r quA comparison of two app

Jorge G. Alvarez-Romero a, *, ScoJohnathan Kool a, e, Jon Brodie b

a Australian Research Council Centre of Excellence for Cb Centre for Tropical Water and Aquatic Ecosystem ReseQLD 4811, Australiac CSIRO Land and Water, GPO Box 1666 Canberra, ACTd f Victoria

ce Divisi

igher accuracy thannge of variation inusing the spatially-g a calibrated SDR.e to explicit repre-e oodplain extentn in SedNet, wouldifferences betweenntly inuenced the

. All rights reserved.

Human-induced changes in ows of nutrients, sediments, andfresh water are major threats to coastal and marine ecosystems

trient enrichmentof coastal and marine waters associated with agricultural fertilizerscreates eutrophic and hypoxic or anoxic conditions that affect thefunctioning of marine ecosystems and the status of biodiversity andhuman health (Diaz and Rosenberg, 2008; Gabric and Bell, 1993).Loads of sediments derived from land clearing, urbanization, andagriculture have signicantly increased in many regions (Walling,2006) and are a major threat to vulnerable ecosystems such ascoral reefs and seagrass beds (Cabaco et al., 2008; Maina et al.,2011). In contrast, damming of rivers reduces the delivery of

* Corresponding author. Tel.: 61 07 4781 6517; fax: 61 07 4781 6722.E-mail addresses: jorge.alvarez.romero@gmail.com, jorge.alvarezromero@jcu.

edu.au (J.G. Alvarez-Romero), Scott.Wilkinson@csiro.au (S.N. Wilkinson), bob.pressey@jcu.edu.au (R.L. Pressey), nban@uvic.ca (N.C. Ban), Johnathan.Kool@ga.

Contents lists availab

Journal of Environm

journal homepage: www.els

Journal of Environmental Management 146 (2014) 164e178gov.au (J. Kool), jon.brodie@jcu.edu.au (J. Brodie).Land-based pollutionSystematic conservation planning

between models. SedNet consistently estimated spatial variations in runoff with hN-SPECT and modeled suspended sediment (TSS) loads mostly within the raobserved loads. N-SPECT overestimated TSS loads by orders of magnitude whendistributed sediment delivery ratio (SDR), but outperformed SedNet when usinDifferences in subcatchments' contribution to pollutant loads were principally dusentation of sediment sinks and particulate nutrients by SedNet. Improving thmodel, and constraining erosion estimates by local data including gully erosioimprove results of this model and help identify effective management responses. Dmodels in the patterns of modeled pollutant supply were modest, but signicaprioritization of subcatchments for management.

2014 Elsevier Ltd

1. Introduction worldwide (Doney, 2010; Tilman et al., 2001). NuIntegrated land-sea planningMarine conservation

assessed their application in selected catchments of the Gulf of California, where managing land-based threats to marine ecosystems is a priority. We found important differences in performanceSchool of Environmental Studies, University oe Geoscience Australia, Environmental Geoscien

a r t i c l e i n f o

Article history:Received 12 December 2012Received in revised form23 June 2014Accepted 5 July 2014Available online 28 August 2014

Keywords:Catchment managementWater qualityhttp://dx.doi.org/10.1016/j.jenvman.2014.07.0070301-4797/ 2014 Elsevier Ltd. All rights reserved.roaches

tt N. Wilkinson c, Robert L. Pressey a, Natalie C. Ban a, d,

oral Reef Studies, James Cook University, Townsville, QLD 4811, Australiaarch (TropWater), Catchment to Reef Research Group, James Cook University, Townsville,

2601, Australia, PO Box 1700 STN CSC, Victoria, BC V8W 2Y2, Canadaon, National Earth and Marine Observations Group, GPO Box 378 Canberra, ACT 2601, Australia

a b s t r a c t

Human-induced changes in ows of water, nutrients, and sediments have impacts on marine eco-systems. Quantifying these changes to systematically allocate management actions is a priority formany areas worldwide. Modeling nutrient and sediment loads and contributions from subcatchmentscan inform prioritization of management interventions to mitigate the impacts of land-based pollu-tion on marine ecosystems. Among the catchment models appropriate for large-scale applications, N-SPECT and SedNet have been used to prioritize areas for management of water quality in coastal-marine ecosystems. However, an assessment of their relative performance, parameterization, andutility for regional-scale planning is needed. We examined how these considerations can inuence thechoice between the two models and the areas identied as priorities for management actions. Wemanagement of wate ality in coastal-marine ecosystems:

Modeling catchment nutrients and sediment loads to inform regionalle at ScienceDirect

ental Management

evier .com/locate/ jenvman

ironmriver-borne sediment and nutrients to the sea (Walling, 2006), butalso has potential adverse effects, signicantly altering biogeo-chemical cycles and species composition in estuarine and marineecosystems (Humborg et al., 2000). Managing such land-basedthreats is increasingly recognized as a crucial component ofmaintaining healthy coastal and marine ecosystems globally(Halpern et al., 2009; Maina et al., 2011).

Regional studies to identify and assess land-based threats tocoastal-marine ecosystems and to explore options to prevent ormitigate these threats are urgently needed (e.g., Burke and Sugg,2006; McKergow et al., 2005). Catchment models are powerfultools to explore management options to improve water quality(Walling et al., 2011) because they link sources of pollutants (e.g.,sediment, nutrients) to affected areas and ecosystems. Further-more, they facilitate identication of dominant processes associ-ated with production and delivery of pollutants (Drewry et al.,2006; Walling et al., 2011), and thus help to determine appro-priate strategies to minimize downstream impacts. Applicationsof catchment modeling for water-quality control include identi-fying erosion hotspots and estimating pollutant loads delivered tothe sea, which can then be incorporated into river plume modelsto assess vulnerability of coastal-marine ecosystems to land-based threats (Alvarez-Romero et al., 2013a; Burke and Sugg,2006). Mapping major sources of pollutants within catchmentsis thus needed to guide management to prevent or reducedownstream impacts (Brodie et al., 2009). Managementactions include implementation of best practices in agricultureand livestock management, water treatment, restoration ofriparian vegetation, protection of erosion-prone areas, andgully stabilization (Thorburn et al., 2013; Wilkinson and Brodie,2011).

Different modeling approaches have been used to estimate thesupply and delivery of sediment and nutrient loads, varying incomplexity, data requirements, and intended spatial and temporalscales of application (see Borah and Bera, 2003; Merritt et al., 2003for reviews on sediment and nutrient export models). Formodeling approaches to be relevant for identifying and assessingland-based threats to coastal-marine ecosystems, several consid-erations are important: intended spatial and temporal scales, dataavailability vs. data requirements of models (climatic, ow, in situwater-quality sampling), existing technical expertise, and con-straints on time and budget (Walling et al., 2011). Commonly, owdata and water-quality sampling or monitoring are spatiallyincomplete (covering only some parts of regions) and temporallyfragmented (spanning limited and/or different time periods indifferent places). Data gaps, together with limited technical ca-pacity, have constrained applications of more complex and real-istic models applicable to regional scales.

Manymodeling approaches have been developed for local, eld-scale applications (Jetten et al., 1999), including event-based ordaily time-step models that are appropriate for short-term andlocal management or ongoing water-quality monitoring (e.g.,SWAT: Arnold and Fohrer, 2005; AGNPS: Young et al., 1989). Thesemodels, however, are generally too data-intensive and detailed tobe applicable for large or multiple catchments. There are fewermodels appropriate for large-scale applications, although severalregional catchment models are pertinent (e.g., N-SPECT: Eslingeret al., 2005; SWIM: Krysanova et al., 2005; SedNet: Wilkinson,2008). SedNet and N-SPECT have been used in extensive applica-tions to estimate end-of-river loads of sediment and nutrients, andto target catchmentmanagement tominimize impacts of terrestrialrunoff on coastal ecosystems (e.g., N-SPECT eMesoamerican Reef:Burke and Sugg, 2006; Madagascar: Maina et al., 2012; SedNet eGreat Barrier Reef: McKergow et al., 2005). Both models are

J.G. Alvarez-Romero et al. / Journal of Envappropriate to estimate long-term annual pollutant supply andend-of-river loads and are also considered to be good compromisesin terms of model scope, data requirements, and performance.

The aim of this study is to compare two prominent regionalcatchment models (N-SPECT: NOAA, 2008; SedNet: Wilkinsonet al., 2004) in terms of their performance, data requirements,ease of implementation, and utility for regional-scale planning.Data required to parameterize these models (as well as availabilityof these data) vary between studies, as do their outputs and ap-plications to identify specic areas for catchmentmanagement. Ourstudy thus aims to examine how these considerations can inuencethe choice between these models and the areas identied as pri-orities for management actions. Here we apply the models tocatchments draining to the Gulf of California, Mexico, where thereis an urgent need to study and incorporate land-based threats tomarine ecosystems into conservation planning (Alvarez-Romeroet al., 2013b). Our study area also has highly variable hydrologicaland climatic conditions and limited data, making it a good test casefor many similar regions around the world. Previous studies haveexamined nutrient enrichment of coastal waters in the Gulf ofCalifornia, but focused on irrigated agriculture (Ahrens et al., 2008),and local scales (Christensen et al., 2006). Our study investigatesthe use of catchment models to estimate runoff at a regional(whole-of-catchment) scale. To our knowledge, our study is therst to apply these two models within the same region and assessthe differences in their outputs, and implications for managementdecisions.

2. Methods and data

We compared twomodels commonly used to prioritize areas formanagement of water quality in coastal-marine ecosystems: N-SPECT and SedNet/ANNEX (hereafter we refer to SedNet whenreferring to both the base model used for modeling sedimentsources and transport and the ANNEX module employed whenmodeling nutrients). We examined the differences in parameteri-zation of the models, as well as the capabilities, performance, anddifferences in their outputs in terms of pollutant supply patternsand loads. We explore the differences in the allocation of man-agement actions based on the outputs of each model and discussthe potential consequences of management decisions, thusproviding criteria to guide managers when selecting and parame-terizing a model.

2.1. Study area

The Gulf of California is a marine ecosystem globally recognizedfor its rich and unique sea life, and is currently threatened by sea-and land-based activities (Lluch-Cota et al., 2007). Human popu-lation density is relatively low in the catchments draining into theGulf of California, but is rapidly increasing, with associated in-creases in threats to the marine environment (Lluch-Cota et al.,2007). While the western coast remains comparatively undis-turbed, many eastern coastal areas are affected by land-basedthreats, including agriculture, urbanization, aquaculture, anddamming (Lluch-Cota et al., 2007; Paez-Osuna and Ruiz-Fernandez,2005). Existing studies in the region show that the impacts of land-based pollution can extend hundreds of kilometers from the coast(Beman et al., 2005).

This study focuses on selected catchments draining into the Gulfof California, including one of the most important agricultural re-gions in Mexico, the Yaqui Valley. The region stands out globally forthe intensive use of agrochemicals and fertilizer (Ahrens et al.,2008). Effects of terrestrial runoff (e.g., associated with the inten-sive use of fertilizers) have been observed in offshore marine areas

ental Management 146 (2014) 164e178 165within the Gulf (Beman et al., 2005). This is particularly relevant

because multiple adjacent coastal-marine areas are considerednational conservation priorities (SEMARNAT, 2006). We selectedthese catchments (Fig. 1) for several reasons: they comprise a va-riety of catchment sizes, fromvery large (e.g., Yaqui River andMayoRiver: 67,629 and 13,303 km2, respectively) to relatively smallcoastal catchments (~50 km2), thus exemplifying the variety oftransport and depositional processes associated with catchmentsize (Prosser et al., 2001); major rivers drain directly into marinemanagement units identied as high and very high priorities formarine conservation (for further information regarding the

regional marine spatial planning exercise dening these marinepriorities see: SEMARNAT, 2006); there is a marked climatic vari-ation and limited stream-gauge data (with drier areas being poorlyrecorded) across the region, providing a test of the models' capa-bilities to represent variations in runoff; and the selected catch-ments include all vegetation types found within the totalwatershed of the Gulf, which is important if a Gulf-wide study is tobe undertaken later. All these characteristics make the study regiona good case to test the application of these two approaches tocatchment modeling.

catitantt in

J.G. Alvarez-Romero et al. / Journal of Environmental Management 146 (2014) 164e178166Fig. 1. Key features of the study area relevant to catchment modeling: A) Geographic loCalifornia, also showing major reservoirs (used in SedNet to estimate changes in polluconservation priority level of adjacent marine management units; C) The strong gradien

to precipitation (Eo/P) ratio; and D) Main vegetation cover/land use classes depicting degreand pasture) potentially associated with spatial patterns of pollutant supply; the percentagon; B) The selected catchments comprising the study area, all draining into the Gulf ofloads due to deposition of sediments and particulate nutrients), gauge stations, and

dryness across the study area, as illustrated by the ratio of potential evapotranspiration

e of human modication of the landscape (e.g., conversion of natural areas to croplande of area occupied by each land use is indicated in brackets.

ironm2.2. Catchment model descriptions and data requirements

N-SPECT and SedNet have common input datasets and similaroutputs but differ in some important aspects (Fig. 2). Both modelsrely on spatial datasets and process conceptualizations. SedNetoutputs (including runoff, erosion, and nutrient supply) are lumpedwithin subcatchments. N-SPECT calculations are performed on acell-by-cell (raster grid pixel) basis, and are hence limited by thecoarse-resolution datasets used here.

Runoff calculations are one fundamental difference between N-SPECT and SedNet. Runoff estimates in N-SPECT are based on theUSDA Soil Conservation Service Curve Number (CN) method, awidely used method for estimating changes in rainfall runoff basedon the amount of precipitation and inltration, which is deter-mined by land use and vegetation cover (hereafter land use), soiltype, surface retention, and impervious surface (USDA, 1986).Although the method was originally designed for single-stormevents and small catchments, it can be scaled to nd averageannual runoff values (NOAA, 2008) and has been used in large-scaleor regional studies (e.g., Burke and Sugg, 2006). In contrast, SedNetrunoff calculations are based on regionalization of the spatial pat-terns in a water balance to predict mean annual runoff (and severalstatistics of daily ow variability) following least-squares re-gressions based on observed ow data from gauge stations andcatchment characteristics, including annual rainfall and potentialevapotranspiration (Wilkinson et al., 2006).

Both models estimate soil loss from hillslope erosion based onthe Revised Universal Soil Loss Equation (RUSLE), which estimatesannual soil loss based on a number of landscape characteristics (i.e.,rainfall erosivity, soil erodibility, length-slope, and vegetationcover) considered to be the main determinants of soil loss (Renardet al., 1997). However, the sediment delivery ratios (SDRs) appliedby both models are quite different. SedNet uses a constant value(default is 0.05) representing a priori knowledge of how erosionestimates relate to sediment delivery. This can be a limitation ofSedNet when empirical data are unavailable or SDRs are highlyvariable across the study region. In contrast, N-SPECT automaticallycreates a spatially-distributed SDR based on an empirical rela-tionship between drainage area (i.e., raster cell size), relief-lengthratio (calculated based on a digital elevation model and ow di-rection grid), and curve number (Williams, 1977). Additionalerosion processes incorporated by SedNet, but not N-SPECT, includegully and river bank erosion.

Calculations of dissolved-nutrient supply in both N-SPECT andSedNet follow the export or runoff coefcient approach (Johnes,1996), which is based on mean runoff concentrations of partic-ular pollutants and nutrients from different land uses. Mean runoffconcentrations are commonly referred to as event mean concen-trations (EMC), but see Bartley et al. (2012) for a discussion ondifferences in methods used to estimate these values. Event meanconcentration coefcients are combined with an estimated runoffvolume to calculate the pollutant supply. In addition, SedNet es-timates the supply and export of particulate nutrients associatedwith hillslope, gully, and bank erosion. For both sediment (sus-pended and bedload material) and particulate nutrients,SedNet also represents reservoirs and oodplains as sinks withinthe river network, and the amount of material actually exportedfrom the river network is net of the upstream sources and sinks.Sinks (SedNet only) include loss of nutrients due to denitricationin reservoirs and lakes, and accumulation or deposition of sedi-ment in river oodplains, reservoirs and lakes, as well as deposi-tion of in-channel bed material. N-SPECT does not considerdepositional or accumulation processes and thus its resultsrepresent the supply of sediment and nutrients; this is a major

J.G. Alvarez-Romero et al. / Journal of Envlimitation of the model.2.3. Model parameterization

In this section, we describe parameterization of models,including key data sources, processing and application for eachmajor step (as described in Fig. 2), indicating, when applicable, datalimitations, adjustments and whether these are used in both oronly one model. A detailed description of data preparation andmodel settings are provided as Supplementary Material.

2.3.1. Catchment and stream network delineationA digital elevation model (DEM) was used for delineating

streams and subcatchments and routing of pollutant transport. Weused a hydrologically-corrected version of the 30 m resolutionASTERGlobal DEM (METI-NASA, 2009), which was resampled usingESRI's ArcMap Spatial Analyst using bicubic interpolation to 240 m.Similar resolutions have been used for other regional catchmentmodeling exercises and provide a good balance between the extentof the study (~600,000 km2) and matching of resolution to that ofother input data.

Bothmodels delineate catchments automatically, based on user-dened parameters to set the subcatchment size. For N-SPECT weused the small subcatchment option, which resulted in 27 sub-catchments with mean size of 3774 km2 (range 385 to 8652 km2).For SedNet, we set the drainage area threshold at 40 km2, whichresulted in 1655 subcatchments, with a mean size of ~60 km2

(STD 39 km2). This ensured that most small coastal catchmentswere included in the SedNet study area. Differences between thestream network delineated using N-SPECT and SedNet were mini-mal, so we used the smaller subcatchments from SedNet to sum-marize the cell-based (i.e., local sources) results of N-SPECT.

2.3.2. Runoff and ow calculationsAnnual precipitation (P) is used in both N-SPECT and SedNet to

calculate runoff. We used the global high-resolution (~1 km) annualprecipitation grid available online from WorldClim (http://worldclim.org), which provides a good resolution for regional-scale catchment modeling. The number of rain days, required byN-SPECT to adjust the initial abstraction (i.e., water retained indepressions, taken up by vegetation, and evaporated or inltrated),was calculated based on the curve number equation (USDA, 1986).The modeled runoff was very sensitive to this parameter, and it wasused to calibrate the model to observed (i.e., gauge data) runoff(Supplementary Material).

The hydrologic soil group (HSG, i.e., the permeability or inl-tration capacity of the soils) is used by N-SPECT, in combinationwith land use, to calculate curve numbers. We assigned HSG classesbased on the classication of soil units proposed by CNA (1987)using a national map of soil units (INIFAP-CONABIO, 1995). Whentwo HSGs were assigned to a particular soil unit, the lowestpermeability (maximum runoff) class was used, as suggested byEslinger et al. (2005). Curve numbers (used in N-SPECT for runoffcalculations) are determined by combining the HSG and land use.Higher curve number values are given for landscapes with moreimpervious cover (e.g., surface soils with high clay content or landswith sparser vegetation). We matched the land use classes of ourstudy area with the Coastal Change Analysis Program (C-CAP) landcover classes (USDA, 1986).

Potential evapotranspiration (PET) is used in SedNet to region-alize mean annual discharge. We used the Global PotentialEvapotranspiration (Global-PET) dataset (available from http://www.cgiar-csi.org). For SedNet we also processed daily andmonthly records from data from gauge stations obtained from theBANDAS hydrometric database (available from http://www.imta.gob.mx). Of the 19 gauges within the study area, only 11 (Fig. 1B)

ental Management 146 (2014) 164e178 167had sufcient data for the model. Our criteria for sufciency were:

J.G. Alvarez-Romero et al. / Journal of Environmental Management 146 (2014) 164e178168

minimum number of years recorded was 20 years, or 15 if no otherstations were available around the area; and at least 10 years ofrecords should be uninterrupted. The temporality of gauge recordsvaried between stations, but was mostly constrained from the

we combined the polygons identied as riparian vegetation fromthe three available maps (1976, 1993, and 2000) of the land-usetime series. We assigned polygons to riparian vegetation whenthese were classied as riparian in the three maps or when theywere identied as riparian in any of the maps and as a differentnatural vegetation type in the other(s). Second, we identiedriparian vegetation as the portions of tree-dominated land-useclasses (2000 maps only) within an 80 m buffer along the stream

left panel (gray area) depicts modeling steps in N-SPECT. The two columns on the rightbox shapes and shadings are used to describe the different elements (i.e., input data, pa-

Table 1Regionalized regression equations used to calculate erosivity us-ing annual precipitation (P).

Regiona Equation

J.G. Alvarez-Romero et al. / Journal of Environmental Management 146 (2014) 164e178 1691960s to early 2000s.

2.3.3. Dissolved pollutants and nutrientsLand use is employed to estimate the pollutant loads in both

models and also to calculate potential runoff (N-SPECT only). Weused the most up-to-date national-scale (1:250,000) land usedataset available (Fig. 1D), which corresponds to the year 2000(SEMARNAT-UNAM, 2002). This dataset was corrected for classi-cation inconsistencies because different methods were used tocreate it (Velazquez et al., 2002).

We calculated pollutant EMC based on a literature review onreported EMC for similar vegetation types, mostly restricted toNorth America, Australia, and South America. Values outside thecommonly observed ranges in EMC for each land use wereexcluded. In particular, outlying values for total nitrogen fromforest land use in areas with high atmospheric deposition of ni-trogen from industrial and high intensity agricultural areas wereexcluded (there was no indication of high deposition in our studyarea). Studies from North and South America (Lewis, 2002; Perakisand Hedin, 2002), Australia (Bartley et al., 2012; Brodie andMitchell, 2005), and modeled data from the USA that accountedfor atmospheric deposition (Smith et al., 2003) provided guidancefor adjusting EMC values to remove the effect of deposition fornatural vegetation land use classes. Values for agricultural landuses were divided into different classes to reect differences infertilizer use (see Supplementary Material for details and selectedEMC values).

2.3.4. Hillslope erosionBoth models use the RUSLE to estimate hillslope erosion based

on rainfall erosivity, soil erodibility, length-slope factor, and vege-tation cover. Rainfall erosivity (R) was regionalized based on pre-cipitation distribution and occurrence characteristics (duration,intensity, number of events) according to historical records avail-able from climatological stations (q.v., Diodato and Bellocchi, 2007;Mikhailova et al., 1997). We used a 1 km annual precipitation grid(Section 2.3.2) to estimate R based on the regionalized erosivitymap and regression equations (Table 1) proposed by Cortes-Torres(1991) and modied by SEMARNAT-UACH (2002). Six of the four-teen erosivity regions described by Cortes-Torres (1991) occurwithin our study area.

Soil erodibility (K) was calculated using the method proposedby FAO (1980), with soil data widely used for erosion studies inMexico (SEMARNAT-UACH, 2002). For soil units with many hori-zons (i.e., soil layers with different characteristics) we calculated aweighted average of K for all horizons combined (q.v., FAO, 1980;NOAA, 2008).

The length-slope (LS) factor is calculated automatically in N-SPECT during the process of delineating catchments and streamnetworks using ArcMap Spatial Analyst.We used the LS grid createdin N-SPECT as input for soil loss (RUSLE) calculations in SedNet.

Vegetation cover (C) is used to solve the soil loss equation toestimate sheet-wash and rill erosion. We used N-SPECT default Cvalues as a reference, but adjusted the values based on a nationalerosion study for Mexico (SEMARNAT-UACH, 2002). Due to large

Fig. 2. Major steps, inputs and outputs of N-SPECT and SedNet/ANNEX models. Thecorrespond to SedNet/ANNEX. Common steps are presented side by side and different

rameters and outputs). Within each step, different font types and colors are used to indiccommon to both models or exclusive to one (depicted in italics and gray or white color forvariations in vegetation cover between different crops, we calcu-lated an average value for all cropland classes based on the pro-portional area covered by dominant crops within the study area.

2.3.5. Gully erosionGullies are potentially important sources of suspended sedi-

ment. In some regions, soil losses from these features can exceedthose originating from hillslope erosion (e.g., Caitcheon et al.,2012), thus signicantly contributing to sediment delivery to thecoast. Despite the signicance of this erosion process in someareas of Mexico (e.g., Mexican Central Highlands: Duvert et al.,2010; Gulf of Mexico's coast: Geissen et al., 2007), a nationalstudy suggests that the contribution of gullies to sediment de-livery is overall of lesser importance than hillslope erosion in ourstudy region (SEMARNAT-UACH, 2002). A local-scale examinationwithin our study area shows that hillslope erosion was about twoorders of magnitude higher than gully erosion (Descroix et al.,2008). Another study covering comparable vegetation types inthe Pacic coast of Mexico also found little presence (and hencecontribution to sediment delivery) of gullies (Cotler and Ortega-Larrocea, 2006). However, the presence of gullies and theircontribution to soil loss and sediment transport could varysignicantly across the region. Despite the capability of SedNet tomodel gully erosion, spatial data on gully density across the studyarea were not available; this is a limitation of our model appli-cation. N-SPECT does not model gully erosion.

2.3.6. River bank erosionSedNet estimates bank erosion as being proportional to

bankfull stream power, but modies this relationship with twofactors: proportion of remnant riparian vegetation and oodplainwidth. A riparian vegetation grid is used to estimate rates of bankerosion based on the assumption that bank erosion decreases asthe proportion of woody riparian vegetation increases (averagedover each subcatchment), while bank erosion is also reducedexponentially where oodplain width is < 100 m, on the basisthat the availability of erodible soil becomes limiting under theseconditions (Wilkinson et al., 2004). We created a consistent ri-parian vegetation map across the study area in three steps. First,

II 3.45552$P 0.006470$P2IV 2.89594$P 0.002983$P2VI 6.68471$P 0.001680$P2V & VII 0.71508$P1. 30751

X 6.89375$P 0.000442$P2a Numerals indicate the original codes assigned by Cortes-

Torres (1991) to the erosivity regions of Mexico.ate differences between the models, i.e., when input data/parameters or outputs areinput or outputs, respectively).

ironmnetwork. Finally, we combined both maps into a single layer thatrepresented the maximum extent of current riparian vegetation.We describe preparation of the oodplain grid in Section 2.3.8.Due to lack of data specic to the region and smaller variation inthe other coefcients used in SedNet to estimate bank erosion(i.e., bank erosion coefcient, sediment bulk density, and pro-portion of ne sediment), we used the default values for ourmodeling exercise. Bank erosion is not included in N-SPECT.

2.3.7. Particulate nutrientsThe percentage of clay and nutrient content (only nitrogen for

our study) of soil units is required to estimate the amount of par-ticulate nutrients associated with soil erosion in SedNet (Wilkinsonet al., 2004). We used the Soil and Terrain Database for LatinAmerica and the Caribbean (SOTERLAC), which provides informa-tion on some of these parameters (Dijkshoorn et al., 2005) and hasbeen used for regional modeling with N-SPECT (Burke and Sugg,2006).

2.3.8. Deposition and loss of sediment and nutrientsReservoirs are a major sink for suspended sediments trans-

ported through rivers (Walling, 2006) and are used in SedNet toestimate sediment deposition based on an updated version of theBrune equation (Wilkinson et al., 2004). The surface area of reser-voirs, along with mean air temperature (also from WorldClimdataset), is also used in SedNet to estimate denitrication. Wemapped 11 reservoirs by combining the national reservoir database(CONAGUA, 2008) and the most recent land use map (i.e., 2000), inwhich water bodies were delineated. Reservoir capacity (GL) wasextracted from the database or estimated using ArcMap 3D Analyst(see Supplementary Material).

Along with reservoirs and lakes, river oodplains are landformswhere substantial volumes of suspended sediments accumulate(Walling et al., 2003) and thus are incorporated into SedNet toestimate sediment deposition. Due to lack of information on thespatial distribution of oodplains, we calculated the multi-resolution valley bottom index (MrVBF) to map at low-lyingareas, which can be related to depth of soil deposit (Gallant andDowling, 2003). We used ArcMap to identify the natural breaks inMrVBF values across the region, and following Wilkinson et al.(2009), we selected those areas with an index value 2.0 as po-tential oodplains. Thismodel can be used as a proxy in the absenceof data, but validation and model renements are advisable toimprove results. Sediment deposition in reservoirs and oodplainsis not considered in N-SPECT.

2.3.9. Sediment delivery ratioDue to the lack of empirical data for our study region we

determined the SDR based on studies undertaken in geographicallyproximate areas and catchments with similar characteristics.Norman (2007) studied awatershed in the U.S.-Mexican border andmodeled a spatially-distributed SDR with a mean of 0.19. A study inthe Gulf of Mexico (Martnez-Menez et al., 2001) calculatedempirical SDR values using measured sediment loads from 10 hy-drometric stations; the median SDR across studied subcatchmentswas 0.09. A large-scale application of SedNet in the Great BarrierReef catchments used an SDR of 0.1, producing results that wereconsistent with measured sediment loads, relative contribution oferosion processes, and observed deposition in oodplains. For ourstudy, using an SDR of 0.1 in SedNet provided a good t betweenmeasured and modeled sediment yields. For N-SPECT, wecompared the results of using the spatially-distributed SDR calcu-lated by the software (median 0.89) and a uniform SDR of 0.1. Usingthe spatially-distributed SDR overestimated sediment yields by

J.G. Alvarez-Romero et al. / Journal of Env170orders of magnitude, so we decided to use the same e uniform eSDR as in SedNet; this allowed us to directly compare the models'outputs.

2.4. Evaluation of performance and outputs of models

2.4.1. Performance of modelsWe summarized and compared differences in performance for

modeled vs. observed runoff and suspended sediments. Wecompared modeled runoff with discharge records from gauge sta-tions (IMTA, 2010). For each station we estimated the mean annualrunoff (based on monthly aggregated data) and the bootstrapped95th (percentile) condence interval (CI) using Pop-Tools' (Hood,2010) Monte Carlo simulation tool (data were resampled withreplacement, 10,000 replicates, a 0.05). Gauge-station data werefurther validated against reported mean annual runoff from Mex-ico's Water Commission (CNA, 2008). Of the 11 gauge stations withgood data in the BANDAS database, 7 stations (9008, 9011, 9018,9063, 9067, 9068 and 9089) also included suspended sedimentrecords (data were limited to a few years of record: median 15),which were summarized as annual sediment loads using monthlyaggregates. To compare modeled and recorded sediment load es-timates, we summarized annual sediment records (reported as m3)using monthly mean volume from data recorded in the gaugestations. To calculate total suspended sediment (TSS) loads (t y1)from gauge station data, we used the typical density of saturatedsediment (1.3 t m3) (O'Connor et al., 2003), and estimated themean and the bootstrapped 95th CI for the mean load following thesame procedure as for runoff. We plotted observed vs. modeleddata to assess agreement and to identify patterns of over- andunder-estimation. We calculated the mean square absolute error(RMSE) across the gauge set as an overall measure of model per-formance (Wilkinson et al., 2014).

2.4.2. Spatial pattern of supply and estimated loads of pollutantsWe summarized similarities and differences between models in

estimates of pollutant load and overall budget. We focused on TSSand nitrogen (dissolved inorganic nitrogen: DIN, and total nitrogen:TN) as examples of typical constituents considered in water qualitymodeling and monitoring.

First, we compared the results based on common outputs forboth models (i.e., supply of TSS and DIN from catchments andsubcatchments). This identied differences in spatial distribution ofsupply of pollutants resulting from inherent differences betweenmodels, rather than from processes not modeled by N-SPECT (e.g.,deposition, denitrication, and particulate nutrients).

Second, we compared results based on the nal outputs of eachmodel (i.e., considering both common outputs and processesmodeled only in SedNet) and quantied differences in end-of-riverpollutant loads, as well as the differences in modeled contributionsfrom specic subcatchments. We used percentage contributionfrom subcatchments to identify the subcatchments that werecontributing most to the region-wide pollutant loads. This relativemeasure was used to directly compare maps from both models andto explore the potential effects when prioritizing catchmentmanagement.

Finally, we assessed the potential consequences of using theoutputs of either model when targeting catchment management toreduce the delivery of pollutants to the sea. We used the decision-support tool Marxan (Ball et al., 2009) to identify subcatchmentsthat could be managed to reduce TSS and DIN loads. Marxan'soptimization algorithm nds solutions (sets of planning units) thatachieve dened targets at the lowest cost possible. We arbitrarilyset our management targets as a reduction of 5%, 10%, 15%, 20%, 25%and 30% of the total load for each pollutant (i.e., six scenarios per

ental Management 146 (2014) 164e178pollutant). The potential contribution of each planning unit

(subcatchment) to the achievement of targets was measured as itsTSS/DIN supply (no reservoirs), thus allowing direct comparison ofthe models. We generated solutions (sets of subcatchments thatachieved the stated targets) for TSS and DIN independently to showthe differences in prioritization for each pollutant. Due to lack ofinformation on economic costs of management actions in the studyarea, we used subcatchment area (km2) as a proxy for managementcosts (hereafter referred to as costs). We generated one hundredsolutions for each scenario and calculated the total cost (i.e.,summed area of selected subcatchments) of each solution and thevariation across solutions. We then used cost to compare the ef-fects of using the models to target management.

3. Results

3.1. Performance of models

Overall, SedNet consistently estimated spatial variations inrunoff with higher accuracy. Inmost cases, SedNet'smodeled valuesfell within the condence intervals for runoff calculated from gaugedata (Fig. 3A). We used the coefcient of efciency (E) to assess the

uniform SDR (0.1) in N-SPECT signicantly improved results (RMSE:102%) and estimated TSS loads with higher accuracy than SedNet.Estimated TSS loads derived using N-SPECT with a uniform SDR fellmostly within the condence intervals around observed values andwithin the expected range of variation. These results, along withthe similarity in the overall patterns of TSS supply between the twomodels (discussed in Section 3.2.1) indicate that the large absolutedifference (overestimation by N-SPECT) can be attributed to thehigh SDR values calculated in N-SPECT (in comparison with theuniform value of 0.1 used in SedNet); see corrected N-SPECT valuesin Fig. 3B. Modeled TSS loads from SedNet were mostly within thecondence intervals, except for a couple of gauge stations (9063and 9089), both corresponding to subcatchments with a dis-proportionally high percentage of modeled oodplain (8.0% and6.6% of the catchment area, respectively) relative to the other gagedsubcatchments (median: 0.9%), which would explain these unex-pected discrepancies. However, given the relatively large discrep-ancies in loads for these two stations, errors in monitored TSS loadscannot be discounted. Variations in estimated TSS loads betweenmodels were in agreement with differences in modeled sedimentsupply described below (Section 3.2).

bank erosion) could have a small e but signicant e inuence

SPECusis es

J.G. Alvarez-Romero et al. / Journal of Environmental Management 146 (2014) 164e178 171Fig. 3. Observed and modeled ow and suspended sediment loads derived using N-sediment yield (due to large differences between observed TSS loads and loads estimatedthe inuence of sediment delivery ratio (SDR) on TSS predictions by presenting N-SPECTperformance or goodness oft of the regionalizationmodels used inSedNet (Wilkinson et al., 2006) and obtained a good t between therunoff coefcient (Rc) and Eo/P (E 0.776). Similarly, E values forbankfull ow (Qbf) and median overbank ow (Qmo) (0.932 and0.829, respectively) indicated a goodt between SedNet predictionsand gauge data. The coefcient of efciency is similar to an R2 andexpresses the proportion of variance of the observed runoff coef-cient explained by the model (Nash and Sutcliffe, 1970). In contrast,many of N-SPECTs modeled runoff values fell outside of the con-dence intervals, generally over-estimating runoff in gaugesrecording lower discharge and under-estimating runoff for gaugestations with higher discharge. The highest discrepancies betweenmodeled and observed runoff (including over- and under-estimation in both models) corresponded to gauges located down-stream of wet areas (e.g., gauges 9067 and 9068, Fig. 1C).

We found important differences in modeled and observed TSSloads for both models (Fig. 3B), but overall performance of SedNetwas better when compared with the default output of N-SPECTderived using a spatially-distributed SDR (RMSE was 202% and737% for SedNet and N-SPECT, respectively). However, using avalues (marked as N-SPECT-b) calculated using a uniform SDR of 0.1. The black line represeinterval for the mean, estimated from gauge data (only available for records before the conT and SedNet models: A) mean annual runoff depths; and B) mean total suspendedng both models, we use a logarithmic scale to visualize estimated loads). Panel B depictstimates of TSS loads using the software's spatially-variable (default) SDR and corrected3.2. Spatial patterns of supply and estimated loads of pollutants

3.2.1. Differences in patterns of pollutant supply and contributionfrom subcatchments

Despite the observable spatial similarity between the twomodels in terms of modeled TSS supply (Fig. 4A and B), the per-centage contribution for individual subcatchments differed notablyacross the study area. Although absolute differences seem smallbetween models (Fig. 4C), these should be interpreted in compar-isonwith the percentage contribution by each subcatchment to thetotal TSS supply across the study area (maximum TSS contributionby any given subcatchment was 1.5%). Overall, the percentage dif-ference between models (calculated as N-SPECT - SedNet) was lowto moderate across most of the catchments (

ironmJ.G. Alvarez-Romero et al. / Journal of Env172when prioritizing catchment management at regional (catchment-wide) scale.

Differences in spatial patterns in estimated DIN supply are moreevident between the two models (Fig. 4D, E) than for TSS supply.The percentage difference between models (calculated as N-SPECTe SedNet) was generally higher (see Fig. 4F) and differences forindividual subcatchments weremore pronounced than for TSS. DINsupply as calculated by SedNet was concentrated strongly ineastern mountainous and wet subcatchments (Fig. 4E, F), while N-SPECT modeled a somewhat less concentrated pattern of supply(Fig. 4D). The more dispersed pattern from N-SPECT included largercontributions from some central, western, and southern catch-ments, in most cases clearly driven by the dominant land uses (e.g.,agriculture and pasture; see Fig. 1C). Neither model systematicallyestimated higher or lower DIN supply across the study area.

Comparing modeled TSS and TN contribution (which in-corporates dissolved inorganic nitrogen, dissolved organic nitro-gen, and particulate nitrogen) between models revealed the stronginuence of sinks (particularly reservoirs) and the input of partic-ulate nutrients (associated with erosion) to the total nutrient load,both of which are incorporated only in SedNet (Fig. 5). The termcontribution is used instead of supply to indicate that SedNetconsiders sinks (e.g., sediment deposition and denitrication) thatmodify subcatchments' supply in terms of their contributions to

Fig. 4. Patterns across subcatchments of supply of suspended sediment (TSS) and dissolved iis represented as the percentage contribution by each subcatchment to the total TSS or DINdifference in percentage TSS supply; and DIN supply estimated by D) N-SPECT; E) SedNet/expose the areas where N-SPECT estimates higher (yellow to red), lower (light blue to cobaANNEX.ental Management 146 (2014) 164e178end-of-river loads. N-SPECT does not consider sinks, so supplyequals contribution. The inuence of large reservoirs was evidentfrom the low TSS contribution estimated by SedNet for most of theYaqui River's upper catchment (yellow to red areas in Fig. 5C) and toa lesser extent for the upper Mayo River catchment (Fig. 5B). Ac-counting for reservoirs in SedNet shifted the distribution of areas ofhigh TSS contribution from the upper to the lower portions of thecatchments compared to N-SPECT (blue areas in Fig. 5C). Theoverall contribution of river-bank erosion estimated in SedNet wasnegligible (~1% of TSS load), explaining the minimal differences inpatterns of TSS supply between the summed (bank and hillslope)and the hillslope-only erosion supply (comparative map not pre-sented). The TN contribution patterns show the importance ofparticulate nutrients associated with erosion to the total load ofnitrogen (Fig. 5D, E), indicated by higher TN supply values in theupper Mayo River catchment estimated by SedNet (blue areas inFig. 5F). Trapping of particulate nutrients in reservoirs is alsoevident from the higher TN contribution values estimated by N-SPECT for the upper Yaqui River and western Mayo River catch-ments (yellow to red areas in Fig. 5F).

Our analysis using Marxan showed that the spatial allocation ofmanagement actions was sensitive to the observed differences inmodeled TSS/DIN supply. The differences in cost between thesolutions based on N-SPECT and SedNet outputs were small, but

norganic nitrogen (DIN), as modeled by N-SPECT and SedNet/ANNEX. In all cases supplysupply across the study area: TSS supply estimated by A) N-SPECT, B) SedNet, and C)

ANNEX, and F) difference in percentage DIN supply per subcatchment. Panels C and Flt) or similar (gray) values for TSS/DIN percentage contribution in relation to SedNet/

ironmJ.G. Alvarez-Romero et al. / Journal of Envcosts were consistently higher when using N-SPECT for targetsabove 10% and 15% reduction in DIN and TSS loads, respectively(Fig. 6). For both pollutants, the cost differences widened as tar-gets increased, indicating that some subcatchments were onlyincluded in solutions based on N-SPECT or SedNet outputs (mostlikely those with the largest difference in percentage contribution)and these subcatchments became more important as the targetsincreased. The differences in costs between models were largerfor DIN than for TSS across all targets; this result is consistent withthe important differences observed in DIN supply patterns.

3.2.2. Differences in estimated end-of river loadsDifferences in the total (end-of-river) loads for both pollutants

were considerable (Table 2), highlighting the more conservativeestimates from SedNet of TSS supply to the stream network. Inaddition to the large differences in estimated TSS loads (up to 3 and2 orders of magnitude when compared with N-SPECTs default andadjusted values, respectively), the relative contributions fromcatchments also varied considerably (Table 2). Major differenceswere notable in loads of the four major catchments (comprising~90% of total study area), where the largest catchment (Yaqui River)

Fig. 5. Patterns of contribution by subcatchments of total suspended sediment (TSS) and totthe percentage contribution by each subcatchment to the total TSS or DIN contribution aestimated by SedNet, including sediment eroded from river banks and deposited in reservoirby N-SPECT (calculated using Event Mean Concentration for TN); E) TN contribution estimatenitrogen (DON) and particulate nitrogen (PN), considering sinks (i.e., DIN denitrication, DONpercentage TN contribution. Panels C and F expose the areas where N-SPECT estimates hpercentage contribution in relation to SedNet.ental Management 146 (2014) 164e178 173was identied as the major contributor to end-of-river loads in N-SPECT, while SedNet estimated that most of the sediments effec-tively delivered to the coast were from the second largest catch-ment (Mayo River; see Table 2). This change in order of importancewas mostly driven by deposition occurring in reservoirs. Differ-ences in overall patterns across catchments of estimated TN loadswere less marked, with the two largest catchments identied byboth models as the main contributors (Table 2). However, threeimportant differences are noteworthy: rst, the contributions fromthese two very large catchments in SedNet was very similar, butslightly larger for the second largest catchment (Mayo); second, N-SPECT estimated total nitrogen load from the Yaqui as three timeslarger than from theMayo (equivalent to 3 times the load estimatedby SedNet); and third, estimates of TN for the two smaller catch-ments were of similar magnitude, but N-SPECT estimated half theload for one and 1.2 times for the other.

4. Discussion

In areas where land-based impacts on coastal-marine ecosys-tems are important (Halpern et al., 2009), a crucial yet neglected

al nitrogen (TN), as modeled by N-SPECT and SedNet/ANNEX. Values are represented ascross the study area: A) TSS contribution estimated by N-SPECT; B) TSS contributions and oodplains; C) difference in percentage TSS contribution; D) TN supply estimatedd by SedNet/ANNEX as the sum of dissolved inorganic nitrogen (DIN), dissolved organicloss in reservoirs and PN deposition in reservoirs and oodplains); and F) difference in

igher (yellow to red), lower (light blue to cobalt) or similar (gray) values for TSS/TN

with higher accuracy and, unlike N-SPECT, its modeled TSS loadswere mostly within the expected range of variation. While accuraterunoff predictions are not required to predict TSS yield (and loadsestimated by N-SPECT can be adjusted using a calibrated SDR),differences in predictions for dissolved nutrients were substantial.In N-SPECT, central, western, and southern catchments dominatedDIN supply (largely driven by land uses), while SedNet predictedhigher supply from the eastern mountainous and wet subcatch-ments; this resulted in differences in the spatial allocation ofmanagement actions to reduce the delivery of DIN to the sea. Themajor differences between the twomodels e for absolute TSS loadse can be attributed to the large disparity between the spatially-variable SDR calculated by N-SPECT (median: 0.89) and theselected SDR used in SedNet (0.1). However, using a uniform SDR of0.1 in N-SPECT resulted in a better t and drew attention to themarked underestimation of SedNet for two subcatchments. Thisunderestimation is likely attributed to oodplain deposition(modeled oodplains cover >6% of both subcatchments, in com-

J.G. Alvarez-Romero et al. / Journal of Environmental Management 146 (2014) 164e178174component of conservation planning is to reduce or mitigate suchimpacts (Alvarez-Romero et al., 2011). We compared two

Fig. 6. Differences in costs associated with the spatial allocation of catchment man-agement based on N-SPECT and SedNet outputs. The black/grey markers represent themean cost (total area of subcatchments selected for management) of 100 solutionsgenerated with Marxan (Ball et al., 2009) using N-SPECT modeled DIN/TSS supply; red/pink correspond to cost of solutions based on SedNet outputs. Error bars are thecondence interval for the mean (a 0.01) based on 100 solutions and lines ttedpower trend lines (R2 > 0.998) across targets (5%e30% reduction in total DIN/TSSsupply). (For interpretation of the references to colour in this gure legend, the readeris referred to the web version of this article).commonly-used catchment models to assess their performance,data requirements, ease of implementation, and utility for regional-scale planning, and found important differences. Here we discussthe implications of our key ndings for conservation practitionersseeking to incorporate land-based impacts into coastal-marineexercises in conservation prioritization.

4.1. Differences in model performance: accuracy and spatialpatterns

We found important differences in performance betweenmodels. SedNet consistently estimated spatial variations in runoff

Table 2End-of-river loads of TSS and TN for the four major catchments.

TSS (kt y1)

Catchment Area (km2) N-SPECTa/b

Yaqui 67,629 83,722/8592Mayo 13,303 29,435/3095Matape 5864 2119/245Cocoraque 1624 901/98

Total major catchment load 116,176/12,031Region-wide load 119,238/12,382

Major catchments prop. 97%

*Numbers in parenthesis correspond to the N-SPECT/SedNet ratio (for a and b N-SPECa Calculated using N-SPECTs default spatially-distributed SDR (median: 0.89).b Adjusted TSS loads estimated by N-SPECT using a uniform SDR (0.1).c TSS load estimated by SedNet includes sediment eroded from river bank and deposid TN load is calculated by N-SPECT using EMC values for TNe TN load is calculated in SedNet as DIN DON PN - DIN denitrication - DON reseparisonwith

ironm4.2. Data requirements and ease of implementation

Overall, data requirements were not considered a major limi-tation to the implementation of either N-SPECT or SedNet in thestudy area, despite the relatively limited data available compared toother regions where the models have been used (e.g., Burke andSugg, 2006; McKergow et al., 2005). In most cases, datasets ofappropriate resolution for region-wide applications were availableor it was possible to model the required data (e.g., locations ofoodplains and riparian vegetation) with readily available tools anddata. However, where used, modeled input data should beconsidered proxies and require validation in subsequent work.Some datasets were not available (e.g., gullies) and likely limit theaccuracy of model outputs. Once the datasets were compiled,running the models was relatively straightforward. However, datapreparation and parameterization demanded more time andexpertise for SedNet than N-SPECT.

The time commitment to locate, assemble, assess and modeldata should not be underestimated. Catchment modeling is a majorundertaking that requires not only substantial time, but also tech-nical expertise in geographic information systems and modeling.Assembling a research or planning group with expertise in areasrelevant to this task (e.g., hydrology, water quality, soil science) cangreatly facilitate data preparation and improve modeling outputs.Also, exploration of intermediate model outputs and comparison ofnal outputs with the results of other studies using the samemodels can help to identify errors in data inputs or parameteriza-tion and explain unexpected results (e.g., very high values). Whiletime consuming, the process of compiling and preparing data canbe important to guide and prioritize collection of further data.Despite these challenges, coastal-marine conservation plannersrecognize the importance of such modeling endeavors (Halpernet al., 2009; Tallis et al., 2008).

4.3. Applicability of outputs to planning

Outputs from both models were appropriate to represent broadspatial patterns of pollutant supply across our study area, butspatial differences in model outputs have implications for coastaland marine conservation planning. For example, spatial patterns ofTSS from subcatchments varied between models, with the outputsof SedNet indicating that TSS loads were highest next to the re-gion's highest marine conservation priority areas. Consequently,these catchments should be prioritized for management of waterquality to minimize land-based impacts on these high-prioritymarine areas (e.g., Klein et al., 2012). However, if N-SPECTs re-sults were used instead, these catchments would not be prioritized.Assuming N-SPECTs outputs are less reliable than those of SedNet,which seems reasonable from our comparisons, prioritization ofcatchment management based on N-SPECT would lead to high-priority marine areas continuing to suffer excessive loads of sedi-ment and nutrients. Alternatively, N-SPECTs predictions could leadto these marine priority areas being excluded as potential marineconservation areas if land-based pollution was seen to be intrac-table (e.g., Tallis et al., 2008).

The observed differences in the spatial distribution of bothTSS and DIN supply between models (Fig. 4C,F/5C,F) had a smalle but signicant e effect on the cost of management whenusing one or the other model. Despite the similarity in regionalpatterns of pollutant supply (Fig. 4A/B and D/E), and the rela-tively small differences in proportional contribution of sub-catchments between the two models (Fig. 4C/F), ourprioritization analysis shows that even small differences canmodify the spatial allocation of catchment management actions

J.G. Alvarez-Romero et al. / Journal of Envand these can have management implications (e.g., in terms ofcosts). We compared priorities for TSS and DIN independently,but concurrent optimization across pollutants will likely result inmore cost-efcient congurations of management actions,particularly if there are spatial correlations between sources ofdifferent pollutants. Therefore, differences in management pri-orities based on N-SPECT or SedNet could be considerably larger.For instance, if TN and TSS are targeted simultaneously, thenumber of subcatchments to be managed (and total managementcost) could be smaller when using SedNet outputs because of thelarge contribution of erosion to TN from particulate nutrients.Differences resulting from incorporating additional processes inSedNet (i.e., sinks and particulate nutrients) are evident in thegeneral patterns of contribution to the region-wide pollutantloads, and thus are reected in larger percentage differencesbetween models. Therefore allocating actions to subcatchmentsfor TSS and TN management using these outputs would be evenmore contrasting (especially if we consider absolute values).These results highlight the need for managers to exercise cautionwhen considering catchment models to allocate actions, espe-cially given that resources for management are limited; and, ifallocated ineffectively, might result in continued degradation ofmarine ecosystems.

Our results also emphasize the need for reliable estimates ofnutrient and sediment contribution, and end-of-river loads inparticular, for marine conservation planning. Ideally load data canbe incorporated into ood-plume models to more fully assessvulnerability of marine areas to land-based threats (e.g., Alvarez-Romero et al., 2013a; Burke and Sugg, 2006), thus allowing man-agers to link pollution sources and affected marine areas.

4.4. Limitations and next steps

Key limitations in our model included limited empirical data tovalidate modeled DIN/TN and, to a lesser extent, estimates of TSS,particularly regarding the actual inuence of reservoirs. Our studyshould thus be interpreted as a comparison of the two modelsunder the limitations imposed by available data and given thecharacteristics of our study area, not as a generic comparison of themodels. Likewise, our results should be regarded as estimatesrequiring further validation.

Catchment modeling in a region with relatively limited datameans that data gaps constrain the results. For example, apparentpatterns of sediment supply might have changed if we had beenable to incorporate other erosion processes (e.g., associated withgullies or mass landslides), which have been identied as impor-tant sources of sediment in some areas of Mexico (Evrard et al.,2010; Geissen et al., 2007). While we expect the inuence of gul-lies in the region would be minor, not having this data in SedNetprevented us from assessing the potential contribution of thisprocess. Studies in other regions have shown that errors in theallocation of sediment to different sources can be signicant (e.g.,gullies were found to be themajor sources of sediments, in contrastto the predicted loads estimated by SedNet: Caitcheon et al., 2012).This means that even if data were available for gullies and bankerosion, results should still be further explored. Therefore, furtherstudies, including mapping of existing gullies (e.g., using remote-sensing) and sediment tracing (Wilkinson et al., 2013) would beuseful to validate pollutant sources, thus improving the spatialallocation and type of management to mitigate erosion and sedi-mentation impacts. Road erosion can be another important sourceof sediments in some regions but is not incorporated in eithermodel; available tools (e.g., SEDMODL a GIS-based sediment pre-diction model: Akay et al., 2008) can help to assess the potentialcontribution of this process in the region and adjust management

ental Management 146 (2014) 164e178 175accordingly.

ironmHigher-quality inputs would improve the outputs of bothmodels. In particular, with a view to improving the accuracy ofcatchment models, priorities for research and monitoring activitiesin the study area should include: improving soil data; mapping andstudying the contribution of gullies to TSS; characterizing localerosion and deposition rates; and monitoring catchment-scale andlong-term nutrient loads associated with rainfall runoff. Our resultsshould thus be regarded as hypotheses to support further investi-gation, for example, to test whether TSS exports are derived pre-dominantly from areas downstream of reservoirs, as predicted bySedNet. The lack of post-damming gauge data stresses the need toreassess the performance of the model based on current TSS loads,particularly if the outputs are to inform prioritization of erosioncontrol to minimize coastal-marine impacts.

Despite data limitations, we used the best available informationand our preliminary estimates are a rst and critical step to identifydata gaps, potential problems, and areas of interest for furtherinvestigation. Commonly, managers will not be able to postponemanagement actions, instead having to make decisions based onlimited information and resources; under these circumstances,recognizing these limitations and the potential consequences iscritical. In addition, managers should consider undertaking asensitivity analysis when parameterizing and calibrating themodels, preparing data inputs and selecting between differentmodels. Provided the levels of uncertainty in each model input arerepresented, sensitivity analysis can help to assess uncertainty inmodel outputs and the potential implications of different man-agement decisions based on limited data conditions (e.g.,Nandakumar and Mein, 1997).

5. Conclusions

Our results indicate that the differences in the predicted pat-terns of pollutant supply varied notably between the two models.Comparing predicted vs. observed data indicates that SedNet pre-dicts TSS loads with a higher degree of accuracy, and generallywithin estimated variations of observed loads, when compared toN-SPECTs default outputs. However, adjusted loads calculated inN-SPECT using a calibrated SDR gave better results and highlightedthe need to revise the modeled oodplains and predicted sedimentdeposition in SedNet. Differences in subcatchments' contribution topollutant loads were principally due to explicit representation ofsediment sinks and particulate nutrients by SedNet. This meansthat the observed limitations of N-SPECT would be emphasized incatchments with reservoirs, and thus direct application of thismodel to estimate TSS loads in regulated catchments is not advis-able or would require further renement. Using different scenariosfor each model would be a useful tool to assess the potential con-sequences of using either model.

We also conclude that, while SedNet predicted spatial variationin TSS loads well, the absence of local data on erosion rates,including gully erosion, means that this preliminary modeling doesnot reliably identify the sediment sources requiring remediation,but rather identies the priority areas of the region where furtherinvestigation can be focused.

Overall, we conclude that N-SPECToutputs are sensitive to someinput parameters, including soil permeability and rain days, and donot account for pollutant-trapping in reservoirs. Another limitationregarding the application of model outputs to management ofnutrients is the lack of functionality to estimate nutrient runoffassociated with irrigation in both N-SPECT and SedNet. This meansthat the signicant contributions to nutrient runoff from coastaland dry parts of the region (e.g., the intensive agricultural system ofthe Yaqui Valley) are likely to be underestimated, as will their po-

J.G. Alvarez-Romero et al. / Journal of Env176tential ecological effects on coastal marine ecosystems. Additionalanalyses are required to address this limitation in these and relatedmodels.

Our study was focused on integrating land-based impacts intocoastal-marine conservation planning. The next steps towards trulyintegrated land-sea planning involve linking land-based impacts tomarine regions (i.e., through river plume models: Alvarez-Romeroet al., 2013a), assessing where the impacts negatively affect sensi-tive and vulnerable coastal and marine habitats or species (Halpernet al., 2009), and prioritizing catchments for management to reduceland-based impacts (Klein et al., 2012). Different strategies toaddress cross-system threats are possible, depending on factorssuch as the uniqueness of the affected marine features and thefeasibility of mitigating land-based and/or marine threats (Alvarez-Romero et al., 2011). The conguration of priority maps associatedwith these strategies can be very different and spatially uncorre-lated, hence there will be tradeoffs. The exploration of spatial op-tions under different management scenarios will provide criticalinformation to guide the planning process. To our knowledge, noprevious study has included all of these steps, so exploring optionsto integrate these elements in coastal and marine planning shouldbe considered a research priority.

Acknowledgments

We thank the following persons for providing information andtechnical advice: Dave Eslinger and Jamieson Carter (U.S. NationalOceanic and Atmospheric Administration (NOAA) Coastal ServicesCenter); Hector Cortes-Torres, Juan Francisco Gomez-Martinezand Pedro Rivera-Ruiz (IMTA); Rafael Hernandez (CIAD); Ste-phen Lewis (ACTFR-JCU); Eusebio Ventura (UQRO); AlejandroGonzalez-Serratos (CONAGUA); Arturo Flores-Martnez andCleotilde Arellano (SEMARNAT); Jose Luis Perez-Damian (INE);and Melanie Kolb (CONABIO). We also thank the following orga-nizations for providing software: NOAA (N-SPECT) and Australia'sCommonwealth Scientic and Industrial Research Organization(CSIRO) Land and Water (SedNet/ANNEX), and datasets: Serviciode Informacion Agroalimentaria y Pesquera (SIAP-SAGARPA);Comision Nacional para el Conocimiento y Uso de la Biodiversidad(CONABIO); Secretara de Medio Ambiente y Recursos Naturales(SEMARNAT); Instituto Nacional de Ecologa (INE-SEMARNAT),and Instituto Nacional de Estadstica, Geografa e Informatica(INEGI). We also thank P. Visconti for support in scripting toanalyze gauge data, Gordon Bailey for providing IT support, andthe High Performance Computing Unit at James Cook Universityfor computational facilities. We thank three anonymous re-viewers, who provided useful comments which improved theoriginal manuscript. JGAR gratefully acknowledges support fromMexico's Consejo Nacional de Ciencia y Tecnologa (CONACYT) andSecretara de Educacion Pblica (SEP), as well as from theAustralian Research Council Centre of Excellence for Coral ReefStudies. RLP and NCB acknowledge the support of the AustralianResearch Council.

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.jenvman.2014.07.007

References

Ahrens, T.D., Beman, J.M., Harrison, J.A., Jewett, P.K., Matson, P.A., 2008. A synthesisof nitrogen transformations and transfers from land to the sea in the YaquiValley agricultural region of northwest Mexico. Water Resour. Res. 44, W00A05.

Akay, A.E., Erdas, O., Reis, M., Yuksel, A., 2008. Estimating sediment yield from a

ental Management 146 (2014) 164e178forest road network by using a sediment prediction model and GIS techniques.Build. Environ. 43, 687e695.

ironmAlvarez-Romero, J.G., Devlin, M., Teixeira da Silva, E., Petus, C., Ban, N.C.,Pressey, R.L., Kool, J., Roberts, J.J., Cerdeira-Estrada, S., Wenger, A.S., Brodie, J.,2013a. A novel approach to model exposure of coastal-marine ecosystems toriverine ood plumes based on remote sensing techniques. J. Environ. Manage.119, 194e207.

Alvarez-Romero, J.G., Pressey, R.L., Ban, N.C., Torre-Coso, J., Aburto-Oropeza, O.,2013b. Marine conservation planning in practice: lessons learned from the Gulfof California. Aquat. Conserv. Mar. Freshw. Ecosyst. 23, 483e505.

Alvarez-Romero, J.G., Pressey, R.L., Ban, N.C., Vance-Borland, K., Willer, C., Klein, C.J.,Gaines, S.D., 2011. Integrated land-sea conservation planning: the missing links.Annu. Rev. Ecol. Evol. Syst. 42, 381e409.

Arnold, J.G., Fohrer, N., 2005. SWAT2000: current capabilities and research oppor-tunities in applied watershed modelling. Hydrol. Process. 19, 563e572.

Ball, I., Possingham, H., Watts, D.J., 2009. Marxan and Relatives: Software for SpatialConservation Prioritization. In: Moilanen, A., Wilson, K., Possingham, H. (Eds.),Spatial Conservation Prioritization Quantitative Methods and ComputationalTools. Oxford University Press, Oxford, UK, pp. 185e195.

Bartley, R., Speirs, W., Ellis, T., Waters, D.K., 2012. A review of sediment and nutrientconcentration data from Australia for use in catchment water quality models.Mar. Pollut. Bull. 65 (4e9), 101e116.

Beman, J.M., Arrigo, K.R., Matson, P.A., 2005. Agricultural runoff fuels large phyto-plankton blooms in vulnerable areas of the ocean. Nature 434, 211e214.

Borah, D., Bera, M., 2003. Watershed-scale hydrologic and nonpoint-source pollu-tion models: review of mathematical bases. Trans. ASAE 46, 1553e1566.

Brodie, J., Lewis, S., Bainbridge, Z., Mitchell, A., Waterhouse, J., Kroon, F., 2009. Targetsetting for pollutant discharge management of rivers in the Great Barrier Reefcatchment area. Mar. Freshw. Res. 60, 1141e1149.

Brodie, J., Mitchell, A., 2005. Nutrients in Australian tropical rivers: changes withagricultural development and implications for receiving environments. Mar.Freshw. Res. 56, 279e302.

Burke, L., Sugg, Z., 2006. Hydrologic Modeling of Watersheds Discharging Adjacentto the Mesoamerican Reef. World Resources Institute,, Washington, DC.

Cabaco, S., Santos, R., Duarte, C.M., 2008. The impact of sediment burial and erosionon seagrasses: a review. Estuar. Coast. Shelf Sci. 79, 354e366.

Caitcheon, G.G., Olley, J.M., Pantus, F., Hancock, G., Leslie, C., 2012. The dominanterosion processes supplying ne sediment to three major rivers in tropicalAustralia, the Daly (NT), Mitchell (Qld) and Flinders (Qld) Rivers. Geo-morphology 151e152, 188e195.

Christensen, L., Riley, W., Ortiz-Monasterio, I., 2006. Nitrogen cycling in an irrigatedwheat system in Sonora, Mexico: Measurements and Modeling. In: NutrientCycling in Agroecosystems, vol. 75, pp. 175e186.

CNA, 1987. Instructivo de hidrologa para determinar la avenida maxima ordinaria.Comision Nacional del Agua (CNA), Mexico, D.F.

CNA, 2008. In: Tecnica, S.G. (Ed.), Cuencas hidrologicas con su disponibilidad mediaanual de aguas superciales publicada. Comision Nacional del Agua, Mexico, DF.

CONAGUA, 2008. Tabla de las principales presas de Mexico indicando capacidad dealmacenamiento y uso, con caractersticas de la presa y serie de volumen tilalmacenado anualmente al 1 de octubre para el periodo 1990e2007. ComisionNacional del Agua (CONAGUA), Subdireccion General de Programacion, Mexico,DF.

Cortes-Torres, H.G., 1991. Caracterizacion de la erosividad de la lluvia en Mexicoutilizando metodos multivariados. In: Colegio de Postgraduados. Montecillo,Mexico, p. 168.

Cotler, H., Ortega-Larrocea, M.P., 2006. Effects of land use on soil erosion in atropical dry forest ecosystem, Chamela watershed, Mexico. Catena 65, 107e117.

Descroix, L., Gonzalez Barrios, J.L., Viramontes, D., Poulenard, J., Anaya, E.,Esteves, M., Estrada, J., 2008. Gully and sheet erosion on subtropical mountainslopes: their respective roles and the scale effect. Catena 72, 325e339.

Diaz, R.J., Rosenberg, R., 2008. Spreading dead zones and consequences for marineecosystems. Science 321, 926e929.

Dijkshoorn, K., Huting, J., Tempel, P., 2005. Update 1:5 million Soil and TerrainDatabase for Latin America and the Caribbean (SOTERLAC, Version 2.0). ISRIC eWorld Soil Information, Wageningen, The Netherlands. Report 2005/01.

Diodato, N., Bellocchi, G., 2007. Estimating monthly (R)USLE climate input in aMediterranean region using limited data. J. Hydrol. 345, 224e236.

Doney, S.C., 2010. The Growing human Footprint on coastal and Open-oceanBiogeochemistry. Science 328, 1512e1516.

Drewry, J.J., Newham, L.T.H., Greene, R.S.B., Jakeman, A.J., Croke, B.F.W., 2006.A review of nitrogen and phosphorus export to waterways: context for catch-ment modelling. Mar. Freshw. Res. 57, 757e774.

Duvert, C., Gratiot, N., Evrard, O., Navratil, O., Nemery, J., Prat, C., Esteves, M.,2010. Drivers of erosion and suspended sediment transport in three head-water catchments of the Mexican Central Highlands. Geomorphology 123,243e256.

Eslinger, D.L., Jamieson Carter, H., Dempsey, E., VanderWilt, M., Wilson, B.,Meredith, A., 2005. The Nonpoint-source Pollution and Erosion ComparisonTool, 1.5.0. NOAA Coastal Services Center, Charleston, SC.

Evrard, O., Nemery, J., Gratiot, N., Duvert, C., Ayrault, S., Lefevre, I., Poulenard, J.,Prat, C., Bonte, P., Esteves, M., 2010. Sediment dynamics during the rainy seasonin tropical highland catchments of central Mexico using fallout radionuclides.Geomorphology 124, 42e54.

FAO, 1980. A provisional methodology for soil degradation assessment. In: FAO.Food and Agriculture Organization of the United Nations (FAO), Rome, Italy.

J.G. Alvarez-Romero et al. / Journal of EnvGabric, A.J., Bell, P.R.F., 1993. Review of the effects of nonpoint nutrient loading oncoastal ecosystems. Aust. J. Mar. Freshw. Res. 44, 261e283.Gallant, J.C., Dowling, T.I., 2003. A multiresolution index of valley bottom atnessfor mapping depositional areas. Water Resour. Res. 39, 1347.

Geissen, V., Kampichler, C., Lopez-de Llergo-Juarez, J.J., Galindo-Acantara, A., 2007.Supercial and subterranean soil erosion in Tabasco, tropical Mexico: devel-opment of a decision tree modeling approach. Geoderma 139, 277e287.

Halpern, B.S., Ebert, C., Kappel, C.V., Madin, E.M.P., Micheli, F., Perry, M.T.,Selkoe, K.A., Walbridge, S., 2009. Global priority areas for incorporating land-sea connections in marine conservation. Conserv. Lett. 2, 189e196.

Hood, G.M., 2010. PopTools Version 3.2.3. Available on the internet. URL. http://www.poptools.org.

Humborg, C., Conley, D.J., Rahm, L., Wulff, F., Cociasu, A., Ittekkot, V., 2000. Siliconretention in river basins: Far-reaching effects on Biogeochemistry and AquaticFood Webs in coastal Marine environments. AMBIO A J. Hum. Environ. 29,45e50.

IMTA, 2010. Banco Nacional de Datos de Aguas Superciales (BANDAS). Inst. Mex.Tecnol. del Agua IMTA Com. Nac. del Agua Mex. DF. http://www.imta.gob.mx/bandas.

INIFAP-CONABIO, 1995. Edafologa de Mexico: Datos vectoriales escalas 1:250,000 y1:1,000,000. Instituto Nacional de investigaciones Forestales y Agropecuarias(INIFAP) & Comision Nacional para el Conocimiento y Uso de la Biodiversidad(CONABIO), Mexico, DF.

Jetten, V., de Roo, A., Favis-Mortlock, D., 1999. Evaluation of eld-scale andcatchment-scale soil erosion models. Catena 37, 521e541.

Johnes, P., 1996. Evaluation and management of the impact of land use change onthe nitrogen and phosphorus load delivered to surface waters: the export co-efcient modelling approach. J. Hydrol. 183, 323e349.

Klein, C.J., Jupiter, S.D., Selig, E.R., Watts, M.E., Halpern, B.S., Kamal, M., Roelfsema, C.,Possingham, H.P., 2012. Forest conservation delivers highly variable coral reefconservation outcomes. Ecol. Appl. 22, 1246e1256.

Krysanova, V., Hattermann, F., Wechsung, F., 2005. Development of the ecohydro-logical model SWIM for regional impact studies and vulnerability assessment.Hydrol. Process. 19, 763e783.

Lewis, W.M., 2002. Yield of nitrogen from minimally disturbed watersheds of theUnited States. Biogeochemistry 57, 375e385.

Lluch-Cota, S.E., Aragon-Noriega, E.A., Arreguin-Sanchez, F., Aurioles-Gamboa, D.,Bautista-Romero, J.J., Brusca, R.C., Cervantes-Duarte, R., Cortes-Altamirano, R.,Del-Monte-Luna, P., Esquivel-Herrera, A., Fernandez, G., Hendrickx, M.E., Her-nandez-Vazquez, S., Herrera-Cervantes, H., Kahru, M., Lavin, M., Lluch-Belda, D.,Lluch-Cota, D.B., Lopez-Martinez, J., Marinone, S.G., Nevarez-Martinez, M.O.,Ortega-Garcia, S., Palacios-Castro, E., Pares-Sierra, A., Ponce-Diaz, G., Ramirez-Rodriguez, M., Salinas-Zavala, C.A., Schwartzlose, R.A., Sierra-Beltran, A.P., 2007.The Gulf of California: review of ecosystem status and sustainability challenges.Prog. Oceanogr. 73, 1e26.

Maina, J., de Moel, H., Vermaat, J.E., Henrich Bruggemann, J., Guillaume, M.M.M.,Grove, C.A., Madin, J.S., Mertz-Kraus, R., Zinke, J., 2012. Linking coral river runoffproxies with climate variability, hydrology and land-use in Madagascar catch-ments. Mar. Pollut. Bull. 64, 2047e2059.

Maina, J., McClanahan, T.R., Venus, V., Ateweberhan, M., Madin, J., 2011. Globalgradients of coral exposure to Environmental stresses and implications for localmanagement. PLoS ONE 6, e23064.

Martnez-Menez, M.R., Lopez-Martnez, R., Hernandez-Fuentes, E., 2001. Relacionsedimentos erosion en la cuenca alta del Papaloapan. In: XI Congreso Nacionalde Irrigacion. Guanajuato, Guanajuato, Mexico.

McKergow, L.A., Prosser, I.P., Hughes, A.O., Brodie, J., 2005. Sources of sediment tothe great barrier reef world Heritage area. Mar. Pollut. Bull. 51, 200e211.

Merritt, W.S., Letcher, R.A., Jakeman, A.J., 2003. A review of erosion and sedimenttransport models. Environ. Model. Softw. 18, 761e799.

METI-NASA, 2009. ASTER global digital elevation model (ASTER GDEM) version 1.0.Earth Remote Sens. Data Anal. Cent. Tokyo, Japan.

Mikhailova, E.A., Bryant, R.B., Schwager, S.J., Smith, S.D., 1997. Predicting rainfallerosivity in Honduras. Soil. Sci. Soc. Am. J. 61, 273e279.

Nandakumar, N., Mein, R.G., 1997. Uncertainty in rainfall-runoff model simulationsand the implications for predicting the hydrologic effects of land-use change.J. Hydrol. 192, 211e232.

Nash, J.E., Sutcliffe, J.V., 1970. River ow forecasting through conceptual models partI d A discussion of principles. J. Hydrol. 10, 282e290.

NOAA, 2008. Nonpoint-source pollution and erosion comparison tool (N-SPECT):technical guide, Ver. 1.5 (NOAA/CSC/RPT 08e5). National Oceanic and Atmo-spheric Administration (NOAA), Coastal Services Center, Charleston, SC.

Norman, L.M., 2007. United States-Mexican border watershed assessment: modelingnonpoint source pollution in Ambos Nogales. J. Borderl. Stud. 22, 79e97.

O'Connor, J.E., Grant, G.E., Haluska, T.L., 2003. Overview of Geology, hydrology,Geomorphology, and sediment budget of the Deschutes river Basin, Oregon. In:O'Connor, J.E., Grant, G.E. (Eds.), A Peculiar River: Geology, Geomorphology, andHydrology of the Deschutes River, Oregon, Water Science and Application. AGU,Washington, DC, pp. 7e29.

Paez-Osuna, F., Ruiz-Fernandez, A.C., 2005. Environmental load of nitrogen andphosphorus from extensive, semiintensive, and intensive shrimp farms inthe Gulf of California ecoregion. Bull. Environ. Contam. Toxicol. 74,681e688.

Perakis, S.S., Hedin, L.O., 2002. Nitrogen loss from unpolluted South Americanforests mainly via dissolved organic compounds. Nature 415, 416e419.

Prosser, I.P., Rutherfurd, I.D., Olley, J.M., Young, W.J., Wallbrink, P.J., Moran, C.J., 2001.

ental Management 146 (2014) 164e178 177Large-scale patterns of erosion and sediment transport in river networks, withexamples from Australia. Mar. Freshw. Res. 52, 81e100.

Renard, K.G., Foster, G.R., Weesies, G.A., McCool, D.K., Yoder, D.C., 1997. PredictingSoil Erosion by Water: a Guide to Conservation Planning with the RevisedUniversal Soil Loss Equation (RUSLE), USDA Handbook. U.S. Department ofAgriculture (USDA), Washington, D.C.

SEMARNAT-UACH, 2002. Evaluacion de la perdida de suelo por erosion hdrica yeolica en la Repblica Mexicana, esc.. 1:1 000 000. Secretaria de Medio Ambi-ente y Recursos Naturales (SEMARNAT) - Universidad Autonoma de Chapingo(UACH), Chapingo, Estado de Mexico, p. 146.

SEMARNAT-UNAM, 2002. Uso de Suelo y Vegetacion 1976, 1993 2000 escala 1:250,000, coberturas preparada para el analisis de cambio de uso del suelo.Instituto Nacional de Ecologa, Secretara del Medio Ambiente y RecursosNaturales (INE-SEMARNAT) & Instituto de Geografa, Universidad NacionalAutonoma de Mexico (IG-UNAM), Mexico, DF.

SEMARNAT, 2006. Programa de Ordenamiento Ecologico Marino del Golfo de Cal-ifornia. Secretaria de Medio Ambiente y Recursos Naturales, Mexico, D.F.

Smith, R.A., Alexander, R.B., Schwarz, G.E., 2003. Natural background concentrationsof nutrients in streams and rivers of the conterminous United States. Environ.Sci. Technol. 37, 3039e3047.

Tallis, H., Ferdana, Z., Gray, E., 2008. Linking terrestrial and marine conservationplanning and threats analysis. Conserv. Biol. 22, 120e130.

Thorburn, P.J., Wilkinson, S.N., Silburn, D.M., 2013. Water quality in agriculturallands draining to the Great Barrier Reef: a review of causes, management andpriorities. Agriculture. Ecosyst. Environ. 180, 4e20.

Tilman, D., Fargione, J., Wolff, B., D'Antonio, C., Dobson, A., Howarth, R., Schindler, D.,Schlesinger,W.H., Simberloff, D., Swackhamer, D., 2001. Forecasting agriculturallydriven global environmental change. Science 292, 281e284.

USDA, 1986. Technical Release: Urban hydrology for small watersheds. In: USDA(Ed.), Natural Resources Conservation Service, Conservation Engineering Divi-sion. United States Department of Agriculture, Washington, DC, p. 164.

Velazquez, A.,Mas, J.F., Palacio, J.L., 2002. Analisis del cambiode usodel suelo. Institutode Geografa, UNAM/Instituto de Ecologia, SEMARNAT, Mexico, D.F, p. 95.

Walling, D.E., 2006. Human impact on land-ocean sediment transfer by the world'srivers. Geomorphology 79, 192e216.

Walling, D.E., Owens, P.N., Carter, J., Leeks, G.J.L., Lewis, S., Meharg, A.A., Wright, J.,2003. Storage of sediment-associated nutrients and contaminants in riverchannel and oodplain systems. Appl. Geochem. 18, 195e220.

Walling, D.E., Wilkinson, S.N., Horowitz, A.J., 2011. Catchment erosion, sedimentdelivery, and sediment quality. In: Wilderer, P. (Ed.), Treatise on Water Science.Oxford Academic Press, Oxford, UK, pp. 305e338.

Wilkinson, C., Brodie, J., 2011. Catchment Management and Coral Reef Conser-vation: a Practical Guide for Coastal Resource Managers to Reduce Damagefrom Catchment Areas Based on Best Practice Case Studies. Global Coral ReefMonitoring Network and Reef and Rainforest Research Centre, Townsville,QLD.

Wilkinson, S., Henderson, A., Chen, Y., Sherman, B., 2004. SedNet User Guide,Version 2, Client Report, CSIRO Land and Water. CSIRO Land and Water, Can-berra, ACT.

Wilkinson, S.N., 2008. Testing the capability of a sediment budget model to targetremediation measures to reduce suspended-sediment yield. In: Schmidt, J.,Cochrane, T., Phillips, C., Elliott, S., Davies, T.R., Basher, L. (Eds.), Sediment Dy-namics in Changing Environments. IAHS Press, Christchurch, New Zealand,pp. 559e566.

Wilkinson, S.N., Dougall, C., Kinsey-Henderson, A.E., Searle, R.D., Ellis, R.J.,Bartley, R., 2014. Development of a time-stepping sediment budget model forassessing land use impacts in large river basins. Sci. Total Environ. 468e469,1210e1224.

Wilkinson, S.N., Hancock, G.J., Bartley, R., Hawdon, A.A., Keen, R.J., 2013. Usingsediment tracing to assess processes and spatial patterns of erosion in grazedrangelands, Burdekin River basin, Australia. Agric. Ecosyst. Environ. 180,90e102.

Wilkinson, S.N., Prosser, I.P., Rustomji, P., Read, A.M., 2009. Modelling and testingspatially distributed sediment budgets to relate erosion processes to sedimentyields. Environ. Model. Softw. 24, 489e501.

Wilkinson, S.N., Young, W.J., DeRose, R.C., 2006. Regionalizing mean annual owand daily ow variability for basin-scale sediment and nutrient modelling.Hydrol. Process. 20, 2769e2786.

Williams, J.R., 1977. Sediment delivery ratios determined with sediment andrunoff models. Eros. Solid Matter Transp. Inland Waters IAHS AISH Publ.122, 168e179.

Young, R.A., Onstad, C.A., Bosch, D.D., Anderson, W.P., 1989. AGNPS: a nonpoint-source pollution model for evaluating agricultural watersheds. J. Soil. WaterConserv. 44, 168e173.

J.G. Alvarez-Romero et al. / Journal of Environmental Management 146 (2014) 164e178178

Modeling catchment nutrients and sediment loads to inform regional management of water quality in coastal-marine ecosystems ...1 Introduction2 Methods and data2.1 Study area2.2 Catchment model descriptions and data requirements2.3 Model parameterization2.3.1 Catchment and stream network delineation2.3.2 Runoff and flow calculations2.3.3 Dissolved pollutants and nutrients2.3.4 Hillslope erosion2.3.5 Gully erosion2.3.6 River bank erosion2.3.7 Particulate nutrients2.3.8 Deposition and loss of sediment and nutrients2.3.9 Sediment delivery ratio

2.4 Evaluation of performance and outputs of models2.4.1 Performance of models2.4.2 Spatial pattern of supply and estimated loads of pollutants

3 Results3.1 Performance of models3.2 Spatial patterns of supply and estimated loads of pollutants3.2.1 Differences in patterns of pollutant supply and contribution from subcatchments3.2.2 Differences in estimated end-of river loads

4 Discussion4.1 Differences in model performance: accuracy and spatial patterns4.2 Data requirements and ease of implementation4.3 Applicability of outputs to planning4.4 Limitations and next steps

5 ConclusionsAcknowledgmentsAppendix A Supplementary dataReferences