the interactions between soil-biosphere-atmosphere (isba...

Geosci Model Dev 10 1621ndash1644 2017wwwgeosci-model-devnet1016212017doi105194gmd-10-1621-2017copy Author(s) 2017 CC Attribution 30 License

The interactions between soilndashbiospherendashatmosphere (ISBA) landsurface model multi-energy balance (MEB) option in SURFEXv8ndash Part 2 Introduction of a litter formulation and model evaluationfor local-scale forest sitesAdrien Napoly1 Aaron Boone1 Patrick Samuelsson2 Stefan Gollvik2 Eric Martin3 Roland Seferian1Dominique Carrer1 Bertrand Decharme1 and Lionel Jarlan4

1CNRM UMR 3589 Meacuteteacuteo-FranceCNRS Toulouse France2Swedish Meteorological and Hydrological Institute Norrkoumlping Sweden3IRSTEA U-R RECOVER Aix en Provence France4Centre drsquoeacutetudes Spatiales de la Biosphegravere (CESBIO) Toulouse France

Correspondence to Adrien Napoly (adrnapolygmailcom)

Received 14 October 2016 ndash Discussion started 27 October 2016Revised 20 March 2017 ndash Accepted 29 March 2017 ndash Published 18 April 2017

Abstract Land surface models (LSMs) need to balance acomplicated trade-off between computational cost and com-plexity in order to adequately represent the exchanges of en-ergy water and matter with the atmosphere and the oceanSome current generation LSMs use a simplified or compos-ite canopy approach that generates recurrent errors in sim-ulated soil temperature and turbulent fluxes In response tothese issues a new version of the interactions between soilndashbiospherendashatmosphere (ISBA) land surface model has re-cently been developed that explicitly solves the transfer ofenergy and water from the upper canopy and the forest floorwhich is characterized as a litter layer The multi-energybalance (MEB) version of ISBA is first evaluated for threewell-instrumented contrasting local-scale sites and sensitiv-ity tests are performed to explore the behavior of new modelparameters Second ISBA-MEB is benchmarked against ob-servations from 42 forested sites from the global micro-meteorological network (FLUXNET) for multiple annual cy-cles

It is shown that ISBA-MEB outperforms the compositeversion of ISBA in improving the representation of soil tem-perature ground sensible and to a lesser extent latent heatfluxes Both versions of ISBA give comparable results interms of simulated latent heat flux because of the similar for-mulations of the water uptake and the stomatal resistanceHowever MEB produces a better agreement with the ob-

servations of sensible heat flux than the previous versionof ISBA for 875 of the simulated years across the 42forested FLUXNET sites Most of this improvement arisesowing to the improved simulation of the ground conductionflux which is greatly improved using MEB especially owingto the forest litter parameterization It is also shown that cer-tain processes are also modeled more realistically (such asthe partitioning of evapotranspiration into transpiration andground evaporation) even if certain statistical performancesare neutral The analyses demonstrate that the shading effectof the vegetation the explicit treatment of turbulent transferfor the canopy and ground and the insulating thermal andhydrological effects of the forest floor litter turn out to beessential for simulating the exchange of energy water andmatter across a large range of forest types and climates

1 Introduction

The land surface model (LSM) is one of the key parame-terization schemes of atmospheric models used for numeri-cal weather prediction and climate simulations It is used tocompute the turbulent fluxes of heat water and momentumalong with the radiative fluxes at the surfacendashatmosphereinterface In addition the current generation of LSMs are

Published by Copernicus Publications on behalf of the European Geosciences Union

1622 A Napoly et al The ISBA-MEB option in SURFEXv8

used to compute flux exchanges of certain chemical species(such as carbon dioxide and biogenic organic volatile car-bon) and emissions of particles (such as aerosols from dust orbiomass burning) The interactions between soilndashbiospherendashatmosphere (ISBA) is part of the SURFace EXternaliseacutee plat-form (SURFEX) developed in recent years at Meacuteteacuteo-France(Masson et al 2013) and a suite of international partnersIt is used for a suit of surface schemes in either offline orcoupled mode with an atmospheric model andor a hydro-logical model ISBA has benefited from continuous improve-ments since its first version (Noilhan and Planton 1989) inparticular for the carbon cycle (Calvet et al 1998 Gibelinet al 2006) soil mass and heat transfer (Boone et al 1999Decharme et al 2011) snowpack processes (Boone andEtchevers 2001 Decharme et al 2016) sub-grid hydrol-ogy (Decharme and Douville 2006) and radiative transferthrough the canopy (Carrer et al 2013)

In order to remain consistent with the aforementioned de-velopments and to respond to both current and future user de-mands a multi-energy balance (MEB) approach has been de-veloped (Boone et al 2017) This improvement is attributedto the representation of the surface as a multi-source model(ie a separation of the canopy and the surface-soil layer)in contrast to the standard ISBA soilndashvegetation compositemodel This new model has been designed to better representcertain processes for forested areas and to incorporate newmodeled processes which can benefit from a more accuraterepresentation of the soilndashvegetation continuum In particu-lar partitioning of the incoming energy into turbulent andground heat fluxes are expected to be more realistic with theexplicit vegetation scheme as well as the partitioning of la-tent heat flux into its different components The carbon cy-cle should also benefits from a more conceptually accuraterepresentation of the surface (Carrer et al 2013) since itis strongly linked to leaf temperature via assimilation of at-mosphere carbon Moreover snowmelt during spring is verysensitive to the snow net radiation budget which is impactedby the presence of tall vegetation and interception and lossby the canopy during the winter season (Rutter et al 2009)

The surface in forested regions beneath the canopy is oftencovered by a layer of dead leaves or needles branches fruitsand other organic material which can be characterized as alitter layer The explicit inclusion of such a layer is neglectedfor the most part in LSMs used in global-scale models or it isonly partly taken into account For example Sakaguchi andZeng (2009) implemented a specific resistance due to litterfor soil evaporation The inclusion of litter-related processeshas been shown to have a substantial impact on hydrolog-ical (Putuhena and Cordery 1996 Guevara-Escobar et al2007) and thermal processes (Andrade et al 2010) Litterhas an important insulating capacity as its thermal diffusiv-ity is small compared to soil (Riha et al 1980) which leadsto a strong impact on the soil temperature diurnal cycle andground conduction flux Changes in this flux can then impactthe energy available for radiative and turbulent fluxes The

presence of litter also alters ground evaporation by cover-ing the soil with a high porosity layer which essentially pre-vents liquid-water capillary rise (Schaap and Bouten 1997)It also constitutes another water interception reservoir forrainfall and canopy drip prior to infiltration and runoff (egGerrits et al 2007) Some models have introduced parame-terizations for litter (Ogeacutee and Brunet 2002 Gonzalez-sosaet al 1999 Wilson et al 2012 Haverd and Cuntz 2010Sakaguchi and Zeng 2009) but the approach can be verydifferent from one to another depending on their complexity(only modifying or adding a ground resistance modeling thelitter using an explicit single or multi-layer model) In addi-tion each of the models were validated against observationsmade for a single well-instrumented site

The goal of the current study is to evaluate the im-pact of a new parameterization of the soilndashlitterndashvegetationndashatmosphere continuum at the local scale In the first part ofthis study an in-depth evaluation for three forest sites inFrance is carried out They have been selected in order torepresent a range of forest types and climates The main goalis to understand the effect of both the explicit canopy layerand the litter layer on the available energy partitioning (la-tent sensible ground) soil temperatures and soil water con-tent The second part of this study consists of a benchmarkstudy using 42 sites from the FLUXNET network (Baldocchiet al 2001) The objective is 2-fold (1) to evaluate the per-formance of the MEB options against the classic compositesoilndashvegetation version of the ISBA model over a wide rangeof forest species and climates and (2) to analyze whetherthe general improvements seen at the three well-instrumentedand documented local-scale sites in France can be extendedto other climates and forest types

This is essential since ISBA is used within the SUR-FEX platform in various configurations at resolutions rang-ing from several kilometers at the regional scale such aswithin the operational mesoscale numerical weather pre-diction model AROME (Seity et al 2011) and the opera-tional distributed hydrological model system SIM (Habetset al 2008) to resolutions of hundreds of kilometers inglobal-scale models such as within the global climate mod-els CNRM-CM51 (Voldoire et al 2013) and CNRM-ESM1(Seacutefeacuterian et al 2016)

2 Model

21 The standard ISBA composite soilndashvegetationmodel

The standard ISBA model uses a single composite soilndashvegetation surface energy budget which means that the prop-erties of the soil and vegetation are aggregated within eachgrid-cell point (Noilhan and Planton 1989 Noilhan andMahfouf 1996) In this study the multi-layer soil diffusion(DF Decharme et al 2011) option is used along with the

Geosci Model Dev 10 1621ndash1644 2017 wwwgeosci-model-devnet1016212017

A Napoly et al The ISBA-MEB option in SURFEXv8 1623

explicit multi-layer snow scheme The prognostic soil tem-perature is represented by Tg for Ng soil layers The snow-pack uses the snow enthalpy as a prognostic variable fromwhich both the snow liquid-water content and temperaturecan be diagnosed for Nn layers The hydrological prognosticvariables are the canopy reservoir for water interception Wr(kg mminus2) the snow liquid-water content Wn for Nn snowlayers and the soil volumetric water content and water con-tent equivalent of frozen water wg and wgf (m3 mminus3) re-spectively for Ngl active hydrological soil layers (whereNgl leNg) Further details describing the soil and snow-pack representations as well as other prognostic variables andmodel assumptions are detailed in Decharme et al (2016)

22 The ISBA-MEB model

The multi-energy balance model ISBA-MEB treats up tothree fully coupled distinct surface energy budgets (ie thesnow surface the bulk vegetation canopy and the groundwhich is characterized as either a soil surface or litter layersee Sect 23) The reader is referred to Boone et al (2017)for an extended description of the various assumptions ofthe MEB approach its full set of governing equations andits numerical aspects Compared to the classic ISBA ap-proach there is one additional prognostic heat storage vari-able which is the vegetation temperature Tv The new hy-drological prognostic variable is the water content equivalentof snow stored within the vegetation canopy Wrn (kg mminus2)

23 Explicit model for litter in ISBA-MEB

Two methods are generally used to represent the effect of alitter layer within LSMs The first method consists of addinga specific ground resistance in order to reduce the soil evapo-ration due to the presence of a litter layer For example Sak-aguchi and Zeng (2009) and Park et al (1998) have used thismethod in their models with the resistance depending on theleaf area index (LAI) of the litter or its thickness The sec-ond method utilizes an explicit model to represent the effectsof the litter The implementation of this method turns outto be quite variable among different LSMs although gener-ally it accounts for both thermal and hydrological effects Inthe current study a model was introduced that simulates thefirst-order effects of a litter layer on the surface energy andwater budgets while minimizing the number of parametersand the complexity To this end a single bulk-layer approachwhich is based on the model of Schaap and Bouten (1997)has been developed The aforementioned study has also in-spired similar approaches by Ogeacutee and Brunet (2002) andHaverd and Cuntz (2010) The method is relatively simple asingle explicit litter layer is inserted between the vegetationlayer and the upper soil layer With this approach three addi-tional state equations have to be solved (i) the litter averagetemperature Tl (K) (ii) the litter liquid-water content Wl(kg mminus2) and (iii) the litter water content equivalent of ice

Wlf (kg mminus2) Two noteworthy assumptions have been madefirst it is assumed that there is no evaporation directly fromthe soil in the presence of litter and second water move-ment by capillarity rise from the soil upwards into litter isneglected The energy budget is computed for the litter layerusing prescribed values of thermal conductivity heat capac-ity and litter thickness The water budget is computed in asimilar manner as for the vegetation canopy layer Finallyfrozen water in the litter is modeled using the same simplefreezing model method as for the soil in ISBA (Boone et al2000) The model is fully described in Appendix A

3 Data

31 Energy balance closure

Energy balance closure is a well-known issue when turbu-lent fluxes are computed with the eddy covariance techniqueThe closure δ is defined as the ratio of turbulent fluxes toavailable energy at the surface

δ =H +LE

RnminusGminus S (1)

where S Rn G H and LE (W mminus2) represent the vegeta-tion heat storage the net radiation ground conduction (heat)flux the sensible heat flux and the latent heat flux respec-tively The basic idea is that closure should be as close aspossible to unity when observations are compared with LSMoutput since such models close the energy budget (to a highdegree) by design As it turns out the mean closure imbal-ance for FLUXNET sites is typically 20 (Wilson et al2002) According to Twine et al (2000) the net radiation isprobably the most accurately measured component of the en-ergy balance and closure can be reasonably done by adjustingH and LE The Bowen ratio closure method is often adoptedto correct the non-closure It takes on the assumption that theBowen ratio (ratio of the sensible to the latent heat fluxes)is well estimated by the eddy covariance system so that theturbulent fluxes can be adjusted while conserving the Bowenratio to ensure closure (Blyth et al 2010 Zheng et al 2014Er-Raki et al 2008) The adjusted sensible and latent heatfluxes can be computed as

Hadj =H + restimesH

H +LE (2)

LEadj = LE + restimesLE

H +LE (3)

where Hadj and LEadj (W mminus2) represent the adjusted sensi-ble and latent heat fluxes respectively and the residual en-ergy is defined as

res= RnminusGminus SminusH minusLE (4)

Note that there is no specific check here that resgt 0 whichis assumed in the correction method But imposing such a

wwwgeosci-model-devnet1016212017 Geosci Model Dev 10 1621ndash1644 2017


Figure 1 The location of the three well-instrumented forested sitesin France

condition has proven to lead to very similar results in paral-lel tests In order to make sure that this assumption did notimpact our results we recomputed the statistics also check-ing that resgt 0 and we found the impact to be negligible

Among the terms in Eq (4) the energy stored in thecanopy S is generally not measured Even though it is rela-tively small compared to the net radiation it has been shownto be non-negligible in some forest canopies (Oliphant et al2004) One possible method which can be used to addressthis issue is to use the simulated residual (Blyth et al 2010)In the current study tests were done using both the modelsimulated S and neglecting it It was found that incorporat-ing this term had very little influence on the adjusted fluxesIn the analysis that follows we neglected this term and boththe original and the adjusted fluxes are shown

32 Local-scale evaluation over France

Three well-instrumented forest sites have been used formodel evaluation that cover a range in climate soils and veg-etation characteristics Their location is shown in Fig 1 andtheir main characteristics are summarized in Table 1 At eachsite measurements of sensible and latent heat fluxes weremade using the eddy covariance technique and the groundheat flux was measured with flux plates Soil temperature(Tg) and volumetric moisture content (wg) profiles were alsoavailable for two of the sites (Le Bray and Barbeau)

The Barbeau site is located in the Barbeau National Forestwhich is approximately 60 km from Paris (484 N 27 EFrance elevation 90 m) This site consists mainly of a de-ciduous broadleaf mature oak forest which has an averageheight of 27 m The climate is temperate with a mean annualtemperature of 107 C and the annual rainfall is 680 kg mminus2

Table 1 Characteristics of the three French forest sites

Site Barbeau Pueacutechabon Bray

Year 2013 2006 2006Localization Paris Montpellier BordeauxMain Sessile Green Maritimevegetation type oak oak pineGrassClimate Temperate Mediterranean MaritimeForest type Deciduous Evergreen Evergreen

broadleaf broadleaf needleleafMean annual temp (C) 107 135 129Annual Rainfall (mm) 680 872 997

The soil texture is loam in the upper soil and clay loam at thelower portion of the soil profile (Preacutevost-Boureacute et al 2009)

The Le Bray site is located about 20 km from BordeauxFrance (447 N minus07 E elevation 62 m) It is a maritimepine forest classified as evergreen needleleaf with an averageheight of 18 m with a dense grass understory which com-prises about half of the total LAI at its maximum The meanannual temperature is 129 C and the mean annual precipi-tation is 997 kg mminus2 The soil is a sandy and hydromorphicpodzol with a layer of compacted sand that starts at a depthbetween 04 and 08 m which constitutes the limit of rootpenetration The site also has a water table whose depth fluc-tuates during the year and at times can reach the surfaceThis is the only site where specific measurements for litterwere made and a near-constant litter thickness of 005 m wasobserved (Ogeacutee and Brunet 2002)

The Pueacutechabon site is located roughly 35 km northwest ofMontpelier in the Pueacutechabon State Forest (437N 35 Eelevation 270 m) Vegetation is largely dominated by a denseoverstory of holm oak (evergreen broadleaf forest) with amean height of 55 m The climate is typical Mediterraneanwith mean annual temperature of 135 C and a mean annualprecipitation of 872 kg mminus2 The soil texture is homogeneousdown to about 05 m depth and it can be described as siltyclay loam The parent rock is limestone (Rambal et al 2014)

33 Benchmark

The second part of the study uses observations from a subsetof the FLUXNET sites to assess systematically or benchmarkthe suite of ISBA versions developed at Meacuteteacuteo-France TheFLUXNET database has been used for LSM evaluation byseveral widely used LSMs (Stoumlckli et al 2008 Blyth et al2010 Ukkola et al 2016) Many sites are available withinthe FLUXNET database however in the current study onlythose sites and years with an energy balance closure of 20 (or less) before adjustment of the turbulent fluxes are re-tained Ground heat flux was assumed to represent 3 ofnet radiation when it is not available (this is close to the av-erage value over the three French sites) After screening 42forested sites remain which give a total of 179 years of ob-servations (see Fig 2 for the locations of the sites used in



the current study) The method used to gap-fill missing at-mospheric forcing data is described in Vuichard and Papale(2015) The three sites from the first part of this study havebeen removed in this analysis

4 Model setup

All the simulations are performed using SURFEXv8 We usethe diffusive soil (DF) option meaning that the soil heat andmass transfers are solved on a multi-layer grid (Decharmeet al 2011) A recently developed multi-layer canopy ra-diative transfer scheme (Carrer et al 2013) has been in-corporated into MEB in order to determine the fraction ofincoming radiation intercepted by the canopy layer trans-mitted to the ground and reflected This model uses struc-tural organization of the canopy and spectral properties ofthe leaves as well as the albedo from the soil and the vege-tation for two different spectral bands (visible and near in-frared) The canopy resistance formulation is based on theA-gs (leaf net assimilation of CO2 ndash leaf conductance to wa-ter vapor) model (Calvet et al 1998 Gibelin et al 2006)which simulates photosynthesis and its coupled to the stom-atal conductance in response to atmospheric CO2 Param-eters of the stomatal resistance scheme are vegetation-typedependent and were chosen according to the ECOCLIMAPdatabase (Faroux et al 2013)

Three simulations were performed for each site in orderto assess the impact of the new canopy and litter layers onthe simulated fluxes soil temperature and soil moisture Themodels were forced with atmospheric data at half-hourlytime steps In the first simulation the reference model (ieusing a single composite vegetation and surface-soil layer)was used and it is hereafter referred to as ISBA The sec-ond simulation with the explicit canopy layer corresponds tothe ISBA model using the MEB approach (referred as MEBherein for simplicity) The last simulation was carried outwith both the explicit canopy layer and the explicit forest lit-ter layer and it is referred to as MEBL

41 Local scale

The pedotransfer functions of Clapp and Hornberger (1978)are used for the soil water flow The soil hydraulic conduc-tivity at saturation the b-exponent and the matric potentialat saturation are computed with the continuous formulationsof Noilhan and Lacarrere (1995) using the textural propertiesat each site The wilting point field capacity and the watercontent at saturation were computed using the observationsof soil volumetric water content andor following the rec-ommendations of the site principal investigators (Table 2)An hydraulic conductivity exponential profile was used tomodel the packing of the soil with depth Soil organic con-tent which affects the hydrological and thermal coefficientsas in Decharme et al (2016) is modeled using the input

data from the Harmonized World Soil Database (HWSDNachtergaele and Batjes 2012) LAI canopy height vegeta-tion type total soil and root depths were set to values foundin the corresponding literature for each site (Table 2) Lit-ter thickness 1zl which is one of the key parameters of thelitter module varies in space and time This study assumesit is constant in time based on estimates made at each siteThis is an important approximation but a literature reviewreveals that measured litter thickness generally varies from 1to 10 cm (see Table 3) Because of the uncertainties relatedto specification of this parameter a series of sensitivity testsare done and the results are presented in Sect 52

At the Le Bray site the water table has a significant influ-ence on the seasonal soil wetness Measurements of its depthwere available and allowed for a simple parameterization tobe developed It consists of a strong relaxation towards sat-uration in soil layers below the observed water table depthThus soil moisture within the saturated zone is very close tothe observed values in the saturated or nearly saturated lay-ers while soil moisture above this zone is permitted to freelyevolve

For each site simulated turbulent heat fluxes were com-pared to both the original and the adjusted-observed valuesassuming that model results should fall inside the area delim-ited by these two curves Moreover since a proper evaluationof each flux component of the energy balance has to be donewith a closed energy budget (in order to be consistent withthe model which imposes closure by design) the scores arecomputed with the adjusted flux values

42 Benchmark

The ECOCLIMAP land cover and the HWSD soil databaseswere used to prescribe most of the needed parameters as wasdone for the local-scale evaluation for the three French sites(in the previous section) This is also consistent with themethod used for spatially distributed offline and fully cou-pled (with the atmosphere) simulations with ISBA Howevernote that soil texture canopy height and vegetation type werechosen in agreement with literature values for each site whereavailable therefore they superseded the ECOCLIMAP val-ues for these parameters The default thickness of the litterlayer is set to 3 cm based on the sensitivity test results pre-sented in Sect 52 and in overall agreement with literaturevalues (Table 3)

Initial conditions can have a significant impact on both thesensible and the latent heat fluxes but they were not knownthus a spin-up period was used for each site Sites were ini-tialized with saturated soil water content conditions and thefirst available year was repeated at least 10 times until a pre-defined convergence criteria was achieved Only the latentand sensible heat fluxes are evaluated since the ground heatflux soil temperature and soil water content were not avail-able at each FLUXNET site



Figure 2 The forested sites from the FLUXNET network Selected sites (shown) have a maximum energy imbalance at or below 20 Thecircles indicate the location of sites retained for this study

Table 2 The main model parameters for each site Literature indicates that values come from studies cited in the text and estimated meansthat values were provided by the principal investigators of each site

Site Barbeau Puechabon Bray Source

Soil parameters

Sand () 41 14 98 literatureClay () 39 40 2 literatureWsat 036ndash048 006 042ndash042 estimatedWfc 015ndash035 0046 017ndash016 estimatedWwilt 005 0018 004ndash003 estimatedSOC 0ndash30 cm (kg mminus2) 5 5 48 HWSDSOC 70ndash100 cm (kg mminus2) 4 55 9 HWSDRoot depth (m) 12 50 08 literatureSoil albedo (visnir) 003013 00502 01018 estimated

Vegetation parameters

LAI (m2 mminus2) 05ndash60 24 20ndash40 literatureHeight (m) 27 55 18 literatureVegetation albedo (visnir) 005020 003017 004017 estimatedVegetation fraction 095 099 095 ECOCLIMAP

Litter parameters

Thickness (m) 003 001 005 estimated

5 Results and discussion

The MEB model has been compared to both observations andthe standard ISBA composite vegetation model for severalwell-instrumented contrasting (in terms of forest type andclimate) sites in France and to a subset of the FLUXNETsites for multiple annual cycles using a benchmarking appli-cation The results of these evaluations and of several sensi-tivity tests are given in this section

51 Evaluation for three well-instrumented forestedsites

511 Net radiation

The total net radiation (Rn) is analyzed in terms of the netlong-wave (LWnet) and shortwave (SWnet) components TheSWnet primarily depends on the prescribed albedo for thesoil αg and the vegetation αv In ISBA the net short-wave radiation is simply defined as SWnet = (1minusαeff)SW darrwhere the effective albedo is simply αeff = vegαv+ (1minusveg)αg and veg is the vegetation cover fraction (which is con-stant in time for forests and varies between 095 and 099 ac-



Table 3 Measured litter thickness reported for various sites

Site Country Main cover Thickness MassReference type (cm) (kg mminus2)

Le Bray France Maritime Pine 5 26Ogeacutee and Brunet (2002) needle constant

Oak Ridge USA deciduous oak 4 06Wilson et al (2012) leaf constant

Kyushu Univ Japan Cjaponica 52 17Sato et al (2004) needle March 2002

Kyushu Univ Japan Ledulis 86 21Sato et al (2004) leaf March 2002

Netherland Douglas fir 5Schaap and Bouten (1997) needle

Russia Taiga 42Vorobeichik (1997) leaf July 1990

Russia Taiga 54Vorobeichik (1997) needle July 1990

CSIR South Africa Eucalyptus grandis 38 23Bulcock and Jewitt (2012) leaves March 2011

CSIR South Africa Acacia mearnsii 18 24Bulcock and Jewitt (2012) needle March 2011

CSIR South Africa Pinus patula 45 33Bulcock and Jewitt (2012) needle March 2011

Tumbarumba Australia Eucalyptus 3Haverd and Cuntz (2010) leaves

Canada White pine 25Wu et al (2014) needle

Ecuador Ocotea infrafoveolata 18 (average)Marsh and Pearman (1997) leaf

Mawphlang India Quercus griffithii 12 (average)Arunachalam and Arunachalam (2000) leaf

Aoyama Japan Pinus thunbergii 45Zhu et al (2003) needle

Seirseminen Finland Spruce 3-5Koivula et al (1999) needle

cording to the forest cover) For the MEB simulations thescheme from Carrer et al (2013) is used which uses the soiland vegetation albedo for two spectral bands (which are ag-gregated to all-wavelength values for ISBA) along with theLAI (the notion of veg is not applicable to MEB) in orderto obtain an estimate of the vegetation optical thickness De-spite these differences results are very similar for the SWnetcalculation as seen in Table 4 The models perform wellwith relatively low values of RMSE (root mean square er-ror) (lt 10 W mminus2) and annual error (AElt 8 W mminus2) for allthree sites

Since MEB explicitly uses the shortwave radiation trans-mitted through the canopy for the ground net radiation com-putation it can be compared with the photosynthetically ac-tive radiation (PAR) measurements below the canopy withinthe Barbeau and Pueacutechabon forests (Fig 3) Comparisonwith MEB results revealed that the transmission coefficientfor the PAR is overestimated The radiative transfer schemeparameter which exerts the most control over this process isthe clumping index and it has been adjusted for these twosites However the shortwave radiation transmission remainsslightly overestimated at the Barbeau forest during the first



Table 4 RMSE (root mean square error) R2 (square correlation coefficient) and AE (annual error) for the three French sites for the threedifferent experiments calculated for SWnet (net shortwave radiation) LWnet (net long-wave radiation)H (sensible heat flux) LE (latent heatflux) and G (ground heat flux)

Site Flux MEBL MEB ISBA

RMSE R2 AE RMSE R2 AE RMSE R2 AE

Bray SWnet 30 10 minus01 30 10 minus01 31 10 05LWnet 71 095 23 60 097 21 71 095 minus02H 524 084 57 525 083 47 631 077 minus21LE 592 054 minus12 608 05 minus15 589 055 08G 96 075 minus33 216 055 minus34 421 036 minus30

Puechabon SWnet 64 097 minus13 64 097 minus13 64 097 minus25LWnet 72 098 minus09 58 099 minus13 93 098 42H 437 095 41 464 094 27 723 087 minus22LE 373 066 minus55 386 062 minus41 413 058 minus73G 158 081 08 292 078 05 569 048 14

Barbeau SWnet 87 096 minus59 87 096 minus59 94 097 minus76LWnet 98 098 minus88 100 099 minus97 87 099 minus72H 541 08 231 638 067 159 511 08 172LE 534 076 minus151 596 066 minus73 486 078 minus102G 47 075 minus32 220 055 minus33 424 036 minus28

three months of the year including this adjustment (Fig 3c)The main reason is that during this period the LAI is verylow but the trunks and branches intercept solar radiation (theforest is relatively old and tall so that this effect can be quitesignificant) which is not taken into account by the modelThis could be done by including a stem area index (SAI) butcurrently the default minimum LAI is used as a proxy for thiseffect

The simulated LWnet depends on the explicit contributionsfrom the soil vegetation and snow in MEB and the com-posite soilndashvegetation layer and snow in ISBA The annualRMSE and AE absolute values are less or equal than 10 and97 W mminus2 respectively over all sites and runs (Table 4) Evenif differences can be noticed between the simulations espe-cially at Pueacutechabon these errors remain relatively small andcomparable at least compared to the errors of the turbulentand conductive heat fluxes (Table 4)

This is due in part to the use of the same values of emis-sivity for soil snow and vegetation in ISBA and MEB Itis also due to the fact that LWdarr is the same for both mod-els Note that another parameter which can have an impacton LWnet is the ratio of the roughness length for momentumto that for heat which is commonly referred to as kBminus1

=

ln(z0z0h) In the default version of ISBA the roughnesslength ratio is defined as z0z0h = 10 (which corresponds tokBminus1

asymp 23) and this value has been determined as a valuewhich works well generally (although for local-scale appli-cations values can be prescribed to range from 1 to on theorder of 1times 102) But note that there is evidence that kBminus1

is actually closer to 1 for tall canopies (eg Yang and Friedl2003) which corresponds to a roughness length ratio of ap-

proximately 27 A ratio of 1 is used for forests within thedefault version of the original two-source model in the RCA(Rossby Centre Regional Atmosphere Model Samuelssonet al 2011) dual-energy budget LSM Thus currently a ra-tio of 1 is used for MEB since it is closer to kBminus1

asymp 1 incontrast to 10 for ISBA The larger roughness ratio in ISBAcan be seen as a way of compensating for an underestimateddiurnal cycle of the surface radiative temperature owing to acomposite soilndashvegetation heat capacity (which is larger thanthat for a bulk vegetation layer as in MEB) This implies thatthe explicit canopy representation in MEB allows us to usea slightly more realistic value of roughness length ratio forforest canopies

512 Latent heat flux

At Le Bray and Pueacutechabon the simulated values of totalLE are relatively close between the three simulations andin fairly good agreement with the adjusted measurementsAbove the forest most of the evapotranspiration comes fromcanopy transpiration (Fig 4) which is modulated to a largeextent for the three sites by the root-zone soil moisture Allthree of the simulations use the same scheme for the soil wa-ter and the stomatal resistance Except with ISBA at Le Braythe simulations tend to slightly underestimate LE (negativeAE) for both these sites and models on an annual basis Onepossible explanation is that the formulation of the stomatalresistance was originally adapted for low vegetation typessuch as crops (Calvet et al 1998) At Barbeau the LE under-estimation (bias) is larger The likely reason is owing to thepresence of a water table that was not explicitly measured



Figure 3 The modeled (thin dashed line) and observed (thickdashed line) incoming shortwave radiation transmitted through thecanopy at (a) Le Bray (b) Pueacutechabon and (c) Barbeau The obser-vations are based on below-canopy PAR measurements The totalincoming shortwave radiation is also plotted (full black line) as areference

but strong evidence for itrsquos existence can be seen in the mea-sured soil water (especially in winter and spring Fig 5e)Without an explicit groundwater parameterization it is notpossible to model this water table accurately and develop-ment of such a scheme was beyond the scope of the currentstudy But these very wet observed soil conditions resulted in

mostly unstressed conditions during this time frame and thishigh level of soil water was not modeled by ISBA or MEB

For the Mediterranean forest at Pueacutechabon most of the netradiation is converted into sensible heat flux (Fig 6a) whichleads to low values of LE in accordance with observations(Fig 6b) However the R2 for LE is still improved by 14 for MEBL compared to ISBA (Table 4) which is related tothe improved sensible heat flux simulation (see Sect 513for more details) The RMSE scores are not very different be-tween the three simulations although the results are slightlyimproved with MEB and MEBL (the RMSE is 7 lowerwith MEB and 10 with MEBL compared to ISBA) Themain difference between the three simulations for this site isthe ground evaporation It is very small for ISBA but this islargely due to the class-dependent prescription of a vegeta-tion fraction veg of 099 (Table 2) This severely limits thebare-soil evaporation for this particular class This high valuewas defined for this class within ISBA to avoid excessivebare-soil evaporation in the composite scheme (amounting toa tuning parameter) resulting from the use of an aggregatedsurface roughness (resulting in a relatively large roughnessfor the ground) and to model in a very simplistic way theshading effect For MEB the veg parameter does not existand instead the partitioning between Eg and Ev is controlledmore by physical processes (largely by the annual cycle ofLAI) The MEB Eg turns out to be significantly larger thanthat for ISBA for this site but MEBL produces an Eg whichis closer to that of ISBA

The two other sites have a higher annual evapotranspira-tion than the Mediterranean site and a larger ground evap-oration especially for ISBA At Le Bray evapotranspira-tion is almost the same for the three simulations (Fig 4a)with a slight underestimation between September and April(Fig 7b) Ground evaporation was measured at this site from14 March to 6 April in 1998 Lamaud et al (2001) leadingto an estimate of the cumulative evaporation from the groundof approximately 72 kg mminus2 during this period For each ofthe three available years of this study (2006ndash2008) MEBLsimulates an evaporation from the litter of around 65 kg mminus2

during this period whereas ISBA evaporates 16 kg mminus2 fromthe ground This shows the ability of the litter model to limitground evaporation during this period

At Barbeau the interpretation of the results is much morecomplex because of the deciduous broadleaf nature of theforest In winter the LAI is very low and latent heat flux isdominated by the ground evaporation In spring when net ra-diation increases but vegetation is not fully developed MEBsignificantly overestimates LE owing to a large ground evap-oration (Fig 8b) in contrast to MEBL which limits surfaceevaporation to 15 times less owing to the litter layer In sum-mer the ISBA simulation appears to better simulate LE (interms of total amount) However taking a closer look at thepartitioning of evapotranspiration reveals that transpirationand evaporation from the canopy reservoir are almost thesame but ground evaporation is very different compared to



Figure 4 The partitioning of latent heat flux for each site and model option into transpiration (black) groundlitter evaporation (white) andevaporation from the canopy (gray) Values for Le Bray (2006) Puechabon (2006) and Barbeau (2013) are shown in panels (a) (b) and (c)respectively

MEB and MEBL During this period both MEB and MEBLsimulations simulate considerably less ground evaporation(about 4 of summer LE) compared to the standard ISBAsimulation which simulates 25 of summer LE as groundevaporation This value is certainly overestimated since atthis time of the year very little energy is expected to reachthe ground (the LAI is around 5 m2 mminus2) For example dur-ing July ISBA and MEBL evaporate 334 and 45 kg mminus2 di-rectly from the ground respectively whereas the total energyreaching the soil is only equivalent to a maximum possibleevaporation of 81 kg mminus2 based on PAR measurements Sec-ond in the composite ISBA scheme the roughness lengthz0 used for ground evaporation is the same as that for tran-spiration and it is computed using an inverse-log averag-ing between soil and vegetation values weighted by the frac-tion cover Noilhan and Lacarrere (1995) The forest cover inISBA uses a fraction cover of 095 thus the effective surfaceroughness is dominated by the vegetation roughness and isnearly 2 orders of magnitude larger than a typical soil rough-ness (001 m which is close to the MEB value) leading to arelatively low aerodynamic resistance Thus the ISBA modelsimulates a reasonable LE which is comparable to MEB ow-ing to a weak transpiration compensating an excessive bare-soil evaporation This is one of the known biases which MEBwas intended to reduce

513 Sensible heat flux

For all of the sites more significant differences occur for thesimulated H compared to LE with the most significant dif-ferences occurring with MEBL (see Table 4) At Le Bray andPueacutechabon the observed H is in better agreement betweenthe MEBL and MEB runs with an average RMSE over thesetwo sites of 677 W mminus2 with the ISBA 495 W mminus2 withMEB and 48 W mminus2 for MEBL In addition the R2 are im-proved with average of 082 for ISBA 089 for MEB and090 for MEBL At Barbeau during summer the MEB andMEBL simulations constantly overestimate H As shownin the previous section LE is underestimated in particularfor MEB and MEBL As net radiation and ground heat flux

(Sect 514) are well simulatedH tends to be overestimatedISBA produces fairly good results despite overestimations ofthe ground heat flux and ground evaporation

The improvement for the two MEB simulations are mainlyrelated to three processes

ndash The use of an explicit canopy layer which interceptsmost of the downward solar radiation thereby reducingthe net radiation at the ground surface This leaves moreenergy available for turbulent fluxes in contrast to ISBAwhich is directly connected to forest floor and can moreeasily propagate energy into the ground by conduction

ndash The use of a lower roughness length ratio (momentumto heat z0z0h) which leads to an increase in the rough-ness length for heat and water vapor and therefore theturbulent fluxes

ndash The presence of a litter layer limits the penetration ofenergy into the ground because of its low thermal dif-fusivity which leads to a reduction of the ground heatflux (amplitude) and thus to more available energy forH and LE

514 Ground heat flux

A substantial effect of both MEB and MEBL is the groundheat flux reduction The explicit representation of the canopyinduces a shading effect due to the leaves stems andbranches so that less energy reaches the ground But in ad-dition the explicit representation of the litter layer also mod-ifies the ground heat flux by acting as a buffer for the top soillayer due to its low thermal diffusivity It can be seen in Fig 9that for each site MEB and MEBL had improved simulationsof G in terms of amplitude and phase with regards to ISBAThe mean AE of the three experiments remains very low (Ta-ble 4) even for the ISBA model This demonstrates that thetwo MEB experiments reduce the amplitude and improve thephasing of this flux while its daily average was reasonablywell represented with the standard ISBA model The meanRMSE averaged over all years and sites is 117 W mminus2 with



Figure 5 The total soil water content calculated over the root depthindicated in Table 2 (left panels a c and e) and the near-surfacevolumetric soil water content (right panels b d and f) at each siteObservations are in black (for Puechabon the black curve corre-sponds to the output from a site-specific calibrated reference modelfrom Lempereur et al (2015) see text) Results for MEBL are inred for MEB in blue and for ISBA in green

the explicit litter 243 W mminus2 without it and 471 W mminus2 us-ing ISBA (the scores are summarized in Table 4)

The ISBA overestimation of the flux amplitude often rep-resents several 10rsquos of W mminus2 during both daytime and night-time (Figs 7c 6c 8c) which reduces the amount of energyavailable for turbulent fluxes For all sites improvements ofthe ground heat flux are the same from one season to anotherwhich means that using a constant thickness for the litterlayer is likely adequate The MEBL overestimates the reduc-tion of G from June to September only for the Le Bray site(Fig 7c) However the presence of a very dense understoryat this site (accounting for half of the LAI over this period)makes the radiative transfer modeling much more compli-cated and could potentially lead to an underestimation of theincoming radiation at the forest floor The understory con-tribution to the shortwave fluxes is simply represented usinga bulk overstoryndashunderstory aggregated LAI In contrast tothe other two sites the effect of shading and litter at Pueacutech-abon are not sufficient to reduce the G significantly (to bemore in line with observed values) no matter what the sea-son The parameters of the radiative transfer scheme havebeen calibrated through the modification of the clumping in-dex so that the energy passing through the canopy is wellmodeled (Fig 3b) Sensitivity tests have been done for whichthe ground albedo and soil texture were modified over realis-tic ranges but this did not resolve the issue and so the over-estimatedG is likely due to other causes (such as the need tobetter adjust the turbulent transfer parameters to better corre-spond to the vegetation at this site) At any rate the MEBL

G is still superior to that simulated using MEB and ISBAfor this site Finally at Barbeau the MEBL simulation pro-duces good results for G (Fig 8c) except for a short periodin winter when the shortwave radiation transmission is alsooverestimated (see Fig 3c)

515 Soil temperature

A good description of the soil thermal characteristics isneeded to model temperatures at different depths along witha correspondingly good estimate of the surface-soil heat fluxThe soil characteristics (thermal conductivity and heat ca-pacity) are calculated based on the input soil texture (sandand clay fractions) and organic matter contained in the soilThe soil temperature simulation statistics in Table 5 werecomputed for the three model configurations using availablemeasurements at each site MEBL and MEB have consid-erably lower RMSE of soil temperatures at different depthsfor each site compared to ISBA The RMSE of the first mea-sured depth of each site (between 004 and 010 m) is 10 to20 lower for MEB compared to the ISBA model and 20to 60 lower between MEBL and ISBA Moreover R2 isalso improved with an average value for all of the sites of096 for MEBL 094 with MEB and 090 for ISBA Thegood agreement between simulated and measured tempera-tures and the improvements with MEB are consistent withthe improvement of the ground heat flux shown in the pre-vious section (Fig 9) Note however that the improvementof the simulated temperature at Pueacutechabon is not as signifi-cant as for the other sites This was expected since as shownin the previous section theG was overestimated for this sitethe RMSE is reduced with the MEB and MEBL experimentsbut it remains high (38 and 33 K respectively) In generalthere are relatively low AE values for each site and soil depth(AElt 13 K) but a seasonally varying AE remains The gen-eral effect of MEB is to reduce the mean annual tempera-ture by 1 to 2 K (resulting in a shift in the annual cycle)while MEBL dampens the annual cycle amplitude Indeedthe litter reduces both the energy loss during the cold seasonand gains in heat during the summer season The compositemonthly average diurnal cycles of the uppermost availablemeasured temperature for each site is shown in Fig 10 Therelatively low AE calculated over the entire year (Table 5)masks a negative AE during the cold season (November toMarch) and a positive AE during the warm season (May toSeptember) At all sites both MEB simulations improve themean daily values during the warm season but only MEBLreduces the negative bias that occurs during the cold seasonSo when looking at the composite diurnal cycles it is obvi-ous that the use of MEBL verses MEB improves the perfor-mance in terms of amplitude and phase all year long even if arelatively small bias remains The ISBA model daytime over-estimation can be as high as 10 K) whereas with the MEBLmodel it does not exceed 3 K



Figure 6 The monthly diurnal cycle composite at Puechabon MEBL is in red MEB in blue ISBA in green measurements are indicatedby a dashed black line and adjusted measurements are represented using a solid black curve As a visual aid the area between the latter twocurves is shaded Ideally model results fall within this area

Figure 7 As in Fig 6 except for Le Bray

516 Soil moisture

The observed and simulated soil water content time seriesfor an annual cycle are displayed in Fig 5 Soil volumet-ric water content measures are available at 30 minute timesteps at Le Bray and Barbeau At Pueacutechabon no half-hourlymeasurements were available but a reference curve whichcorresponds to a model-derived interpolation of discrete soilwater storage measurement (Lempereur et al 2015) has

been added to the figure A fairly good overall agreementbetween simulated and observed total soil water content isfound Moreover the three experiments lead to very similarresults which is consistent with the fact that evapotranspi-ration has been found to be the least impacted flux betweenISBA MEB and MEBL

More significant differences can be seen in terms of the topsoil water content In particular for the Le Bray and Barbeau



Figure 8 As in Fig 6 except for Barbeau

Figure 9 The Taylor diagram for ground heat fluxG at Le Bray (a) Puechabon (b) and Barbeau (c) MEBL is represented by a circle MEBby a square ISBA by a triangle and measurements are indicated using a star Three scores are represented in each diagram the correlationcoefficient (R) is the angle of the polar plot the normalized standard deviation (NSD) is the radial distance from the origin and the normalizedcentered root mean square error (cRMSE) is proportional to the distance with the reference (star)

forests at the end and beginning of the year respectivelyISBA and MEB simulate unrealistic drops in the liquid soilwater content (Fig 5b and f) These drops are caused by theunderestimation of the soil temperature during these periodswhich makes the water freeze These drops are not presentin the MEBL simulation In fact the near-surface soil tem-perature stays above 0 C owing to an insulating effect fromthe litter so that the water remains liquid as seen in the mea-surements At Barbeau all the three model approaches un-derestimate the total soil moisture during the first two thirdsof the year (Fig 5e) which is mainly related to the presenceof a water table which is not explicitly modeled MEB andMEBL are more sensitive than ISBA to this process sincethey simulate more transpiration than ISBA (although MEBL

is slightly wetter since Eg is always less in MEBL comparedto MEB)

52 Sensitivity tests

There is uncertainty with respect to the definition of severalkey model parameters which are not usually available in theobservations so that several sensitivity tests have been un-dertaken Three parameters have been tested (i) the extinc-tion coefficient for the long-wave transmission through thecanopy (Boone et al 2017) (ii) the leaf geometry whichmodulates the solar radiation transmitted through the canopy(Carrer et al 2013) and (iii) the litter thickness which con-trols the impact of litter on both the hydrological and thermalregimes (see Appendix A)



Table 5 The root mean square error RMSE square correlation coefficient R2 and annual error AE computed with available soil tempera-tures of each site for the three different experiments

Site Depth MEBL MEB ISBA

(m) RMSE R2 AE RMSE R2 AE RMSE R2 AE

Bray 004 19 098 minus10 29 096 minus13 36 093 01032 23 097 minus09 28 097 minus12 30 096 02

10 30 094 minus11 35 091 minus12 38 089 02

Puechabon 010 33 091 minus09 38 092 minus12 42 089 09

Barbeau 004 14 098 minus04 28 093 minus09 39 089 05008 14 098 minus04 27 093 minus09 37 091 05016 14 097 minus05 25 094 minus09 32 092 04032 13 098 minus06 22 095 minus10 27 094 03

For all three of the sites the long-wave transmission coef-ficient has a very weak influence on each of the simulationsfor a reasonable range of values (03 to 07 05 being the de-fault value which is based on an value used by Verseghy et al(1993) The RMSE calculated with observed and simulatedtotal long wave up radiation varies less than 2 W mminus2 overthis range for the 3 sites on an annual basis

The litter layer thickness values tested range between 001and 010 m based on values from the literature (Table 3) TheMEBL default values based on recommendations from thesite principal investigators are shown in Table 1 The G isthe most sensitive to these changes and RMSE values forthese tests are shown on Fig 11 for Le Bray and BarbeauNote that these tests were not done for Pueacutechabon since Gwas overestimated despite the litter layer (which implies thatoptimizing litter thickness for these site to improve fluxeswould likely result as a compensating error related to anotherprocess) Minimum errors values are reached for thicknessesabove approximately 003 m which is consistent with the rec-ommended site values and those from several other studies(see Table 3) Note that these thickness values also producethe lowest H and LE RMSE (not shown)

The clumping index was tested since it is the key pa-rameter of the radiative transfer controlling the transmis-sion and absorption of incoming shortwave radiation throughthe canopy (Carrer et al 2013) Three configurations weretested for the MEBL configuration (i) the default value (ii) aclumping index increased by 50 (a more closed canopy)and (iii) an index decreased by 50 This parameter al-ters the computation of the upwelling shortwave radiationand the values of the turbulent and ground heat flux sinceit changes the partitioning of available energy between thecanopy and the litter layer However the H and LE RMSEonly changed by 3 to 8 depending on site thus for now thedefault clumping index values are used

In summary owing to these sensitivity tests the defaultextinction coefficient for long-wave transmission is retainedas a constant value of 05 A default constant value of litter

thickness is defined as 003 m since it is both not widely ob-served and it tends to be a threshold for which the effect oflitter onG becomes notable and positive (and errors begin toslowly increase slightly above this threshold) for several con-trasting sites One could even define an slightly larger valuebased on Fig 11 (004 m for example) but a first-order ap-proximation is required as opposed to a tuned value basedon just two sites Also the layer should be relatively thinto insure a robust diurnal cycle (as opposed to using valueson the high side from these tests or based on the literature)The default clumping index used in the shortwave radiativetransfer led to an overestimation of the below-canopy radia-tion for the two sites where measurements of below-canopyPAR were available and slight improvements were found byadjusting this parameter (to values still within itrsquos realisticrange) But in fact the impact of tuning the clumping in-dex was found to have only a relatively small impact on thesimulated turbulent heat fluxes with the MEBL simulation sothat the default value is considered to be robust enough fornow As a final note if measured values are available or canbe readily determined for the long-wave transmission coef-ficient litter thickness and shortwave radiation transmission(for example at local-scale well-instrumented sites for stud-ies oriented towards detailed process analysis) they can beused to replace the aforementioned default values

53 Worldwide skill score assessment

The comparison of modeling results with field measurementsfrom over 42 FLUXNET sites (Fig 2) reveals that the resultsfrom the benchmark are consistent with those from the threelocal French sites For improved clarity only the MEBL andthe ISBA models are compared here since the previous eval-uation showed the consistent improvement of using the litteroption when using MEB for forests

MEBL and ISBA generally performed well in simulatingtheRn but both tended to slightly underestimate daytime val-ues with a mean bias of 10 W mminus2 corresponding to approxi-



Figure 10 Monthly average soil temperature (K) diurnal cycle (at004 m soil depth) composites at Le Bray (a) Puechabon (b) andBarbeau (c) MEBL is in red MEB in blue ISBA in green and theobservations are shown in black

mately 9 of the average net radiation Note that the Rn wasbetter simulated for the three local-scale sites (Sect 51) butthis results in part because the benchmark uses values fromthe ECOCLIMAP database (which likely introduces some er-ror relative to using observed values of albedo for example)Scatter plot of the statistical metrics (RMSE R2 and AE) ob-tained for the sensible and the latent heat fluxes are shown inFigs 12 and 13 respectively Each black dot corresponds to

Figure 11 The G RMSE computed at Le Bray and Barbeau fordifferent values of the litter thickness

the average score over all available years for a given site Theshaded area corresponds to the sites and years for which theMEBL model outperforms the ISBA model and the dashedlines represent the average over all sites Finally the percent-age of sites for which MEBL was superior to ISBA for aparticular score is given in the title of each panel within theaforementioned figures

The sensible heat flux AE (Fig 12c) is quite similar forthe two model approaches as evidenced by the relatively lowscatter (most of the values are withinminus20 to 20 W mminus2) Thisis consistent with previous conclusions for the detailed local-scale run analysis (Table 4) and confirms that the influence ofthe multi-source model is more significant for diurnal cyclesthan for daily average results However a very significantimprovement of the sensible heat flux R2 is observed andfor 935 of the sites for MEBL This shows the ability ofMEBL as demonstrated in Sect 51 to better represent thephase of turbulent fluxes owing to the explicit representationof the canopy layer and especially the more realistic lowervalue of the heat capacity of vegetation with respect to ISBA

The amplitude of the sensible heat flux diurnal cycle isalso improved as suggested by the improved RMSE values(Fig 12a) over 905 of the sites This better amplitudeperformance is related to both the canopy shading and lit-ter insulation effects which lower the diurnal amplitude ofthe ground heat flux and provide more energy for the tur-bulent fluxes The recent so called PLUMBER experimentBest et al (2015) found that for H most of the LSMs theystudied (including ISBA) were not able to beat the one vari-able nonlinear regression that uses instantaneous downwardshortwave radiation and observed surface fluxes The resultsin this study present a possible reason for this (since most ofthe LSMs do not include an explicit litter layer) therefore werecommend that the ground heat flux should be better inves-tigated and could be contributing to the generally poor per-



Figure 12 Scatter plots of RMSE R2 and AE for sensible heat flux (H ) for each site and year The abscissa corresponds to the ISBAmodel and the ordinate to the MEBL model Points falling within the gray shaded area imply a better statistical score for MEBL Each dotcorresponds to one site The intersection of the dashed lines corresponds to the average over all 42 sites The percentages at the top of eachpanel correspond to the percent of sites that have better statistics for MEBL

formance of simulating H among LSMs for the FLUXNETsites

The latent heat flux LE results between the two mod-els are more similar than for H Based on the results fromSect 51 this was expected since the vapor flux is mainlycontrolled by the stomatal resistance parameterization andsoil hydrology (which are the same for MEBL and ISBA)Fig 13a shows that the differences in RMSE between MEBLand ISBA are relatively small between the two simulationsHowever an improvement in the R2 coefficient is obtained(Fig 13b) with 733 of the sites showing an improvementwith MEBL Note that a negative AE is obtained for mostof the sites This result is also consistent with the local eval-uation over the three French sites (Table 4 and as shown inFig 4) Since both the ISBA and MEBL approaches havesimilar errors in this respect it is assumed that this is likelycaused by an aspect of ISBA which is common among thetwo models such as root-zone water uptake hydrology orstomatal resistance for example

In a general a very good consistency is found betweenthe analysis in Sect 51 for highly instrumented sites and themore general benchmark application in this section This isa positive result for a model which is designed for weatherforecast and climate simulations at a global scale

As mentioned in Sect 51 significant improvement is ob-tained with MEB compared to ISBA and even more im-provement with MEBL An example of the improvement inH and LE between MEB and MEBL is shown on Fig 14 Butbecause of the consistently best behavior with MEBL versesISBA compared to MEB MEBL has become the default op-tion for forests

Finally it is of interest to determine if the results are condi-tioned by certain key physiographic parameters notably theLAI The differences in RMSE between MEBL and ISBAare shown in Fig 15 for different ranges of LAI as box plotsEach box is plotted using the 10 25 50 75 and 90 per-centiles of the normalized difference in RMSE calculated

over 1-month periods for each site that satisfies the closurecondition adopted for this study (ie less than 20 im-balance) In Fig 15a the improvement of RMSE is signif-icant for forested sites with a relatively low LAI (LAIlt 2)which represents 19 of the cases moderate for mediumLAI (2ltLAIlt 4) 46 of the cases and weak for high LAI(LAIgt 4) 35 of the cases This shows that the gains ofthe multi-source energy budget are more significant when thecanopy is relatively sparse or open (corresponding to a rela-tively low LAI) so that the surface fluxes have significantcontributions from both the ground and the canopy Whenthere is a medium LAI range the same conclusion applies toa lesser extent Thus as LAI becomes large the MEBL andISBA results converge (since the fluxes are dominated by thecanopy) Figure 15 is consistent with previous results in thatthe simulated LE is similar between MEBL and ISBA so thatRMSE differences are low on average

6 Conclusions

This study is the second of a set of two papers which describethe introduction of the new multi-energy balance (MEB)approach within the interactions between soilndashbiospherendashatmosphere (ISBA) model as part of the SURFEX platformTwo new explicit bulk layers have been implemented onefor the vegetation canopy and the other for a litter layer Thispaper describes a two-part local-scale offline evaluation ofboth the bulk canopy scheme (MEB) and the combined bulkcanopyndashlitter layer (MEBL) approaches and the results arealso compared to the standard composite vegetation ISBAmodel The model parameterization governing the litter layeris also presented The evaluation is done by investigating theability of the models to simulate the fluxes above (sensibleand latent heat flux) and below (ground heat flux) differentforest canopies and the ability of the different approaches toreproduce the observed soil temperatures and soil water con-tent



Figure 13 As in Fig 12 except for latent heat flux (LE)

In the first part of this study an evaluation over three well-instrumented forested sites in France was done using ob-served forest characteristics and turbulent radiative and heatconduction (ground flux) measurements with a particular at-tention paid to the energy balance closure issue The mid-latitude forest sites were contrasting in terms of both vegeta-tion type (needleleaf broadleaf) and climate (Mediterraneantemperate) In the second part of this study a statistical evalu-ation was done using the framework of a benchmark platformbased upon 42 sites scattered throughout the world from theFLUXNET network

In terms of the model evaluation for the three Frenchforested sites the standard ISBA model was found to un-derestimate the amplitude of the sensible heat flux H andoverestimate that of the ground heat flux G the latter ofwhich is in agreement with an overestimation of the soil tem-perature amplitude Also latent heat flux is generally wellsimulated despite being slightly underestimated The simu-lation with the two new explicit layers greatly reduced theG and soil temperature errors owing mainly to shading ef-fect of the canopy layer and the low thermal diffusivity ofthe litter layer The relatively low thermal conductivity of thelitter greatly reduced the ground heat flux and soil tempera-ture amplitudes in addition to modifying their phase in bet-ter agreement with the measurements the RMSE decreasedfrom 471 to 100 W mminus2 on average As a result more en-ergy was available for the turbulent heat fluxes Since latentheat flux was generally water limited most of this excessenergy went into H the average RMSE for the three well-instrumented sites decreased from 621 to 50 W mminus2 withthe new parameterizations This result might have impor-tance beyond those for ISBA since the results of the recentPLUMBER (Best et al 2015) LSM inter-comparison andevaluation experiment highlighted the general poor perfor-mance of LSM-estimated H (including ISBA) compared toa subset of the FLUXNET observations Since net radiationand latent heat fluxes were generally reasonably well simu-lated this implies that one of the main sources of error inH islikely related to the ground heat flux simulation based on theenergy budget equation This study showed that at least for

ISBA (using the MEBL option) the inclusion of litter lead tosignificantly improved ground conduction which resulted inbetter sensible heat fluxes

In terms of temporal dynamics the main differences in la-tent heat flux between ISBA and MEBL occur during springfor the deciduous forest site where the litter layer acts tosignificantly limit soil evaporation whereas ISBA and MEB(without explicit litter) overestimate evapotranspiration dueto strong ground evaporation (owing to a relatively low LAIcombined with large incoming radiation and generally lowto unstressed soil conditions) And despite the overall sim-ilar total annual evapotranspiration simulated by ISBA andMEBL the partitioning between canopy evapotranspirationEv and ground evaporationEg is more realistic with MEBLthan ISBA This is mainly related to the conceptual designof the composite vegetation which uses a single aggregatedroughness length for both vegetation and soil (which leadsto a slight underestimate of canopy roughness but a signif-icant overestimate of ground roughness and thus relativelylarge Eg compared to the explicit bulk canopy approach ofMEB) In addition the structure of MEB enables the modelto directly use the radiation transmitted through the canopywhen computing the ground surface energy budget Measure-ments from two of the three well-instrumented forest sites in-dicate that MEBL simulates the downwelling shortwave ra-diation relatively well and this sets a theoretical limit on theamount of energy available for the turbulent fluxes at the sur-face (which again is often lower than the energy availablefor Eg in the composite ISBA scheme) The more physicallybased vegetation canopy in MEB also implies that there is nolonger a dependence on the ISBA parameter veg (vegetationcover fraction) which was constant for forests and tuned (tovalues between 095 and 099) to prevent excessive Eg

The main conclusions made from analysis of the threewell-instrumented sites were found to be consistent with theresults of a statistical benchmark analysis over a subset of42 forested sites from the FLUXNET network The sensi-ble heat flux RMSE was improved for 875 of these siteswith the new parameterizations (MEBL) The selected sitesencompassed a wide range of climate and several different



Figure 14 Scatter plots of RMSE for sensible heat flux H (toppanel) and latent heat flux LE (bottom panel) for each site and yearThe abscissa corresponds to the MEB model and the ordinate to theMEBL model Points falling within the gray shaded area imply abetter statistical score for MEBL Each dot corresponds to one siteThe intersection of the dashed lines corresponds to the average overall 42 sites The percentages at the top of each panel correspond tothe percent of sites that have better statistics for MEBL

forest land-cover types thus it is a necessary test before im-plementing MEBL in regional to global-scale applicationsThe benchmark also showed that the impact of the explicittreatment of the canopy and litter layer was more signifi-cant for relatively open canopies (low to medium LAI val-ues) whereas for closed canopies (high LAI) all three of themodel approaches simulation results converge This is notoverly surprising since in the limit as a canopy becomes talland quite dense the composite scheme resembles a vegeta-tion canopy (the soil contribution becomes significantly less)But again ISBA tends to simulate considerably more bare-soil evaporation in the peak growing season (maximum LAI)

Figure 15 Box plots of improvement in RMSE between MEBL andISBA Each RMSE value is computed with over a month-long pe-riod that satisfies the energy budget closure condition The sensibleheat flux (H ) is shown in panel (a) and the latent heat flux (LE) isin panel (b) The box plots contain 6 13 19 27 21 and 14 of allconsidered months (values) for ranges in LAI increasing from 0 to1 m2 mminus2 to 5ndash8 m2 mminus2 respectively

than MEBL so there is error compensation which is maskedto a large extent when looking at the total evapotranspiration

In terms of prospectives evaluation of MEBMEBL is on-going and offline spatially distributed applications within theSIM chain (SAFRAN-ISBA-MODCOU Habets et al 2008)and at the global scale are being tested and compared toISBA along with an examination of the improvements ob-tained in high latitudes (owing in part to a more detailed rep-resentation of snow interception processes) Also additionallocal-scale tests are being done for both tropical (since therewere relatively few such sites in FLUXNET) and semi-aridforests (owing to the impact of MEB for the latter as shownherein) Finally experiments are being prepared for fully



coupled landndashatmosphere simulation tests in both mesoscaleand climate model applications at Meacuteteacuteo-France

Code availability The MEB code is a part of the ISBA LSMand is available as open source via the surface modeling plat-form called SURFEX which can be downloaded at httpwwwcnrm-game-meteofrsurfex SURFEX is updated at a relativelylow frequency (every 3 to 6 months) and the developments pre-sented in this paper are available starting with SURFEX version80 If more frequent updates are needed or if what is required is notin Open-SURFEX (DrHOOK FALFI formats GAUSSIAN grid)you are invited to follow the procedure to get a SVN account andto access real-time modifications of the code (see the instructions atthe previous link)



Appendix A Description of the litter layer model

Forest litter is represented using a single model layer whichgenerally ranges in thickness from 001 to 010 m and in theabsence of ancillary data the default value is 003 m Whenthis option is active an additional layer is added to the soilfor the thermal and energy budget computations with litter-specific thermal properties This means that the numericalsolution method is identical to that presented in the com-panion paper by Boone et al (2017) In terms of hydrologyan additional reservoir is added which uses a relatively sim-ple bucket-type scheme with a litter-specific maximum waterstorage capacity The model physics and governing equationsare reviewed herein

A1 Prognostic equations

The energy budget for the snow-free litter layer can be ex-pressed as

ClpartTl

partt= Rnl minus Hl minus LEl minus Gg1 + Lf8l (A1)

where Tl is the litter temperature (K) 1zl (m) is the thick-ness of the litter layer and Cl (J Kminus1 mminus2) is the effectiveheat capacity of the litter Rnl Hl LEl Gg1 represent thenet radiation sensible heat flux latent heat flux and groundconduction flux from the litter layer respectively Note thatwhen litter is present Rnl Hl LEl correspond to the groundfluxes in Boone et al (2017) The additional terms added ow-ing to a snow cover are described in Boone et al (2017) andare therefore not repeated here

The liquid-water content of the litter layer evolves follow-ing

partWl

partt= PlminusElminusDlminus8l

(0ltWl ltWlmax

) (A2)

where Pl is the sum of the rates of the rainfall passing throughthe canopy canopy drip and ground-based snowmelt El rep-resents the litter evaporation rate Dl is the drainage ratefrom the litter to the soil (all in kg mminus2 sminus1) The maxi-mum liquid-water content in the litter reservoir is definedas Wlmax = wlmax1zl ρw (kg mminus2) The default value forthe maximum holding capacity of the litter layer wlmax is012 m3 mminus3 (Putuhena and Cordery 1996) All water in ex-cess of this maximum value is then partitioned between in-filtration and surface runoff by the ISBA soil hydrologicalmodel The liquid-water equivalent of ice contained in thelitter layer is governed by

partWlf

partt=8lminusElf (A3)

where Elf represents the sublimation of ice contained withinthe litter layer

A2 Phase change

The phase change rate 8l (kg mminus2 sminus1) is defined as

8l =1τice

δf min

[ρiCi1zl (Tlminus Tf)

Lf Wlf

]+ (1minus δf) min

[ρiCi1zl (Tfminus Tl)

Lf Wl

] (A4)

where Lf represents the latent heat of fusion (J kgminus1) ρi isthe density of ice (here defined as 920 kg mminus3) the freez-ing point temperature is Tf = 27315 K and Ci is the specificheat capacity of ice (2106times103 J Kminus1 kgminus1) The delta func-tion δf = 1 if energy is available for melting (ie TlminusTf gt 0)otherwise it is δf = 0 τice is a parameter which representsthe characteristic timescale for phase changes (Giard andBazile 2000) The updated temperature is first computedfrom Eq (A1) with 8l = 0 then the phase change is com-puted as an adjustment to Tl Wl and Wlf as in Boone et al(2000)

A3 Energy fluxes

It is assumed that litter below the canopy is spatially homo-geneous so that it intercepts all of the incoming radiationThus the net radiation Rnl for the litter layer is the same thatfor the first soil layer in the standard model

Rng = SWnet g+LWnet g (A5)

where SWnet g and LWnet g correspond to the net shortwaveand long-wave radiation (W mminus2) as in Boone et al (2017)Note that currently the soil emissivity and albedo values areused for the litter for spatially distributed simulations pend-ing the development of global datasets of these parametersfor litter or the development of an appropriate model to es-timate them For local-scale simulations the values can bedefined based on observations

The below-canopy sensible heat fluxHl (W mminus2) is com-puted the same way that for the top soil layer in the ISBAmodel as

Hl = ρa(Tlminus Tc)

Rag-c (A6)

where Rag-c is the aerodynamic resistance between theground and the canopy air space which is based on Choud-hury and Monteith (1988) Tl and Tc (J kgminus1) are thermo-dynamic variables which are linearly related to temperature(Boone et al 2017) The latent heat flux is partitioned be-tween evaporation and sublimation in the litter layer

LEl = (1minusplf) LEl + plf LElf (A7)



where plf is the fraction of frozen water in the litter layer and

LEl =Lv ρa

[hul qsat (Tl) minus qc

]Rag-c

(A8a)

LElf =Ls ρa

[hulf qsat (Tl) minus qc

]Rag-c

(A8b)

where the specific humidity of the canopy air space is repre-sented by qc The specific humidity at saturation over liquidwater is represented by qsat (kg kgminus1) Note it would be moreaccurate to use the specific humidity at saturation over ice inEq (A8b) but this complicates the linearization and this ef-fect is neglected for now (Boone et al 2017) The surfacehumidity factors for liquid and frozen water are representedby hul and hulf respectively They are computed as the rela-tive humidity following (Noilhan and Planton 1989)

hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)

Note that hulf is computed by replacingWl andWlmax by thevalues for the liquid-water equivalent ice content The max-imum liquid holding capacity is modified for ice following(Boone et al 2000)

Finally the ground conduction flux (W mminus2) between thelitter layer and the underlying soil is computed as

Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)

where λl and λg1 are the litter and first soil layer thermalconductivities respectively and 1zg1 is the thickness of thefirst soil layer

A4 Water interception and fluxes

The water intercepted by the litter layer corresponds to thesum of the rain passing through the canopy and the drip fromthe canopy

Pl = (1minus σv)Pr + Dc (A11)

Note that when snow is present melt from the snowpack isalso included in this term (Boone et al 2017) The fractionof the total rainfall Pr (kg mminus3 sminus1) intercepted by the canopyis modulated by σv which depends on the LAI of the canopyFinally the canopy drip is represented by Dc Further detailson canopy interception and drip are given in Boone et al(2017) The drainage from the litter is simply computed as inNoilhan and Mahfouf (1996)

Dl =max(0WlminusWlmax

)(A12)

and the amount ofDl which can infiltrate into the soil is lim-ited by Darcyrsquos law with any residual contributing to surfacerunoff Note that for simplicity a gravitational drainage-typeformulation is not used for litter but rather a tipping bucketfollowing Ogeacutee and Brunet (2002)

A5 Thermal properties

The litter thermal conductivity λl (W mminus1 Kminus1) is computedaccording to De Vries (1963) as

λl = 01+ 003(

Wl

ρw1zl

) (A13)

The effective heat capacity of the litter Cl (J mminus2 Kminus1) iscomputed using

Cl =1zl ρldCld + WlCw + WlfCi (A14)

where the specific heat capacity of liquid water is Cw =

4218times 103 (J Kminus1 kgminus1) The dry density of the litter is de-fined as ρld Ogeacutee and Brunet (2002) used a value of dry lit-ter density of 45 kg mminus3 for a pine forest Meekins and Mc-Carthy (2001) measured a litter density of 46 kg mminus3 in a de-ciduous forest and Kostel-Hughes et al (1998) estimated val-ues varying between 27 and 38 kg mminus3 for oak forests Cur-rently ECOCLIMAP does not distinguish between differenttypes of deciduous trees thus by default ρld is assigned avalue of 45 kg mminus3 As a proxy for the specific heat of litterwe use the specific heat capacity of organic material fromFarouki (1986) which is Cld = 1926times103 J kgminus1 Kminus1 Cur-rently constant values for ρld and Cld are used for spatiallydistributed applications or on the local scale unless observa-tional data are available

Appendix B Statistical scores

These three scores are defined respectively as

R =(xmminus xm)(xominus xo)

σmσoSD=

σm

σo

cRMSE=

radic[(xmminus xm)minus (xominus xo)]2

σm (B1)

where the overbar represents the average xm and xo are themodeled and observed datasets respectively and σ is thestandard deviation defined as

σ =

radic(xminus x)2 (B2)

The centered root mean square error (cRMSE) differencebetween MEBL and ISBA is defined as

1cRMSE =cRMSE(ISBA)minus cRMSE(MEBL)

min[cRMSE(ISBA) cRMSE(MEBL)]times 100 (B3)

where a positive value corresponds to an improvement usingMEBL



Competing interests The authors declare that they have no conflictof interest

Acknowledgements This work is a contribution to the ongoingefforts to improve the SURFEX platform The authors would liketo thank the teams of the three French eddy flux towers involved inthis study Le Bray Barbeau and Pueacutechabon for making their dataavailable and to all of the contributors to the FLUXNET databaseThis study was fully funded by a Meacuteteacuteo-France grant

Edited by S ValckeReviewed by two anonymous referees

References

Andrade J A V Abreu F M G D and Madeira M A V In-fluence of litter layer removal on the soil thermal regime of apine forest in a Mediterranean climate Rev Bras Ciecircnc Solo34 1481ndash1490 2010

Arunachalam A and Arunachalam K Influence of gap size andsoil properties on microbial biomass in a subtropical humid for-est of north-east India Plant Soil 223 187ndash195 2000

Baldocchi D Falge E Gu L Olson R Hollinger D RunningS Anthoni P Bernhofer C Davis K Evans R Fuentes JGoldstein A Katul G Law B Lee X Malhi Y MeyersT Munger W Oechel W Paw K Pilegaard K Schmid HValentini R Verma S Vesala T Wilson K and Wofsy SFLUXNET A new tool to study the temporal and spatial variabil-ity of ecosystem-scale carbon dioxide water vapor and energyflux densities B Am Meteorol Soc 82 2415ndash2434 2001

Best M Abramowitz G Johnson H Pitman A Balsamo GBoone A Cuntz M Decharme B Dirmeyer P Dong J EkM Guo Z Haverd V van den Hurk B Nearing G PakB Peters-Lidard C Santanello J Stevens L and VuichardN The plumbing of land surface models benchmarking modelperformance J Hydrometeorol 16 1425ndash1442 2015

Blyth E Gash J Lloyd A Pryor M Weedon G P and Shut-tleworth J Evaluating the JULES land surface model energyfluxes using FLUXNET data J Hydrometeorol 11 509ndash5192010

Boone A and Etchevers P An intercomparison of three snowschemes of varying complexity coupled to the same land-surfacemodel Local scale evaluation at an Alpine site J Hydrometeo-rol 2 374ndash394 2001

Boone A Calvet J and Noilhan J Inclusion of a third soil layerin a land surface scheme using the force-restore method J ApplMeteorol 38 1611ndash1630 1999

Boone A Masson V Meyers T and Noilhan J he influence ofthe inclusion of soil freezing on simulations by a soil-vegetation-atmosphere transfer scheme J Appl Meteorol 9 1544ndash15692000

Boone A Samuelsson P Gollvik S Napoly A Jarlan LBrun E and Decharme B The interactions between soilndashbiospherendashatmosphere land surface model with a multi-energybalance (ISBA-MEB) option in SURFEXv8 ndash Part 1 Model de-scription Geosci Model Dev 10 843ndash872 doi105194gmd-10-843-2017 2017

Bulcock H H and Jewitt G P W Modelling canopy and litter in-terception in commercial forest plantations in South Africa usingthe Variable Storage Gash model and idealised drying curvesHydrol Earth Syst Sci 16 4693ndash4705 doi105194hess-16-4693-2012 2012

Calvet J-C Noilhan J Roujean J-L Bessemoulin P Ca-belguenne M Olioso A and Wigneron J-P An interactivevegetation SVAT model tested against data from six contrastingsites Agr Forest Meteorol 92 73ndash95 1998

Carrer D Roujean J-L Lafont S Calvet J-C Boone ADecharme B Delire C and Gastellu-Etchegorry J-P Acanopy radiative transfer scheme with explicit FAPAR for the in-teractive vegetation model ISBA-A-gs Impact on carbon fluxesJ Geophys Res 188 888ndash903 2013

Choudhury B J and Monteith J L A four-layer model for theheat budget of homogeneous land surfaces Q J Roy MeteorSoc 114 373ndash398 1988

Clapp R and Hornberger G Empirical equations for some soilhydraulic properties Water Resour Res 14 601ndash604 1978

Decharme B and Douville H Introduction of a sub-grid hydrol-ogy in the ISBA land surface model Clim Dynam 26 65ndash782006

Decharme B Boone A Delire C and Noilhan J Local evalu-ation of the Interaction between Soil Biosphere Atmosphere soilmultilayer diffusion scheme using four pedotransfer functions JGeophys Res 116 doi1010292011JD016002 2011

Decharme B Brun E Boone A Delire C Le Moigne Pand Morin S Impacts of snow and organic soils parameteri-zation on northern Eurasian soil temperature profiles simulatedby the ISBA land surface model The Cryosphere 10 853ndash877doi105194tc-10-853-2016 2016

De Vries D A Thermal properties of soils Physics of plant envi-ronment 1963

Er-Raki S Chehbouni A Hoedjes J Ezzahar J Duchemin Band Jacob F Improvement of FAO-56 method for olive orchardsthrough sequential assimilation of thermal infrared-based esti-mates of ET Agr Water Manage 95 309ndash321 2008

Farouki T O (Ed) Thermal Properties of Soils vol 11 Clausthal-Zellerfeld Germany Trans Tech 1986

Faroux S Kaptueacute Tchuenteacute A T Roujean J-L Masson VMartin E and Le Moigne P ECOCLIMAP-IIEurope atwofold database of ecosystems and surface parameters at 1 kmresolution based on satellite information for use in land surfacemeteorological and climate models Geosci Model Dev 6 563ndash582 doi105194gmd-6-563-2013 2013

Gerrits A M J Savenije H H G Hoffmann L and Pfister LNew technique to measure forest floor interception ndash an applica-tion in a beech forest in Luxembourg Hydrol Earth Syst Sci11 695ndash701 doi105194hess-11-695-2007 2007

Giard D and Bazile E Implementation of a new assimilationscheme for soil and surface variables in a global NWP modelMon Weather Rev 128 997ndash1015 2000

Gibelin A-L Calvet J C Roujean J-L Jarlan L andLos S O Ability of the land surface model ISBA-A-gsto simulate leaf area index at the global scale Compari-son with satellites products J Geophys Res 111 d18102doi1010292005JD006691 2006

Gonzalez-sosa E Braud I Jean-Louis T Michel V Pierre Band Jean-Christophe C Modelling heat and water exchanges of



fallow land covered with plant-residue mulch Agr Forest Mete-orol 97 151ndash169 1999

Guevara-Escobar A Gonzalez-Sosa E Ramos-Salinas M andHernandez-Delgado G D Experimental analysis of drainageand water storage of litter layers Hydrol Earth Syst Sci 111703ndash1716 doi105194hess-11-1703-2007 2007

Habets F Boone A Champeaux J Etchevers P FranchisteguyL Leblois E Ledoux E Moigne P L Martin E MorelS Noilhan J Segui P Q Rousset-Regimbeau F and Vi-ennot P The SAFRAN-ISBA-MODCOU hydrometeorologi-cal model applied over France J Geophys Res 113 D06113doi1010292007JD008548 2008

Haverd V and Cuntz M SoilndashLitterndashIso A one-dimensionalmodel for coupled transport of heat water and stable isotopes insoil with a litter layer and root extraction J Hydrol 388 438ndash455 2010

Koivula M Punttila P Haila Y and Niemelauml J Leaf litter andthe small-scale distribution of carabid beetles (Coleoptera Cara-bidae) in the boreal forest Ecography 22 424ndash435 1999

Kostel-Hughes F Young T P and Carreiro M M Forest leaf lit-ter quantity and seedling occurrence along an urban-rural gradi-ent Urban Ecosys 2 263ndash278 doi101023A10095367068271998

Lamaud E Ogeacutee J Brunet Y and Berbigier P Validation ofeddy flux measurements above the understorey of a pine forestAgr Forest Meteorol 106 187ndash203 2001

Lempereur M Martin-StPaul N K Damesin C Joffre R Our-cival J-M Rocheteau A and Rambal S Growth durationis a better predictor of stem increment than carbon supply ina Mediterranean oak forest implications for assessing forestproductivity under climate change New Phytol 207 579ndash5902015

Marsh D and Pearman P Effects of habitat fragmentation on theabundance of two species of Leptodactylid frogs in an Andeanmontane forest Conserv Biol 11 1323ndash1328 1997

Masson V Le Moigne P Martin E Faroux S Alias AAlkama R Belamari S Barbu A Boone A Bouyssel FBrousseau P Brun E Calvet J-C Carrer D Decharme BDelire C Donier S Essaouini K Gibelin A-L Giordani HHabets F Jidane M Kerdraon G Kourzeneva E LafaysseM Lafont S Lebeaupin Brossier C Lemonsu A MahfoufJ-F Marguinaud P Mokhtari M Morin S Pigeon G Sal-gado R Seity Y Taillefer F Tanguy G Tulet P VincendonB Vionnet V and Voldoire A The SURFEXv72 land andocean surface platform for coupled or offline simulation of earthsurface variables and fluxes Geosci Model Dev 6 929ndash960doi105194gmd-6-929-2013 2013

Meekins J F and McCarthy B C Effect of environmen-tal variation on the invasive success of a nonindigenousforest herb Ecol Appl 11 1336ndash1348 doi1018901051-0761(2001)011[1336EOEVOT]20CO2 2001

Nachtergaele F and Batjes N Harmonized world soil databaseFAO Rome Italy 2012

Noilhan J and Mahfouf J-F The ISBA land surface parameteri-sation scheme Global Planet Change 13 145ndash159 1996

Noilhan J and Planton S A simple parameterization of land sur-face processes for meteorological models Mon Weather Rev117 536ndash549 1989

Noilhan J and Lacarrere P GCM grid-scale evaporation frommesoscale modeling J Climate 8 206ndash223 1995

Ogeacutee J and Brunet Y A forest floor model for heat and moistureincluding a litter layer J Hydrol 255 212ndash233 2002

Oliphant A Grimmond C Zutter H Schmid H Su H-BScott S Offerle B Randolph J and Ehman J Heat storageand energy balance fluxes for a temperate deciduous forest AgrForest Meteorol 126 185ndash201 2004

Park H-T Hattori S and Tanaka T Development of a numericalmodel for evaluating the effect of litter layer on evaporation JForest Res 3 25ndash33 1998

Preacutevost-Boureacute N C Ngao J Berveiller D Bonal D DamesinC Dufrecircne E Lata J-C Le Dantec V Longdoz B PontonS Soudani K and Epron D Root exclusion through trenchingdoes not affect the isotopic composition of soil CO2 efflux PlantSoil 319 1ndash13 2009

Putuhena W M and Cordery I Estimation of interception capac-ity of the forest floor J Hydrol 180 283ndash299 1996

Rambal S Lempereur M Limousin J M Martin-StPaul NK Ourcival J M and Rodriacuteguez-Calcerrada J How droughtseverity constrains gross primary production(GPP) and its parti-tioning among carbon pools in a Quercus ilex coppice Biogeo-sciences 11 6855ndash6869 doi105194bg-11-6855-2014 2014

Riha S J McInnes K Childs S and Campbell G A finite ele-ment calculation for determining thermal conductivity Soil SciSoc Am J 44 1323ndash1325 1980

Rutter N Essery R Pomeroy J Altimir N Andreadis KBaker I Barr A Bartlett P Boone A Deng H Douville HDutra E Elder K Ellis C Feng X Gelfan A GoodbodyA Gusev Y Gustafsson D Hellstroumlm R Hirabayashi Y Hi-rota T Jonas T Koren V Kuragina A Lettenmaier D LiW-P Luce C Martin E Nasonova O Pumpanen J PylesR D Samuelsson P Sandells M Schaumldler G ShmakinA Smirnova T G Staumlhli M Stoumlckli R Strasser U SuH Suzuki K Takata K Tanaka K Thompson E VesalaT Viterbo P Wiltshire A Xia K Xue Y and YamazakiT Evaluation of forest snow processes models (SnowMIP2) JGeophys Res 114 D06111 doi1010292008JD011063 2009

Sakaguchi K and Zeng X Effects of soil wetness plant litterand under-canopy atmospheric stability on ground evaporation inthe Community Land Model (CLM35) J Geophys Res 114D01107 doi1010292008JD010834 2009

Samuelsson P Jones C Willeacuten U Ullerstig A GollvikS Hansson U Jansson C Kjellstroumlm E Nikulin Gand Wyser K The Rossby Centre Regional Climate ModelRCA3 Model description and performance Tellus A 63 1ndash3doi101111j1600-0870201000478x 2011

Sato Y Kumagai T Kume A Otsuki K and Ogawa S Ex-perimental analysis of moisture dynamics of litter layers ndash theeffects of rainfall conditions and leaf shapes Hydrol Process18 3007ndash3018 2004

Schaap M and Bouten W Forest floor evaporation in a denseDouglas fir stand J Hydrol 193 97ndash113 1997

Seacutefeacuterian R Gehlen M Bopp L Resplandy L Orr J C MartiO Dunne J P Christian J R Doney S C Ilyina T Lind-say K Halloran P R Heinze C Segschneider J Tjiputra JAumont O and Romanou A Inconsistent strategies to spinup models in CMIP5 implications for ocean biogeochemical



model performance assessment Geosci Model Dev 9 1827ndash1851 doi105194gmd-9-1827-2016 2016

Seity Y Brousseau P Malardel S Hello G Beacutenard P BouttierF Lac C and Masson V The AROME-France convective-scale operational model Mon Weather Rev 139 976ndash9912011

Stoumlckli R Lawrence D Niu G-Y Oleson K Thornton PYang Z-L Bonan G Denning A and Running S Use ofFLUXNET in the Community Land Model development J Geo-phys Res 113 G01025 doi1010292007JG000562 2008

Twine T E Kustas W Norman J Cook D Houser P Mey-ers T Prueger J Starks P and Wesely M Correcting eddy-covariance flux underestimates over a grassland Agr Forest Me-teorol 103 279ndash300 2000

Ukkola A M Pitman A J Decker M De Kauwe M GAbramowitz G Kala J and Wang Y-P Modelling evapotran-spiration during precipitation deficits identifying critical pro-cesses in a land surface model Hydrol Earth Syst Sci 202403ndash2419 doi105194hess-20-2403-2016 2016

Verseghy D L McFarlane N A and Lazare M CLASS ndash ACanadian land surface scheme for GCMs II Vegetation modeland coupled runs Int J Climatol 13 347ndash370 1993

Voldoire A Sanchez-Gomez E y Meacutelia D S B DecharmeC C Seacuteneacutesi S Valcke S Beau I Alias A Cheval-lierM Deacutequeacute M Deshayes J Douville H E Fernandez G MMaisonnave E Moine M P Planton S Saint-Martin DS S Tyteca S Alkama R Belamari S A Braun L Cand Chauvin F The CNRM-CM51 global climate model de-scription and basic evaluation Clim Dynam 40 2091ndash2121doi101007s00382-011-1259-y 2013

Vorobeichik E On the methods for measuring forest litter thick-ness to diagnose the technogenic disturbances of ecosystemsRuss J Ecol 28 230ndash234 1997

Vuichard N and Papale D Filling the gaps in meteorological con-tinuous data measured at FLUXNET sites with ERA-Interim re-analysis Earth Syst Sci Data 7 157ndash171 doi105194essd-7-157-2015 2015

Wilson K Goldstein A Falge E Aubinet M Baldocchi DBerbigier P Bernhofer C Ceulemans R Dolman H FieldC Ibrom A Law B Kowalski A Meyers T Moncrieff JMonson R Oechel W Tenhunen J Valentini R and VermaS Energy balance closure at FLUXNET sites Agr Forest Me-teorol 113 223ndash243 2002

Wilson T Meyers T Kochendorfer J Anderson M and HeuerM The effect of soil surface litter residue on energy and carbonfluxes in a deciduous forest Agr Forest Meteorol 161 134ndash147 2012

Wu H Peng C Moore T R Hua D Li C Zhu Q Peichl MArain M A and Guo Z Modeling dissolved organic carbon intemperate forest soils TRIPLEX-DOC model development andvalidation Geosci Model Dev 7 867ndash881 doi105194gmd-7-867-2014 2014

Yang R and Friedl M A Determination of roughness lengths forheat and momentum over Boreal forests Bound-Lay Meteorol107 581ndash603 2003

Zheng D Van Der Velde R Su Z Booij M J Hoekstra A Yand Wen J Assessment of roughness length schemes imple-mented within the Noah land surface model for high-altitude re-gions J Hydrometeorol 15 921ndash937 2014

Zhu J Matsuzaki T Lee F and Gonda T Effect of gap size cre-ated by thinning on seedling emergency survival and establish-ment in a coastal pine forest Forest Ecol Manag 182 339ndash3542003


Abstract

Introduction

Model

The standard ISBA composite soil--vegetation model

The ISBA-MEB model

Explicit model for litter in ISBA-MEB

Data

Energy balance closure

Local-scale evaluation over France

Benchmark

Model setup

Local scale

Benchmark

Results and discussion

Evaluation for three well-instrumented forested sites

Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests

Worldwide skill score assessment

Conclusions

Code availability


Appendix A1 Prognostic equations

Appendix A2 Phase change

Appendix A3 Energy fluxes

Appendix A4 Water interception and fluxes

Appendix A5 Thermal properties


Competing interests

Acknowledgements

References


used to compute flux exchanges of certain chemical species(such as carbon dioxide and biogenic organic volatile car-bon) and emissions of particles (such as aerosols from dust orbiomass burning) The interactions between soilndashbiospherendashatmosphere (ISBA) is part of the SURFace EXternaliseacutee plat-form (SURFEX) developed in recent years at Meacuteteacuteo-France(Masson et al 2013) and a suite of international partnersIt is used for a suit of surface schemes in either offline orcoupled mode with an atmospheric model andor a hydro-logical model ISBA has benefited from continuous improve-ments since its first version (Noilhan and Planton 1989) inparticular for the carbon cycle (Calvet et al 1998 Gibelinet al 2006) soil mass and heat transfer (Boone et al 1999Decharme et al 2011) snowpack processes (Boone andEtchevers 2001 Decharme et al 2016) sub-grid hydrol-ogy (Decharme and Douville 2006) and radiative transferthrough the canopy (Carrer et al 2013)

In order to remain consistent with the aforementioned de-velopments and to respond to both current and future user de-mands a multi-energy balance (MEB) approach has been de-veloped (Boone et al 2017) This improvement is attributedto the representation of the surface as a multi-source model(ie a separation of the canopy and the surface-soil layer)in contrast to the standard ISBA soilndashvegetation compositemodel This new model has been designed to better representcertain processes for forested areas and to incorporate newmodeled processes which can benefit from a more accuraterepresentation of the soilndashvegetation continuum In particu-lar partitioning of the incoming energy into turbulent andground heat fluxes are expected to be more realistic with theexplicit vegetation scheme as well as the partitioning of la-tent heat flux into its different components The carbon cy-cle should also benefits from a more conceptually accuraterepresentation of the surface (Carrer et al 2013) since itis strongly linked to leaf temperature via assimilation of at-mosphere carbon Moreover snowmelt during spring is verysensitive to the snow net radiation budget which is impactedby the presence of tall vegetation and interception and lossby the canopy during the winter season (Rutter et al 2009)

The surface in forested regions beneath the canopy is oftencovered by a layer of dead leaves or needles branches fruitsand other organic material which can be characterized as alitter layer The explicit inclusion of such a layer is neglectedfor the most part in LSMs used in global-scale models or it isonly partly taken into account For example Sakaguchi andZeng (2009) implemented a specific resistance due to litterfor soil evaporation The inclusion of litter-related processeshas been shown to have a substantial impact on hydrolog-ical (Putuhena and Cordery 1996 Guevara-Escobar et al2007) and thermal processes (Andrade et al 2010) Litterhas an important insulating capacity as its thermal diffusiv-ity is small compared to soil (Riha et al 1980) which leadsto a strong impact on the soil temperature diurnal cycle andground conduction flux Changes in this flux can then impactthe energy available for radiative and turbulent fluxes The

presence of litter also alters ground evaporation by cover-ing the soil with a high porosity layer which essentially pre-vents liquid-water capillary rise (Schaap and Bouten 1997)It also constitutes another water interception reservoir forrainfall and canopy drip prior to infiltration and runoff (egGerrits et al 2007) Some models have introduced parame-terizations for litter (Ogeacutee and Brunet 2002 Gonzalez-sosaet al 1999 Wilson et al 2012 Haverd and Cuntz 2010Sakaguchi and Zeng 2009) but the approach can be verydifferent from one to another depending on their complexity(only modifying or adding a ground resistance modeling thelitter using an explicit single or multi-layer model) In addi-tion each of the models were validated against observationsmade for a single well-instrumented site

The goal of the current study is to evaluate the im-pact of a new parameterization of the soilndashlitterndashvegetationndashatmosphere continuum at the local scale In the first part ofthis study an in-depth evaluation for three forest sites inFrance is carried out They have been selected in order torepresent a range of forest types and climates The main goalis to understand the effect of both the explicit canopy layerand the litter layer on the available energy partitioning (la-tent sensible ground) soil temperatures and soil water con-tent The second part of this study consists of a benchmarkstudy using 42 sites from the FLUXNET network (Baldocchiet al 2001) The objective is 2-fold (1) to evaluate the per-formance of the MEB options against the classic compositesoilndashvegetation version of the ISBA model over a wide rangeof forest species and climates and (2) to analyze whetherthe general improvements seen at the three well-instrumentedand documented local-scale sites in France can be extendedto other climates and forest types

This is essential since ISBA is used within the SUR-FEX platform in various configurations at resolutions rang-ing from several kilometers at the regional scale such aswithin the operational mesoscale numerical weather pre-diction model AROME (Seity et al 2011) and the opera-tional distributed hydrological model system SIM (Habetset al 2008) to resolutions of hundreds of kilometers inglobal-scale models such as within the global climate mod-els CNRM-CM51 (Voldoire et al 2013) and CNRM-ESM1(Seacutefeacuterian et al 2016)

2 Model

21 The standard ISBA composite soilndashvegetationmodel

The standard ISBA model uses a single composite soilndashvegetation surface energy budget which means that the prop-erties of the soil and vegetation are aggregated within eachgrid-cell point (Noilhan and Planton 1989 Noilhan andMahfouf 1996) In this study the multi-layer soil diffusion(DF Decharme et al 2011) option is used along with the









3 Data



δ =H +LE

RnminusGminus S (1)


Hadj =H + restimesH

H +LE (2)


H +LE (3)



















33 Benchmark





4 Model setup



41 Local scale





42 Benchmark








Soil parameters




Litter parameters





511 Net radiation



































=










































516 Soil moisture


















10 30 094 minus11 35 091 minus12 38 089 02




























6 Conclusions



























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References








3 Data



δ =H +LE

RnminusGminus S (1)


Hadj =H + restimesH

H +LE (2)


H +LE (3)



















33 Benchmark





4 Model setup



41 Local scale





42 Benchmark








Soil parameters




Litter parameters





511 Net radiation



































=










































516 Soil moisture


















10 30 094 minus11 35 091 minus12 38 089 02




























6 Conclusions



























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References















33 Benchmark





4 Model setup



41 Local scale





42 Benchmark








Soil parameters




Litter parameters





511 Net radiation



































=










































516 Soil moisture


















10 30 094 minus11 35 091 minus12 38 089 02




























6 Conclusions



























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References



4 Model setup



41 Local scale





42 Benchmark








Soil parameters




Litter parameters





511 Net radiation



































=










































516 Soil moisture


















10 30 094 minus11 35 091 minus12 38 089 02




























6 Conclusions



























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References





Soil parameters




Litter parameters





511 Net radiation



































=










































516 Soil moisture


















10 30 094 minus11 35 091 minus12 38 089 02




























6 Conclusions



























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References

































=










































516 Soil moisture


















10 30 094 minus11 35 091 minus12 38 089 02




























6 Conclusions



























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References


































516 Soil moisture


















10 30 094 minus11 35 091 minus12 38 089 02




























6 Conclusions



























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References














10 30 094 minus11 35 091 minus12 38 089 02




























6 Conclusions



























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References

















6 Conclusions



























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References

























ClpartTl




partWl


(0ltWl ltWlmax

) (A2)


partWlf



A2 Phase change


8l =1τice

δf min


Lf Wlf

]+ (1minus δf) min


Lf Wl

] (A4)


A3 Energy fluxes






Rag-c (A6)






LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References



LEl =Lv ρa


]Rag-c

(A8a)

LElf =Ls ρa


]Rag-c

(A8b)


hul =12

[1minus cos

(π

Wl

Wlmax

)] (A9)



Gg1 =Tl minus Tg1

(1zlλl) +(1zg1λg1

) (A10)







)(A12)




λl = 01+ 003(

Wl

ρw1zl

) (A13)








σmσoSD=

σm

σo

cRMSE=


σm (B1)


σ =











References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References





References










































































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References
















































Abstract

Introduction

Model


The ISBA-MEB model


Data



Benchmark

Model setup

Local scale

Benchmark



Net radiation

Latent heat flux

Sensible heat flux

Ground heat flux

Soil temperature

Soil moisture

Sensitivity tests


Conclusions

Code availability








Competing interests

Acknowledgements

References

the interactions between soil-biosphere-atmosphere (isba...

Documents