Enhancing the Archive System - ECMWF · ECMWF Newsletter No. 120 – Summer 2009 1 Contents ... Each single cartridge will be able to hold up to one terabyte of (uncompressed) data

ECMWF Newsletter No. 120 – Summer 2009

1

ContentsEditorial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

News

Changes to the operational forecasting system. . . . . . . . . . . 2

New items on the ECMWF website . . . . . . . . . . . . . . . . . . . . 2

71st Council session on 25–26 June 2009 . . . . . . . . . . . . . . . 3

EUMETNET’s ‘Oslo Declaration’ . . . . . . . . . . . . . . . . . . . . . . . 3

Operational assimilation of Indian radiosondes . . . . . . . . . . . 4

Assimilation of IASI in NWP . . . . . . . . . . . . . . . . . . . . . . . . . . 5

Goodbye GEMS – Hello MACC. . . . . . . . . . . . . . . . . . . . . . . . 6

Ocean waves at ECMWF . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

ECMWF Annual Report for 2008. . . . . . . . . . . . . . . . . . . . . . . 7

Forecast Products Users’ Meeting, June 2009. . . . . . . . . . . . 8

Diagnostics of data assimilation system performance . . . . . 9

Meteorology

EPS/EFAS probabilistic flood prediction forNorthern Italy: the case of 30 April 2009 . . . . . . . . . . . . . . . 10

NEMOVAR: A variational data assimilation systemfor the NEMO ocean model . . . . . . . . . . . . . . . . . . . . . . . . . 17

Improvements in the stratosphere andmesosphere of the IFS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

The direct assimilation of cloud-affectedinfrared radiances in the ECMWF 4D-Var . . . . . . . . . . . . . . . 32

General

Peer reviewed publications in 2008 . . . . . . . . . . . . . . . . . . . 39

ECMWF Calendar 2009 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

ECMWF publications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

Index of past newsletter articles. . . . . . . . . . . . . . . . . . . . . . 41

Useful names and telephone numbers within ECMWF . . . . 43

Publication policy

The ECMWF Newsletter is published quarterly. Its purpose is tomake users of ECMWF products, collaborators with ECMWFand the wider meteorological community aware of new devel-opments at ECMWF and the use that can be made of ECMWFproducts. Most articles are prepared by staff at ECMWF, butarticles are also welcome from people working elsewhere,especially those from Member States and Co-operating States.The ECMWF Newsletter is not peer-reviewed.

Editor: Bob Riddaway

Typesetting and Graphics: Rob Hine

Any queries about the content or distribution of the ECMWFNewsletter should be sent to [email protected]

Contacting ECMWF

Shinfield Park, Reading, Berkshire RG2 9AX, UK

Fax: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .+44 118 986 9450

Telephone: National . . . . . . . . . . . . . . . . . .0118 949 9000

International . . . . . . . . . . . .+44 118 949 9000

ECMWF website . . . . . . . . . . . . . . . . . .http://www.ecmwf.int

EDITORIAL

Enhancing the Archive System

The ease of access to data is fundamental to the successof ECMWF and in particular to the productivity of itsresearch staff. The ECMWF archive system combines alarge, comprehensive collection of meteorological data witha high-performance data management and delivery system.This unrivalled combination has become one of ECMWF’smost important assets.

The capability offered by the archive system is bothvalued and strongly supported by the ECMWF Council. Atits June 2009 session, the Council authorised the Directorto enter into an eight-year call-off contract with Sun Micro -systems Ltd for the provision of automated tape librariesand tape drives. By early 2010, three SL8500 automatictape libraries (ATLs) will have been installed. The librarieswill be interconnected, with each library having fifty tapedrives and being capable of holding 10,000 tape cartridges.Each single cartridge will be able to hold up to one terabyteof (uncompressed) data. Additional ATLs and tape driveswill be installed in subsequent years to keep pace with thegrowth of the archive. The equipment procured under thiscontract will replace the ageing STK tape silos. Themigration of the data from the old to the new system willbe transparent from a user’s point of view.

ECMWF’s archive system has always been highly appre -ci ated by its users; from the early days, when the amountof data was measured in tens of gigabytes to today, whereover ten petabytes of data is archived. In the beginning thearchive consisted of files residing on magnetic tape and itwas up to the individual user to keep track of the data onthese tapes. Very early on it was recognised that thisparadigm was not sustainable and so MARS (Meteo ro -logical Archival and Retrieval System) was created. MARSwas the first system in NWP centres worldwide that usedmeta-data to enable users to access data using a languagetailored towards meteorological terms (e.g. temperature,pressure and grids) rather than reading a particular set offiles from a particular set of tapes. ECFS (ECMWF FileSystem) was designed to complement MARS, appearing tothe user as a remote general-purpose file system.

The Convention establishing ECMWF specifies theobjec tives of the Centre: these objectives include specifi -cally the collection and storage of meteorological data. Thearchitecture of the data handling system makes it relativelyeasy to support new data types and enables an efficientmanagement of large data volumes. The various reanalysisdata sets, the TIGGE archive and YOTC (Year of TropicalConvection) data are examples of information that is verywidely used by research communities located throughoutthe world.

Registered users from the ECMWF Member States andCo-operating States are given comprehensive access toECMWF’s archive. As such, the archive held at ECMWF isnot only valuable to scientists residing at ECMWF but it isalso an important asset to the entire European atmosphericcommunity.

Walter Zwieflhofer

EDITORIAL


2

NEWS

Changes to the operational forecasting system

New items on the ECMWF website

DAVID RICHARDSON

New cycle (Cy35r3)A new cycle of the ECMWF forecastand analysis system, Cy35r3, isundergoing pre-operational testing.Changes for this cycle include:� Addition of GOME-2 total columnozone data from the METOP satellite.� Assimilation of cloud-affectedradiances for infra-red instruments.

� Improved assimilation of satellitechannels that are sensitive to the landsurface.� Assimilation of total columnwater vapour data from the MERISinstrument.� Variational bias correction forozone satellite data.� Improved quality control (usingHuber norm) of conventionalobservations.

� Improved background-errorstatistics for humidity, new humidityformulation in 4D-Var.� Weak-constraint 4D-Var taking intoaccount systematic model errors inthe stratosphere.� Non-orographic gravity wave scheme.� New trace gas climatology.� Further revision of the snow scheme.� Revised stochastic physics (modelperturbations) for EPS.

ANDY BRADY

New model simulated satelliteimagesA new product is available from ourdeterministic forecast – simulatedinfra-red and water vapour geo -station ary satellite images from theT799 deterministic forecast. Theproduct is available out to five days.� www.ecmwf.int/products/forecasts/

d/charts/medium/deterministic/simulated/sim/

New wave forecast chartsA new page is available for our oceanwave products including an extendedset of verification products for thewave forecasts. A comparison withother operational forecast centres isalso available.� www.ecmwf.int/products/forecasts/

wavecharts/

Funding and Member States’contributionsThe scale of contribution of eachMember State and Co-operating Stateis according to its Gross NationalIncome (GNI) and is revised every 3years. The contributions to the ECMWFBudget have been updated for 2009.� www.ecmwf.int/about/funding/

T-PARC – THORPEX Pacific AsianRegional CampaignT-PARC is a two-phase multi-nationalfield experiment. The first andprimary phase (Summer T-PARC) ranfrom August to October 2008. Thesecond phase (Winter T-PARC) ranfrom mid-January through to March2009. In both phases the geographicalfocal point is the North Pacific andsurrounding countries. DuringSummer T-PARC the main meteo ro -logical focus was tropical cyclonesand extra-tropical transition – mostactivity therefore was in the westernNorth Pacific region. In Winter T-PARCdownstream impacts over NorthAmerica are more relevant so activitythen covers the whole of the NorthPacific region, but with less emphasison the tropics.� www.ecmwf.int/research/

WMO_projects/TPARC/index.html

ECMWF/EUMETSAT NWP-SAFWorkshop on the ‘Assimilation ofIASI in NWP’The ECMWF/EUMETSAT NWP-SAFWorkshop on the ‘Assimilation of IASIin NWP’ was held from 6 to 8 May2009. The presentations are available.� www.ecmwf.int/newsevents/meetings/

workshops/2009/IASI_data/index.html

ECMWF Workshop on ‘Diagnosticsof data assimilation systemperformance’The Workshop on ‘Diagnostics of dataassimilation system performance’ washeld from 15 to 17 June 2009. Thepresentations are available.� www.ecmwf.int/newsevents/meetings/

workshops/2009/Diagnostics_DA_System_Performance/

MACC scientific and technicaltraining workshopECMWF will host the first Scientificand technical training workshop forthe MACC (Monitoring AtmosphericComposition and Climate) projectfrom 11 to 15 January 2010.� www.ecmwf.int/newsevents/meetings/

workshops/2010/MACC/

SAFER (Service and Applicationsfor Emergency Response)In the frame of GMES initiative(Global Monitoring for Environmentand Security), the SAFER project aimsat implementing preoperationalversions of the Emergency ResponseCore Service. SAFER will reinforceEuropean capacity to respond toemergency situations.� www.ecmwf.int/research/EU_projects/

SAFER/


3

NEWS

71st Council session on 25–26 June 2009

MANFRED KLÖPPEL

On the invitation by its President,Dr Adérito Vicente Serrão fromPortugal, the ECMWF Council held its71st session in Ponta Delgada in theAzores on 25–26 June 2009.

Besides several decisions made onfinancial and staff matters (such asadoption of Reports from theCo-ordinating Committee onRemuneration), the main results ofthis session were as follows.� Amendments to the Convention.Good progress had been made withthe ratification process withinMember States. If such progress ismaintained, the amended Conventioncould come into force by the end of2009.� Co-operation Agreements. TheCouncil unanimously authorised theDirector to conclude a Co-operationAgreement with the Government ofBulgaria and to renew the Co-operationAgreement with the African Centre ofMeteorological Application forDevelopment (ACMAD).� Extension of an existing licenceagreement. The Council unanimouslyapproved the extension of theexisting licence agreement betweenECMWF and EUROBRISA, allowingBrazil’s Centro de Previsão de Tempo eEstudos Climáticos (CPTEC) to receivereal-time, multi-model, seasonalforecasts from ECMWF until June 2012.

� GMES (Global Monitoring forEnvironment and Security). TheCouncil agreed that ECMWF shouldcontinue to play an active role ascoord inator of and main contributorto the atmosphere network, as fore -seen in the European Commission’scommunication issued in November2008.� Registered Partnership. The Counciladopted the same text as alreadyadopted by the Council of Europe onregistered partnerships, to be imple -mented at ECMWF under a newArticle of the Staff Regulations.� Financial Matters. The Council tooknote of the Auditor’s Report regardingthe financial year 2008 and gavedischarge to the Director in respect ofthe implementation of the budget for2008. The Council also considered pre -liminary estimates of Member States’contributions to the 2010 budget.

� Data Handling System. The Councilunanimously authorised the Directorto enter into a contract for therenewal of the automated tape librarywith Sun Microsystems Ltd., and on acall-off basis for 8 years from the startof the contract.� Oslo Declaration. The Councilunanimously adopted the ‘OsloDeclaration’ as a guideline forformulating ECMWF’s data policy.� Use of ECMWF Products. TheCouncil unanimously adopted therevised standard licence agreementfor use of ECMWF Products and therevised ‘Rules governing the distri bu -tion and dissemination of ECMWF real-time products’.

Prof Erland Källén, who took up hispost as Head of Research at ECMWF on6 July 2009, attended the Councilgiving a presentation about ‘Using re-analyses to understand arctic warming’.

EUMETNET’s ‘Oslo Declaration’

BOB RIDDAWAY

EUMETNET is a network of 26European National MeteorologicalServices (NMHSs). It provides aframework for organising co-operative programmes involvingbasic meteorological activities. At ameeting in Oslo on 26–27 March theEUMETNET Council reviewed thecurrent policies concerning

meteorological data. This led to theadoption of the ‘Oslo Declaration’which encapsulates the points onwhich a common position wasagreed during the meeting.

At its meeting on 25–26 June 2009,the ECMWF Council adopted the‘Oslo Declaration’ as a guideline forformulating ECMWF’s data policy.

The following is the ‘OsloDeclaration’.

At their Meeting in Oslo, Norway on26–27 March 2009 to discuss datapolicy matters, the directors of theEUMETNET National MeteorologicalServices, recognising that:– Significant investments andsustained funding are necessary atnational level to maintain and developthe basic components of the EuropeanMeteorological Infrastructure requiredto fulfil the general needs of the


4

NEWS

public, the decision makers and theEuropean economy;– National governments can meetthis basic requirement under differentfunding policies associated to differenteconomic models for NMHS andvariations in data policies;– Although their NMHS operate underdifferent official mandates andeconomic models and accordingly rundifferent activity portfolios, all arecommitted to provide the best possiblecontribution to protection of life andproperty under their Public WeatherService mission, directed at nationalauthorities and the general public;– The internet and the webtechnologies have changed radicallythe expectations from the generalpublic and the access to and possiblesharing and use of data, products andservices;– The ongoing development of theWMO Information System will takeadvantage of these innovativetechnologies;– The legal framework provided bythe EU including the INSPIRE and PSIDirectives places new expectations onthe way the NMHSs provide theirserviceshave reached consensus on thefollowing:1) Data policies of the EUMETNET

members, ECMWF and EUMETSAT andthe related development of on-lineservices, should:– facilitate direct access to basicmeteorological data and products ona non discriminatory basis, under well-documented licensing conditions, incompliance with relevantinternational and national regulationsand with WMO resolutions;– be harmonized as far as possible toachieve greater transparency andimprove accessibility for re-userswithin the constraints of the legalframeworks and respective economicmodels of the NMHS, taking intoaccount relevant discussions in thecontext of EUMETNET, ECMWF,ECOMET and EUMETSAT bodies.2) The evolution of the basic output ofNMHS, as a result of advances inscience and in observation andnumerical weather prediction andtechnology, should lead to regularreviews at NMHS level, and, whenappropriate, to progressive expansionsof:– their set of “Essential” data andproducts made available on a free andunrestricted basis;– their catalogue of data andproducts licensed for re-use by theprivate sector, under the PSI directivewhere applicable.

3) In the same spirit, NMHS shouldalso consider adapting their licensingconditions with the objective ofdelivering more information per priceunit, if financial constraints allow.4) The EUMETNET members shouldcontinue to develop, individually andjointly through EUMETNETprogrammes, on line services givinguser- friendly access to thesecatalogued data and products, inorder to facilitate further the re-use ofpublic meteorological informationand to maximise the societal benefits.5) In the spirit of the INSPIRE directive,the EUMETNET members shouldexpand the set of graphical productsaccessible free of charge to the publicfor general purposes, through theviewing services of their respective websites, whilst continuing to license asappropriate the underlying digitaldata and products.6) In the context of GMES,meteorological data and productsrequired by the operators of the GMESCore Services to deliver new, nonmeteorological public good services,could be licensed to them at noinformation cost to fulfil theirproduction duties, subject tocommitment from EU and nationalauthorities to fund these Core Serviceson a sustainable basis.

ANTONIO GARCIA-MENDEZ

For many years, temperature reportsfrom Indian radiosondes havegenerally been blacklisted in theECMWF operational data assimilationbecause of their inconsistency andrelatively low quality. The IndianMeteorological Department has nowbegun a programme to update itsradiosonde network to use the FrenchModem MK2K GPSonde. The firstphase of this plan can be seen in thetable.

The first five sites were upgraded inMarch to April 2009. The observationsfrom these stations were carefullymonitored at ECMWF during the

Operational assimilation of Indian radiosondes

Station name Identifier Location Date commissioned

Thiruvananthpuram 43371 8.79°N, 76.57°E 9 March 2009

Mohanbari 42314 27.79°N, 95.01°E 9 April 2009

Chennai 43279 12.59°N, 80.11°E 13 April 2009

Port Blair 43333 11.39°N, 92.44°E 16 April 2009

Minicoy 43369 8.16°N, 73.03°E 23 April 2009

Goa 43192 15.79°N, 73.49°E 12 May 2009

Hyderabad 43128 17.77°N, 78.28°E 5 May 2009

Visakhapatnam 43150 17.92°N, 83.18°E 18 May 2009

Patna 42492 25.86°N, 85.06°E 21 May 2009

Srinagar 42027 34.55°N. 74.50°E 24 May 2009


5

NEWS

following weeks to assess the perform -ance using the new equipment. In allcases the quality change was remark -able and the ECMWF operationalmonitoring showed that these all nowconsistently provide good qualitytemperature soundings. The numberand quality of humidity and windobservations also improved, enablingthe ECMWF analysis system to usethese data more extensively than thedata from the previous radiosondes.

On 12 May, these five sites wereremoved from the ECMWF blacklistand their observations are nowassimilated in the operational system.Five more sites were upgraded duringMay. After a short period of moni -toring that confirmed the quality andconsistency of the new systems, thesesites were also removed from theblacklist and included in the assimi -lation. As more stations are upgraded,they too will be assimilated aftershort periods of monitoring.

Indian RS stations June 2009

BlacklistedProposedUnblacklisted

20°N

30°N

80°E

90°E

42027

42492

4312843150

43192

42314

4327943333

43369

43371Standard deviation (°C)

Pres

sure

(hPa

)

1000

850

500

300

200

100

50

2010

50

Bias (°C)0 1 2 3

Beforeupgrade

4 5 6 7 0 2 4-4 -2

Afterupgrade

The Indian radiosonde network. Five stationswere upgraded to use Modem M2K2 GPSsondes in March to April and were assimi-lated at ECMWF from 12 May (green dots);five more stations were upgraded duringMay and assimilated at ECMWF from 16June (blue dots). The remaining stations inthe Indian network have not yet been upgraded(red dots), and temperatures from these sitesare not used at ECMWF.

Upgrade of Indian radiosondes. The figureshows the statist ics for a station in theextreme South of India (Thiruvananthpuram)that was upgraded on 9 March 2009. Shownare the vertical statistics for standard devi-ation ( left) and bias (r ight) between thetemperature observations and the ECMWFanalysis for periods before the upgrade (31January to 8 March) and after the upgrade(9 March to 23 May).

TONY MCNALLY

A Workshop on the assimilation ofIASI in NWP, co-sponsored by theEUMETSAT NWP-SAF (SatelliteApplication Facility) and organized byECMWF, was held at ECMWF from 6 to8 May 2009.

IASI (Infrared Atmospheric Sound -ing Interferometer) was launchedonboard MetOp in October 2006 andis the first hyperspectral instrumentcarried by an operational satellite. Bymeasuring atmospheric radiation inmany thousands of different channelsIASI provides unprecedented informa -tion on temperature, humidity andatmospheric composition. IASI radi -ances have been incorporated intoECMWF’s operational forecastingsystem since June 2007.

The 40 participants at theworkshop represented operationalNWP centres, satellite data providersand academic institutions. Thepurpose of the workshop was to:� Gather experts to assess theprogress that has been made in the

exploitation of IASI and identify keyareas for future development.� Provide training through a sharingof experiences among data providersand data users.

The workshop followed a format ofone and a half days of lectures afterwhich the participants broke out intothree parallel working groups.

In the group dealing with IASIinstrument validation and radiativetransfer modelling there was aconsensus that the IASI spectra were

calibrated extremely accurately andthat the procedures that led to thisexceptional performance should bedocumented and applied to futuremissions. The important role ofvalidation field campaigns, NWPradiance monitoring and inter-comparison activities such as CSICS(Global Space-Based IntercalibrationSystem) was highlighted. It was notedthat the high quality of the IASIobservations placed very stringentrequirements upon the accuracy of

Assimilation of IASI in NWP


6

NEWS

Goodbye GEMS – Hello MACC

ADRIAN SIMMONS

The EU-funded GEMS project ranfrom 1 March 2005 to 31 May 2009.Operation and improvement of thesystems developed during GEMS iscontinuing in a new EU-fundedproject MACC – Modelling Atmos phe -ric Composition and Climate.

The EU-funded GEMS projectstarted on 1 March 2005 under the

leadership of the late Tony Hollings -worth, and came to an end on 31 May2009. GEMS achieved its objectives ofbuilding and demonstrating theoperation of global and regionalanalysis and modelling systems thatmonitor and predict variations inatmospheric constitu ents that forceclimate change and influence airquality, ultraviolet radiation and solarenergy resources. In total, 32 partner

organizations were involved, withECMWF develop ing and operat ing thecore global component and support -ing the regional component with datahandling, validation and display ofresults.

Most of the partners in GEMS willcontinue in MACC, but the project willincrease in size to 48 partners asactivities are included from a relatedESA-funded project called PROMOTE

0

0.5

1

1.5

2

2.5

3

3.5

4

00.10.170.250.320.390.460.540.610.680.750.830.93

July 2003 July 2005Total column carbon monoxide

July 2003 July 2005Aerosol optical depth at 550 nm

radiative transfer models. In particu -lar the group stressed the need forlong-term support of improve mentsin spectroscopy and line-by-linemodelling.

The fact that all NWP centres usingIASI data reported significant impactson forecasting skill was noted by theworking group dealing with IASIassimilation. Indeed some centresassess the impact of IASI as beinggreater than that of any other singlesatellite instrument. It was noted thatthe speed at which IASI was imple -mented in a number of operationalsystems was partly due to the very

good experience gained with AIRSdata. The group acknowledged thatIASI was still significantly under usedand suggested that develop mentsaimed at treating data over land andcloud affected radiances should leadto future improvements.

The application of IASI data to theestimation of trace gas and otherchemical species was dealt with bythe working group on environmentalmonitoring. During the workshop thewealth of information provided byIASI for constituent retrieval wasdocumented. The group noted thatthis activity was the most demanding

on data quality and stressed that anyfuture degrada tion to achievecompression for data transfer shouldbe avoided.

More detailed reports from theworkshop will appear in the forth -coming proceedings, but in the mean -time the presentations are availablefor viewing on:� www.ecmwf.int/newsevents/

meetings/workshops/2009/IASI_data/In conclusion ECMWF would like to

thank EUMETSAT for their supportand thank the participants for makingthe workshop and extremelyinteresting and successful event.

GEMS aerosol and carbon monoxide analyses. These analyses show large differences over north-eastern Asia between July 2003 andJuly 2005, due to much greater fire activity over Siberia in 2003. MACC will provide both daily analyses and forecasts and a reanalysisfor 2003–2010.


7

NEWS

0 4824 1089672 144 168 192 216 240Forecast range (hours)

0.4

0.6

0.8

0.2

1.8

1

1.6

1.4

1.2

Root

mea

n sq

uare

err

or (m

)

1992–19931994–19951996–19971997–19981999–20002001–20022003–20042005–2006

PETER JANSSEN

Prediction of the maximum waveheight of extreme events is oftremendous benefit to the marineworld. A considerable amount ofprogress in wave modelling has beenachieved in the past 20 years or soand there are definite reasons forfurther development. A global waveprediction model has been inoperations at ECMWF since 1992.Since then, the original WAMforecasting system has undergonesignificant enhancements including:� Assimilation of altimeter waveheight data.� Two-way interaction of wind andwaves.� Increase in spatial resolution from330 km to 25 km.

In addition there has been amassive improvement in the ability

to forecast the driving surface windfields. These developments haveresulted in a significant improvementin the accuracy of the wave heightforecasts. In addition the ECMWFforecasts of wave height outperformthose from other operational centres.

More detail about these develop -ments along with information aboutthe ensemble prediction of oceanwaves, extreme wave forecasting andthe use of satellite observations canbe found in the leaflet ‘Ocean waves atECMWF’. The leaflet can be found at:� www.ecmwf.int/about/

information_leaflets/Ocean-waves-English.pdfAlso information about wave fore -

casting is available from a new websitethat has been launched recently: see‘Ocean wave forecast’ at:� www.ecmwf.int/products/forecasts/

d/charts/

Ocean waves at ECMWF

Improvement in the accuracy of wave height forecasts. The improvement in accuracyis illustrated by the reduction in root mean square error of forecast wave height verifiedagainst independent observations from buoys for winters (October to March) since 1992/93.The forecasts are from 12 UTC each day. These results can be interpreted as indicating again of more than two days in the accuracy of the forecasts.

(Protocol Monitoring for the GMESService Element: Atmosphere), whichconcludes later in 2009.

MACC will adopt a more operationalapproach to the running of its systemsand have a stronger focus on theinterface with users and downstream

service providers. It will run from 1June 2009 until late in 2011, when itshould be ready to emerge as a fullyoperational atmos pheric service underthe European Union’s GMES (GlobalMonitoring for Environment andSecurity) programme.

Pending the establishment of acapability for product display anddelivery from the MACC web pages,the GEMS web pages are being used topresent the results from MACC’soperation of global and regionalsystems.

ECMWF AnnualReport for 2008

BOB RIDDAWAY

The ECMWF Annual Report 2008 hasbeen published. It is intended to be animportant way of maintaining theflow of information to all the peopleand institutions that have an interestin ECMWF.

The report draws attention to someof the key events of 2008 that affectedoperational activities.� Revised production schedule. Arevised production schedule was intro -duced with the result that productshave become available from 10 to 15minutes earlier than before, depend -ing on the type of product. 29 January.� Development of a unifiedforecasting system. The MonthlyForecasting System was integratedwith the medium-range VariableResolution Ensemble PredictionSystem (VarEPS). 11 March.� Availability of ERA-Interimproducts. ERA-Interim reanalysisproducts for the years 1989–1998were made available for research andeducation with a latitude/longituderesolution of 1.5°. 10 April.� Provision of marine products forWMO Members. A range of marineforecast products to support severeweather marine forecasting was madeavailable to WMO Members. 15 April.� Assimilation of GPS radiooccultation. GPS radio occultationdata from the GNSS Receiver forAtmospheric Sounding (GRAS)instrument onboard EUMETSAT’sMetOp satellite was activelyassimilated at ECMWF. 28 May.� Implementation of IFS Cycle 33r1. Anew IFS cycle was implemented withimportant changes to the wave model,


8

NEWS

some retuning of the representationof physical processes and improveduse of satellite data. 3 June.� Monitoring altimeter wind andwave data from the Jason-2. ECMWFstarted to receive and monitoraltimeter wind and wave data fromthe Jason-2 satellite. 9 September.� Implementation of IFS Cycle 35r1.A new IFS cycle was introduced withseveral changes to the representationof physical processes, such asimprove ments to the handling ofmelting snow and a new sea surfacetemperature analysis product.30 September.

� Usage of the HPCF. Member States’usage of the ECMWF High Perform -ance Computer Facility resourcesreached a new milestone with 100million units being exceeded in 2008.22 December.

As well as describing how researchand operational activities combine tobring high quality research throughto enhancing operational capabilities,the Annual Report also includesinformation about the following.� Performance of the forecastingsystems. The performance of theforecasting systems remained at ahigh level throughout the year withseveral severe weather events success -fully forecast several days in advance.� High Performance ComputingFacility (HPCF). The existing HPCF,one of the most powerful in Europe,continued to provide a very goodservice to users. In additionpreparations have been made for theinstallation of the new super -computer provided by IBM UK Ltd. � Workshops and training courses. Avariety of workshops were held thathelp shape the research and develop -ment activities at ECMWF. AlsoECMWF contributed to the develop -ment of meteorological expertisewithin Europe by running a variety oftraining courses.

DAVID RICHARDSON

ECMWF organizes an annual meetingof users of its medium range andextended range products. Thepurposes of the meetings are to:� Review the development of theoperational system and to discussfuture developments includingforecast products.� Give users of the ECMWF forecaststhe opportunity to discuss theirexperience with and to exchangeviews on the use of the medium-range and extended-range products

The 2009 Forecast Product UsersMeeting took place at ECMWF from 10to 12 June, with 42 participants repre -senting 17 National Meteoro logical

Services and 9 commercial customersin the Member States and Co-operatingStates.

Two new cycles of the operationalassimilation and forecasting systemhave been implemented since thelast meeting. These changes and theimpact on the forecasts werepresented and included severalmodifications to the physics, such asa revised snow scheme, a new seasurface temperature analysis product,and direct 4D-Var assimilation all-skyof microwave imagers. Plans forfuture developments of the ECMWFforecasting system were alsopresented. These include:� A substantial increase in horizontalresolution during 2009 for the data

assimilation, deterministic forecastand EPS.� Revision of the initial perturbationsand stochastic physics for the EPS.� Further improvements to the dataassimilation and model physics.

The Member States and Co-opera -ting States representatives presentedtheir use of ECMWF products andtheir experience with the forecasts inthe past year. They were generallyvery happy with the quality of theforecasts. It was emphasised thatECMWF products are widely used inthe forecast offices, and are especiallyuseful in providing guidance for earlywarnings of potential high-impactand extreme events. Users noted theimprove ments to the snow forecasts

Forecast Products Users’ Meeting, June 2009

� Collaborative researchprogrammes. ECMWF participated ina number of collaborative researchprogrammes run by the EuropeanUnion and WMO. The outcome ofthese programmes helps improveglobal forecasting, develop new tech -nologies and improve atmosphericmonitoring. � Amended Convention. By the endof 2008, 13 Member States hadnotified acceptance of the amend -ments to the Convention. Also theCouncil authorised the start ofnegotiations with a view to increasingthe number of States that have fullmembership of ECMWF.

Dominique Marbouty, ECMWFDirector, outlined the key develop -ments in his foreword to the AnnualReport and finished by stating that:

“The last year has been verysuccessful and the foundations havebeen laid to build upon the pastachievements of ECMWF in thecoming years. As in the past, thesuccess of ECMWF will continue to bebased on the quality of the staff andthe fruitful relationship betweenECMWF and its Member States.”

The Annual Report can bedownloaded from:� www.ecmwf.int/publications/

annual_report/


9

NEWS

in the last winter, but also highlightedthe issues about the snow cover andnight-time temperature biases thataffected Central Europe in the pastwinter. However, an extreme coldspell affecting Germany had beenwell predicted two weeks in advance.

New products that have beenimplemented this year includesimulated infra-red and water vapourgeostationary satellite images fromthe deterministic forecast. MemberStates confirmed their requirementsfor the new products that are indevelopment at ECMWF, includingtracking of extra-tropical cyclonic

features, strike probability forecastsfor tropical cyclones that developduring the forecast (for bothmedium-range and monthly forecast)and revision of the EPS clustering.Representatives presented somerequests for additional products,mainly to provide more directinformation on weather parameterssuch as visibility, cloud base heightand low level winds.

The forecast products displayed onthe ECMWF web site are increasinglyused during the forecasters’ routineduties in the Member States. Therewas confirmation of their require -

ments for the ECMWF web to providea highly available service and toenhance interactivity, including newfeatures such as zoom, pan andoverlay of charts, and more choice ofparameters for EPSgrams. ECMWF’splans for a re-engineered web serviceto provide such forecast products forthe Member States and Co-operatingStates were reviewed and discussed.

The presentations and summaryfrom the meeting are available on theECMWF website:� www.ecmwf.int/newsevents/meetings/

forecast_products_user/Presentations2009/

Diagnostics of data assimilation system performance

CARLA CARDINALI

The Workshop on ‘Diagnostics of dataassimilation system performance’ washeld at ECMWF from 15 to 17 June2009.

Over the last years data assimi la -tion schemes have evolved intocomplicated systems with millions ofdegrees of freedom and handlingmassive amounts of observations.Effective monitoring of these systemsis required and emerging techniques

are now rapidly developing at mostNWP centres. The review of thevarious methodologies and theireffectiveness in diagnosing theimpact of observations in NWP wassuitably timely.

The workshop considered allavailable techniques to measure theperformance of the NWP assimilationsystem. The impact of observationson the assimilation and the short-range forecast has been assessedthrough different diagnostic tech -

niques. In particular, method ologiesthat evaluate the analysis optimalitywere presented.

Over the last few years, adjointtechniques have also been used toassess the impact of observations onthe short-range forecast. Thismethod ology was described andcompared with the more commonly,so far used, Observation SystemExperiment technique. The use ofadjoint technique to measure theimpact in the short-range forecast ofthe other analysis input parameterswas also presented. Interestingpresentations showed differentmethodologies to assess model bias,for example through re-analysisexperiments or projection of analysisincrements and short-range forecasterror onto the most important modesof the numerical model.

Workshop lectures were byinvitation only. However, a call forposter was introduced to give theopportunity to the younger scientiststo participate and illustrate theirwork. The workshop proceeding willbe issued before the end of the year.

The presentations given at theworkshop along with the posters canbe found at:� www.ecmwf.int/newsevents/meetings/

workshops/2009/Diagnostics_DA_System_Performance/

OZONEGOESMETEOSATAMSUBSSMI-TCWVSSMI

AIRSAMSUA

GPS RO

HIRSSCATMODIS-WVMODIS-IR

METEOSAT-WVMETEOSAT-VISMETEOSAT-IRGOES-WVGOES-VISGOES-IR

GSM-WVGSM-VISGSM-IR

PILOTTEMPAIREPDRIBUSYNOP

OZONEGOESMETEOSATAMSUBSSMI-TCWVSSMIAIRSAMSUAHIRSSCATMODIS-WVMODIS-IRMETEOSAT-WVMETEOSAT-VISMETEOSAT-IRGOES-WVGOES-VISGOES-IRPILOTTEMPAIREPDRIBUSYNOP

Forecast error contribution (kJ kg–1)-50-150

-150

-30 -10 0 10

Forecast error contribution (kJ kg–1)-50 -30 -10 0 10

Summer 2006 (15 June – 15 July) Winter 2007 (5 January – 12 February)

Monitoring ECMWF’s data assimilation system. Contribution of the various componentsof the observing system to the 24-hour forecast error for a period in the summer of 2006 (leftpanel) and a period in the winter of 2007 (right panel). Negative (positive) values are asso-ciated with a decrease (increase) of forecast error due to different component of the observingsystem. Globally the observations contribute to decrease in the forecast error.


10

METEOROLOGY

ROBERTO BUIZZA, FLORIAN PAPPENBERGER,PETER SALAMON, JUTTA THIELEN, AD DE ROO

by the hydrological model to predict river dischargelevels. EFAS probabilistic forecasts use data from theECMWF EPS as initial and boundary weather conditions.

The ECMWF EPSThe EPS has changed several times since its imple-mentation in 1992 (see Palmer et al., 2007 for moredetails). Since 11 March 2008, the ECMWF EPS runstwice-a-day, at 00 and 12 UTC, with a variable resolution(Buizza et al., 2006 and Vitart et al., 2008). It includes51 members: one starting from unperturbed initialconditions (defined by interpolating the high-resolutionT799L91 analysis to the ensemble resolution) and 50members starting from perturbed initial conditions.

The perturbed initial conditions are constructed asa linear combination of the perturbations with thefastest growth over a 48-hour period, provided by singu-lar vectors computed with a T42L62 model and a totalenergy norm. The perturbed forecasts are integratedwith a stochastic scheme designed to simulate the effecton forecast error of random model errors due to uncer-tainties in the parametrized physical processes (thestochastic physics).

The 00 UTC EPS is run at T399L62 resolution fromday 0 to day 10 with persisted sea-surface-tempera tureanomalies, and then at T255L62 resolution from day 10to day 15 (day 32 every Thursday at 00 UTC) coupledwith an ocean model. Every Thursday the ensemble isextended to 32 days to cover the monthly forecastrange. The 12 UTC EPS has the same configuration asthat run at 00 UTC, but uses persisted sea-surface-temperature anomalies also between day 10 and 15,instead of a coupled ocean model.

The JRC EFASThe European Flood Alert System (EFAS) was launchedin 2003 by the European Commission with the aim toincrease preparedness for floods in trans-nationalEuropean river basins (Thielen et al., 2009; Bartholmeset al., 2009). The system being developed has two mainobjectives:� To complement European Member States activities

on flood preparedness and to achieve longer earlywarning times.

� To provide the European Commission with anoverview of ongoing and expected floods in Europefor improved international aid and crisis managementin the case of large transnational flood events thatmight need intervention on an international level.

The EFAS prototype that is currently run operationallyis set up for the whole of Europe on a 5-km grid. Twicedaily it provides the national hydrological centres with

EPS/EFAS probabilistic flood prediction forNorthern Italy: the case of 30 April 2009

AFFILIATIONSRoberto Buizza, Florian Pappenberger, ECMWF, Reading, UKPeter Salamon, Jutta Thielen, Ad de Roo, JRC, Institute forEnvironment and Sustainability, Ispra, Italy

ENSEMBLE hydrological predictions generated by theEuropean Union Joint Research Centre European FloodAlert System (JRC EFAS) driven by the ECMWFEnsemble Prediction System (EPS) have been used toassess the risk of flooding of the Po’ river at the end ofApril 2009. This case illustrates the added value of usingprobabilistic flood predictions to signal the possibleoccurrence of flooding, and confirms statistically-basedresults published in the scientific literature. It shows thatthe key advantage of ensemble prediction systems,compared to systems that rely on one single forecast,is that they can be used not only to identify the mostlikely outcome, but also to assess the probability ofoccurrence of extreme/rare events.

Medium-range ensemble prediction systems are todaypart of the operational suite at many meteorologicalcentres. Nine centres (in Australia, Brazil, Canada,China, England, Japan, Korea and the USA; see, forexample, Buizza et al., 2008) run global, medium-rangeensemble prediction systems, and many regional centresare running limited area ensemble prediction systems(e.g. in Australia, England, France, Germany, Italy,Norway and Spain). The past decade has seen anincreased use of ensemble forecasts; see, for example,the work done within the DEMETER project on estab-lishing the utility of coupled multi-model ensembleforecasts, in particular in the agriculture and healthsectors; www.ecmwf.int/research/demeter/.

Another area where the value of an ensembleapproach has been widely recognized is hydrology,which has seen several institutions developing and test-ing ensemble-based flood prediction systems that useensemble weather forecasts as initial and boundaryconditions (see, for example, the work done within theHEPEX project, Thielen et al., 2008). One of thesehydrological ensemble systems for flood prediction isEFAS, developed and successfully implemented by JRCin Ispra, Italy.

Probabilistic prediction of severe water levelconditions with EPS/EFAS

The EPS/EFAS flood prediction system runs twice-a-dayat JRC using forecasts of the weather variables required


11

METEOROLOGY

medium-range ensemble flood forecasting informa-tion. In addition, when a high probability for floodingis forecast, the EFAS partners are alerted by e-mail andadvised to monitor the development of the situationusing the EFAS information system. Currently, fore-casts with lead times of 3 to 10 days are achieved throughthe incorporation of ensemble and single forecasts:� Two deterministic weather forecasts from ECMWF

and the German Weather Service (DWD).� Two sets of ensembles: the medium-range EPS from

ECMWF (51 members, 10 days) and the shorterrange COSMO-LEPS (16 members, 5 days) providedby the Regional Meteorological Service of Emilia-Romagna (ARPA-SIM), Bologna, Italy.

EPS/EFAS probabilistic predictions for28–30 April 2009

To illustrate the value of EPS/EFAS probabilistic floodpredictions in weather-related risk management, thesevere meteorological and hydrological conditions thataffected Northern Italy at the end of April 2009 are nowdiscussed. For this example, only precipitation fore-casts generated by the ECMWF EPS and thecorresponding EPS-driven EFAS flood predictions havebeen used.

April 2009 was a rather wet period for Northern Italy.Towards the end of the month, river levels were quitehigh due to springtime rain and the beginning of thesnow melting (Northern Italy witnessed very heavy snow-falls during winter 2008/09). On 29 April, Italiannewspapers started listing the various damages due tothe very high river levels caused by the heavy precipi-tation that hit this region on the 27th and 28th. It isinteresting to read what was reported about this eventin a local Italian news paper (VareseNews from Ispra).

Figures 2 and 3 show some of the consequences ofthe high river levels: in Alessandria, one of the maincities of Piemonte, over Northwestern Italy, some areaswere flooded and some of the bridges were closed dueto the extremely high level of the river Tanaro (Figure2). In Piacenza, on 30 April one of the Po’ river bridgescollapsed causing accidents and casualties (Figure 3).The collapse has been linked to the water pressureassociated with the extremely high levels of the Po’.

Precipitation forecastsRiver flooding was due to the heavy precipitation thataffected this region between 27 and 29 April. On the26th, the upper-level atmospheric flow was characterizedby a deep trough positioned over Spain with a south-west-northeast axis that started funnelling hot, moist airtowards Northwest Italy. The surface flow was charac-terized by a weak cyclonic circulation centred on theBalearic Islands. During the subsequent 48 hours, theupper-level trough moved slowly towards Italy with itsaxis gradually tilting from southwest-northeast towardssouth-north, and then towards a southeast-northwestdirection, while at the surface the cyclonic circulationintensified. This upper level and surface circulationcaused heavy precipitation over Italy, especially overthe Po’ Valley.

Figure 4 shows the areas of heavy precipitation (48-hour accumulated precipitation in excess of 15 mm)given by the 0–48 hour forecast produced by the ECMWFhigh-resolution model, and the corresponding observed

Verese News, 29 April 2009.

Piena del Po: confermate le previsioni del Ccr

La piena del Po di questi giorni era stata prevista eannunciata dal Centro Comune di Ricerca (CCR, alsonamed Joint Research Centre, JRC) di Ispra da venerdìscorso. Lo comunica lo stesso centro comunitario cherende nota l’attività svolta da EFAS, il sistema europeodi allerta sulle alluvioni ospitato all’interno dello stessoCCR. Il 24 aprile i responsabili di EFAS hanno inviato lanotifica di allerta sulla piena del Po all’IstitutoSuperiore per la Protezione e la Ricerca Ambientale, alleautorità competenti per le acque del Po e alleorganizzazioni nazionali partner per il grande fiumepadano. Le previsioni di EFAS infatti prevedevanoun’alta probabilità di alluvioni del Po’ a partire dal 27aprile, con un picco d’onda tra il 28 e il 29. L’avviso èstato emesso con 4–5 giorni di anticipo, il che confermail grande potenziale di questo sistema di avvertimentopreventivo. A partire da venerdì 24 inoltre i partneritaliani dell’EFAS hanno potuto seguire l’andamentodelle previsioni, aggiornate ogni 12 ore sull’area adaccesso riservato del sito Internet del sistema.

Figure 1 Ext ract f rom VareseNews o f 29 Apr i l 2009 (seehttp://www3.varesenews.it/varese/).

Po’ river flood: the JRC forecast is confirmed

JRC predicted and communicated last Friday the highlevel reached by the Po river during these days. Thenews comes from JRC, which runs EFAS, the EuropeanFlood Alert System. On the 24th of April, EFAS person-nel sent an alert message to the Italian ProtectionAgency, the Po’ river Authority and other partnerauthorities. EFAS was predicting a high probabilitythat the Po river would reach flood levels from the27th of April, with peak levels between the 28th and the29th. The warning was issued 4–5 days before theevent, confirming the great potential of the EFAS-based warning system. Since Friday the 24th, EFASpartners could follow the sequence of EFAS predictions,updated every 12 hours, from the JRC web site, via thearea that gives them restricted access to the EFAS fore-casts (see Figure 1 for the original Italian article).

This newspaper article indicates that users are becom-ing increasingly familiar with the notion of a proba bilisticforecast, and that decisions are made based on proba-bilistic forecasts.


12

METEOROLOGY

precipitation. These two maps show that the whole of Italywas affected by heavy precipitation (note that the figureshave shading starting from 15 mm, thus precipitationvalues between 1–15 mm are not shown), with very highvalues affecting Northwest Italy. Over this region, theobserved precipitation and the 48-hour forecast showvalues between 30–60 mm, with a small area with valuesbetween 60–120 mm. Note, however, that care must betaken when comparing these with single, grid-pointobservations, since these two fields are defined on a0.25° grid, and thus represent average values at this scale.

Hereafter, attention will focus on the probabilisticprediction of 48-hour accumulated precipitation exceed-ing these two values: 15 and 30 mm.

Figure 5 shows the EPS probabilistic forecasts of 48-hour accumulated precipitation exceeding 15 and 30mm issued on the 25th (T+48 to T+96 hour) and 23rd(T+96 to T+144 hour). The corresponding forecastsissued on the 21st (T+144 to T+192 hour) and 19th(T+192 to T+240 hour) of April are given in Figure 6.

The EPS forecasts issued on the 23rd (Figure 5a)give probabilities of 40%–100% of precipitation exceed-ing both thresholds in the area where these amountswere observed. The probabilities issued two days earlier,on the 23rd (Figure 5b), are still correctly located and

similar to those for the 15 mm threshold, but lowerfor the 30 mm threshold.

Forecast probabilities issued four days earlier, onthe 21st (Figure 6a), are also correctly positioned forthe 15 mm threshold, although they are now lower,20%–40%, while the probabilities for the 30 mm thresh-old are positioned to the west of the area where thisamount of rain was observed. This is an indication ofa too slow propagation of the trough in the EPS fore-casts. The forecast probabilities issued on the 19th(Figure 6b) give a 10%–20% probability that precipi-tation in excess of 15 mm could affect the Po’ Valley,and signal that there is a 2–10% probability that precip-itation could exceed 30 mm.

Although it is not possible to judge the quality of theEPS probabilistic precipitation forecasts from one caseonly (for this we refer the reader to published literature,see e.g. Pappenberger & Buizza (2008) and Bartholmes etal. (2009)), these maps indicate that the EPS gave goodprobabilistic forecasts up to T+192 hour, with a weakersignal also present at T+240 hour.

Flood forecastsEFAS gave the first indications of a possible floodingin the Po’ river basin in the forecast issued at 12 UTC

La Stampa, 29 April 2009.

La Regione Piemonte «Stato di emergenza dopo lapiena». In tutto il Piemonte ci sono stati allagamenti esmottamenti di strade. Situazione difficile oltre che adAlessandria anche nel Cuneese, nell’Astigiano e in Vald’Ossola. La presidente Mercedes Bresso ha chiesto lostato di emergenza. Allerta per il Po’ invece dallaProtezione civile dell’Emilia-Romagna, dove la piena e’attesa per oggi. La piena in Piemonte, dovrebbe avereuna portata stimata sui 7.000 metri cubi al secondo.

La Repubblica, 30 April 2009.

Crolla un’arcata, precipitano 4 macchine. Grave unautomobilista ricoverato in rianimazione. Piena delPo’, crolla ponte a Piacenza Le auto finiscono in acqua:4 feriti Piena del Po’, crolla ponte a Piacenza. Un’arcataha ceduto alla furia della piena del fiume e l’asflato siè piegato verso l’acqua trascinando quattro auto. Lastrada si è piegata in una sorta di “v”.

Figure 2 Alessandria, 29 April 2009. The Tanaro river reaches extremely high levels (photo: A Contaldo, Photonews, available from http://torino.repub-blica.it/)

Figure 3 Piacenza, 30 April 2009. The collapse of one of the bridges across the Po’ (photo: P Caridi, WordPress, available fromhttp://peppecaridi2.wordpress.com/).


13

METEOROLOGY

on 21 April. However, the number of ECMWF EPSmembers (and also of the single forecasts from theDWD and ECMWF high-resolution systems) exceedingalert thresholds were not persistent until the 00 UTCforecast from the 24th. On that date, both single high-resolution forecasts and the probabilistic forecastssimulated discharges exceeding the EFAS high-levelthreshold throughout the Po’ catchment, and confirmedthe earlier indications of a possible flooding from theprevious forecasts. Hence, a flood alert was sent on the24th to the Italian national authorities, informing themabout the high risk of flooding in the Po’ basin from

27 April until 2 May. The alert suggested that theyshould keep on monitoring the situation carefully usingthe subsequent forecasts on the EFAS web interface aswell as on their national systems.

Figure 7 shows the spatial distribution of the numberof ensemble members of the ECMWF EPS forecast(started at 00 UTC on 24 April) simulating dischargeswhich exceed the high-level threshold. EFAS predictedincreased, though not extreme, discharges for almostall tributaries within the Po’ catchment. The increasedflows were principally caused by the predicted heavyprecipitation and only to a minor extent by snowmelt

45°N

50°N

5°E 10°E 15°E

15 30 60 120 240 mm

45°N

50°N

5°E 10°E 15°E

a 48-hour accumulated precipitation b Observed accumulated precipitation

45°N

50°N

5°E 10°E 15°E

45°N

50°N

5°E 10°E 15°E

a T+48 to T+96 hour forecasts issued on 25 April

45°N

50°N

5°E 10°E 15°E

2 10 20 30 60 100 %

45°N

50°N

5°E 10°E 15°E

b T+96 to T+144 hour forecasts issued on 23 April

Probability (rain >15 mm) Probability (rain >30 mm)


Figure 5 (a) T+48 to T+96 hour forecastprobability of occurrence of precipitation inexcess of 15 mm (left) and 30 mm (right)issued on 25 April. (b) The correspondingT+96 to T+144 hour forecast probabilityissued on 23 April. The shading indicatesthe probabil ity intervals specif ied in thelegend.

Figure 4 (a) The 0-48 hour forecast of accu-mulated precipitation in excess of 15 mmfrom the ECMWF high-resolut ion model(T799L91) issued at 12 UTC on 27 April andvalid between 12 UTC on the 27th and 29th.(b) The 48-hour accumulated observed precip-itation, computed by interpolating onto aregular grid the observations from SYNOPstations. The shading indicates the precipi-tation intervals specified in the legend.


14

METEOROLOGY

processes, despite the exceptionally large snow accu-mulations in the Southern Alps. However, the increasein discharge for the majority of tributaries resulted insignificantly high flows in the main part of the Po’.

The temporal development of the EFAS/EPS forecastsfor reporting points close to Alessandria (upper Po’),Piacenza (lower Po’) and Ferrara is illustrated in Figure 8.

The probabilistic forecast indicated that the highestrisk of flooding would occur around 28 and 29 April inthe upstream part of the Po’, and around 29 April to 1May for the downstream part. However, the start of theflooding on the 27th in the upper Po’ and on the 28thin the lower Po’ showed a varying number of ensemblemembers simulating discharges above the thresholdlevel even with a lead time of two days. This was mostlikely caused by the temporal resolution of the EFASsimulations and/or the different initial conditions.

Figure 9 illustrates the added value of using proba-bilistic forecasts. Neither of the EFAS forecasts driven bythe single, high-resolution forecasts (ECMWF T799L91and DWD) could capture the likely high increase ofdischarge from the 27th onwards. Although both singleforecasts lie within the uncertainty range predicted bythe probabilistic forecasts, it was not until the 00 UTCforecast on the 26th that a significant exceedance ofthe alert level was predicted (not shown here).

Unfortunately, no discharge values are yet availablefor the location shown in Figure 9. Nevertheless, theobserved water levels at a reporting station close toAlessandria (Figure 10) confirm the predicted peak onthe 28th, although the EFAS forecasts simulate a longer-lasting exceedance of high alert levels. Note that alertlevels between EFAS and national/local authoritiescannot be compared directly as they were derived

51 ensemble members abovehigh threshold

1 ensemble member abovehigh threshold

a b c

45°N

50°N

5°E 10°E 15°E

45°N

50°N

5°E 10°E 15°E

a T+144 to T+192 hour forecasts issued on 21 April

45°N

50°N

5°E 10°E 15°E

2 10 20 30 60 100 %

45°N

50°N

5°E 10°E 15°E

b T+192 to T+240 hour forecasts issued on 19 April



Figure 6 (a) The T+144 to T+192 hourforecast probability of occurrence of precip-itation in excess of 15 mm (left) and 30 mm(right) issued on the 21 April. (b) The corre-sponding T+192 to T+ 240 hour forecastprobability issued on 19 April. The shadingindicates the probability intervals specifiedin the legend.

Figure 7 Number of ensemble members based on the ECMWF EPSforecast from 00 UTC on 24 April simulating discharges whichexceed the EFAS high alert level for the Po river basin. The circlesdenote the locations of reporting points as used in Figure 8.


15

METEOROLOGY

differently. Overall, the comparison of the forecastswith observations at other locations reported by ARPA-SIM after the flooding occurred indicated that EFASpredicted the high risk of flooding with a lead time ofthree to four days.

EPS/EFAS future changes

One of the key advantages of ensemble predictionsystems compared to systems that rely on one single fore-cast is that they can be used not only to identify the mostlikely outcome, but also to assess the probability ofoccurrence of extreme/rare events. The EPS/EFASdischarge prediction for Alessandria illustrated thatthis was actually the case for the situation that occurredon 28–30 April.

Work is progressing to continuously improve theECMWF EPS and the JRC EFAS systems.

It is planned that ECMWF will introduce three keymodifications to the EPS that are expected to furtherimprove its quality.� The scheme used to simulate the effect of model

errors is going to be upgraded.� The methodology followed to generate the EPS initial

conditions is going to be modified to include anensemble data assimilation system.

� The EPS resolution during the first 10 days will beincreased from 50 to about 30 km. In parallel to thischange the resolution of the single high-resolutionand of the data-assimilation system will also beincreased.

These changes are expected to further improve thequality of EPS probabilistic forecast. Results from thiswork will be reported in due course.

In the near future EFAS is expected to improve itsperformance significantly through the assimilation ofdata from higher-density station networks, which are nowcollected in real time through the EU-FLOOD-GIS datacollection system running at the JRC. This will improvethe calibration of the system, the determination of crit-ical thresholds and the calculation of initial conditions.

While case studies will remain an important part ofthe verification of the skill of the system to predictmedium to major floods, a newly developed skill scoremodule is now being implemented pre-operationally tomonitor the skill of the probabilistic forecasts at regu-lar intervals. In addition to the constant improvementof EFAS as an early flood warning system in Europe, itis also investigating which other products from EFAScould be made available to the partners and researchersin general (e.g. European soil moisture maps, low flowcalculations or flood monitoring).

Furthermore, feasibility studies are currently ongo-ing to test if EFAS methodologies can be transferred toAfrican basins, where climatology is different than inmid-latitude Europe, and where an early warning systemneeds to satisfy different requirements. If successful, anextension to a global application could be envisaged.

Forecast day 22 23 24 25 26 27 28 29 30 01 02 03 04

2009042200 2 21 22 16 11 4

2009042212 34 38 35 30 22 13

2009042300 44 47 46 38 30 24

2009042312 45 48 48 46 40 33 24

2009042400 8 44 46 45 45 40 32 25

2009042412 31 40 38 36 31 21 16 15

2009042500 6 29 31 23 21 10 6 4 3

2009042512 42 45 40 37 34 25 21 15 9

Forecast day 22 23 24 25 26 27 28 29 30 01 02 03 04

2009042200 19 18 4 1

2009042212 21 32 21 10 3

2009042300 4 38 43 27 10

2009042312 44 48 42 22 6 3

2009042400 23 44 46 43 23 1 1

2009042412 36 41 39 28 6 2 5

2009042500 8 43 40 25 4

2009042512 44 47 42 34 7 2

Forecast day 23 24 25 26 27 28 29 30 01 02 03 04 05

2009042200 20 26 28 20

2009042212 17 37 40 37 18

2009042300 13 43 46 40 29

2009042312 35 42 44 40 31 11

2009042400 31 46 45 44 35 14

2009042412 51 51 51 49 41 20 4

2009042500 3 47 51 51 51 42 15 3

2009042512 40 51 51 51 49 29

b Piacenza

c Ferrara

a Alessandria

Figure 8 Temporal evolution of the ECMWF EPS ensemble membersexceeding the EFAS high alert level for a location close to (a)Alessandria, (b) Piacenza and (c) Ferrara. The column on the left denotesthe date and time of the forecast. The colour scale indicates the numberof ensemble members above the high alert level (see also Figure 7),and can be used to compute the probability of occurrence of high-alert conditions.


16

METEOROLOGY

FURTHER READINGBartholmes, J.C., J. Thielen, M.H. Ramos & S. Gentilini,2009: The European Flood Alert System EFAS – Part 2:Statistical skill assessment of probabilistic and deterministicoperational forecasts. Hydrol. Earth Syst. Sci., 13, 141–153.Buizza, R., Y.-Y. Park, M. Leutbecher & F. Pappenberger,2008: Predictability studies using TIGGE data. ECMWFNewsletter No. 116, 16–20.Buizza, R., J.-R. Bidlot, N. Wedi, M. Fuentes, M. Hamrud,G. Holt, T. Palmer & F. Vitart, 2006: The ECMWF VariableResolution Ensemble Prediction System (VAREPS).ECMWF Newsletter No 108, 14–20.Palmer, T.N., R. Buizza, M. Leutbecher, R. Hagedorn,T. Jung, M. Rodwell, F. Vitart, J. Berner, E. Hagel,A. Lawrence, F. Pappenberger, Y.-Y. Park, L. van Bremen &I. Gilmour, 2007: The ECMWF Ensemble PredictionSystem: recent and on-going developments. A paperpresented at the 36th Session of the ECMWF Scientific

Advisory Committee. ECMWF Tech. Memo. No 540.Pappenberger, F. & R. Buizza, 2008: The skill of ECMWFprecipitation and temperature predictions in the Danubebasin as forcings of hydrological models. ECMWF Tech.Memo. No. 558.Thielen, J., J. Schaake, R. Hartman & R. Buizza, 2008:Aims, Challenges and Progress of the HydrologicalEnsemble Prediction Experiment (HEPEX): a summary ofthe 3rd HEPEX workshop held in Stresa 27–29 June 2007.Atmos. Sci. Lett., 9, 29–35.Thielen J., J. Bartholmes, M.-H. Ramos & A. de Roo, 2009:The European Flood Alert System-Part 1: Concept anddevelopment. Hydro. Earth Syst. Sci., 13, 125–140.Vitart, F., A. Bonet, M. Balmaseda, G. Balsamo, J.-F. Bidlot,R. Buizza, M. Fuentes, A. Hofstadler, F. Molteni &T.N. Palmer, 2008: Merging VAREPS with the monthlyforecasting system: a first step towards seamless prediction.ECMWF Newsletter No. 115, 35–44.

24/04 25/04 26/04 27/04 28/04 29/04Date

30/04 01/05 02/05 03/05

1600

Dis

char

ge (m

3 s–1

)1400

1200

1000

800

600

400

200

0

24/04 25/04 26/04 27/04 28/04 29/04Date

30/04 01/05 02/05 03/05

8

Obs

erve

d w

ater

leve

l (m

etre

s) 7

6

5

4

3

2

1

0

Piena straordinaria

Piena ordinaria

Figure 9 Hydrograph for the reporting point close to Alessandria (see also top panel of Figure 7) for the 00 UTC forecast on the 24 April.The black dots denote the discharge as simulated based on observed data. The black and brown solid lines denote the predicted dischargeusing the single high-resolution forecasts from the DWD and ECMWF, respectively. Note that the DWD forecast ends at T+168 hour. TheBox-Whisker plots denote the predicted discharges using the ECMWF EPS forecasts. The background colours correspond to the low (green),medium (yellow), high (red) and severe (purple) alert levels.

Figure 10 Daily averaged observed water levels at a reporting point close to Alessandria. Green and red solid lines denote the ordinaryand extraordinary alert levels (piena ordinaria and piena straordinaria) as used by the Piemonte regional authorities.


17

METEOROLOGY

KRISTIAN S. MOGENSEN,MAGDALENA ALONSO BALMASEDA, ANTHONY WEAVER,

MATTHEW MARTIN, ARTHUR VIDARD

THE OCEAN-ATMOSPHERE coupled model used formonthly and seasonal forecasts requires the modellingand initialization of the three-dimensional global ocean.To this end, ECMWF has been using the HOPE(Hamburg Ocean Primitive Equation) ocean modeland an Optimal Interpolation (OI) data assimilationsystem since the beginning of the coupled-model fore-casting activities in 1995. However, with the consolidationof a large part of the European ocean modelling activ-ities around NEMO (Nucleus for European Modellingof the Ocean; Madec, 2008), it was decided that ECMWFshould adopt NEMO as the ocean component of itscoupled forecasting system.

NEMO is a comprehensive ocean modelling frame-work consisting of a dynamical ocean model (OPA), seaice model (LIM) and biochemistry models. ECMWF isprimarily concerned with the ocean component of theNEMO system, but is planning to investigate the use ofthe LIM sea ice model in the future. The next opera-tional seasonal forecast system (System 4) will use theatmospheric model from the Integrated Forecast System(IFS) linked to NEMO via the OASIS coupler that is usedto exchange relevant fields. The same coupled modelswill also be used for monthly forecasting within theECMWF Ensemble Prediction System (EPS).

Part of the transition from HOPE to NEMO involvessetting up a new system for providing ocean initialconditions. Rather than adopting the current OIscheme used by HOPE, it was decided to pursue thedevelopment of a variational-based assimilation system(NEMOVAR) in collaboration with CERFACS (Euro -pean Centre for Research and Advanced Training inScientific Computation) and the Met Office. Otherresearch institutes, such as INRIA/LJK (Institut Nationalde Recherche en Informatique et Automatique/Labora toire Jean Kuntzmann) based in Grenoble, arealso making significant contributions to NEMOVAR.Though we usually refer to this system as NEMOVAR,at present it only applies to the dynamical ocean part

(OPA) of the NEMO system. This article describes theimplementation plan and current status of NEMOVAR.

Development of NEMOVAR

The basis for the NEMOVAR developments is OPAVAR(Weaver et al., 2005), a variational assimilation systemdeveloped for a previous version of OPA. The OPAVARsystem has been used for several studies as well asEuropean research projects such as DEMETER, ENACTand ENSEMBLES. It supports 4D-Var as well as 3D-Varsince it includes the tangent-linear and adjoint codesof the previous version of OPA. It has mostly been usedfor low-resolution (~2°) studies of the global ocean. Itis based on an incremental algorithm with outer andinner ‘loops’, similar to that of the IFS 4D-Var. Theseconcepts will be discussed later.

OPAVAR has three major drawbacks.� It is based on an obsolete version of OPA that is no

longer maintained as part of NEMO.� It is not straightforward to adapt to different ocean

model resolutions.� It cannot be run using distributed memory paral-

lelization.These points are severe limitations when porting thecode to future computer architectures and for higherresolution applications.

For these reasons it was decided to developNEMOVAR. In the following we summarize the keyfeatures of OPAVAR and describe the implementationof NEMOVAR.

The OPAVAR legacyOPAVAR is able to assimilate observations from sub-surface profiles of temperature and salinity, along-tracksea level anomaly (SLA) observations from satellitealtimeters, and sea surface temperature (SST) maps. Ithas a multivariate background-error formulation fortemperature, salinity, horizontal components of veloc-ity, and sea surface height. The multivariate formulationcaptures the local relationship between temperature andsalinity obtained by imposing approximate preserva-tion of water masses, an equation of state relatingtemperature and salinity to density, geostrophic adjust-ment of the horizontal components of velocity, andlocal relationships between sea surface height and verti-cal density profiles. The details can be found in Weaveret al. (2005) and references therein.

Like most assimilation systems, OPAVAR uses a mech-anism where observations within a given time windoware collected to compute an analysis. The analysis isobtained by combining observations with a first-guess

AFFILIATIONSKristian S. Mogensen, Magdalena Alonso Balmaseda,ECMWF, Reading, UKAnthony Weaver, CERFACS, Toulouse, FranceMatthew Martin, Met Office, Exeter, UKArthur Vidard, INRIA/LJK, Grenoble, France

NEMOVAR: A variational data assimilationsystem for the NEMO ocean model


18

METEOROLOGY

(background) field, i.e. the previous analysis propa-gated to the current analysis time using the model.The details are given in Box A.

Development of the current version of NEMOVARFor the development of NEMOVAR the following taskswere identified:� Development of observation operators.� Development of an assimilation system that is based

on 3D-Var.� Implementation of an online, automatic system for

the quality control of real-time observations.� Development of a system to produce an ensemble of

analyses to provide estimates of the uncertainty in theocean initial conditions.

� Development of the tangent-linear and adjointmodels of the ocean component of NEMO.

� Development of an assimilation system that is basedon 4D-Var.The first four tasks are needed for the initial opera-

tional implementation of NEMOVAR at ECMWF, whichin addition will require other developments (see later).Some of the tasks are not completely independent.The 3D-Var system needs the observation operators inorder to compute the cost function to be minimized.However, other components such as the quality control

of observations and the NEMO tangent-linear andadjoint models are independent and allow for paralleldevelopments by different groups.

The observation operators in NEMO are central toNEMOVAR but are also valuable for model validationsince they allow online comparison with a variety ofdatasets. Presently, observation operators have beendeveloped for temperature and salinity profiles, SLAfrom satellite altimeters, SST (both from maps andfrom individual measurements), and currents (e.g.from the TAO (Tropical Atmosphere Ocean) dataset).The observation operator for currents is a new devel-opment that is not available in OPAVAR.

The NEMOVAR 3D-Var system has reached a fairlymature state. The first assimilation experiments forextended periods have been performed by assimilatingtemperature and salinity data from the quality controlledEN3 dataset provided by the Met Office as part ofENSEMBLES (an EU-funded project that aims toconstruct scenarios of future climate change with ensem-ble simulations of Earth-system models). This newdataset includes corrections to expendable bathyther-mographs (XBTs) to compensate for errors in theirfall-rates. Some of the results from these experimentswill be discussed below.

Work is ongoing to improve the specification of theobservation-error and background-error covariancesand the models used to represent them in NEMOVAR.In addition, work is being carried out to implement amultivariate model bias correction scheme similar to theone that exists in the current HOPE/OI (System 3). Thebasic modules for assimilation of SLA and SST dataare in place, and testing is underway.

The work on the tangent-linear and adjoint of theNEMO model is in an advanced state and developmentof a prototype 4D-Var system will start later this year.

The quality control system used by the Met Office(Ingleby &Huddleston, 2007) for their operational oceanforecasting system and for producing the ENSEMBLESdatasets has been implemented in the NEMOVARsystem. The system is currently used for temperature andsalinity profiles and will be extended to include SLA andSST data.

Results from some preliminary experiments will nowbe described.

Setup of the experiments

The next ECMWF seasonal forecasting system, based onNEMO, will use a 1° global configuration. This config-uration, known as ORCA1, has been chosen for theinitial experimentation with NEMOVAR.

The impact of assimilation is assessed by comparinga 3D-Var experiment with an experiment without dataassimilation (the control). Both experiments are forcedwith daily atmospheric data from the ERA-Interimreanalysis. They start with the same initial conditions,obtained after 20 years of spin-up using daily climato-logical data from ERA-Interim. The model SST is

Box AVariational algorithm implemented in OPAVARand NEMOVAR

The variational algorithm implemented in OPAVAR,which has been migrated to NEMOVAR, consists ofthe following steps:� An initialization phase which computes the initial

(background) model trajectory and the misfitsbetween the observations and this trajectory. Thismodel integration uses the model state at theend of the previous cycle as the starting point.

� One or several outer loops consisting of:– An inner loop that minimizes a quadratic cost

function to produce a set of increments(changes to the model field). This cost func-tion depends on the trajectory and the misfits.

– An update of the trajectory and misfits,obtained by applying the increments to themodel. The updated misfits and trajectory canthen be used as input for another minimizationif multiple outer loops are required.

The analysis is formed by adding the resulting incre-ments to the first guess. This can be performedeither directly or by using Incremental AnalysisUpdates (IAU) where the increments are appliedgradually over several time steps. The IAU approachis typically used in 3D-Var to reduce spurious high-frequency adjustment processes resulting from theinitialization procedure.


19

METEOROLOGY

strongly relaxed to external SST analyses (NOAA SSTOIv2), and there is a three-year time-scale damping tothe climatology of temperature and salinity from theWorld Ocean Atlas. Rather than using the NEMO icemodel, these experiments prescribe the ice cover. Theexperiments start from an initial state on January 1989and continue until the end of 2006.

For the control experiment, the model equivalent ofthe temperature and salinity observations werecomputed for diagnostic purposes only, whereas forthe 3D-Var experiment both the temperature and salin-ity data were assimilated. An assimilation window of 10days was used. The 3D-Var experiment was performedwith a single outer loop and a fixed set of 40 iterationsused for minimization in the inner loop. With thesettings used for this experiment the computationalcost of the inner loop is roughly equivalent to a 10-daymodel integration. This means that two thirds of thecomputer time is spent on integrating the model for the3D-Var system since two model integrations are needed:one for the background trajectory initialization andone for the application of the analysis increment. Fordifferent configurations this cost ratio might change.

Statistics of the model fit to the observations

Figure 1 shows the root-mean-square (rms) error andmean error (or bias) of the model for (a) potentialtemperature and (b) salinity observations as a func-tion of depth for the global ocean and for a region inthe central Equatorial Pacific (NINO3.4: 5°S–5°N,170°–120°W). Shown are the values for the controlexperiment and for both the outer and inner loops ofthe 3D-Var experiment. The biases and rms errors forthe 3D-Var experiment are clearly reduced comparedto the control experiment.

The bias between the analysis and the observationsin the inner loop is very close to zero at all depths. Theouter loop results indicate the fit between the modelbackground and observations in the assimilation windowbefore the analysis is performed. If either the modelbackground or observations are biased, the bias in theouter loop will not be zero – this is clearly the casefrom Figure 1. As expected the rms error of the innerloop departures is lower than that of the outer loop sincewe are fitting the observations in the analysis. A largediscrepancy between the outer and inner loop fits to theobservations is an indication that the model is havingdifficulty in retaining the information brought by assim-ilation. Figure 1 shows that this is the case for the centralregional of the Equatorial Pacific.

The rms error and bias statistics only give part of thepicture. Figure 2 shows histograms of the temperatureand salinity departures (model minus observations) forthe global ocean and for the central Equatorial Pacificfor the two experiments. The histograms reveal that forthe NEMO model (both with and without assimilation)there is a large warm tail compared to the observationsat a depth of 98 metres. In the deeper ocean (below

approximately 450 metres), there is a large cold tail inthe histograms (not shown), indicating that the thermo -cline in the model is too diffuse. The histograms for theinner loop are very well centred, but in the outer loopa tail towards the warm side develops. Similarly, for salin-ity the histogram for the outer loop has a distributionthat is too salty.

The surface temperature is well constrained by thestrong relaxation to the external OIv2 SST analyses. Itis found that the biases in salinity are largest at thesurface since there is no such constraint on surfacesalinity. The results also suggest that the assimilation isless effective in the central Equatorial Pacific comparedto the global ocean. In the central Equatorial Pacific therms error of temperature and salinity grows rapidlyfrom the inner loop to the outer loop (i.e. the analysisand forecasts are inconsistent). In the case of salinity,

10

102

103

-0.5 0 0.5 1 1.5 2 2.5

Dep

th (m

etre

s)

Bias and rms error (°C)

Global Ocean NINO 3.4 Region

Global Ocean NINO 3.4 Region

a Potential temperature

b Salinity

10

102

103

-0.5 0 0.5 1 1.5 2 2.5

Dep

th (m

etre

s)

Potential Temperature (°C)

10

102

103

Dep

th (m

etre

s)

Bias and rms error (psu)

10

102

103

Dep

th (m

etre

s)

Bias and rms error (psu)-0.2 0 0.2 0.4 0.6 0.8 1 1.2 -0.05 0 0.05 0.1 0.15 0.2 0.25

Control Outer Inner

Figure 1 Vertical profiles of the rms error (thick lines) and bias (thinlines) between model and observations for (a) potential tempera-ture (°C) and (b) salinity (practical salinity units, psu) for the globalocean using all observations (left panels) and for observations in thecentral Equatorial Pacific (NINO3.4) region only (right panels). In thepanels, ‘control’ corresponds to the control experiment, ‘outer ’corresponds to the outer loop of the 3D-Var experiment, and ‘inner ’to the inner loop of the 3D-Var experiment. The observations are takenfrom the EN3 dataset.

102

103

104

105

106

-6 -4 -2 0 2 4 6 -6 -4 -2 0 2 4 6

Num

ber o

f obs

erva

tions

Departure (°C)

10

102

103

104

105

Num

ber o

f obs

erva

tions

Departure (°C)

10

102

103

104

105

106

-2 -1 0 1 2

Num

ber o

f obs

erva

tions

Departure (psu)

10

102

103

104

-0.8 -0.4 0 0.4 0.8

Control Outer Inner

Num

ber o

f obs

erva

tions

Departure (psu)

GlobalOcean

GlobalOcean

a Potential temperature at 98 metres

NINO 3.4Region

NINO 3.4Region

b Salinity at 98 metres


20

METEOROLOGY

the information assimilated into the system in a givencycle is lost in the next cycle (see Figure 1b).

Geographical distribution of errors

To investigate the geographical distribution of theerrors, the temporal mean of the increments for temper-ature (Figure 3a) and salinity (Figure 3b) at a depth of98 metres are considered. These results show severalinteresting features. The largest increments occur in thearea of the western boundary currents and in the trop-ical ocean (within 10° of the equator). The assimilationtries to compensate for an inaccurate position of theboundary currents which the ocean model, with a reso-lution of 1°, is unable to represent correctly. Especiallynoticeable is the dipolar structure of the increments inthe Gulf Stream separation area. The assimilation istrying to increase the gradient across the Gulf Streamby adding cool/fresh water to the north of the currentand warm/salty water to the south.

In the tropics the story is different. The ocean modelhas enough resolution to resolve the equatorial dynam-ics, and therefore the source of model error is likely to

reside somewhere else. To investigate this in greaterdetail, a cross-section of the temperature increments atthe equator appears in Figure 4a, together with thetemperature differences between the control and assim-ilation experiments in Figure 4b. In the Pacific theassimilation is trying to steepen the east-west gradient ofthe thermocline (deepening it in the west and shallow-ing it in the east). In this experiment, equatorial currentsare not improved by assimilating temperature and salin-ity data (not shown). In the western Pacific they caneven be degraded. This aspect should be improved withthe introduction of the model bias correction scheme.

The east-west dipolar structure of the assimilationincrement contrasts with the mean differences betweencontrol and assimilation in Figure 4b, which does notshow any east-west dipole: in all the basins, the controlexperiment is too warm above the thermocline and toocold below it. These results demonstrate that the effectof the assimilation is non-local. How the increments inFigure 4a produce the differences shown in Figure 4b isnot obvious. It could be achieved by a variety of processes.For instance, in the Equatorial Atlantic, direct coolingby the assimilation in the western part of the basin, andsubsequent advective transport of the negative incre-ments via the equatorial undercurrent, could lead to abasin-wide cooling. Another possible reason is the effecton the vertical mixing: the warm increment in the east-ern part of the Indian Ocean around 100 metres (justabove the thermocline) could increase the stratification,resulting in reduced vertical mixing of heat down to the

Figure 2 Distribution of model-observation departures at 98 metresdepth for (a) potential temperature (°C) and (b) salinity (psu) for theglobal ocean using all observations (left panels) and for observationsin the central Equatorial Pacific (NINO3.4) region only (right panels).In the panels ‘control’ corresponds to the control run, ‘outer ’ corre-sponds to the outer loop of the 3D-Var experiment and ‘inner ’ to theinner loop of the 3D-Var experiment. The observations are takenfrom the EN3 dataset.

80°N

60°N

40°N

20°N

0°

20°S

40°S

60°S

80°S

340°E300°E260°E220°E180°E140°E100°E60°E20°E

340°E300°E260°E220°E180°E140°E100°E60°E20°E

80°N

60°N

40°N

20°N

0°

20°S

40°S

60°S

80°S

b Salinity increments at 98 metres

a Potential temperature increments at 98 metres

-0.33 -0.29 -0.25 -0.21 -0.17 -0.13 -0.09 -0.05 -0.01 0.03 0.07

-1.45 -1.25 -1.05 -0.85 -0.65 -0.45 -0.25 -0.05 0.15 0.35

Figure 3 Temporal mean of (a) potential temperature increments(°C) and (b) salinity increments (psu) at 98 metres for all 10-day windows.


21

METEOROLOGY

a Mean potential temperature increments

100°E 160°W 60°WLongitude

100°E 160°W 60°WLongitude

300

200

100

0

Dep

th (m

etre

s)

-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5

(°C)

(°C)

b Potential temperature difference between control and 3D-Var experiments

300

200

100

0

Dep

th (m

etre

s)

-2.0 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2

Indian Ocean Pacific Ocean Atlantic Ocean

Indian Ocean Pacific Ocean Atlantic Ocean

Figure 4 Equatorial cross-section of (a) the temporal mean of all10-day window potential temperature increments (°C ) and (b) thedifferences in mean potential temperature (°C ) between the controland the 3D-Var experiments. The three ocean basins (Indian, Pacific,Atlantic) are illustrated. The averages have been computed for theperiod of 1989 to 2006. The different spatial structures of the assim-ilation increment and mean difference demonstrate the non-local effectof the assimilation.

thermocline. A more careful diagnostic study of thedifferent processes involved in the reduction of the errorcan help to diagnose model errors.

Conclusions and outlook

The preliminary results with the NEMOVAR 3D-Varsystem are encouraging, but also highlight limitationswith the system. Though the assimilation system workscorrectly according to the assumptions of no model orobservation bias, we have demonstrated this is not avalid assumption. The present HOPE OI assimilationsystem accounts for model bias and the initial res

Documents

Enhancing the Archive System - ECMWF · ECMWF Newsletter No. 120 – Summer 2009 1 Contents ... Each single cartridge will be able to hold up to one terabyte of (uncompressed) data