boehm, J.; Werl, B.; Schuh, H. Troposphere Mapping Functions for GPS and Very Long BaselineInterferometry from European Centre for Medium-Range Weather Forecasts Operational Analysis Data.J. Geophys. Res. Solid Earth 2006, 111, 1–9. [CrossRef]

30. Niell, A.E. Improved Atmospheric Mapping Functions for VLBI and GPS. Earth Planets Space 2000,52, 699–702. [CrossRef]

31. Saastamoinen, J. Atmospheric Correction for the Troposphere and Stratosphere in Radio Ranging Satellites.In Geophysical Monograph Series; Henriksen, S.W., Mancini, A., Chovitz, B.H., Eds.; American GeophysicalUnion: Washington, DC, USA, 1972; pp. 247–251. [CrossRef]

32. Dee, D.P.; Uppala, S.M.; Simmons, A.J.; Berrisford, P.; Poli, P.; Kobayashi, S.; Andrae, U.; Balmaseda, M.A.;Balsamo, G.; Bauer, P.; et al. The ERA-Interim Reanalysis: Configuration and Performance of the DataAssimilation System. Q. J. R. Meteorol. Soc. 2011, 137, 553–597. [CrossRef]

33. Ricchiazzi, P.; Yang, S.; Gautier, C.; Sowle, D. SBDART: A Research and Teaching Software Tool forPlane-Parallel Radiative Transfer in the Earth’s Atmosphere. Bull. Am. Meteorol. Soc. 1998, 79, 2101–2114.[CrossRef]

34. WMO. Report of the GCOS Reference Upper-Air Network Implementation Meeting; Report GCOS-121; WorldMeteorological Organization (WMO): Lindenberg, Germany, 2008.

35. McClatchey, R.A.; Fenn, R.W.; Selby, J.E.A.; Volz, F.E.; Garing, J.S. Optical Properties of the Atmosphere, 3rd ed.;AFCRL Environment Research Papers; AFCRL: Cambridge, MA, USA, 1972.

36. Marty, C. Downward Longwave Irradiance Uncertainty under Arctic Atmospheres: Measurements andModeling. J. Geophys. Res. 2003, 108. [CrossRef]

37. Kendall, M.G.; Stuart, A.; Ord, J.K. The Advanced Theory of Statistics in 3 Volumes 1 1; OCLC: 1071028235;Griffin: London, UK, 1994.

38. Mann, H.B. Nonparametric Tests Against Trend. Econometrica 1945, 13, 245. [CrossRef]39. Sen, P.K. Estimates of the Regression Coefficient Based on Kendall’s Tau. J. Am. Stat. Assoc. 1968, 63, 1379.

[CrossRef]40. Bennouna, Y.S.; Torres, B.; Cachorro, V.E.; Ortiz de Galisteo, J.P.; Toledano, C. The Evaluation of the Integrated

Water Vapour Annual Cycle over the Iberian Peninsula from EOS-MODIS against Different Ground-BasedTechniques. Q. J. R. Meteorol. Soc. 2013, 139, 1935–1956. [CrossRef]

A.6. Artıculo 6 111

Remote Sens. 2020, 12, 1307 12 of 12

41. Román, R.; Antón, M.; Cachorro, V.; Loyola, D.; Ortiz de Galisteo, J.; de Frutos, A.; Romero-Campos, P.Comparison of Total Water Vapor Column from GOME-2 on MetOp-A against Ground-Based GPSMeasurements at the Iberian Peninsula. Sci. Total Environ. 2015, 533, 317–328. [CrossRef]

42. Ortiz de Galisteo, J.P.; Bennouna, Y.; Toledano, C.; Cachorro, V.; Romero, P.; Andrés, M.I.; Torres, B. Analysisof the Annual Cycle of the Precipitable Water Vapour over Spain from 10-Year Homogenized Series of GPSData. Q. J. R. Meteorol. Soc. 2014, 140, 397–406. [CrossRef]

43. Ning, T.; Elgered, G. Trends in the Atmospheric Water Vapour Estimated from GPS Data for DifferentElevation Cutoff Angles. Atmos. Meas. Tech. Discuss. 2018, 1–25. [CrossRef]

44. Chen, B.; Liu, Z. Global Water Vapor Variability and Trend from the Latest 36 Year (1979 to 2014) Data ofECMWF and NCEP Reanalyses, Radiosonde, GPS, and Microwave Satellite. J. Geophys. Res. Atmos. 2016,121, 11,442–11,462. [CrossRef]

45. Vicente-Serrano, S.M.; Azorin-Molina, C.; Sanchez-Lorenzo, A.; Morán-Tejeda, E.; Lorenzo-Lacruz, J.;Revuelto, J.; López-Moreno, J.I.; Espejo, F. Temporal Evolution of Surface Humidity in Spain: Recent Trendsand Possible Physical Mechanisms. Clim. Dyn. 2014, 42, 2655–2674. [CrossRef]

46. Moratiel, R.; Soriano, B.; Centeno, A.; Spano, D.; Snyder, R. Wet-Bulb, Dew Point, and Air TemperatureTrends in Spain. Theor. Appl. Climatol. 2017, 130, 419–434. [CrossRef]

47. Kvalevåg, M.M.; Myhre, G. Human Impact on Direct and Diffuse Solar Radiation during the Industrial Era.J. Clim. 2007, 20, 4874–4883. [CrossRef]

48. Wild, M. Global Dimming and Brightening: A Review. J. Geophys. Res. 2009, 114. [CrossRef]49. Wild, M. Decadal Changes in Radiative Fluxes at Land and Ocean Surfaces and Their Relevance for Global

Warming. Wiley Interdiscip. Rev. Clim. Chang. 2016, 7, 91–107. [CrossRef]50. R Core Team. R: A Language and Environment for Statistical Computing; R Foundation for Statistical Computing:

Vienna, Austria, 2016.51. Wickham, H. Tidyverse: Easily Install and Load the ’Tidyverse’; R package version 1.2.1; R Software Inc.: Miami,

FL, USA, 2017.52. Grolemund, G.; Wickham, H. Dates and Times Made Easy with lubridate. J. Stat. Softw. 2011, 40, 1–25.

[CrossRef]53. Kassambara, A. Ggpubr: ’ggplot2’ Based Publication Ready Plots; R package version 0.2.4; R Software Inc.:

Miami, FL, USA, 2019.54. Zeileis, A.; Grothendieck, G. zoo: S3 Infrastructure for Regular and Irregular Time Series. J. Stat. Softw. 2005,

14, 1–27. [CrossRef]55. Pohlert, T. Trend: Non-Parametric Trend Tests and Change-Point Detection; R package version 1.1.2;

R Software Inc.: Miami, FL, USA, 2020.

c© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open accessarticle distributed under the terms and conditions of the Creative Commons Attribution(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

112 Apendice A. Artıculos

Apendice B

Abreviaturas y acronimos

AEMet Agencia Estatal de MeteorologıaAIRS Atmospheric Infrared SounderAMC-DOAS DOAS con masa optica corregidaAMF factor de masa opticaCF Fraccion de cubierta nubosa

DOAS Espectrografıa de Absorcion Diferencial OpticaERA Reanalisis del Centro Europeo de Previsiones Meteo-

rologicas a Plazo MedioEUREF Regional Reference Frame Sub-Commission for EuropeGFZ Centro de Investigacion Aleman de GeocienciasGNSS Global Navigation Satellite SystemGOME-2 Global Ozone Monitoring Instrument - 2GPS Global Positioning SystemGRUAN GCOS Reference Upper-Air NetworkHR Tasa de calentamientoI InteriorIQR Rango intercuartılicoIR InfrarrojoIWV Vapor de agua integradoLW Onda largaMBE Sesgo medioMODIS Moderate Resolution Imaging SpectroradiometerMS Mar Mediterraneo

114 Apendice B. Abreviaturas y acronimos

NIR Infrarrojo cercanoOMI Ozone Monitoring InstrumentPWV Vapor de agua precipitableRMSE Error cuadratico medioSBDART Santa Barbara’s Disort Radiative TransferSCIAMACHY SCanning Imaging Absorption SpectroMeter for At-

mospheric CHartographYSD Desviacion estandarSEVIRI Spinning Enhanced Visible and Infrared ImagerSTD Retraso troposferico inclinadoSW Onda cortaSZA Angulo solar cenitalUTC Tiempo universal coordinadoWMO Organizacion Meteorologica MundialWVEFF Eficiencia radiativa del vapor de aguaWVMR Razon de mezcla del vapor de aguaWVRE Efecto radiativo del vapor de aguaZHD Retraso hidrostatico cenitalZTD Retraso troposferico cenitalZWD Retraso humedo cenitalNA Atlantico Norte

Apendice C

Erratas de Artıculos

En el Artıculo A.2, la Figura 10 es incorrecta. En ella aparece repetidala Figura 9 de ese mismo artıculo, cuando la correcta serıa la Figura C.1.

▰

▰

▰

▰

▰

▰

▰ ▰▰

▰▶

▶

▶▶

▶

▶▶

▶

◆

◆

◆ ◆

◆

◆

◆ ◆

◼

◼◼

◼

◼

◼

◼◼

◼

◼

35

40

45

50

55

0.25 0.50 0.75CF

IQR

[δ(%

)]

Satellite

◼

▰

◆

▶

AIRS (CF ≠ 0)

GOME-2 (CF ≠ 0)

MODIS-Aqua (CF ≠ 0)

MODIS-Terra (CF ≠ 0)

Figura C.1. Grafica que deberıa aparecer en la Figura 10 del Artıculo A.2.