land surface assimilation march 2001 · land surface assimilation march 2001 by jean-françois...
TRANSCRIPT
Land surface assimilation
1 Meteorological Training Course Lecture Series
ECMWF, 2002
Land surface assimilationMar ch 2001
By Jean-François Mahfouf and Pedro Viterbo
Table of contents
1 Introduction
2 Design of land surface parametrizations
2.1 General features
2.2 Surface fluxes
3 Introduction to land surface assimilation
4 Simple land surface initialisation methods
5 Soil moisture initialisation using SYNOP observations
5.1 The ECMWF method
5.2 Possible improvements
6 Other techniques to initialise soil moisture
6.1 Methods based on precipitation data
6.2 On-site observations and methods based on satellite imagery
7 Initialisation of other land surface quantities
7.1 Snow mass
7.2 Deep soil temperatures
7.3 Vegetation properties
8 Conclusions
REFERENCES
Abstract
This lecturenoteprovidesa review of techniquesrecentlydevelopedfor initialising the prognosticvariablesof land surfaceparametrizationsin numericalweatherpredictionmodels.The importanceof soil moistureinitialisation is emphasizedsincethe evolution of the boundarylayer is very sensitive to its specificationandthe associatedtime scalesaremuchlongerthanthoseof mediumrangeforecasts.The analysisof snow massis alsodescribed,usingthe ECMWF methodasan illustrativeexample.Differentmethodsanddataavailablefor theinitialisationof otherslowly varyingcomponentsat thesurface,suchassoil temperature, vegetation fraction, leaf area index and albedo, are described at the end.
1INTRODUCTION
Theimportanceof landsurfaceprocesseshasbeenrecognisedfor alongtimeby theclimatemodellingcommunity,
in orderto describein aconsistentwayall thecomponentsof thewaterandenergy cycleover longperiodsof time.
Land surface assimilation
2 Meteorological Training Course Lecture Series
ECMWF, 2002
As aconsequenceavarietyof schemeshavebeendevisedrangingfrom thesimplebucketmodelof Manabe(1969)
to the complex soil-vegetation SiB (Simple Biosphere) model ofSellers et al. (1986).
Although a numberof sensitivity studieshave shown that land surfaceprocessescanalsoaffect mediumrange
weatherforecasts(Rind 1982;RowntreeandBolton 1983;Yehet al. 1984;Rowell andBlondin 1990;Beljaarset
al. 1996),the parametrizationof the interactionsbetweencontinentalsurfacesandthe lower atmosphereis still
rather crude in most numerical weather prediction models.
Recentfield experiments,suchasHAPEX-MOBILHY 86,FIFE87or BOREAS94,haveenabledthedevelopment
andthevalidationof landsurfaceschemesdescribingthemostimportantprocessesgoverningthewater, heatand
momentumexchangeswhile remainingsimple enoughto be includedin operationalweatherforecastsmodels
(NoilhanandPlanton1989;ViterboandBeljaars1995).Suchparametrizationscanimprovesignificantlythequal-
ity of the forecastsof weatherelements(Bougeaultet al. 1991;Lanzinger1995).Initialisationof theprognostic
soil variablesatglobalscaleis animportantissuegiventheverydifferenttimescalesof evolutionbetweentheat-
mosphericandthesoil systems.Realisticschemesappearto besensitive to thespecificationof initial soil temper-
aturesandwatercontents(JacqueminandNoilhan1990;Bouttieret al. 1993a,1993b).Theincreasein realismof
otherphysical parametrizations(clouds,radiation,convection)in numericalweatherpredictionmodelshasalso
madeerrorsin thesurfacerepresentationeasierto identify. Theimportanceof positive feedbacksbetweenthesur-
faceandtheatmospheremustbeunderlined: realisticmechanismsshouldberepresented(which is not possible
whensurfaceboundaryconditionsarefixed),but spuriousones,resultingfrom systematicerrorsin therepresenta-
tion of somecomponentsof theenergy andwatercycles,shouldberemovedwith anappropriateinitialisationpro-
cedure.
In thefollowing, wewill startin Section2 by describingthebasicfeaturescommonto mostsurfaceschemes.Spe-
cial featuresof thelandsurfaceassimilationareidentifiedin Section3, in termsof dataavailability andits relation
to modelvariables.Section4 reviewstheproblemsof simplesurfaceinitializationprocedures.Methodsfor thein-
itialisation of soil water in numericalmodelsusingnear-surfacetemperatureandhumidity observationsarede-
scribedin Section5, while Section6 explainsthedrawbacksof on-sitesurfaceobservationsof soil moistureand
discussestheneedfor remotesensing.Finally, Section7 reviewstheinitialisationof otherslowly varyinglandsur-
face variables, such as snow mass, deep soil temperatures and vegetation characteristics.
2DESIGN OF LAND SURFACE PARAMETRIZA TIONS
2.1 General features
Theaimof landsurfaceschemesis to computetemperatureandspecifichumidityat thelowerboundaryof atmos-
phericmodels.Thesetwovariablesarerequiredin theestimationof heat,waterandmomentumexchangesbetween
the continental surfaces and the lower atmosphere.
The surface temperature is derived as the solution of the surface energy balance written as:
(1)
whereRn is thenetradiationflux convertedin latentheatflux , sensibleheatflux andgroundheatflux .
The ground heat fluxG depends upon deep soil temperatures and soil thermal properties.
The surface specific humidity can be written formally as :
�sk
�n
��� � �+ +=
��� � �
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 3
(2)
where representsthevalueof specifichumidityat thelowestmodellayer. Thequantities and (described
with moredetailsin the next paragraph)dependuponsoil moisture which is obtainedby solving the surface
water budget :
(3)
whereP is the precipitation flux,E the total evaporation flux andR the surface runoff.
Thisbrief descriptionshowsthatalandsurfaceschememustmanageat leasttwo prognosticequationsfor thetem-
perature and soil moisture in the soil. Heat and water transfers are governed by the following diffusion laws :
(4)
(5)
Thediffusivity D andconductivity K coefficientsarenon-linearfunctionsof soil moisturecontent.Thediffusion
equationsaregenerallydiscretizedover2 to 4 layersin orderto dealwith timescalesrangingfrom daysto months.
2.2 Surface fluxes
Thelink betweensoil andatmosphericvariablesis providedthroughtheexpressionof thesurfacefluxes,usually
basedon Monin-Obukhov theory;thecrucialvariablehereis theevaporationflux becauseits magnitudedepends
explicitly uponsurfaceproperties.Recentlandsurfaceschemesrepresentdifferentlythegrid boxfractionscovered
by: a) baresoil, with evaporationcontrolledby soil moisturein a shallow top soil layer;b) vegetation,with tran-
spirationcontrolledby soil moisturein the root zoneaffecting the magnitudeof stomatalresistance;c) snow or
interceptionreservoir, evaporatingat thepotential(maximum)rate.Thesetwo lastcomponentsrequiretheexist-
ence of model prognostic equations for snow mass and interception reservoirs.
Evaporation from bare soil can be written as (seeMahfouf and Noilhan 1991 for a review) :
(6)
Theefficiency of theturbulenttransfersis accountedfor throughtheaerodynamicresistance , while thecontrol
by thesurfacesoil moisture is representedby thesurfacerelative humidity [ and in
Equation(2)]. A typical variationof is presentedon the left panelof Fig. 1 . Evaporationtakesplaceat
the potential rate above a threshold value (field capacity) (defined from soil texture) up to the saturation .
Transpiration from vegetation canopy writes similarly:
(7)
Theanalogywith Equation(2) leadsto and . Watertransfersfrom the
�s
� �sat�
sk( ) �L×+=
�L
� θ
dθd------ � �
–�
–=
∂�
s
∂ ---------∂
∂�------ ��∂�
s
∂�--------- –
∂θ∂ ------
∂∂�------ θ
∂θ∂�------
∂ � θ
∂�----------+=
�g ρ
��� �sat�
sk( ) ���–�a
---------------------------------------=
�a
θs
���θs( )
� ���= 0=���
θs( )θf c θsat
�tr ρ
�sat
�sk( ) ���–�
c
�a+
--------------------------------=
� �a
�c
�a+( )⁄= �
c
�c
�a+( )⁄=
Land surface assimilation
4 Meteorological Training Course Lecture Series
ECMWF, 2002
root zoneto theatmospheredependbothonbiologicalandphysicalcontrols.Plantslimit theirwaterlossesin un-
favorableenvironmentalconditionsdeterminedby soil moisturein therootzone,atmosphericwatervapourdeficit,
solarradiation,air temperatureandcarbon-dioxideconcentration.A typical dependency of thecanopy resistance
with soilmoistureisshownontherightpanelof Fig.1 . Below athresholdvalueoftendefinedasthepermanent
wilting point , it is assumedthat plantsareunableto pumpwaterfrom the root zoneto the stomatalcells,
correspondingto arapiddecreaseof transpiration(increasein ). As for baresoils,it if oftenassumedthatabove
the field capacity the plant transpiration is not controlled by soil moisture.
Figure 1Surface relative humidityhu as a function of the surface volumetric water content (left) and canopy
resistanceRc as a function of the mean volumetric water content (right). (FromMahfouf 1991).
3INTRODUCTION TO LAND SURFACE ASSIMILA TION
Themajorproblemof landsurfaceassimilationis the lack of routineobservationsof soil moistureandsoil tem-
perature.This is speciallytruein thecaseof soil moisture,wherecurrentmethodscannotprovideglobalcoverage
routinely(seeSubsection6.1 for moredetails).Furthermore,soil moistureobservationsshow largevariability in
smallspatialscales(seee.g.WetzelandChang1988);not all scalesareof relevancefor theatmosphere,andthe
assimilationmethodhasto take thatinto account.For soil temperaturetheclimatenetwork exists,with acoverage
similar to theSYNOPstationsandwith mostof thestationsperformingobservationsat leastdaily; unfortunately,
thoseobservationsarenot exchangedroutinelyat thetime of measurement,so in practicethey canonly beused,
in delayed mode, for verification purposes.
Thenatureandavailability of theobservationsimposestheuseof proxyvariablesfor soil moisture.Theamountof
waterin the root zoneimpactson theevaporative fraction,which in itself determines,for a givenamountof net
radiation,middaysummerscreenlevel temperatureandhumidity (seeprevioussectionandthereview by Bettset
al. 1996).On theotherhand,the time evolution of soil waterdependson therainfall intensityandon its timing.
Threemaintypesof datahavebeenusedin thepastto infer soil moisture:(a)Screenlevel atmospherictemperature
andhumidity; (b) Rainfall rates;and(c) Radiometricsurfacetemperature(infrared,microwave).Notethat,since
thoseobservationsalreadyrepresent,to acertainextent,theimpactof soil moistureontheatmosphereabove,they
have already filtered out the smaller scales in soil water.
Therearetwo aditionaldifficulties in the assimilationof land surfaceobservations.First, theseobservationsare
non-linearlyrelatedto soil moistureandsoil temperature(throughtheequationsof thelandsurfacescheme).Sec-
ondly, thestatisticsof forecasterrors,usedto spatiallydistributethelocal incrementsin atmosphericanalyses,are
�c
θpwp �c
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 5
not known for soil variables.
4SIMPLE LAND SURFACE INITIALISA TION METHODS
Dueto thedifficultiespresentedabove,soil variablesareinitialisedempiricallyin mostoperationalforecastsmod-
els.
A first methodfor initialising soil variablesis to settheanalysedsoil variablesto climatologicalvalueswithout
usingany modelinformation(throughthefirst-guess).In practice,this is doneby addinga relaxationto climatol-
ogyto theprognosticequationsof thelandsurfacescheme(Blondin1991).Theunderlyingideaof this termis that
simplesurfaceschemescanonly describecorrectlythevariableshaving a shorttime scaleevolution, like thesur-
facesoil temperature,andthatvariableshaving a time scalelongerthana few daysmustbeprescribed.Problems
arisefrom thespecificationof climatologicalsoil valueswhicharebasedonindirectatmosphericobservationsand
very simplifed transfermodels.Theseproducts,like thesoil moistureclimatologyof Mintz andSerafini(1992),
haveahigh level of uncertaintyasshown by ViterboandBeljaars(1995).Anotherpotentialproblemis thatwith a
relaxationof deepvariablestowardsclimatology, seasonalanomaliescannotbe forecasted.An exampleof such
deficiency hasbeenshown by Beljaarset al. (1996)with two versionsof the ECMWF model for the 1993US
floods.
Theotherapproachto initialisesoil variablesis to settheanalysedvaluesto thefirst-guess.In thatcontext, anab-
soluteconfidenceis givento thelandsurfaceschemefor evolving its prognosticvariablesandnocontrolexiststo
prevent thelandsurfaceschemefrom drifting to anunrealisticstate.Suchdrifts canoccurthroughpositive feed-
backswith theatmosphere,in situationswheretheschemeexperiencessystematicerrorsin theatmosphericforcing
(toomuchradiation,toomuchrainfall, ...) or from amisrepresentationof somelandsurfaceprocesses.Thesecond
point canbecheckedwith stand-alonesimulationswheretheatmosphericforcing is prescribed.An exampleof a
drift of thecurrentECMWF landsurfaceschemewithin thedataassimilationsystemis describedin thenext sec-
tion.
5SOIL MOISTURE INITIALISA TION USING SYNOP OBSERVATIONS
5.1 The ECMWF method
Thelandsurfaceschemedevelopedby Viterbo andBeljaars(1995)wasintroducedoperationallyin August1993
andall soil prognosticvariableswereinitialized to first-guessvalues.Oneof themaindifferenceswith respectto
thepreviousECMWFlandsurfacescheme(Blondin1991)is theabsenceof climatologicalrelaxationfor deepsoil
temperatureandwatercontent.During May-June1994,thesoil reservoirs weredrying out, leadingto surfaceair
temperatureerrorsincreasinglypositive,andin comparisonwith othermodels,suchastheGermanWeatherServ-
ice(constrainedby climatologicalsoil moisture),forecastskill wasdeteriorating(ViterboandCourtier1995).The
downwarddrift of soil moistureappearsto belinkedto excessincomingsolarradiationprimarily causedby under-
predictionof clouds.Furthermore,theexcessive warmingat thelower troposphereaffectedtheforecastperform-
ance,asshown in Fig. 2 . The500hPageopotentialday2 forecastaveragedoverEuropeis presentedfor April to
June 1994. The ECMWF forecast (dashed line) has a positive bias from late April onwards.
Land surface assimilation
6 Meteorological Training Course Lecture Series
ECMWF, 2002
Figure 2Bias of the 500 hPa geopotential height, averaged over Europe, for the day 2 ECMWF (dashed) and
DWD (solid) forecast.
Analysisincrementsof specifichumidityat thelowestmodellevel duringMay 1994show positivevaluesoverEu-
ropewith maximareaching2.5g/kg.Theatmosphericanalysistriesto compensatefor themodelbiasby moisten-
ing andcoolingthelower atmosphere.Therefore,thelow level humidity incrementscanbeusedto identify areas
wherethesoil is toodry. Knowing theanalysisincrementof specifichumidity , thecorrectionof soil
moisture to be applied the root zone is assumed to be proportional :
(8)
with ∆ t = 6 hours,andthesubscriptsaandf referto analysisandforecastvalues,repsectively. Therelaxationco-
efficientD is constantin spaceandtimeandcorrespondsto aspecifichumidityanalysisincrementof 1.5g/kgfill-
ing 150mm of waterin thesoil in 9 days.Thefractionof vegetation in theabove formulaguaranteesthatthe
schemeis notactiveoverdeserts.No incrementsareproducedin thepresenceof snow, andtheanalysedsoil mois-
turecontentsθa arelimited by thefield capacityandpermanentwilting point thresholds.Theintegratedsoil water
incrementsaredistributedin eachof thethreesoil layersfollowing themodelrootextraction.Thisnudgingscheme
wasimplementedoperationallyin December1994.It suppliessoil moistureto maintainevaporationin areasof ex-
cessive radiationat thesurface,caused,amongotherfactors,by insufficient cloudcover. Fig. 3 comparestheday
3 forecasterrorsin screenlevel daytime temperatureandhumidity, averagedoverEurope.Threeexperimentsare
compared:Operations,FWDC,whereno initialisationof soil wateris appliedandthediagnosticcloudschemeis
used;IWDC, initial soil watermethodusingthetechniquedescribedabove,diagnosticcloudscheme,and;IWPC,
initial soil watermethodandprognosticcloudscheme.Theerrorsin bothtemperatureandhumidity aredramati-
cally reducedwhenthe initialization of soil wateris used,andthey arereducedeven furtherwhenthe radiation
forcing at thesurfaceis improvedby theuseof theprognosticcloudscheme.Notethatthesoil waterwasresetto
field capacityat3 July, in aquickeffort to correctthelargesystematicerrorsin themodel;thisexplainsthemuch
reducedoperationalerrorsafterthatdate.Thereductionof nearsurfacewarmdry biasremovesthebiasat thetrop-
osphericgeopotential,impactingfavourablyon theroot meansquare(rms)of thetroposphericgeopotentialover
EUROPE LAT 35.000 TO 75.000 LON -12.500 TO 42.500MEAN ERROR FORECAST
500hPa GEOPOTENTIAL FORECAST VERIFICATION
M’�
MA = 5 Day Moving Average�
1�APRIL�
1994�3� 5� 7� 9� 11 13 15 17 19 21 23 25 27 29 1�
MAY� 3� 5� 7� 9� 11 13 15 17 19 21 23 25 27 29 31 2�JUNE
4 6! 8" 10 12 14 16 18 20 22 24 26 28 30 2�-15
-10
-5
0
5
10
15
20
ECMWF 12Z T+ 48 MA
OFFNB 12Z T+ 48 MA#
∆ � �a�
f–=
∆θ θa θf–=
θa θf– $ v ∆ � a�
f–( )=
$ v
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 7
land areas.Examplesfor the geopotentialrms at 500 hPa over Europeandat 200 hPa over North Americaare
shown in Fig. 4.
Figure 3Averaged European bias (model minus observations) in the 2m temperature (left) and the humidity
(right) for theday3 forecastverifying at12UTC. FWDC:Control;IWDC: Nudgingof water;IWPC:Nudgingof
water and prognostic cloud scheme.
Similar techniquesareusedoperationallyat Météo-France(Coiffier et al. 1987)andat theCanadianMeteorolog-
ical centreMailhot et al. 1997).Recently, Yanget al. (1994)haveproposedto improve this methodby usingboth
informationsof temperatureandspecifichumidity. previousforecasterrorswith coefficientsdependingonvegeta-
tion type.Unfortunatelythemethodproposedis biased,becauseacorrectionof soil moistureis appliedevenif the
forecast of atmospheric low level parameters is perfect; this may lead to a long term drift.
Figure 4Averagedgeopotentialrootmeansquareerrorfor the20 forecastswith initial datesbetween940601and
940620, for 500 hPa Europe (top) and 200 hPa North America (bottom); experiment names as in Fig 3. For all 3
experiments the root mean square is computed against the operational analysis.
5.2 Possible improvements
Mahfouf (1991)andBouttieretal. (1993a,1993b)haveproposedanoptimalinterpolationschemefor theassimi-
lation of soil moistureusinginformationof both temperatureandrelative humidity at two metres,which canbe
formaly written:
1-Jun% 11-Jun% 21-Jun% 1-Jul% 11-Jul& 21-Jul& 31-Jul& 10-Aug'-1.0
0.0
1.0
2.0
3.0
4.0
5.0
(K)(
2m Temperature Europe)T + 72*
FWDCIWDCIWPC
1-Jun% 11-Jun% 21-Jun% 1-Jul% 11-Jul& 21-Jul& 31-Jul& 10-Aug'-2.5
-1.5
-0.5
0.5
(g/k
g)
Specific Humidity Europe+-,T + 72
0. 1/ 20 31 42 53 64 75 86 97 10Forecast Day80
20
40
60
80
100
120
140
160
180M9 DATE1=940601/... DATE2=940601/... DATE3=940601/...
AREA=EUROPE TIME=12 MEAN OVER 20 CASESROOT MEAN SQUARE ERROR FORECAST
500 hPa GEOPOTENTIALFORECAST VERIFICATION
FWDC
IWDC
IWPC
0. 1/ 20 31 42 53 64 75 86 97 10Forecast Day80
20
40
60
80
100
120
140
160
180M9 DATE1=940601/... DATE2=940601/... DATE3=940601/...
AREA=N.AMER TIME=12 MEAN OVER 20 CASESROOT MEAN SQUARE ERROR FORECAST
200 hPa GEOPOTENTIALFORECAST VERIFICATION
FWDC
IWDC
IWPC
Land surface assimilation
8 Meteorological Training Course Lecture Series
ECMWF, 2002
(9)
Theoptimalcoefficientsα andβ minimisetheanalysisvarianceandarerelatedto theforecasterrorstatistics.They
aremodeldependentandthesuccessof themethoddependsontheiraccurateestimation.Mahfouf(1991)hasused
a Monte-Carlotechniquewith a one-columnmodel,wheresoil moistureis perturbedrandomlyin a rangeof pos-
siblevalues.Oneconclusionof this studyis that thecoefficientsα andβ stronglydependuponthediurnalcycle
(informationon soil moisturefrom atmosphericparameterscanbeextractedmoreeasilyduringdaytime in clear
sky conditions)anduponthevegetationcover (over baresoil, correctionsareappliedto thesuperficialreservoir
andwhenthevegetationcover is importantsoil correctionsareappliedover thewholeroot zone).Bouttieret al.
(1993a)proposeda first parametrizationof the optimumcoefficients,recentlygeneralizedby Giard andBazile
(2000).In orderto beused,this initialisationmethodrequiresananalysisof temperatureandrelative humidity at
two metres(Navascues1997);theanalysisincrementsshouldbezeroin thosesituationswhereparametersin the
boundarylayerarenot informativeaboutsoil moisture,e.g.strongadvection,andlow radiative forcingat thesur-
face.
Oncetheoptimalcoefficientsarederived,thesequentialassimilationcaneasilybeimplementedin currentopera-
tionaldataassimilationsystems;however, it assumeslinearrelationshipsbetweenatmosphericincrementsandcor-
rectionsto be appliedin the soil which is not a goodapproximationfor mostof the physical parametrizations.
Anotheroptionis thevariationalmethod,which seemsa priori moresuitableto theanalysisof soil moisturedue
to thenon-linearitiesof theproblemandto theimportanceof thetime distribution of observations(surfacevaria-
blesarestronglyaffectedby thediurnalcycle).Mahfouf (1991)andmorerecentlyCalliesetal. (1998)useda1D-
Var approachto estimatetheinitial soil moistureof a one-columnmodelthatbestfit observationsof temperature
andrelative humidity duringa diurnalcycle.Thevariationalmethodwasappliedby Rhodinet al. (1999)to a re-
gionalweatherforecastmodelover a five-dayspringperiod.Theoptimal soil moistureminimisesthe following
cost-function:
(10)
In theaboveformula , , and , represent,respectively, thescreenlevel temperatureandrelativehu-
midity, andtheirassumedobservationalerrors,andthesubscriptsoi andfi representtheobservationi andthefore-
castvalueinterpolatedto thepoint i; thesummationisdoneoverthetotalnumberof observationobservationpoints,
N. Mahfouf(1991)hasvalidatedthetwo methodsdescribedabovewith aone-columnversionof theMétéo-France
forecastmodelusingdatafrom theHAPEX-MOBILHY 1986field experiment.Datafrom thiscampaignprovided
at variouslocationssimultaneousinformationaboutsoil moisturecontentandlow-level atmosphericparameters,
i.e.temperature,relativehumidity, windspeed.Whenusingobservationsof temperatureandrelativehumidity, both
the sequentialandvariationaltechniqueconverge towardsthe neutronprobeestimatesof soil moisture,starting
from arbitrary initial values of soil moisture (Fig. 5).
θa θf– α�
a
�f–( ) β
�:�a
�;�f–( )+=
<θ( )
�o= �
f =–
σ�---------------------- 2
�;�o= �;�
f =–
σ >@?---------------------------------- 2
+= 1=
A∑=
� �B�σ� σ >-?
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 9
Figure 5Evolution of the surface (left panel) and root (right panel) soil moisture contents during the sequential
assimilationon4-5July1986.Dottedcurvesindicatetheobservedvalues.Thesiteis HAPEX-MOBILHY. (From
Mahfouf 1991).
Mahfouf (1991)assumedthatthereis a priori no usefulinformationin thefirst-guesswhich is certainlyincorrect
with anoperationalmodelof someskill; in realisticapplicationsa backgroundtermshouldbeaddedto thecost-
function.
Resultsshow that for clear-sky situationsbothmethodsretrieve soil moisturecontentscloseto eachotherandto
theobservations.Thevariationalmethodis moreefficientbut non-linearitiesof theproblemmaketheefficiency of
theconvergencedependenton the initial startof theminimisation.Whenthe fractionof vegetationis large,soil
moisturein therootzoneis retrievedmoreaccuratelythansurfacesoil moisture.Theexaminationof thecost-func-
tion (Fig. 6 ) showstheexistenceof asecondaryminimum(correspondingto ambiguityin theresponse,thesame
surfaceevaporationcanbeobtainedwith totally differentswatercontentsin thesoil reservoirs)aswell asaplateau
wheresurfaceevaporationis notsensitiveto modificationsin soil moisture(abovefield capacity, evaporationis as-
sumed to take place at a potential rate, therefore it is no more controlled by the surface).
Land surface assimilation
10 Meteorological Training Course Lecture Series
ECMWF, 2002
Figure 6Variations of the cost function with soil moisture for 4-5 July 1986. (A) represents the dry guess
( ), (B) represents the moist guess ( ), and the square is the searched state (reference).
(FromMahfouf 1991).
6OTHER TECHNIQ UES TO INITIALISE SOIL MOISTURE
6.1 Methods based on precipitation data
Two methodshave beenintroducedto initialise soil moisturefrom precipitationdata,bothof themrequiringthe
availability of measurementsover large areasandan algorithmto performa precipitationanalysisbeforehand.
They have only been applied over areas with a good observational coverage like the US or the UK.
Thefirst techniqueis anuncoupledinitialisationwherea landsurfaceschemeis forcedwith conventionalmeteor-
ologicalobservations(temperature,humidity, wind,radiationandprecipitation)toprovideanestimateof soilmois-
ture.Feasibilitystudieshave beenundertakenin variouslimited areamodelsby Smithet al. (1994),Macpherson
(1996)andMitchell (1994).Theimprovementof theforecastsof screenlevel humidity whensoil wateris initial-
izedwith sucha method,usingtheUK Met Office waterbudgetscheme,is shown on Fig. 7 (from MacPherson
1996) for all UK stations.
θ 0.30θsat= θ 0.70θsat=
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 11
Figure 7Screenlevel relativehumidityverificationof forecastsfrom 28June199500UTC, with differentinitial
soil moisture:Operational(OP)with freecycling moisture,climatological(CLIM); MORECS(S)andMORECS
(R) refer to a smoothed and raw version, respectively, of the soil moisture initialisation using oberved
precipitation. (FromMacPherson 1996).
Anothertechniquemakesuseof bothobservedprecipitationratesandmodelfirst-guess.Assuminga rainfall rate
increment over a 6-hour period:
(11)
This quantityis convertedin soil moistureincrementby usingthetangentlinearmodelof a soil moisturebudget
scheme:
(12)
whereE is themeanevaporationrateandθ themeansoil moisturecontentduringthe6-hourassimilationperiod
(trajectory).Then,thesoil moistureincrement∆ θ is addedto thesuperficialreservoir. Studieshave beenunder-
takenatECMWFby Vasiljevic (1989,personalcommunication)andatUKMO by JonesandMacpherson(1995).
Sucha methodis sensitive to thespecificationof surfacerun-off andcanconvergeslowly whenbiasesin theroot
zone are large.
6.2 On-site observations and methods based on satellite imagery
Existingtechniquesfor groundbasedobservationsof soil moisture(seerecentreviews in Schulinetal. 1992;Wei
1995)aretimeconsumingandnormallyrequirehumanintervention.Therepresentativenesserrorof theon-sitees-
timatesis bestavoidedby deploying several instrumentswithin a relatively small area( 100m2), increasingthe
costof themeasurements.In spiteof its problems,on-sitesoil moisturedataarevery usefulfor regionalesimates
for climaticstudies,essentialto closethewaterbudgetin large-scalehydrologicexperiments(CuencaandNoilhan
1991;Goutorbeet al. 1989;Mahfouf 1990)andto calibrateremotesenseretrieval techniques(Georgakakosand
Baumer 1996).
Thereisnoprospectof obtainingreal-timeglobalestimatesof soilmoisturebasedonexistingtechnologyof ground
basedinstruments.For this reason,severalalgorithmshavebeendevelopedto infer soil moisturefrom satelliteob-
3C
6D
9E
12 1815
fForecast time (hours)
2G
0 CH
LIM
10
5I0J
15
-5
-10
-150J
mea
n er
ror
(%)
20
fForecast time (hours)
15
10
5I
0J
3C
6D
9E
12 15 18
rms
erro
r (%
)
MORECS(S)MORECS(R)OK
P
∆� � a � f–
∆ ------------------=
d∆Θd-----------
∆θθ
-------�
– ∆�+=
Land surface assimilation
12 Meteorological Training Course Lecture Series
ECMWF, 2002
servations,althoughnoneof themis currentlyusedin anoperationaldataassimilationsystem.Threetypesof tech-
niqueshave beenproposed(seereviews in Paloscia1996; Schulin et al. 1992; Wei 1995) basedon infrared
measurements,passive microwave and, more recently, active microwave (radar) instruments.
In theinfraredchannels,thesensitivity of thediurnalcycle of surfacetemperatureto soil moisturehasbeenused
to definemethodsbasedon the observed changeson the infraredskin temperature(which avoid the problemof
absolutecalibrationof thesatellitesensor).For reviewsof applicationsseeCarlson(1991),SchmuggeandBecker
(1991),andSchulinetal. (1992).Geostationarysatellitesallow for abettertemporalsampling(Wetzeletal. 1984
; McNideretal.1994).Thesemethodscanonly beappliedin clearsky conditionsbut provideaninformationabout
soil moisturein therootzoneovervegetatedareas.Bastianssen(1995)developedrecentlya techniqueto estimate
regionalevaporationoverheterogeneousterrain,basedonaseparateestimateof theevaporationof unstressedpix-
els,basedon potentialevaporation,andfully stressedpixels,basedon the infrareddiurnal cycle technique.The
evaporationof theremainingcloud-freepixelscanbeobtainedby interpolationbetweenthewetpixelsandthedry
pixels.vandenHurk etal. (1997)hasappliedthis techniqueto initialize thesoil waterof a limited areamodelover
the Iberianpeninsula.Themodelsoil moistureis the linearizedsolutionof a variationalproblemthatminimizes
the difference between model and satellite estimates of evaporative fraction ).
Microwavechannelscanbeusedto infer soil moisturedueto theimportantvariationsof thedielectricconstantof
asoil with volumetricwatercontentfor frequenciesbetween1 and5 GHz(SchmuggeandJackson1994).Passive
microwavetechniquesusethefactthatsoil emissivity changeswith its watercontent.In activemicrowavesensors
(radar)thesignalis emittedby anartificial sourceandtheintensityof thebackscatteredradiation,afterreflection
by thesurface,is measured.Thereflectivity of thesoil changeswith its watercontents,hencetheintensityof the
reflectedsignalcanberelatedto thesoil moisture.Activemicrowavesystemsallow, for thesamewavelength(same
maximumpenetrationdepth),a finer horizontalresolution,becausethegroundcanbescannedwith anangularly
confinedbeam.Oneof thedrawbacksof microwaveretrievalsis thatthesurfaceemissivity/reflectivity is alsosen-
sitiveto thesurfaceroughnessandthewatercontentsof thevegetationcanopy. Nevertheless,it appearsthatsimple
estimatesof surfaceroughnessof broadvegetationclassesaresufficienttocorrectthesoilmoistureestimate(Njoku
andEntekhabi1996).Ontheotherhand,theoppacityof thevegetationlayerincreaseswith its watercontent,mak-
ing thecorrectionsdueto vegetationincreasinglyunreliablefor moistsoils.Perhapsthemajordrawbackof micro-
wave estimatesis thedepthof penetrationof thesignal,limited to thetop layerof thesoil (2 to 10 cm,depending
on thewavelength).However, for specificsoil hydrologicalandatmosphericconditions,thesoil watercontentsof
theroot layer is correlatedwith thetop soil water. Recentstudiesshow that it is possibleto infer, in a physically
consistentway, thewholeprofileof soil waterfrom its valuesat thetoplayer(e.g.Njoku andEntekhabi1996;Cal-
vet et al. 1997).
Wehaveshown in thissectionthatground-basedestimatesof soil moisture,althoughveryimportantfor calibration
purposesandin intensive field efforts, cannotgive a nearreal-timeglobalstimateof soil moisture.Satelliteesti-
matescanachieveglobalcoverage,but arelimited to clear-sky conditions(infraredchannels)or senseonly thetop
few centimetresof soil (microwavechannels).Thefuturereliesonphysicallybasedestimatesof soil moisturefrom
a combinationof satellitemeasurementsandmodelshort-termforecasts,usinga variationaltechniquein orderto
find the soil water contents that fits best the satellite signal.
7INITIALISA TION OF OTHER LAND SURFACE QUANTITIES
7.1 Snow mass
Themethodsusedto analysesnow massarerelatively crudewhencomparedto analysisschemesfor theatmos-
phericvariables(Lorenc1986)andcould easilybe improved. In this section,the ECMWF snow analysisis de-
� � ���+( )⁄
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 13
scribed. The methodology used in other operational centres is very similar.
Figure 8Snow depth(cm) for January(top)andApril (bottom).Theclimatologyusedin theECMWFanalysisis
shown on the left, differences to the US Air Force climatology is shown on the right. The linear snow density
formulaof Verseghy(1991),varyingbetween188kg m-3 and450kg m-3, is usedto convert thesnow massvalues
of the analysis climatology into snow depth.
Every 6 hours a snow analysis is performed in three steps:
20ON
40ON
60L
ON
80M
ON
180O135OWN
90O
OW
45P
OWN
0Q
O45P
OER
90O
OE
135OE
snow depth (cm) EC AN climate 01S
20ON
40ON
60L
ON
80M
ON
180O135OWN
90O
OW
45P
OWN
0Q
O45P
OER
90O
OE
135OE
snow depth (cm) (EC AN - USAF) climate 01
20ON
40P
ONT
60L
ON
80M
ONT
180O135OWN
90O
OWN
45OW 0Q
O45P
OER
90O
OER
135OE
snow depth (cm) EC AN climate 04S
20ON
40P
ONT
60L
ON
80M
ONT
180O135OWN
90O
OWN
45OW 0Q
O45P
OER
90O
OER
135OE
snow depth (cm) (EC AN - USAF) climate 04
Land surface assimilation
14 Meteorological Training Course Lecture Series
ECMWF, 2002
1) Snowfall analysis:From SYNOPreportsof temperatureandprecipitationrate,a snowfall rate is
estimatedat the observation points, and then a spatial interpolation is done by a successive
correction method.
2) Snow massbackgroundfield: A very simplesnow modelevolution is usedto build thebackground
from the analysedsnowfall , the previous analysisof snow mass(persistence) , a snow
climatological mass , and an empirical melting function M basedon the 2m temperature
forecast:
(13)
or in a discretized form:
(14)
3) 3) Snow massanalysis:a sucessive correctionmethodis appliedto thesnow massincrements(the
differencebetweenthebackgroundfield in (2) andtheobservationsof snow depth,suitablyscaled
assuming a fixed snow density of 250 kg m-3).
Themainshortcomingsarethatinformationsfrom themodel(first-guess)arenottakeninto accountin theanalysis
processandthat thesnow massclimatology, towardswhich theanalysisis relaxedandin fact thedominantterm
in datavoid regions,is ratherpoor. Thesnow climatology(BrankovicandvanMaanen1985,BvM85) hasbeen
constructedby runningto equilibriumanempiricalsnow massmodelforcedby precipitationmonthlyclimateval-
uesandadjustedfor meltingusingatemperatureclimatology;noticethatnosnow observationswereusedandthat
thespatialresolutionis ratherpoor( ). Fig. 8 shows BvM85 climatologyfor JanuaryandApril, andcom-
paresits valueswith the USAir Force(USAF)climatology(FosterandDavy 1988),thelattertakinginto
considerationanextensivecollectionof regionalandglobalsurveysof groundbasedsnow depthobservations.Note
that,sincetheUSAFis aclimatologyof snow depth,BvM85 snow massvalueshave to beconvertedusinganem-
pirical specificationof snow density:Eq(48)of Verseghy(1991),thattakesintoaccountthepackingof snow under
its own weight,wasused.It is clearthat in January, theBvM85 valuesextendtoo far southin Asia,overestimate
thesnow depthin southSiberia,MongoliaandeasternEurope,northwesternandeasternCanada,andunderesti-
matethe snow depthin northernSiberia,Iraq, Iran, westernEuropeand CentralCanada.BvM85 valuesover
GreenlandaremuchlargerthanUSAF, partlybecausetheUSAFrepresentsaclimatologyof seasonalsnow depth,
while BvM85 snow massvalueshave beenassignedthearbitrarily high valueof 10 m, "to avoid unrealisticmelt-
ing" (BvM85). In springBvM85 overestimatesthesnow depthvirtually everywhere,but for theeuropeanvalues
(e.g. the Alps and Pyrenees), where a small underestimation is detected.
MeanmonthlyJanuaryandApril snow depthvaluesof theECMWFreanalysisareshown in Fig. 9 , togetherwith
theirdifferenceto theUSAFclimatology. Theoverallpatternsin theright panelsof Figs.8 and9 areverysimilar,
indicatingthat theBvM85 climateinfluencesstronglythereanalysisvalues.Nevertheless,it is alsoclearthat the
informationfrom observationsin Januaryis effectively used(valuesof top panelof Fig. 9 aresmallerthanthe
correspondingvaluesin Fig. 8 ), especiallyfor EuropeandwesternAsia.Substantialerrorsin theBvM85 clima-
tology in springleadto largeanomaliesof the reanalysismeanvalues.As a further illustrationof the reanalysis
informationbroughtby theuseof observations,themonthlydeviationsof snow massfrom theBvM85 valuesis
shown ontheleft panelsof Fig.10 , while theright panelsshow thestandarddeviationof themonthlymeanvalues
of analysedsnow mass.Sincewearedealingexclusively with snow massvalues,therewasnoneedto usetheden-
sity relationin thisfigure.ThenortheasternAsianvaluesof snow in winteraresimilar to BvM85, andthestandard
deviationis small.A separateanalysis(Corti etal.2000)showsthataverylargenorthernAsianareacorresponding
Ug � s
UpU
c
dU
d------- � s
V–
Uc
U–
τ----------------+=
UXW1 λ–( )
Up λ
Uc ∆ Y� s
V–( )+ +=
5° 5°×1° 1°×
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 15
to theformerSoviet Union hasa datacover very inhomogeneousin time: thenumberof observationsfrom 1979
to 1992are10to 30timessmallerthanin 1993.Valuesfor themonthof May (notshown) displaylargerdifferences
to theclimateandlargervariability in northernlatitudes.Thisis dueto interannualvariability in theratesof melting
implictely usedin thesnow analysisalgorithm(Eq.(14)), andcontrolledby thefirst-guesstwo-metretemperature.
Figure 9AsFig. 8, but for the reanalysis mean January value (top left) and the mean April value (bottom left).
Right panels show the difference between these values and the US Air Force climatology.
20OZN
40OZN
60ON
80OZN
180O135OW
90OW
45OW 0O45OE
90OE
135O[E
snow depth (cm) EC AN climate 01\
20ON
40ON
60ON
80ON
180O135OW
90OW
45OW 0O45OE
90OE
135O[E
snow depth (cm) (EC AN - USAF) climate 01
20ON
40OZN
60OZN
80OZN
180O135OW
90OW
45OW 0OZ 45O
ZE
90OZE
135OE
snow depth (cm) EC AN climate 04\
20ON
40ON
60ON
80ON
180O135OW
90OZW
45OZW 0O
45OE
90OE
135OE
snow depth (cm) (EC AN - USAF) climate 04
Land surface assimilation
16 Meteorological Training Course Lecture Series
ECMWF, 2002
Figure 10Snow mass(mm)differencebetweenthereanalysismeanmonthlyvaluesandtheanalysisclimatology
(left), and standard deviation of the monthly mean reanalysis values (top row, January; bottom row, April).
20OZN
40ON
60ON
80OZN
180OZ
135OZW
90OW
45OW 0O45OE
90OE
135O[E
snow mass (mm) (ERA - EC AN) climate 01
20ON
40ON
60ON
80ON
180O135OW
90OW
45OW 0O45OE
90OE
135O[E
std snow mass (mm) ERA 01
20ON
40OZN
60OZN
80OZN
180O135OW
90OW
45OW 0OZ 45OE
90OE
135OE
snow mass (mm) (ERA - EC AN) climate 04
20ON
40ON
60ON
80ON
180O135OW
90OW
45OZW 0O
45OE
90OE
135OE
std snow mass (mm) ERA 04
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 17
Figure 11Dailysnow mass(mm)valuesfrom 1 August1981to 31July1991for apointover theFrenchAlps (47
N, 6 E, altitude1100m). Thesolid line representsaproxy for thetruth,estimatedwith Martin (1996)model(see
text for more details), the dashed line shows reanalysis values.
Theanalysisalgorithmhasbeenfurtherevaluatedby Martin (1996),andresultsfor snow massaredisplayedin Fig.
11 . Thevaluesof the reanalysis(labelledERAin thepicture)arecomparedwith thoseproducedby anoff-line
integrationof adetailed,physicallybasedsnow model,CROCUS(Brunetal. 1989;1992),forcedby ERA values
of screenlevel temperature,wind speedandhumidity, precipation,andsemi-empiricalderivedvaluesfor incoming
radiation.Thepictureshows daily values,from 1 August1981to 31July 1991,for apoint in theFrenchAlps (47
N 6 E), wherethemodelelevationis similar to thestationheight(1100m). Thereanalysissnow massvaluescap-
turewell the interannualvariability, with maximumvaluesfor theyears81/82and83/84,andminimafor 82/83,
88/89and89/90.However, thereanalysisvaluesaresystematicallytoohighfor theaccumulationperiod,with ERA
overshootingCROCUSvaluesat times.Themeltingperiodcomestooearly, althoughthereanalysismeltingrates
aresometimessimilar to valuesfrom CROCUS.Thesnow analysis(step3 above) requiresthespecificationof a
snow densityto convert theobserveddepthof snow in equivalentwaterdepths(massvalues)which arethequan-
tities managedby thelandsurfacescheme.A fixedvalueof 250kg m-3 is assumedwhich appearsto betoo high
for freshsnow, explainingtheoverestimationduringtheaccumulationperiod.In thereanalysisvalues,therelaxa-
tion coefficient towardstheBvM85 climatologyis λ = 0.02which correspondsto a time scaleof τ = 12.5days.
PoorspatialresolutionovertheAlps mightresultin lackof dataandarelaxationto theunderestimatedclimatolog-
ical values (see bottom right panel ofFig. 8 over the Alps).
Possibleimprovementsof this techniquecouldbeto useanupdatedsnow massclimatologybasedondirectobser-
vationsandavailableataresolutionon1 degreecompatiblewith thecurrentECMWFmodelresolution(Sellerset
al. 1996).Thelandsurfaceschemecouldbeimprovedto provideamorereliablefirst-guess.With theintroduction
of thesnow densityasa prognosticvariable(Douville et al. 1995)it shouldbepossibleto performananalysisin
termsof snow depths,theobservedground-basedvariable,insteadof snow waterequivalent(sincesnowfall can
Aug]Sep^ Oct_
Nov` Dec^Jana Feb^
Marb Aprc
MayJun]
Jul]
100
200
300
(mm
)
100
200
300
(mm
)
100
200
300
(mm
)
100
200
300
(mm
)
100
200
300(m
m)
Aug]Sep^ Oct_
Nov` Dec^Jana Feb^
Marb Aprc
MayJun]
Jul]
CROCUSERA
81/82d
82/83d
83/84d
84/85d
85/86d
86/87d
87/88d
88/89d
89/90d
90/91e
Land surface assimilation
18 Meteorological Training Course Lecture Series
ECMWF, 2002
easilybeconvertedknowing densityof freshsnow). Remotelysensedinformationshouldbeused,sinceweekly,
andmorerecentlydaily, snow cover chartsarepreparedby NOAA/NESDIS (BruceRamsay, personalcommuni-
cation,seealsoSellersetal.1996)basedonacombinationof visible,infraredandmicrowaveimagery. Microwave
valuescouldbeusedto obtainsnow massandsnow density, usingthedependency of thesignalon frequency and
polarization(seereview in Hallikainen1996a;1996b);theavailablealgorithmsaremoreaccuratefor dry snow,
andhaveproblemsfor wetsnow or whentherearelargevariationsof theroughnessin thefield of view. Thevalues
of ECMWF analysisof snow massshouldbe routinelycomparedwith the US Air Forcedaily analysisof snow
depth, based on ground observations and satellite imagery.
7.2 Deep soil temperatures
Figure 12Soil temperature profiles averaged for a number of stations over northern Germany for 10 April 1997.
The solid (red) line represents the mean observed profile at 14 UTC, the dashed (blue) line represents the
interpolated short-term forecast values verifying at 15 UTC.
Sincedeepsoil temperaturesevolveover timescalesmuchlongerthanshortandmediumrangeforecasts,thelack
of initialisationfor thesevariablescanleadto potentialdrifts astheoneobservedfor soil moisturein theECMWF
model.ExperienceatECMWFduringthewinter95/96hasshown thatthisproblemcanoccurin winter, whenthe
atmosphericforcing over thesurfaceis weaker, andthethermalinfluenceform thesoil deeperlayersis moreim-
portant.Thecombinationof amisrepresentationof somephysicalprocessesin thelandsurfacescheme(soil water
freezing),excessive radiative cooling,andinsufficient downwardsheattransferin very stableboundarylayersled
-200
-190
-180
-170
-160
-150
-140
-130
-120
-110
-100
-90
-80
-70
-60
-50
-40
-30
-20
-10
0
soil
dep
th [
cm]
f
-200
-190
-180
-170
-160
-150
-140
-130
-120
-110
-100
-90
-80
-70
-60
-50
-40
-30
-20
-10
00g 2h 4i 6j 8k 10 12 14
soil temperature [C]l
0g 2h 4i 6j 8k 10 12 14
Germany (10001-10999) 137 stationsFC: 97041012 t + 3h / OBS: 97041014
Vertical soil temperature profiles
OBS MODEL
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 19
to excessivecoolingof thesoil, andlargenegativebiasesin screenlevel temperature.Nevertheless,theproblemis
lesscritical thanthesummersoil moisturedrift sincein winter thestablestructureof theboundarylayerdecouples
the atmosphere from the surface below and prevents surface errors from propagating into the mid-troposphere.
Unlike soil moisture,routinemeasurementsof profilesof temperaturein thesoil areperformedby mostmeteoro-
logical offices.Unfortunately, only very few of thesemeasurementsaretransmittedin theGTS,andthereforeno
globalor evencontinentalcoverageexistsin realtime.A handfulof europeancountriessendsdaily valuesto EC-
MWF, andthoseareusedregularly for validationandmonitoringthemodelvalues.An examplein Fig.12 , where
anaverageprofileof soil temperaturefor stationsin northernGermany is comparedwith thecorrespondingmodel
values,for a recentdate.Themodeldeeptemperaturesshow acoldbias,but it is interestingto notethatthemodel
gradientis compareswell with observations.Thismight indicatereasonablygoodsoil thermalproperties(they de-
terminetheverticalgradient),but comparatively pooratmosphericforcing in thepreviousweeks(determiningthe
rateof coolingof thewholesoil slab).Theresultsareconsistentwith thewell known systematicunderestimation
of longwave downwardsurfaceradiationof theECMWF modelandothermodels(Wild et al. 1995;Garrattand
Prata 1996).
Theonly existing methodfor initialising soil temperaturehasbeenproposedby Coiffier et al. (1987)whereana-
lysedincrementsat2 m arereporteddirectly in thesoil.Thissub-optimalmethodcouldbeimprovedby anoptimal
interpolation following the methodology ofMahfouf (1991).
7.3 Vegetation properties
Xueetal. (1996)have recentlyshown thatthespecificationof theseasonalvariationsof vegetationpropertiescan
haveasignificantimpacton thesimulationof monthlymeantemperaturesnearthesurfaceover theNorthAmeri-
cancontinent.Thesesonalevolution of thevegetationcanbeof importanceover areaswith agriculturalpractices
(mid-latitudecontinents,tropical regions).Vegetationpropertiesarecurrentlyeitherfixedspatially, asin theEC-
MWF model(Viterbo andBeljaars1996),or canvary from monthto monthaccordingto correspondencetables
(Mahfoufetal. 1995).In anoperationalcontext, theuseof satellitedatalikeGlobalVegetationIndexes(GVI) (see
Gutman1994)couldbeabetterwayto captureseasonallityandinter-annualvariability of thevegetation.Thema-
jor difficulty is to relatesatellitereflectancesto input parametersof the land surfacescheme(leaf areaindex,
albedo,vegetationcover,...).However, variationaltechniquesalreadyusedfor theretrieval of atmosphericvertical
profiles from satellite radiances appear promising (Eyre et al. 1993).
8CONCLUSIONS
During thelastfive years,theinitialisationof prognosticanddiagnosticsurfacevariableshasbeenrecognisedas
animportantissuefor numericalweatherprediction.Weaknessesof initialisationsbasedeitheronfirst-guessonly
or climatologyhave beenclearlyidentified.As a consequence,soil moistureis currentlyinitialisedin variousop-
erationalweathercentresusingsub-optimalanalysismethodsandobservationsfrom SYNOPreports(ECMWF,
UKMO, CMC, Météo-France,HIRLAM). Thesetechniquescouldbebeimprovedby usinganalysismethodsal-
readytestedfor atmosphericvariableslikeoptimuminterpolationin thesequentialframework and4D-Var (which
is ratherappealingby betteraccountingfor thenon-linearitiesof theproblemaswell asthetemporaldistribution
of observations).
Theotherlandsurfaceprognosticvariablesarestill crudelyinitialised.Concerningsnow massanalysisatECMWF,
thecurrentschemecouldbeimprovedby usingamorerecentclimatologyandamorerealisticobservationoperator
if snow density can be computed by the land surface parametrization.
Areasnotyetexploredconcerntheinitialisationof deepsoil temperaturesfor which thetimescaleof evolution is
Land surface assimilation
20 Meteorological Training Course Lecture Series
ECMWF, 2002
alsomuchlongerthanshortrangeforecasts,andthespecificationof vegetationpropertieshaving aseasonalcycle
(suchasthealbedo,thevegetationcoveror theleafareaindex). Remotesensedatathatcouldbeusefulin thatcon-
text include quantities such as satellite skin temperatures or the normalized difference vegetation indices (NDVI).
Finally, theexamplesshown in thepaperillustratethefactthattheinitializationandparametrizationmethodshave
to bedevelopedtogether. Realisticmodelsareneededto supportindirectmeasurementstechniques,e.g.,forecast
snow densityallows a betteruseof snow depthmeasurementsto initialise snow mass.On theotherhand,refine-
mentsin thelandsurfaceparametrizationusedby numericalmodelsmightbeneededin orderto extractfrom new
measurement techniques its full information content.
REFERENCES
Bastianssen, W.G.M. 1995Regionalizationof surfacefluxdensitiesandmoisture indicators in compositeterrain -
A remotesensingapproach under clear skiesin Mediterraneanclimates.Phd Thesis,Agricultural University,
Wageningen, The Netherlands, 273 pp.
Beljaars, A.C.M., P. Viterbo, M.J.Miller, and A.K. Betts1996Theanomaloussensitivity overUSA duringJuly
1993: sensitivity to land surface parameterization.Mon. Wea. Rev., 124,362–383.
Betts, J.R., J.H.Ball, A.C.M. Beljaars,M.J.Miller and P. Viterbo1996Thelandsurface-atmosphereinteraction:
A review based on observational and global modeling perspectives.J. Geophys. Res., 101D,7209–7225.
Blondin, C. 1991Parametrizationof landsurfaceprocessesin numericalweatherprediction.Landsurfaceevapo-
ration: Measurement and parametrization., T.J. Schmugge and J.C. André, Eds, Springer, 31–54.
Bougeault, P., B. Bret, P. Lacarrère,and J.Noilhan1991An experimentwith anadvancedsurfaceparametriza-
tion in a meso-beta-scale. Part II: The 16 June 1986 simulation.Mon. Wea. Rev., 119,2374–2392.
Bouttier, F., J.-F. Mahfouf,and J.Noilhan1993aSequentialassimilationof soil moisturefrom atmosphericlow-
level parameters. Part I: Sensitivity and calibration studies.J. Appl. Meteorol., 32,1335–1350.
Bouttier, F., J.-F. Mahfouf,and J.Noilhan1993bSequentialassimilationof soil moisturefrom atmosphericlow-
level parameters. Part II:Implementation in a mesoscale model.J. Appl. Meteorol., 32,1351–1364.
Brankovic, C.,and J.vanMaanen1985TheECMWFclimatesystem.ECMWFTech. Mem.109, ECMWF, Read-
ing, UK, 51 pp. + figs.
Brun, E., P. David, M. Sudul,and G. Brunot1992A numericalmodelto simulatesnow -cover stratigraphy for
operational avalanche forecasting.J. Glaciology, 38, 13–22.
Brun, E., E.Martin, Y. Simon, C.Gendre,and C.Coleou1989An energyandmassmodelof snow coversuitable
for operational avalanche forecast.J. Glaciology, 35, 333–342.
Callies, U., A. Rhodin,and D.P. Eppel1998A casestudyonvariationalsoil humidityanalysisfrom atmospheric
observations.J. Hydrol., 212–213, 95–108.
Calvet, J.-C., J.Noilhan,and P. Bessemoulin1998Retrieving theroot-zonesoil moisturefrom surfacesoil mois-
ture or temperature estimates: A feasibility study based on field measurements.J. Appl. Meteorol., 37, 371–386.
Carlson, T.N. 1991Recentadvancesin modelingthe infraredtemperatureof vegetationcanopies.Land surface
evaporation: Measurement and parametrization., T.J. Schmugge and J.C. André, Eds, Springer, 349–358.
Coiffier, J., Y. Ernie, J.-F. Geleyn, J. Clochard, J. Hoffman,and F. Dupont1987Theoperationalhemispheric
modelat theFrenchmeteorologicalservice.J. Met.Soc.Japan, SpecialIssueonShortandMediumRangeNumer-
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 21
ical Weather Prediction,337–345.
Corti, S., F. Molteni, and C. Brankovic 2000Predictabilityof snow-depthanomaliesoverEurasiaandassociated
circulation patterns.Q. J. R. Meteorol. Soc., 126,241–262.
Cuenca, R.H.,and J.Noilhan1991Useof soil moisturemeasurementsin hydrologicalbalancestudies.Landsur-
face evaporation: Measurement and parametrization., T.J. Schmugge and J.C. André, Eds, Springer, 287–299.
Douville, H., J.-F. Royer, and J.-F. Mahfouf 1995A new snow parameterizationfor the Météo-Franceclimate
model. Part I: validation in stand-alone experiments.Climate Dynamics, 12,21–35.
Eyre, J.R., G.A. Kelly, A.P. McNally, E. Anderson,and A. Persson1993Assimilationof TOVS radianceinfor-
mation through one-dimensional variational analysis.Q. J. R. Meteorol. Soc., 119,1427–1463.
Foster, D.J.,and R.D.Davy 1988Globalsnow depthclimatology. USAFETAC/TN-88/006,ScottAir ForceBase,
Illinois, 48 pp.
Garratt, J.R.,and A.J. Prata1996Downwelling longwave fluxesat continentalsurfaces-A comparisonof obser-
vationswith GCM simulationsandimplicationsfor theglobal land-surfaceradiationbudget.J. Climate, 9, 646–
655.
Georgakakos, K.P., and O.W. Baumer1996Measurementandutilizationof on-sitemoisturedata.J. Hydrol., 184,131–152.
Giard, D., and E.Bazile2000Implementationof anew assimilationschemefor soil andsurfacevariablesin aglo-
bal NWP model.Mon. Wea. Rev., 128,997–1015.
Goutorbe, J.-P., J.Noilhan, C. Valancogne,and R.H.Cuenca1989Soil moisturevariationsduringHAPEX-MO-
BILHY. Ann. Geophys., 7, 415–426.
Gutman, G.G. 1994Global dataon land surfaceparametersfrom NOAA AVHRR for usein numericalclimate
models.J. Climate, 7, 669–680.
Hallikainen, M. 1996aPhysicalbasisfor microwave remotesensingof seaice andsnow. Radiationandwaterin
the climate system., T.J.E. Raschke, Ed, Springer, 489–525.
Hallikainen, M. 1996bRetrieval of seaice andsnow parametersfrom microwave radiometerandradardata.Ra-
diation and water in the climate system., T.J.E. Raschke, Ed, Springer, 527–549.
Jacquemin, B., and J. Noilhan1990Sensitivity studyandvalidationof a landsurfaceparametrizationusingthe
HAPEX-MOBILHY data set.Bound.-Lay. Meteor., 52,93–134.
Jones, C.D., and B. Macpherson1995A hydrologycorrectionsschemefor themesoscalemodelusingobserved
precipitation rates. Forecasting Research Tech. Rep. 151, Meteorological Office, Bracknell, UK.
Lanzinger, A. 1995The verificationof weatherparameters.Proc. SeminarECMWF, 5-9 September1994, EC-
MWF, Reading, UK, 373–383.
Lorenc, A. C. 1986 Analysis methods for numerical weather prediction.Q. J. R. Meteorol. Soc., 112,1177–1194.
Macpherson, B. 1996 Initialisation of soil moisturein the operationalmesoscalemodel.ForecastingResearch
Tech. Rep. 183, Meteorological Office, Bracknell, UK.
Mahfouf, J.-F. 1990A numericalsimulationof thesurfacewaterbudgetduringHAPEX-MOBILHY. Bound.-Lay.
Meteor., 53,201–222.
Mahfouf, J.-F. 1991Analysisof soil moisturefrom near-surfaceparameters:A feasibilitystudy. J. Appl.Meteorol.,
30,1534–1547.
Land surface assimilation
22 Meteorological Training Course Lecture Series
ECMWF, 2002
Mahfouf, J.-F., A.O. Manzi, J.Noilhan, H. Giordani,and M. Déqué1995ThelandsurfaceschemeISBA within
theMétéo-FranceclimatemodelARPEGE.Part I: Implementationandpreliminaryresults.J. Climate, 8, 2039–
2057.
Mahfouf, J.-F., and J.Noilhan1991Comparativestudyof variousformulationsof evaporationfrombaresoil using
in-situ data.J. Appl. Meteorol., 30,1354–1365.
Mailhot, J., R. Sarrazin,B. Bilodeau, N. Brunet,and G. Pellerin1997Developmentof the35-kmversionof the
operational regional forecast system.Atmosphere-Ocean, 35,1–28.
Manabe, S.1969Climateandtheoceancirculation.I. Theatmosphericcirculationandthehydrologyof theearth’s
surface.Mon. Wea. Rev., 97,739–774.
Martin , E. 1996 Validation of the ECMWF Re-analyses.Alpine validationswith the snow model CROCUS.
Météo-France CNRM/CEN Internal Report, 57pp.
McNider, R.T., A.J.Song, D.M. Casey, P.J.Wetzel, W.L. Crosson,and T.N. Carlson1995Towardsadynamic-
thermodynamicassimilationof satellitesurfacetemperaturein numericalatmosphericmodels.Mon. Wea.Rev.,
122,2784–2803.
Mintz, Y., and Y.V. Serafini1992A globalmonthlyclimatologyof soil moistureandwaterbalance.ClimateDyn.,
8 13–27.
Mitchell, K.E. 1994GCIPinitiativesin operationalmesoscalemodelinganddataassimilationat NMC. Proceed-
ings of the Fifth Symp. on global change studies, Nashville, Tenn., Amer. Meteor. Soc., 192–198.
Navascues, B. 1997Analysisof 2 metertemperatureandrelative humidity. HIRLAM Tech.rep.28,26 pp.Nor-
rkoping.
Noilhan, J.,and S.Planton1989A simpleparameterizationof landsurfaceprocessesfor meteorologicalmodels.
Mon. Wea. Rev., 117,536–549.
Njoku, E.G., andD. Entekhabi 1996 Passive microwave sensing of soil moisture.J. Hydrol., 184,101–129.
Paloscia, S.1996Remotesensingof soil moisture.Radiationandwaterin theclimatesystem., T.J.E.Raschke,Ed,
Springer, 361–400.
Rhodin, A., F. Kucharski, U. Callies, D.P. Eppel,and W. Wergen1999Variationalsoil moistureanalysisfrom
screen-level atmosphericparameters:Applicationto a short-rangeweatherforecast.Q. J. R.Meteorol. Soc., 125,2427–2448.
Rind, D. 1982The influenceof groundmoistureconditionsin North Americaon summerclimateasmodeledin
the GISS GCM.Mon. Wea. Rev., 110,1487–1494.
Rowell, D.P., and C.Blondin1990Theinfluenceof soil wetnessdistributiononshort-rangerainfall forecastingin
West African Sahel.Q. J. R. Meteorol. Soc., 116,1471–1485.
Rowntree, P.R.,and J.A.Bolton1983Simulationof theatmosphericresponseto soil moistureanomaliesoverEu-
rope.Q. J. R. Meteorol. Soc., 109,501–526.
Schmugge, T.J.,and F. Becker 1991Remotesensingobservationsfor themonitoringof land-surfacefluxesand
waterbudgets.Landsurfaceevaporation: Measurementandparametrization., T.J.SchmuggeandJ.C.André,Eds,
Springer, 337–347.
Schmugge, T., and T.J.Jackson1994Mappingsurfacesoil moisturewith microwave radiometers.Meteorol. At-
mos. Phys., 54,213–223.
Land surface assimilation
Meteorological Training Course Lecture Series
ECMWF, 2002 23
Schulin, R., H. Fluhler, H.M. Selim, B. Sevruk, and P.J.Wierenga 1992Soil moisture.Snowcover measure-
mentsandarealassessmentof precipitationandsoil moisture, B. Sevruk, Ed.,Op.HydrologyRep.No. 35,WMO
No. 749, Part III, 219-283.
Sellers, P.J., B.W. Meeson,J.W. Closs, J.Collatz, F.E.Corprew, D.Dazlich, F.G.Hall, Y. Kerr, R. Koster, S.
Los, K. Mitchell, J.M.P. McManus, D.M. Myers, K.-J.Sun,and P. Try 1996TheISLSCPinitiativeI-globaldata
sets:Surfaceboundaryconditionsandatmopshericforcingsfor land-atmospheremodels.Bull. Amer. Meteor. Soc.,
77,1987–2005.
Sellers, P.J., Y. Mintz, Y.C.Sud,and A. Dalcher1986A SimpleBiosphere(SiB) modelfor usewhin generalcir-
culation models.J. Atmos. Sci., 43,505–531.
Smith, C.B., M.N. Lakhtakia, W.J.Capehart,and T.N. Carlson1994Initializationof soil-watercontentin region-
al-scale atmospheric prediction models.Bull. Amer. Meteor. Soc., 75,585–593.
Taconet, O., R. Bernard,and D. Vidal-Madjar1986Evapotranspirationover anagriculturalregion usinga sur-
face flux/temperature model based on NOAA-AVHRR data.J. Climate and Applied Meteor., 25284–307.
vandenHurk, B., W. Bastiaanssen,H. Pelgrum,and E.vanMeijgaard1997A new methodologyfor assimilation
of initial soil moisturefields in weatherpredictionmodelsusingMETEOSAT andNOAA. J. Appl. Meteor. 361271–1283.
Verseghy, D.L. 1991CLASS-ACanadianlandsurfaceschemefor GCMs.I. Soil model.Int. J.Climatol., 11111–
133.
Viterbo, P., and A.C.M. Beljaars1995An improvedlandsurfaceparametrizationschemein theECMWF model
and its validation.J. Climate, 8, 2716–2748.
Viterbo, P., and P. Courtier1995Theimportanceof soil waterfor medium-rangeweatherforecasting.Implications
for dataassimilation.Worshopon Imbalanceof SlowlyVaryingComponentsof PredictableAtmosphericMotions,
W.M.O., Beijing, China, March 1995.
Wang, J.R.,1992An overview of measurementsof soil moistureandmodelingof moistureflux in FIFE.J. Geo-
phys. Res., 97D,18955–18959.
Wei, M., Ed. 1984Soil moisture:Reportof a workshopheldat Tiburon,California25-27January1994.NASA
Conference Publication 3319.
Wetzel, P.J., D. Atlas,and R.H.Woodward1984Determiningsoil moisturefrom geosynchronoussatelliteinfra-
red data: A feasibility study. J. Appl. Meteorol., 23,375–391.
Wetzel, P.J.,andJ.T. Chang1988Evapotranspirationfrom nonuniformsurfaces:A first approachfor short-term
numerical weather prediction.Mon. Wea. Rev., 116,600–621.
Wild, M., A. Ohmura,H. Gilgen,and E. Roeckner1995Validationof generalcirculationmodelradiativefluxes
using surface observations.J. Climate, 8, 1309–1324.
Xue, Y., M.J.Fennessy, and P.J.Sellers1996Impactof vegetationpropertiesonU.S.summerweatherprediction.
J. Geophys. Res., 101D,7419–7430.
Yang, R., M.J.Fennessy, and J.Shukla1994Theinfluenceof initial soil wetnessonmediumrangesurfaceweath-
er forecasts.Mon. Wea. Rev., 122471–485
Yeh, T.C., R.T. Wetherald,and S. Manabe1984Theeffect of initial soil wetnesson theshort-termclimateand
hydrology change-A numerical experiment.Mon. Wea. Rev., 112474–490.