Testing water ﬂuxes and storage from two hydrology ......to miscalculate partitioning of evapotranspiration (ET) into transpiration (T) and bare soil evaporation (E), with models

Hydrol Earth Syst Sci 24 5203ndash5230 2020httpsdoiorg105194hess-24-5203-2020copy Author(s) 2020 This work is distributed underthe Creative Commons Attribution 40 License

Testing water fluxes and storage from two hydrology configurationswithin the ORCHIDEE land surfacemodel across US semi-arid sitesNatasha MacBean1 Russell L Scott2 Joel A Biederman2 Catherine Ottleacute3 Nicolas Vuichard3 Agnegraves Ducharne4Thomas Kolb5 Sabina Dore6 Marcy Litvak7 and David J P Moore8

1Department of Geography Indiana University Bloomington IN 47405 USA2Southwest Watershed Research Center United States Agricultural Department Agricultural Research ServiceTucson AZ 85719 USA3Laboratoire des Sciences du Climat et de lrsquoEnvironnement LSCEIPSL CEA-CNRS-UVSQ Universiteigrave Paris-SaclayGif-sur-Yvette 91191 France4UMR METIS Sorbonne Universiteacute CNRS EPHE Paris 75005 France5School of Forestry Northern Arizona University Flagstaff AZ 86011 USA6Hydrofocus Inc Davis CA 95618 USA7Department of Biology University of New Mexico Albuquerque NM 87131 USA8School of Natural Resources and the Environment University of Arizona Tucson AZ 85721 USA

Correspondence Natasha MacBean (nlmacbeangmailcom)

Received 7 November 2019 ndash Discussion started 4 December 2019Revised 1 September 2020 ndash Accepted 24 September 2020 ndash Published 10 November 2020

Abstract Plant activity in semi-arid ecosystems is largelycontrolled by pulses of precipitation making them partic-ularly vulnerable to increased aridity that is expected withclimate change Simple bucket-model hydrology schemes inland surface models (LSMs) have had limited ability in ac-curately capturing semi-arid water stores and fluxes Recentmore complex LSM hydrology models have not been widelyevaluated against semi-arid ecosystem in situ data We hy-pothesize that the failure of older LSM versions to repre-sent evapotranspiration ET in arid lands is because simplebucket models do not capture realistic fluctuations in upper-layer soil moisture We therefore predict that including adiscretized soil hydrology scheme based on a mechanisticdescription of moisture diffusion will result in an improve-ment in model ET when compared to data because the tem-poral variability of upper-layer soil moisture content bettercorresponds to that of precipitation inputs To test this pre-diction we compared ORCHIDEE LSM simulations from(1) a simple conceptual 2-layer bucket scheme with fixed hy-draulic parameters and (2) an 11-layer discretized mechanis-tic scheme of moisture diffusion in unsaturated soil based onRichards equations against daily and monthly soil moisture

and ET observations together with data-derived estimatesof transpiration evapotranspiration TET ratios from sixsemi-arid grass shrub and forest sites in the south-westernUSA The 11-layer scheme also has modified calculations ofsurface runoff water limitation and resistance to bare soilevaporation E to be compatible with the more complex hy-drology configuration To diagnose remaining discrepanciesin the 11-layer model we tested two further configurations(i) the addition of a term that captures bare soil evaporationresistance to dry soil and (ii) reduced bare soil fractionalvegetation cover We found that the more mechanistic 11-layer model results in a better representation of the dailyand monthly ET observations We show that as predictedthis is because of improved simulation of soil moisture inthe upper layers of soil (top sim 10 cm) Some discrepanciesbetween observed and modelled soil moisture and ET mayallow us to prioritize future model development and the col-lection of additional data Biases in winter and spring soilmoisture at the forest sites could be explained by inaccuratesoil moisture data during periods of soil freezing andor un-derestimated snow forcing data Although ET is generallywell captured by the 11-layer model modelled TET ratios

Published by Copernicus Publications on behalf of the European Geosciences Union

5204 N MacBean et al Testing water fluxes and storage from two hydrology configurations

were generally lower than estimated values across all sitesparticularly during the monsoon season Adding a soil re-sistance term generally decreased simulated bare soil evapo-ration E and increased soil moisture content thus increas-ing transpiration T and reducing the negative bias betweenmodelled and estimated monsoon TET ratios This negativebias could also be accounted for at the low-elevation sites bydecreasing the model bare soil fraction thus increasing theamount of transpiring leaf area However adding the baresoil resistance term and decreasing the bare soil fraction bothdegraded the model fit to ET observations Furthermore re-maining discrepancies in the timing of the transition fromminimum TET ratios during the hot dry MayndashJune periodto high values at the start of the monsoon in JulyndashAugustmay also point towards incorrect modelling of leaf phenol-ogy and vegetation growth in response to monsoon rains Weconclude that a discretized soil hydrology scheme and associ-ated developments improve estimates of ET by allowing themodelled upper-layer soil moisture to more closely match thepulse precipitation dynamics of these semi-arid ecosystemshowever the partitioning of T from E is not solved by thismodification alone

1 Introduction

Semi-arid ecosystems ndash which cover sim 40 of the Earthrsquosterrestrial surface and which include rangelands shrublandsgrasslands savannas and seasonally dry forests ndash are inzones of transition between humid and arid climates and arecharacterized by sparse patchy vegetation cover and lim-ited water availability Moisture availability in these ecosys-tems is therefore a major control on the complex interac-tions between vegetation dynamics and surface energy wa-ter and carbon exchange (Biederman et al 2017 Haverdet al 2016) Given the sensitivity to water availability semi-arid ecosystem functioning may be particularly vulnerable toprojected changes in climate (Tietjen et al 2010 Maestreet al 2012 Gremer et al 2015) IPCC Earth system model(ESM) projections and observation-based datasets indicatethese regions will likely experience more intense warmingand droughts increases in extreme rainfall events and agreater contrast between wet and dry seasons in the future(IPCC 2013 Donat et al 2016 Sippel et al 2017 Huanget al 2017)

To simulate the impact of climate change on semi-aridecosystem functioning it is essential that the land sur-face model (LSM) component of ESMs accurately repre-sent semi-arid water flux and storage budgets (and all asso-ciated processes) In the last 2 to 3 decades LSM groupshave progressively updated their hydrology schemes fromthe more simplistic ldquobucketrdquo-type models included in earlierversions (Manabe 1969) The resulting schemes typically in-clude more physically based representations of vertical diffu-

sion of water in unsaturated soils (Clark et al 2015) In ad-dition to increasing the complexity of soil hydrology severalstudies have attempted to address the issue that models tendto miscalculate partitioning of evapotranspiration (ET) intotranspiration (T ) and bare soil evaporation (E) with modelssystematically underestimating TET ratios (Wei et al 2017Chang et al 2018) One such mechanism that models haveintroduced is an evaporation resistance term that reduces therate of water evaporation from bare soil surfaces (Swensonand Lawrence 2014 Decker et al 2017) The developmentof these more mechanistic soil hydrology schemes shouldmean that LSMs better capture high temporal frequency toseasonal and long-term temporal variability of water storesand fluxes However it is not always apparent that increasingmodel complexity provides more accurate representations ofreality (as encapsulated by observations of different vari-ables at multiple spatio-temporal scales) Further increasingmodel complexity comes at a cost of increased computationalresources and unknown parameters Therefore it is imper-ative that we test models of increasing complexity againstmultiple types of observations at a variety of sites represent-ing different ecosystem types

New-generation LSM water flux and storage estimateshave been extensively tested at multiple scales from the sitelevel to the globe (Abramowitz et al 2008 Dirmeyer 2011Guimberteau et al 2014 Mueller and Senerviratne 2014Best et al 2015 Ukkola et al 2016b Raoult et al 2018Scanlon et al 2018 2019) Modelndashdata biases are observedacross all biomes however a key finding common to thesestudies is that models do not capture seasonal to inter-annualwater stores and fluxes well during dry periods andor atdrier sites (Mueller and Senerviratne 2014 Swenson andLawrence 2014 De Kauwe et al 2015 Best et al 2015Ukkola et al 2016a Humphrey et al 2018 Scanlon et al2019) Mueller and Senerviratne (2014) showed that CMIP5models overestimated multiyear mean daily ET in many re-gions with the strongest bias in dryland regions (particu-larly western North America) Likewise Grippa et al (2011)and Scanlon et al (2019) demonstrated that LSMs underes-timate seasonal amplitude of total water storage in semi-arid(and tropical) regions However compared to more mesicecosystems semi-arid ecosystem LSM water flux and stor-age simulations have rarely been tested extensively againstin situ observations apart from a few exceptions (Hogueet al 2005 Abramowitz et al 2008 Whitley et al 2016Grippa et al 2017) Whitley et al (2016) compared carbonand water flux simulations from six LSMs at five OzFlux sa-vanna sites Their study highlighted two key deficiencies inmodelling water fluxes (i) modelled C4 grass T is too lowand (ii) models with shallow rooting depths typically under-estimate woody plant dry season ET As part of a modelinter-comparison for western Africa (the AMMA LSM In-tercomparison Project ndash ALMIP) LSM water storage fluxesrunoff and land surface temperature were evaluated againstin situ and remote sensing data in the Malian Gourma region

Hydrol Earth Syst Sci 24 5203ndash5230 2020 httpsdoiorg105194hess-24-5203-2020

N MacBean et al Testing water fluxes and storage from two hydrology configurations 5205

of the central Sahel (Boone et al 2009 De Kauwe et al2013 Lohou et al 2014 Grippa et al 2011 2017) Thesestudies highlight that temporal characteristics of water stor-age and fluxes in this monsoon-driven semi-arid region arecaptured fairly well by models however the studies alsopoint to various model issues including difficulties in sim-ulating bare soil evaporation response to rainfall events (Lo-hou et al 2014) underestimation of dry season ET (Grippaet al 2011) the need for greater water and energy exchangesensitivity to different vegetation types and soil characteris-tics (De Kauwe et al 2013 Lohou et al 2014 Grippa et at2017) and overestimation of surface runoff (Grippa et al2017) How models prescribe or predict leaf area index (LAI)has also been highlighted as a driver of hydrological modelndashdata differences (Ukkola et al 2016b Grippa et al 2017)

The aim of this study was to contribute a new LSM hydrol-ogy model evaluation in a semi-arid region not previouslyinvestigated the monsoon-driven semi-arid south-westernUnited States (hereafter the SW US) The density and diver-sity of research sites in the SW US provide a rare opportunityto test an LSM across a range of semi-arid ecosystems Thesemi-arid SW US has also been identified as one of the keyregions of global landndashatmosphere coupling (Koster et al2004) and the most persistent climate change hotspot in theUS (Diffenbaugh et al 2008 Allen 2016) Expected futuresoil moisture deficits in this region will result in strong at-mospheric feedbacks with consequent high temperature in-creases (Senerivatne et al 2013) and a potential weakeningof the terrestrial biosphere C sink (Berg et al 2016 Greenet al 2019) Several studies based on model predictionsinstrumental records and paleoclimatic data analyses havesuggested that over the coming century the risk of more se-vere multi-decadal drought in the SW US will increase con-siderably (Ault et al 2014 2016 Cook et al 2015) In factmodels suggest that a transition to drier conditions is alreadyunderway (Seager et al 2007 Archer and Predick 2008Seager and Vecchi 2010) Investigating how well LSMs cap-ture hydrological stores and fluxes in this region thereforeprovides a crucial test for how well models can produce ac-curate global climate change projections

Here we tested the ability of the ORCHIDEE (ORganiz-ing Carbon and Hydrology in Dynamic EcosystEms) LSMto simulate multiple water-flux- and storage-related vari-ables at six SW US semi-arid Ameriflux eddy covariancesites spanning forest and shrub- and grass-dominated ecosys-tems (Biederman et al 2017) We tested two versions of theORCHIDEE LSM with hydrological schemes of differingcomplexity (1) a simple 2-layer conceptual bucket scheme(hereafter 2LAY) with constant water-holding capacity (deRosnay and Polcher 1998) and (2) an 11-layer mechanisticscheme (hereafter 11LAY) based on the Richards equationwith hydraulic parameters based on soil texture (de Ros-nay et al 2002) Besides the change in the soil hydrol-ogy between the 2LAY and 11LAY versions several otherhydrology-related processes have also been modified due to

increases in the complexity of the model These modifica-tions are described further in Sect 22 and summarized inTable 2 The 2LAY scheme was used in the previous CMIP5runs whereas the 11LAY scheme is the default scheme inthe current version of ORCHIDEE that is used in the ongo-ing Coupled Model Intercomparison Project (CMIP6) simu-lations (Ducharne et al 2020)

Our analyses were organized as follows First we evalu-ated how changing from the conceptual 2LAY bucket modelto the physically based 11LAY soil hydrology scheme ndash andall associated modifications ndash has influenced the high tempo-ral frequency and seasonal variability of semi-arid ecosys-tem soil moisture ET (and its component fluxes) runoffdrainage and snow massmelt Although there have beenmany previous studies comparing simple bucket schemes vsmechanistic multi-layer hydrology we include such a com-parison in the first part of our analysis for the following rea-sons (a) the simple bucket schemes were the default hydrol-ogy in some CMIP5 model simulations and these simula-tions are still being widely used to understand ecosystem re-sponses to changes in climate (b) variations on the simplebucket schemes are still implemented by design in varioustypes of hydrological models (Bierkens et al 2015) (c) therehave not yet been extensive comparisons of these two typesof hydrology model for semi-arid regions and especially notfor the SW US and (d) so that the 2LAY scheme can serveas a benchmark for the 11LAY scheme Second we eval-uated the temporal dynamics of the 11LAY model againstobservations at three specific soil depths (shallow le 5 cmmid 15ndash20 cm deep ge 30 cm) to assess whether the physi-cally based discretized scheme accurately captures moisturetransport down the soil profile Note that when evaluatingthe 11LAY model soil moisture against observations our pri-mary focus was on the temporal dynamics ndash rather than theabsolute magnitude ndash given the difficulty of comparing ab-solute values of volumetric water content between the mod-els and the data (see Sect 232 for more details) Thereforein the modelndashdata comparison we scale the observations tothe 11LAY model simulations via linear CDF matching Fi-nally having evaluated the standard (default) 11LAY modelagainst in situ semi-arid water stores and fluxes a novelcomponent of our study was to investigate whether someof the site-scale semi-arid LSM hydrology model discrep-ancies outlined above (eg underestimation of C4 grass T weak dry season ET and therefore low TET ratios ET is-sues related to incorrect representation of leaf area and over-estimation of surface runoff) are improved with recent OR-CHIDEE hydrology model developments Where the modeldoes not capture observed patterns we investigated whichmodel processes or mechanisms in the 11LAY scheme mightbe responsible for remaining modelndashdata discrepancies Inparticular we assessed the impact of (a) decreasing the baresoil fraction (thus increasing leaf area) and (b) includingthe optional bare soil resistance term in the 11LAY scheme(Ducharne et al 2020) Given the sparsely vegetated nature

httpsdoiorg105194hess-24-5203-2020 Hydrol Earth Syst Sci 24 5203ndash5230 2020

of the low-elevation semi-arid grass- and shrub-dominatedsites in our study we hypothesized that inclusion of this termmay counter any dry season ET underestimate Throughoutwe explored whether there are any discernible differencesacross sites due to elevation and vegetation composition

Section 2 describes the sites data model and methodsused in this study Sect 3 details the results of the two-partmodel evaluation (as outlined above) and Sect 4 discusseshow future studies may resolve remaining model issues inorder to improve LSM hydrology modelling in semi-arid re-gions

2 Methods and data

21 South-western US study sites

We used six semi-arid sites in the SW US that spanned arange of vegetation types and elevations (Biederman et al2017) The entire SW US is within the North AmericanMonsoon region therefore these sites typically experiencemonsoon rainfall during July to October preceded by a hotdry period in May and June Table 1 describes the dom-inant vegetation species and soil texture characteristics ateach site together with the observation period The fourgrass- and shrub-dominated sites (US-SRG US-SRM US-Whs and US-Wkg) are located at low elevation (lt1600 m)in southern Arizona with mean annual temperatures be-tween 16 and 18 C (Biederman et al 2017) These foursites are split into pairs of grass- and shrub-dominated sys-tems US-SRG (C4 grassland site) and US-SRM (mesquite-dominated site) are located at the Santa Rita ExperimentalRange sim 60 km south of Tucson AZ whilst US-Whs (cre-osote shrub-dominated site) and US-Wkg (C4 grassland site)are located at the Walnut Gulch Experimental Watershedsim 120 km to the south-east of Tucson AZ Moisture avail-ability at these low-elevation sites is predominantly driven bysummer monsoon precipitation however winter and springrains also contribute to the bi-modal growing seasons at thesesites (Scott et al 2015 Biederman et al 2017) The US-Fuf(Flagstaff Unmanaged Forest) and US-Vcp (Valles CalderaPonderosa) sites are at higher elevations (2215 and 2501 m)Both high-elevation sites experience cooler mean annualtemperatures of 71 and 57 C respectively and are dom-inated by ponderosa pine (Anderson-Teixeira et al 2011Dore et al 2012) The high-elevation forested sites havetwo annual growing seasons with available moisture com-ing from both heavy winter snowfall (and subsequent springsnowmelt) and summer monsoon storms US-Fuf is locatednear the town of Flagstaff in northern AZ whilst US-Vcp islocated in the Valles Caldera National Preserve in the JemezMountains in northernndashcentral New Mexico Groundwaterdepths across all sites are typically tens to hundreds of me-ters Flux tower instruments at all six sites collect half-hourlymeasurements of meteorological forcing data and eddy co-

variance measurements of net surface energy and carbon ex-changes (see Sect 231)

22 ORCHIDEE land surface model

221 General model description

The ORCHIDEE LSM forms the terrestrial component ofthe French IPSL ESM (Dufresne et al 2013) which con-tributes climate projections to IPCC Assessment ReportsORCHIDEE has undergone significant modification sincethe ldquoAR5rdquo version (Krinner et al 2005) which was used torun the CMIP5 (Coupled Model Inter-comparison Project)simulations included in the IPCC 5th Assessment Report(IPCC 2013) The model code is written in Fortran 90 Herewe use ORCHIDEE v20 that is used in the ongoing CMIP6simulations ORCHIDEE simulates fluxes of carbon waterand energy between the atmosphere and land surface (andwithin the sub-surface) on a half-hourly time step In uncou-pled mode the model is forced with climatological fields de-rived either from climate reanalyses or site-based meteoro-logical forcing data The required climate fields are 2 m airtemperature rainfall and snowfall incoming longwave andshortwave radiation wind speed surface air pressure andspecific humidity

Evapotranspiration ET in the model is calculated as thesum of four components (1) evaporation from bare soilE (2) evaporation from water intercepted by the canopy(3) transpiration T (controlled by stomatal conductance)and (4) snow sublimation (Guimberteau et al 2012b) Thereare two soil hydrology models implemented in ORCHIDEEone based on a 2-layer (2LAY) conceptual model the otheron a physically based representation of moisture redistribu-tion across 11-layers (11LAY) In this study the soil depthfor both schemes was set to 2 m based on previous studiesthat tested the implementation of the soil hydrology schemes(de Rosnay and Polcher 1998 de Rosnay et al 2000 2002)Further modifications to the model have been made since theimplementation of the 11LAY scheme to augment the in-creased complexity in the 2LAY scheme runoff occurredwhen the soil reached saturation whereas in the 11LAYscheme surface infiltration runoff and drainage are treatedmore mechanistically based on soil hydraulic conductivity(see Sect 222) In the 2LAY scheme there was an im-plicit resistance to bare soil evaporation based on the depthof the dry soil for the bare soil plant functional type (PFT)In the 11LAY scheme there is an optional bare soil evapo-ration resistance term based on the relative soil water con-tent of the first four soil layers based on the formulationof Sellers et al (1992) ndash (see Sect 223) Both resistanceterms aim to describe the resistance to evaporation exertedby a dry mulch soil layer Similarly the calculation of mois-ture limitation on stomatal conductance has changed In the2LAY version moisture limitation depended on the dry soildepth of the upper layer whereas in the 11LAY version the

ndashse

ndash201

08ndash2

ndash201

07ndash2

Table 2 Summary of differences between 2LAY and 11LAY model versions All other parameters and processes in the model includingthe PFT and soil texture fractions (Table 1) the vegetation and bare soil albedo coefficients (Sect 221) and the multi-layer intermediate-complexity snow scheme (Sect 225) are the same in both versions

Model process Model version

2LAY 11LAY

Soil moisture(Sect 222)

2-layer bucket scheme ndash upper layer variable to10 cm depth and can disappear

1D Richards equation describing moisture dif-fusion in unsaturated soils

Maximum water-holding (field) capacitySect 222)

Constant (150 kgmminus2) for all soil types Derived using van Genuchten (VG) relation-ships for characteristic matric potentials andvary with soil texture

Runoffdrainage(Sect 222)

When soil moisture exceeds field capacity5 partitioned as surface runoff and 95 asgroundwater drainage

Calculated soil hydraulic conductivity deter-mines precipitation partitioning into infiltrationand runoff Drainage in form Of free gravita-tional flow at bottom of soil

Bare soil evaporation resistance(Sect 223)

Based on depth of dry soil for bare soil PFT Notoptional ndash included by default

Empirical equation based on relative water con-tent of the 1st four layers Optional ndash not in-cluded by default

Empirical plant water stress function β(Sect 224)

Based on dry soil depth of upper layer Based on plant water availability for root wateruptake throughout soil column

E and T over vegetated grid cell fraction(Sect 221)

Only T occurs Both T and E occur over effective vegetatedand effective bare soil fraction respectivelyCalculation of effective fractions based on LAI(BeerndashLambert approach)

limitation is based on plant water availability for root wa-ter uptake throughout the soil column Finally in the 2LAYscheme there is no E from the vegetated portion of the gridcell (only T ) whereas in the 11LAY scheme both E and Toccur (see Sect 221) The main differences between the twoORCHIDEE configurations used in this study are describedin the sections below and are summarized in Table 2

In ORCHIDEE a prognostic leaf area is calculatedbased on phenology schemes originally described in Bottaet al (2000) and further detailed in MacBean et al (2015 ndashAppendix A) The albedo is calculated based on the aver-age of the defined albedo coefficients for vegetation (onecoefficient per PFT) soil (one value for each grid cell re-ferred to as background albedo) and snow weighted by theirfractional cover Snow albedo is also parameterized accord-ing to its age which varies according to the underlying PFTThe albedo coefficients for each PFT and background albedohave recently been optimized within a Bayesian inversionsystem using the visible and near-infrared MODIS white-skyalbedo product at 05times 05 resolution for the years 2000ndash2010 The prior background (bare soil) albedo values wereretrieved from MODIS data using the EU Joint ResearchCenter Two Stream Inversion Package (JRC-TIP)

As in most LSMs all vegetation is grouped into broadPFTs based on physiology phenology and for trees thebiome in which they are located In ORCHIDEE by defaultthere are 12 vegetated PFTs plus a bare soil PFT The 13 PFTfractions are defined for each grid cell (or for a given site as

in this study) in the initial model set-up and sum to 10 (un-less there is also a ldquono biordquo fraction for bare rock ice and ur-ban areas) Independent water budgets are calculated for eachldquosoil tilerdquo which represent separate water columns within agrid cell In the 2LAY scheme soil tiles directly correspondto PFTs therefore a separate water budget is calculated foreach PFT within the grid cell In the 11-layer scheme thereare three soil tiles one with all tree PFTs sharing the samesoil water column one soil column with all the grass and cropPFTs and a third for the bare soil PFT Therefore three sep-arate water budgets are calculated one for the forested soiltile one for the grass and crop soil tile and one for the baresoil PFT tile (Ducharne et al 2020 see Sect 222 to 225for details on the hydrology calculations) In the two-layerscheme there is no E from the vegetated tiles (only transpi-ration) In the 11-layer scheme both T and E occur in thevegetated (forest and grasscrop) soil tiles T occurs for eachPFT in the ldquoeffectiverdquo vegetated sub-fraction of each soiltile which increases as LAI increases whereas E occurs atlow LAI (eg during winter) over the effective bare soil sub-fraction of each soil tile Note that the bare soil sub-fractionof each vegetated soil tile is separate from the bare soil PFTtile itself The effective vegetated sub-fraction is calculatedusing the following equation that describes attenuation oflight penetration through a canopy f jv = f j (1minuse(minuskextLAIj ))where f j is the fraction of the grid cell covered by PFT j

(ie the unattenuated case) f jv is the fraction of the effec-tive sub-fraction of the grid cell covered by PFT j and kext

is the extinction coefficient and is set to 10 The effectivebare soil sub-fraction of each vegetated soil tile f jb is equalto 1minus f jv The total grid cell water budget is calculated byvegetation fraction weighted averaging across all soil tiles(Guimberteau et al 2014 Ducharne et al 2020) Soil tex-ture classes and related parameters are prescribed based onthe percentage of sand clay and loam

222 Soil hydrology

Two-layer conceptual soil hydrology model

In the ldquoAR5rdquo version of ORCHIDEE used in the CMIP5 ex-periments the soil hydrology scheme consisted of a concep-tual two-layer (2LAY) so-called ldquobucketrdquo model based onChoisnel et al (1995) The depth of the upper layer is vari-able up to 10 cm and changes with time depending on thebalance between throughfall and snowmelt inputs and out-puts via three pathways (i) bare soil evaporation limited bya soil resistance increasing with the dryness of the topmostsoil layer (ii) root water extraction for transpiration with-drawn from both layers proportionally to the root densityprofile and (iii) downward water flow (drainage) to the lowerlayer If all moisture is evaporated or transpired or if the en-tire soil saturates the top layer can disappear entirely Threeempirical parameters govern the calculation of the drainagebetween the two layers which depends on the water contentof the upper layer and takes a non-linear form so drainagefrom the upper layer increases considerably when the wa-ter content of the upper layer exceeds 75 of the maximumcapacity (Ducharne et al 1998) Transpiration is also with-drawn from the lower layer via water uptake by deep rootsFinally runoff only occurs when the total soil water contentexceeds the maximum field capacity set to 150 kgmminus2 asin Manabe (1969) It is then arbitrarily partitioned into 5 surface runoff to feed the overland flow and 95 drainageto feed the groundwater flow of the routing scheme (Guim-berteau et al 2012b) which is not activated here

Eleven-layer mechanistic soil hydrology model

The 11LAY scheme was initially proposed by de Rosnayet al (2002) and simulates vertical flow and retention ofwater in unsaturated soils based on a physical descriptionof moisture diffusion (Richards 1931) The scheme im-plemented in ORCHIDEE relies on the one-dimensionalRichards equation combining the mass and momentum con-servation equations but is in its saturation form that usesvolumetric soil water content θ (m3 mminus3) as a state variableinstead of pressure head (Ducharne et al 2020) The twomain hydraulic parameters (hydraulic conductivity and dif-fusivity) depend on volumetric soil moisture content definedby the Mualemndashvan Genuchten model (Mualem 1976 vanGenuchten 1980) The Richards equation is solved numer-ically using a finite-difference method which requires the

vertical discretization of the 2 m soil column As describedby de Rosnay et al (2002) 11 layers are defined the top layeris sim 01 mm thick and the thickness of each layer increasesgeometrically with depth The fine vertical resolution nearthe surface aims to capture strong vertical soil moisture gra-dients in response to high temporal frequency (sub-diurnalto a few days) changes in precipitation or ET De Rosnayet al (2000) tested a number of different vertical soil dis-cretizations and decided that 11 layers was a good compro-mise between computational cost and accuracy in simulat-ing vertical hydraulic gradients The mechanistic represen-tation of redistribution of moisture within the soil columnalso permits capillary rise and a more mechanistic represen-tation of surface runoff The calculated soil hydraulic con-ductivity determines how much precipitation is partitionedbetween soil infiltration and runoff (drsquoOrgeval et al 2008)Drainage is computed as free gravitational flow at the bottomof the soil (Guimberteau et al 2014) The USDA soil tex-ture classification provided at 112 resolution by Reynoldset al (2000) is combined with the look-up pedotransferfunction tables of Carsel and Parrish (1988) to derive therequired soil hydrodynamic properties (saturated hydraulicconductivity Ks porosity van Genuchten parameters resid-ual moisture) while field capacity and wilting point are de-duced from the soil hydrodynamic properties listed aboveand the van Genuchten equation for matric potential by as-suming they correspond to potentials of minus33 and minus150 mrespectively (Ducharne et al 2020) Ks increases exponen-tially with depth near the surface to account for increased soilporosity due to bioturbation by roots and decreases exponen-tially with depth below 30 cm to account for soil compaction(Ducharne et al 2020)

The 11LAY soil hydrology scheme has been implementedin the ORCHIDEE trunk since 2010 albeit with variousmodifications since that time as described above and in thefollowing sections The most up-to-date version of the modelis described in Ducharne et al (2020) Similar versions ofthe 11LAY scheme have been tested against a variety ofhydrology-related observations in the Amazon basin (Guim-berteau et al 2012a 2014) for predicting future changes inextreme runoff events (Guimberteau et al 2013) and againsta water storage and energy flux estimates as part of ALMIPin western Africa (as detailed in Sect 1 ndash drsquoOrgeval et al2008 Boone et al 2009 Grippa et al 2011 2017)

223 Bare soil evaporation and additional resistanceterm

The computation of bare soil evaporationE in both versionsis implicitly based on a supply and demand schemeE occursfrom the bare soil column as well as the bare soil fraction ofthe other soil tiles (see Sect 221) In the 2LAY version Edecreases when the upper layer gets drier owing to a resis-tance term that depends on the height of the dry soil in thebare soil PFT column (Ducoudreacute et al 1993) In the 11LAY

versionE proceeds at the potential rateEpot unless the watersupply via upward diffusion from the water column is limit-ing in which case E is reduced to correspond to the situationin which the soil moisture of the upper four layers is at wilt-ing point However since ORCHIDEE v20 (Ducharne et al2020) E can also be reduced by including an optional baresoil evaporation resistance term rsoil which depends on therelative water content and is based on a parameterization fit-ted at the FIFE grassland experimental site at Konza PrairieField Station in Kansas (Sellers et al 1992)

rsoil = exp(8206minus 4255W1) (1)

where W1 is the relative soil water content of the first fourlayers (22 cm ndash Table S1 in the Supplement) W1 is calcu-lated by dividing the mean soil moisture across these layersby the saturated water content The calculation for E thenbecomes

E =min(Epot(1+ rsoilra)Q) (2)

where Epot is the potential evaporation ra the aerodynamicresistance Q the upward water supply from capillary diffu-sion through the soil and rsoil the soil resistance to this up-ward exfiltration In all simulations the calculation of ra in-cludes a dynamic roughness height with variable LAI basedon a parameterization by Su et al (2001) By default in the11LAY version there is no resistance (rsoil = 0) Note thatthere is no representation of below-canopy E in this versionof ORCHIDEE given there is no multi-layer energy budgetfor the canopy Note also that the same roughness is used forboth the effective bare ground and vegetated fractions

224 Empirical plant water stress function β

The soil moisture control on transpiration is defined by anempirical water stress function β Whichever the soil hy-drology model β depends on soil moisture and on the rootdensity profile R(z)= exp(minuscjz) where z is the soil depthand cj (in mminus1) is the root density decay factor for PFT j In both model versions for a 2 m soil profile cj is set to 40for grasses 10 for temperate needleleaved trees and 08 fortemperate broadleaved trees In 11LAY a related variable isnroot(i) quantifying the mean relative root density R(z) ofeach soil layer i so that

sumnroot(i)= 1

In the 2LAY version β is calculated as an exponentialfunction of the root decay factor cj and the dry soil heightof the topmost soil layer (hd

β = exp(minuscj h

dt) (3)

In 11LAY β is rather based on the available moisture acrossthe entire soil moisture profile and is calculated for eachPFT j and soil layer i and then summed across all soil layers(starting at the second layer given no water stress in the firstlayer ndash a conservative condition that prevents transpiration

T from inducing a negative soil moisture from this very thinsoil layer)

β(j)=

11sumi=2

nroot(i)

middotmax

(0(Wiv minusWwpt)(WminusWwpt

) ))) (4)

where Wi is the soil moisture for that layer and soil tile inkgmminus2 Wwpt is the wilting point soil moisture and W isthe threshold above which T is maximum ndash ie above thisthreshold T is not limited by β W is defined by

W =Wwpt+p(WfcminusWwpt) (5)

where Wfc is the field capacity and p defines the thresholdabove which T is maximum p is set to 08 and is constantfor all PFTs This empirical water stress function equationmeans that in 11LAY β varies linearly between 0 at the wilt-ing point and 1 at W which is smaller than or equal to thefield capacity LSMs typically apply β to limit photosynthe-sis (A) via the maximum carboxylation capacity parameterVcmax or to the stomatal conductance gs via the g0 or g1 pa-rameters of the Ags relationship or both (De Kauwe et al2013 2015) In ORCHIDEE there is the option of applyingβ to limit either Vcmax or gs or both In the default configu-ration used in CMIP6 β is applied to both (based on resultsfrom Keenan et al 2010 Zhou et al 2013 2014) thereforethis is the configuration we used in this study

225 Snow scheme

ORCHIDEE contains a multi-layer intermediate complexitysnow scheme that is described in detail in Wang et al (2013)The new scheme was introduced to overcome limitations ofa single-layer snow configuration In a single-layer schemethe temperature and vertical density gradients through thesnowpack which affect the sensible latent and radiative en-ergy fluxes are not calculated The single-layer snow schemedoes not describe the insulating effect of the snowpack orthe links between snow density and changes in snow albedo(due to aging) in a physically mechanistic way In the newexplicit snow scheme there are three layers that each have aspecific thickness density temperature and liquid water andheat content These variables are updated at each time stepbased on the snowfall and incoming surface energy fluxeswhich are calculated from the surface energy balance equa-tion The model also accounts for sublimation snow settlingwater percolation and refreezing Snow mass cannot exceeda threshold of 3000 kgmminus2 Snow age is also calculated andis used to modify the snow albedo Default snow albedo coef-ficients have been optimized using MODIS white-sky albedodata as per the method described in Sect 221 Snow frac-tion is calculated at each time step according to snow massand density following the parametrization proposed by Niuand Yang (2007)

23 Data

231 Site-level meteorological and eddy covariancedata and processing

Meteorological forcing and eddy covariance flux data foreach site were downloaded from the AmeriFlux data por-tal (httpamerifluxlblgov last access 5 November 2020)Meteorological forcing data included 2 m air temperatureand surface pressure precipitation incoming longwave andshortwave radiation wind speed and specific humidity Torun the ORCHIDEE model we partitioned the in situ precip-itation into rainfall and snowfall using a temperature thresh-old of 0 C The meteorological forcing data were gap-filled following the approach of Vuichard and Papale (2015)which uses downscaled and corrected ERA-Interim data tofill gaps in the site-level data Eddy covariance flux datawere processed to provide ET from estimates of latent en-ergy fluxes ET gaps were filled using a modified look-uptable approach based on Falge et al (2001) with ET pre-dicted from meteorological conditions within a 5 d movingwindow Previous comparisons of annual sums of measuredET with site-level water balance measurements at a few ofthese sites show an average agreement within 3 of eachother but could differ by minus10 to +17 in any given year(Scott and Biederman 2019) Estimates of TET ratios werederived from Zhou et al (2016) for the forested sites andboth Zhou et al (2016) and Scott and Biederman (2017)for the more water-limited low-elevation grass- and shrub-dominated sites Zhou et al (2016) (hereafter Z16) usededdy covariance tower gross primary productivity (GPP) ETand vapour pressure deficit (VPD) data to estimate TETratios based on the ratio of the actual or apparent underly-ing water use efficiency (uWUEa) to the potential uWUE(uWUEp) uWUEa is calculated based on a linear regressionbetween ET and GPP multiplied by VPD to the power 05(GPPtimesVPD05) at observation timescales for a given sitewhereas uWUEp was calculated based on a quantile regres-sion between ET and GPPtimesVPD05 using all the half-hourlydata for a given site Scott and Biederman (2017) (hereafterSB17) developed a new method to estimate average monthlyTET from eddy covariance data that was more specificallydesigned for the most water-limited sites The SB17 methodis based on a linear regression between monthly GPP and ETacross all site years One of the main differences between theZ16 and SB17 methods is that the regression between GPPand ET is not forced through the origin in SB17 becauseat water-limited sites it is often the case that ET 6= 0 whenGPP= 0 (Biederman et al 2016) The Z16 method also as-sumes the uWUEp is when TET= 1 which rarely occurs inwater-limited environments (Scott and Biederman 2017) Inthis study TET ratio estimates are omitted in certain wintermonths when very low GPP and limited variability in GPPresults in poor regression relationships

Figure 1 Comparison of the 2LAY vs 11LAY mean daily hy-drological stores and fluxes (i) evapotranspiration (ET mmdminus1 ndasha) (ii) total soil moisture (SM kgmminus2) in the upper 10 cm ofthe soil (b) (iii) total column (0ndash2 m) SM (c) (iv) surface runoff(mmdminus1 d) (v) drainage (mmdminus1 e) and (vi) total runoff (sur-face runoff plus drainage ndash f) Error bars show the SD for ET andSM and the 95 confidence interval for runoff and drainage Forsoil moisture the absolute values of total water content for the up-per layer and total 2 m column are shown for both model versionsie the simulations have not been re-scaled to match the temporaldynamics of the observations (as described in Sect 232) there-fore soil moisture observations are not shown Observations areonly shown for ET

232 Soil moisture data and processing

Daily mean volumetric soil moisture content (SWC θ m3 mminus3) measurements at several depths were obtained di-rectly from the site PIs For each site Table 3 details thedepths at which soil moisture was measured Soil moisturemeasurement uncertainty is highly site and instrument spe-cific but tests have shown that average errors are gener-ally below 004 m3 mminus3 if site-specific calibrations are madeGiven the maximum depth of the soil moisture measurementsis 75 cm (and is much shallower at some sites) we cannotuse these measurements to estimate a total 2 m soil columnvolumetric SWC Instead we only used these measurementsto evaluate the 11LAY model (and the 2LAY upper-layer soil

Table 3 Soil moisture measurement depths (and corresponding model layer in brackets ndash see Table S1)

US-SRM US-SRG US-Whs US-Wkg US-Fuf US-Vcp

Soil moisture depths 25ndash5 cm (5)15ndash20 cm (7)60ndash70 cm (9)

25ndash5 cm (5)15ndash20 cm (7)75 cm (9)

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

moisture ndash calculated for 0ndash10 cm) because unlike the 2LAYmodel with the 11LAY version of the model we have modelestimates of soil moisture at discrete soil depths Howeverseveral factors mean that we cannot directly compare abso-lute values of measured vs modelled soil SWC even though11LAY has discrete depths First site-specific values for soil-saturated and residual water content were generally not avail-able to parameterize the model (see Sect 24) instead thesesoil hydrology parameters are either fixed (in 2LAY) or de-rived from prescribed soil texture properties (in 11LAY ndash seeSect 222) Therefore we may expect a bias between themodelled and observed daily mean volumetric SWC Secondwhile the soil moisture measurements are made with probesat specific depths it is not precisely known over which depthranges they are measuring SWC Therefore with the excep-tion of Fig 1 in which we examine changes in total watercontent between the two model versions for the remaininganalyses we do not focus on absolute soil moisture valuesin the modelndashdata comparison Instead we focus solely oncomparison between the modelled and observed soil mois-ture temporal dynamics To achieve this we removed anymodelndashdata bias using a linear cumulative density function(CDF) matching function to re-scale and match the mean andSD of soil moisture simulations to that of the observations foreach layer where soil moisture is measured using the follow-ing equation

θModCDF =σθObs(θModminus θMod)

σθMod+ θObs (6)

Raoult et al (2018) found that linear CDF matching per-formed nearly as well as full CDF matching in capturing themain features of the soil moisture distributions therefore forthis study we chose to simply use a linear CDF re-scalingfunction Note that while we do compare the re-scaled 2LAYupper-layer soil moisture (top 10 cm) and 11LAY simula-tions at certain depths to the observations (see Sect 31) wecannot compare the total column soil moisture given our ob-servations do not go down to the same depth as the model(2 m) Also note that because of the reasons given above wechose to focus most of the modelndashdata comparison on in-vestigating how well the (re-scaled) 11LAY model capturesthe observed temporal dynamics at specific soil depths (seeSect 32)

24 Simulation set-up and post-processing

All simulations were run for the period of available site data(including meteorological forcing and eddy covariance fluxdata ndash see Sect 231 and Table 1) Table 1 also lists (i) themain species for each site and the fractional cover of eachmodel PFT that corresponds to those species (ii) the maxi-mum LAI for each PFT and (iii) the percent of each modelsoil texture class that corresponds to descriptions of soil char-acteristics for each site ndash all of which were derived from theassociated site literature detailed in the references in Table 1The PFT fractional cover and the fraction of each soil textureclass are defined in ORCHIDEE by the user The maximumLAI has a default setting in ORCHIDEE that has not beenused here instead values based on the site literature wereprescribed in the model (Table 1) Note that ORCHIDEEdoes not contain a PFT that specifically corresponds to shrubvegetation therefore the shrub cover fraction was prescribedto the forested PFTs (see Table 1) Due to the lack of avail-able data on site-specific soil hydraulic parameters acrossthe sites studied we chose to use the default model valuesthat were derived based on pedotransfer functions linking hy-draulic parameters to prescribed soil texture properties (seeSect 222) Using the default model parameters also allowsus to test the default behaviour of the model

At each site we ran five versions of the model (1) 2LAYsoil hydrology (2) 11LAY soil hydrology with rsoil flag notset (default model configuration) (3) 11LAY soil hydrologywith rsoil flag not set and with reduced bare soil fraction (in-creased C4 grass cover) (4) 11LAY soil hydrology with thersoil flag set (therefore Eq 2 activated) and (5) 11LAY soilhydrology with the rsoil flag set and with reduced bare soilfraction Tests 3 and 5 (reduced bare soil fraction) are de-signed to account for the fact that grass cover is highly dy-namic at intra-annual timescales at the low-elevation sitesand therefore during certain seasons (eg the monsoon) thegrass cover will likely be higher than was prescribed in themodel based on average fractional cover values given in thesite literature The C4 grass cover was therefore increased tothe maximum observed C4 grass cover under the most pro-ductive conditions (100 cover for the Santa Rita sites and80 cover for the Walnut Gulch sites) A 400-year spinupwas performed by cycling over the gap-filled forcing datafor each site (see Table 1 for period of available site data)to ensure the water stores were at equilibrium Followingthe spinup transient simulations were run using the forcing

data from each site Daily outputs of all hydrological vari-ables (soil moisture ET and its component fluxes snowpacksnowmelt) the empirical water stress function β LAI andsoil temperature were saved for all years and summed or av-eraged to derive monthly values where needed For certainfigures we show the 2009 daily time series because that wasthe only year for which data from all sites overlapped and acomplete year of daily soil moisture observations was avail-able To evaluate the two model configurations we calcu-lated the Pearson correlation coefficient between the simu-lated and observed daily time series for both the upper-layersoil moisture (with the model re-scaled according to the lin-ear CDF matching method given in Sect 232) and ET Wealso calculated the RMSE mean absolute bias and a measureof the relative variability α between the modelled and ob-served daily ET The latter is calculated as the ratio of modelto observed SDs (α = σm

σo) based on Gupta et al (2009) All

model post-processing and plotting was performed using thePython programming language (v2715) (Python SoftwareFoundation ndash available at httpwwwpythonorg last access2 November 2020) the NumPy (v1161) (Harris et al 2020)numerical analysis package and Matplotlib (v202) (Hunter2007) and Seaborn (v090) (Waskom et al 2017) plottingand data visualization libraries

3 Results

31 Differences between the 2LAY and 11LAY modelversions for main hydrological stores and fluxes

Increasing the soil hydrology model complexity between the2LAY and 11LAY model versions does not result in a uni-form increase or decrease across sites in either the simulatedupper-layer (top 10 cm) and total column (2 m) soil mois-ture (kgmminus2) (Fig 1b and c also see Fig S1 in the Sup-plement for complete daily time series for each site) Thelargest change between the 2LAY and 11LAY versions inthe upper-layer soil moisture were seen at the high-elevationponderosa forest sites (US-Fuf and US-Vcp ndash Figs 1 and S1aand b) In the 2LAY simulations the upper-layer soil mois-ture is similar across all sites whereas in the 11LAY simu-lations the difference between the high-elevation forest sitesand low-elevation grass and shrub sites has increased At US-Fuf both the upper-layer and total column soil moisture in-crease in the 11LAY simulations compared to 2LAY whichcorresponds to an increase in mean daily ET (Fig 1a) awayfrom the observed mean and a decrease in total runoff (sur-face runoff plus drainage ndash Fig 1f) In contrast at US-Vcpwhile there is an increase in the upper-layer soil moisturethere is hardly any change in the total column soil mois-ture The higher upper-layer soil moisture at US-Vcp causesa slight increase in mean ET (and ET variability) that bettermatches the observed mean daily ET and a decrease in totalrunoff Note that changes in maximum soil water-holding ca-

pacity are due to how soil hydrology parameters are definedIn 2LAY a maximum capacity is set to 150 kgmminus2 across allPFTs whereas in 11LAY the capacity is based on soil textureproperties and is therefore different for each site

At the low-elevation shrub and grass sites (US-SRM US-SRG US-Whs and US-Wkg) the differences between thetwo model versions for both the upper-layer and total columnsoil moisture are much smaller (Fig 1) Correspondingly thechanges in mean daily ET and total runoff are also marginal(although the mean total runoff is lower at Walnut GulchUS-Wkg and US-Whs) Across all sites both model versionsaccurately capture the overall mean daily ET (Fig 1) AtSanta Rita (US-SRM and US-SRG) the 11LAY soil mois-ture is marginally lower than 2LAY whereas at the WalnutGulch sites the 11LAY moisture is higher

As described above at all sites there is either no changebetween the 2LAY and 11LAY simulations (Santa Rita) ora decrease in total runoff (surface runoff plus drainage ndashFig 1f) Across all sites excess water is removed as drainagein the 2LAY simulations with little to no runoff (Fig S1andashf3rd panel) whereas in the 11LAY simulations excess waterflows mostly as surface runoff with more limited drainage(Fig S1andashf 2nd panel) This is explained by the fact thatin the 2LAY scheme the drainage is always set to 95 ofthe soil excess water (above saturation) and runoff can ap-pear only when the total 2 m soil is saturated However the11LAY scheme also accounts for runoff that exceeds the in-filtration capacity which depends on the hydraulic conduc-tivity function of soil moisture (Horton runoff) This meansthat when the soil is dry the conductivity is low and morerunoff will be generated In the 11LAY simulations the tem-poral variability in total runoff (as represented by the er-ror bars in Fig 1) has also decreased As just described in11LAY the total runoff mostly corresponds to surface runoff(Fig S1andashf) The lower drainage flux (and higher surfacerunoff) in the 11LAY simulations corresponds well to thecalculated water balance at US-SRM (Scott and Biederman2019) The 11LAY limited drainage is also likely to be thecase at US-Fuf given that nearly all precipitation at the siteis partitioned to ET (Dore et al 2012) In general all thesesemi-arid sites have very little precipitation that is not ac-counted for by ET at the annual scale (Biederman et al 2017Table S1)

Across all sites the magnitude of temporal variabilityof the total column soil moisture (represented by the errorbars in Fig 1c) only increases slightly between the 2LAYand 11LAY model versions In the upper layer (top 10 cm)the soil moisture temporal variability again only increasesmarginally between 2LAY and 11LAY for the high-elevationforest sites (Fig 1b error bars) however the magnitude ofvariability decreases considerably in the 11LAY model forthe low-elevation shrub and grass sites (also see Fig S2 inthe Supplement) At all sites the 2LAY upper-layer soil mois-ture simulations fluctuate considerably between field capac-ity and zero throughout the year including during dry periods

Table 4 Model evaluation metrics comparing the 2LAY and 11LAY daily upper-layer soil moisture (re-scaled via linear CDF matching)and daily ET simulations to observations across the whole time series (where data are present ndash see Fig S2) Metrics include correlationcoefficient (R) root mean squared error (RMSE) mean absolute bias and a measure of the relative variability α between the model andthe observations The mean absolute bias=modelminus observations therefore a negative value represents a mean model underestimation ofobserved ET α = σm

σo(see Sect 24 with ldquoidealrdquo values approaching 1)

Site Modelversion

Upper-layer(0ndash10 cm)soil moisture R

ET R ET RMSE(mmdminus1)

ET mean bias(mmdminus1)

ET relativevariability α

US-Fuf 2LAY 030 036 104 minus008 10811LAY 078 076 086 038 133

US-Vcp 2LAY 027 026 139 minus054 07911LAY 037 059 102 minus027 082

US-SRM 2LAY 052 053 084 minus003 07011LAY 085 084 053 minus007 087

US-Whs 2LAY 056 054 068 minus003 06711LAY 090 085 043 minus002 089

US-SRG 2LAY 048 052 102 001 07011LAY 067 088 057 minus011 090

US-Wkg 2LAY 046 062 063 0 07111LAY 076 09 037 minus001 107

with no rain These fluctuations are due to the fact that in thetwo-layer bucket scheme the top layer can disappear entirely(see Sect 222) In 11LAY however the temporal dynamicsof the upper-layer moisture simulations correspond more di-rectly to the timing of rainfall events (see Fig 2 bottom panelfor an example at three sites in 2009 and Fig S2 for the com-plete time series for each site) This results in a much betterfit of the 11LAY model to the temporal variability seen in theobservations (Figs 2 and S2) This improvement in upper-layer soil moisture temporal dynamics is also indicated bythe strong increase in correlation at all sites between the re-scaled modelled and observed 11LAY upper-layer soil mois-ture compared to 2LAY (increases in R ranged from 01 to048 ndash Table 4) Note that not only is the upper high fre-quency temporal variability therefore arguably more realisticin the 11LAY version but the finer-scale discretization of theuppermost soil layer in this version will also allow a mucheasier comparison with satellite-derived soil moisture prod-ucts that can only ldquosenserdquo the upper few centimeters of thesoil (Raoult et al 2018)

A major and important consequence of the changes in theupper-layer soil moisture temporal dynamics is a consider-able improvement across all sites in the 11LAY-simulateddaily ET (Fig 2andashc second panels which shows 2009 forthree sites Fig S2andashf show the complete time series forall sites) Across all sites the 11LAY RMSE between dailymodelled and observed ET has decreased in comparison to2LAY and the correlation has increased by a fraction of 03to 04 (Table 4) With the exception of US-Vcp the meanabsolute daily ET modelndashdata bias has increased slightly be-

tween the 2LAY and 11LAY versions (Table 4) which is dueto the fact that the 2LAY version both underestimates andoverestimates ET in the spring and summer respectively re-sulting in a smaller mean absolute bias (Fig S3 in the Sup-plement) However the 11LAY model only slightly under-estimates mean daily ET at most sites except at US-Fuf Inboth model versions the biases correspond to less than 10 of the mean daily ET across all low-elevation sites At thehigh-elevation sites the 11LAY bias corresponds to sim 20 of the mean daily ET ndash an increase (decrease) compared to2LAY at US-Fuf (US-Vcp) The ratio of modelled to ob-served SD in ET α is also provided as a measure of relativevariability in the simulated and observed values (Table 4)With the exception of US-Fuf α values tend closer to 10in the 11LAY simulations compared to 2LAY ndash highlightingagain that the 11LAY version does a better job of captur-ing the daily variability The higher ET modelndashdata bias andα at US-Fuf is mostly due to model discrepancies in spring(Fig S2a) which we discuss further in Sect 33 As previ-ously discussed the increase in 11LAY model upper-layermoisture content at the high-elevation forest sites (Figs 1band 2a bottom panel have resulted in an increase in E andT at those sites which in turn results in a lower ET RMSEbetween the model and the observations (Table 4 and seeFigs 2 and S2 2nd panel) if not a decrease in the mean ETbias for US-Fuf (Table 4 and Fig 1) At the low-elevationshrub and grass sites the improvement in ET is also relatedto changes between the two versions in the calculation ofthe empirical water stress function β (Figs 2 and S2 5thpanel) which acts to limit both photosynthesis and stomatal

Figure 2 Comparison of daily time series (for 2009) of upper-layer soil moisture surface water fluxes and related variables between the2LAY (green curve) and 11LAY (blue curve) simulations Changes between the two versions are shown for three sites representing the mainvegetation types left column high-elevation tree-dominated site (US-Fuf) middle column low-elevation mesquite shrub-dominated site(US-SRM) right column low-elevation C4 grass site (US-SRG) At each site top panel LAI 2nd panel ET compared to observations (blackcurve) 3rd panel bare soil evaporation 4th panel transpiration 5th panel empirical water limitation function (β) that scales photosynthesisand stomatal conductance bottom panel model soil moisture (re-scaled via linear CDF matching) expressed as volumetric soil water content(SWC) in the uppermost 10 cm of the soil compared to observations (black curve) Precipitation is shown in the grey lines in the bottompanel for each site (Note full time series across all years are shown for all site in Fig S2andashf) Light brown shaded zones show periods ofmaximum plant water limitation (β) at Santa Rita and consequent troughs in T and SWC

conductance (therefore T ) during periods of moisture stress(Sect 224) With the new calculation in the 11LAY ver-sion (see Sect 224) we see a stronger more rapid decreasein β (increased stress) during warm dry periods that corre-spond to strong reductions in T (light brown shaded zonesin Fig 2) Aside from T and E the other ET components(interception and sublimation) did not change much betweenthe two hydrology schemes (results not shown) thereforethese terms are not contributing to improvements betweenthe 2LAY and 11LAY versions

The improvement in daily ET temporal dynamics re-sults in an 11LAY mean monthly ET that is also well cap-tured by the model throughout the year including both thewarm dry MayndashJune period followed by monsoon summerrains particularly for low-elevation grass and shrub sites(Figs 3 and S3) As previously discussed the improvedhigher monthly ET in the 11LAY version during the pe-riod of maximum productivity (ie the spring and summerfor the high-elevation sites and the summer monsoon for the

low-elevation sites ndash Fig 3) is likely due to the increase inplant-available water (Figs 1b and c and S1) Despite theimprovement in the 11LAY temporal variability at the high-elevation forest sites there is still a bias in the mean monthlyET magnitude between the 11LAY model and observationsat US-Fuf there is a distinct overestimation of ET during thespring (Fig S3a) whereas at US-Vcp there is a noticeableunderestimation of ET during the spring and monsoon pe-riods (Fig S3b) We will return to these remaining 11LAYET modelndashdata discrepancies in Sect 33 after having evalu-ated the 11LAY soil moisture against observations at differ-ent depths

32 Comparison of 11LAY soil moisture againstobservations at different depths

Figure 4 compares model vs observed daily volumetric soilwater content time series for 2009 at three different depths(see Fig S4 in the Supplement for the full time series ateach site) The complete model time series were re-scaled

Figure 3 Evapotranspiration (ET) monthly mean seasonal cyclecomparing the 2LAY (green curve) and 11LAY (blue curve) sim-ulations with observations (black curve) Individual site simula-tions have been averaged over the high-elevation tree-dominatedsites (a) and across all the low-elevation grass- and shrub-dominatedsites (b) Units are millimeters per month (mmmonthminus1)

via linear CDF matching to remove modelndashobservation bi-ases (see Sect 232) however the linear CDF matching pre-serves the mean and SD of the temporal variability As seenin Sect 31 and Fig 2 (bottom row showing upper 10 cm soilmoisture) in Fig 4 the high-frequency temporal variabilityof the 11LAY soil moisture in the uppermost layer almostperfectly matches the observed one particularly at the low-elevation shrub- and grass-dominated sites (US-SRM US-SRG US-Whs US-Wkg) At most of the low-elevation sitesthe soil moisture drying rates in the upper 20 cm of soil arewell captured by the model with the small exception of theSanta Rita sites between January and March in which themodel appears to dry down at a faster rate than observed(Fig 4 US-SRM and US-SRG top and middle rows)

In contrast the temporal mismatch between the obser-vations and the model in the uppermost layer is higher atthe forest sites The US-Fuf and US-Vcp 11LAY simula-tions appear to compare reasonably well with observationsin the upper 2 cm of the soil from June through to the end ofNovember (end of September in the case of US-Vcp) (Fig 4)However in some years the model appears to overestimatethe SWC at both sites during the winter months (positivemodelndashdata bias) and underestimate the observed SWC dur-ing the spring months (negative modelndashdata bias) particu-larly at US-Fuf Although US-Fuf and US-Vcp are semi-aridsites their high elevation means that during winter precipi-tation falls as snow therefore these apparent model biasesmay be related to (i) the ORCHIDEE snow scheme (ii) in-correct snowfall meteorological forcing andor (iii) incorrectsoil moisture measurements under a snowpack During theearly winter period the model soil moisture increases rapidlyas the snowpack melts and is replenished by new snowfallwhereas the observed soil moisture response is often slower(Fig 5a and b light blue shaded zones) This often coincides

with periods when the soil temperature in the model is be-low 0 C (Fig 5b) suggesting that in the field soil freezingmay be negatively biasing the soil moisture measurementsAn alternative explanation is that ORCHIDEE overestimatessnow cover (and therefore snowmelt and soil moisture) at theforest sites because it assumes that snow is evenly distributedacross the grid cell whereas in reality the snow massdepthis lower under the forest canopy than in the clearings

At US-Fuf it appears that the model melts snow quiterapidly after the main period of snowfall (Fig 5a light greenshaded zones) Once all the snow has melted the model soilmoisture also declines however the observed soil moistureoften remains high throughout the spring ndash causing a negativemodelndashdata bias (Fig 5a) Unlike US-Fuf a similar nega-tive modelndashdata bias at US-Vcp often coincides with periodswhen snow is still falling although the amount is typicallylower (Fig 5b light green shaded zones) however the modeldoes not always simulate a high snow mass during these pe-riods These periods coincide with rising surface tempera-ture above 0 C Although snow cover mass or depth datahave not been collected at these sites snow typically re-mains on the ground until late spring after winters with heavysnowfall suggesting the continued existence of a snowpackand slower snowmelt that replenishes soil moisture until latespring when all the snow melts Therefore the lack of a sim-ulated snowpack into late spring could explain the negativemodelndashdata soil moisture bias To test the hypothesis that themodel melts or sublimates snow too rapidly thereby limitingthe duration of the snowpack and also allowing surface tem-peratures to rise we altered the model to artificially increasesnow albedo and decrease the amount of sublimation how-ever these tests had little impact on the rate of snowmelt orthe duration of snow cover (results not shown) Aside frommodel structural or parametric error it is possible that there isan error in the meteorological forcing data Rain gauges mayunderestimate the actual snowfall amount during the periodswhen it is snowing (Rasmussen et al 2012 Chubb et al2015) If the snowfall is actually higher than is measured itmay in reality lead to a longer lasting snowpack than is es-timated by the model To test this hypothesis we artificiallyincreased the meteorological forcing snowfall amount by afactor of 10 and re-ran the simulations Although this artifi-cial increase is likely exaggerated the result was an improve-ment in the modelled springtime soil moisture estimates atUS-Fuf (Fig S5 in the Supplement) However the same testincreased the positive modelndashdata bias in the early winterat US-Fuf and degraded the model simulations at US-VcpThis preliminary test suggests that inaccurate snowfall forc-ing estimates may play a role in causing any negative modelndashdata bias spring soil SWC but more investigation is neededto accurately diagnose the cause of the springtime negativemodelndashdata bias

Overall there is a decrease in the model ability to captureboth high frequency and seasonal variability with increasingsoil depth At all sites the temporal dynamics of the deep-

Figure 4 Daily simulated volumetric soil water content (SWC ndash m3 mminus3) in 2009 (re-scaled via linear CDF matching) compared to obser-vations at each site for three depths (upper middle lower) in the soil profile The soil depths and their corresponding model layers are givenin Table 3 Precipitation is shown in the grey lines in the bottom panel for each site

est observations are not well represented in the model (Fig 4bottom row for each site) At the high-elevation forest sites(US-Fuf and US-Vcp) the model does not capture the re-sponse of observed soil moisture in the deepest layer to sum-mer storm events In contrast at the low-elevation shrub andgrass sites the 11LAY SWC is far too dynamic in the deep-est layer The smoother model temporal profile at depth atthe forest sites compared to the sites with higher grass frac-tion is likely related to impact of rooting depth on expo-nential changes in Ks towards the surface (see Sect 222)As the forests have deeper roots the increase in Ks startsfrom a lower depth in the soil profile than the more grass-dominated sites which in turn allows for a quicker infiltra-tion of moisture to deeper layers The higher Ks at depthalso allows for a higher drainage and therefore decreasedsoil moisture temporal variability However this descriptionof the model behavior does not explain the modelndashdata dis-crepancies The poor modelndashdata fit at lower depths may berelated to the discretization of the soil column with a geomet-ric increase in internode distance Therefore the soil layerthicknesses increase substantially beyond sim 2ndash4 cm (7th and8th soil layers ndash Table S1) For the deeper soil moisture ob-servations it is therefore harder to match the depth of theobservations with a specific soil layer Alternatively it ispossible that the model description of a vertical root den-sity profile which is used to calculate changes in Ks withdepth is too simplistic for semi-arid vegetation that typically

has extensive lateral root systems that are better adapted forwater-limited environments It is also possible that assign-ing semi-arid tree and shrub types to temperate PFTs as wehave done in this study in the absence of semi-arid-specificPFTs has resulted in a root density decay factor that is tooshallow In contrast to temperate forests semi-arid trees andshrubs also often have deep taproots for accessing ground-water Finally changes in soil texture that may occur in real-ity with depth in the soil could alter hydraulic conductivityparameters in the model however hydraulic conductivityonly changes (exponentially) with depth owing to soil com-paction (see Sect 222) In addition semi-arid region soilsoften have a higher concentration of rock and gravel (Grippaet al 2017) ndash neither of which are represented in the OR-CHIDEE soil texture classes

33 Remaining discrepancies in ET and its componentfluxes

Despite the improvement in seasonal ET temporal dynamicsin the 11LAY model particularly the timing of the reduc-tion during the dry season key modelndashdata discrepancies inET remain during spring (MarchndashApril) and monsoon (JulyndashSeptember) periods (i) at US-Fuf the 11LAY ET is overes-timated during the spring and early summer (Fig S3a) (ii) atUS-Vcp the model underestimates ET for much of the grow-ing season likely due to low LAI values in the earlier andlater years of the simulation (Fig S3b) (iii) at US-SRM the

Figure 5 (a) US-Fuf and (b) US-Vcp 11LAY (blue curve) daily time series (2007ndash2010) of model (re-scaled via linear CDF matching)vs observed volumetric soil water content (middle panel SWC ndash m3 mminus3) (black curve) compared to simulated snow mass (top panel)and soil temperature from the corresponding 2 cm soil thermal layer (bottom panel) Snowfall is also shown as grey lines in the SWC timeseries In the bottom panel the grey horizontal dashed line shows a 0 C threshold Light blue shaded zones show periods where the modeloverestimates the observations light green shaded zones show periods where the model underestimates the observations

11LAY model overestimates springtime ET (in contrast toother low-elevation monsoon sites) (Fig S3c) and (iv) the11LAY model still slightly underestimates peak monsoon ETat the low-elevation shrub sites (US-SRM and US-Whs ndashFig S3c and d) as seen in a previous semi-arid model evalu-ation study (Grippa et al 2011)

The model overestimate in spring ET at US-Fuf could berelated to the snowfall issues that are causing the model tounderestimate spring soil moisture during the same period(Figs 4 and 5 and see Sect 32) The lack of a persistentsnowpack in the model during this period can explain thepositive bias in spring ET because in reality the presence of

snow would suppress bare soil evaporation As discussed inSect 32 to accurately diagnose this issue we would needfurther information on snow mass or depth Further supportfor the suggestion that modelled spring E is overestimatedcomes from comparing the model with estimated TET ratios(Fig 6) Although both E and T increase in the US-Fuf (andUS-Vcp) 11LAY simulations (compared to 2LAY ndash Fig S3aand b) due to the increase in upper-layer soil moisture (aspreviously described in Sect 31 and Figs 2 and S2a and b)the stronger increase in 11LAY E compared to T resultedin lower 11LAY TET ratios across all seasons (Fig S3aand b) While the model captures the bimodal seasonality at

Figure 6 Comparison of modelled and data-derived estimates ofmean monthly TET ratios for each site Forest site (US-Fuf andUS-Vcp) TET estimates are derived using the method of Zhouet al (2016 ndash Z16 ndash green curve) Monsoon low-elevation grass-and shrub-dominated site TET estimated are based on both Zhouet al (2016) and Scott and Biederman (2017 ndash SB17 ndash orangecurve) Blue curves show the model ratios at each site Please seeSect 231 for details on methods for data-derived TET estimates

the forested sites as seen in the Z16 data-derived estimates(Fig 6) the magnitudes of model TET ratios appear to betoo low in all seasons given the 100 tree cover at these siteswith a maximum LAI of sim 24 Whilst low spring 11LAYTET ratios at US-Fuf may be due to overestimated E as aresult of higher soil moisture and underestimated snow coverthe generally low bias in TET ratios across all seasons atboth US-Fuf and US-Vcp may also point to the issue that nobare soil evaporation resistance term is included in the de-fault 11LAY version This may explain why the model TETratios do not increase as rapidly as estimated values at thestart of the monsoon (Fig 6) However discrepancies in thetiming of TET ratio peak and troughs between the modeland data-derived estimates at the forested sites could alsobe due to the fact that evergreen PFTs have no associatedphenology modules in ORCHIDEE instead changes in LAIare only subject to leaf turnover as a result of leaf longevitywhich may be an oversimplification

At US-SRM the modelled spring TET ratio overesti-mates the Z16 estimate and underestimates the SB17 esti-mate (Fig 6) The current state of the art is that differentmethods for estimating TET typically compare well in termsof seasonality but differ in absolute magnitude thereforethe uncertainty in data-derived estimates of TET magni-tude during the spring at US-SRM makes it difficult to gleanany information on whether T or E (or both) are responsi-

ble for the 11LAY overestimate of modelled springtime ET(Fig S3c) If the SB17 method is more accurate then it isprobable that modelled springE at this site is too high (TETunderestimated) again potentially due to the lack of the baresoil evaporation resistance term in the default 11LAY con-figuration However if the Z16 estimate is accurate then it islikely that spring T is overestimated at US-SRM potentiallydue to an overestimate in LAI The modelndashdata bias in springmean monthly ET appears to correlate well with modelledspring mean LAI at US-SRM (Fig S6 in the Supplement) Ifmodel LAI at US-SRM is too high during the spring it is im-possible to determine whether the shrub or grass LAIs are in-accurate without independent accurate estimates of seasonalleaf area for each vegetation type which are not available atpresent however in the field the spring C4 grass LAI is typi-cally half that of its monsoon peak ndash a pattern not seen in themodel (Fig S6)

During the monsoon at the low-elevation grass- and shrub-dominated sites both data-derived estimates of TET agreeon the seasonality and while different in magnitude bothare higher than the model TET values (Fig 6) Given thisagreement both sets of estimated values can help to diagnosewhy the 11LAY model also underestimates monsoon peakET at the low-elevation shrub sites (US-SRM and US-Whsndash Fig S3c and d) The underestimate in modelled monsoonTET ratios across all grassland and shrubland sites could beeither because T is too low or E is too high At the shrublandsites (US-SRM and US-Whs) both monsoon ET and TETare underestimated therefore for these sites it is plausiblethat the dominant cause is a lack of transpiring leaf area Aswas the case for spring ET at US-SRM monsoon modelndashdataET biases are better correlated with LAI at shrubland sitescompared to grassland sites (Fig S8 in the Supplement) Incontrast at the grassland sites (US-SRG and US-Wkg) mon-soon ET is well approximated by the 11LAY model thusthe underestimate in TET ratios suggests that both the tran-spiration is too low and the bare soil evaporation too highFurthermore although 11LAY does capture the decrease inET during the hot dry period of May to June at the grass andshrub sites (which is a significant improvement compared to2LAY ndash see Sect 31) the 11LAY TET ratios are slightlyout of phase with the estimated values Both data-derived es-timates agree that TET ratios at all grass and shrub sites de-cline in June during the hottest driest month (as expected)however the model TET ratios reach a minimum 1 monthlater in July (Fig 6) This 1-month lag in model TET ra-tios is apparent despite the fact that the ET minimum is ac-curately captured by the model (Figs 3b and S3cndashf) Themodelled TET ratios also do not increase as rapidly as bothestimates during the wet monsoon period (JulyndashSeptember)which can be explained by the fact that the model E at thestart of the monsoon increases much more rapidly than mod-elled T Taken together these results suggest that LAI is notincreasing rapidly enough after the start of monsoon rains(see Fig S7 in the Supplement) resulting in negatively bi-

Figure 7 Monthly mean seasonal cycle for ET TET ratios T andE averaged across all low-elevation grass- and shrub-dominatedsites comparing the default 11LAY simulations (blue curve) with asimulation in which bare soil fraction is decreased C4 grass coverincreased (yellow curve) ET is compared to observations (blackdashed curve) and TET ratios are compared to the data-derivedestimates from Scott and Biederman (2017 ndash orange dashed curve)and Zhou et al (2016 ndash green dashed curve) Units are millimetersper month (mmmonthminus1)

ased TET ratios in July Meanwhile the increase in avail-able moisture from monsoon rains potentially coupled with alack of bare soil evaporation resistance in the default 11LAYversion is causing a positively biased model E that compen-sates for the lower T These compensating errors result inaccurate ET simulations The underestimate in modelled leafarea during the monsoon could either be (i) incorrect timingof leaf growth for either grasses or shrubs and an underes-timate of peak LAI andor (ii) due to the fact that the staticvegetation fractions prescribed in the model do not allow foran increase in vegetation cover during the wet season (ie themodel lacks the ability to grow grass in interstitial bare soilareas)

We attempted to explore both the hypotheses that couldexplain discrepancies in model ET and TET ratios (incorrectT due to lack of transpiring leaf area at low-elevation grassand shrub sites or overestimated E across all sites) with twofurther tests These final tests and their results are describedin the following section

34 Testing decreased bare soil cover and the additionof the 11LAY bare soil resistance term

To further investigate the possibility that summer ET andTET ratios are underestimated at low-elevation sites be-cause of a lack of transpiring leaf area we reduced the baresoil fraction and increased C4 grass fraction to the maximumobserved C4 grass cover under the most productive condi-tions This decrease in bare soil fraction increased ET andTET ratios during the monsoon period at all low-elevationgrass- and shrub-dominated sites and also increased ET dur-ing spring at the Santa Rita sites (Fig S9 in the Supplement

mean across low-elevation sites in Fig 7) However althoughthe TET ratios reduced the negative model biases in thesummer monsoon period when compared to the data-derivedestimates the model now overestimated ET in all seasons(Figs 7 and S9) Furthermore the spring ET modelndashdata biasat US-SRM was further exacerbated by the decrease in baresoil fraction (Fig S9) and the mean estimated TET ratiosacross all low-elevation grass and shrub sites were a closermatch to the original 11LAY version (Fig 7) Finally whilethe decrease in the bare soil fraction (increase in C4 grasses)may have partially accounted for the negative bias in TETratios at the start of the monsoon the changes did not correctthe phase discrepancy between the estimated and modelledTET seasonal trajectories the estimated TET still declinedto a minimum in June (as expected during the hot dry pe-riod) whereas the model declined 1 month later Putting thelatter points together this new test gives further weight tothe suggestion put forward in Sect 33 that the model is notcapturing the correct increase in leaf area at the start of themonsoon ndash ie the problem is not just that there is a lack inthe overall amount of transpiring leaf area (or a too high baresoil fraction) ndash due to issues with the model phenology for in-dividual PFTs andor its ability to capture dynamic changesin seasonal vegetation cover

As described in Sect 33 the remaining model ET issues(and its component fluxes) in both high-elevation forest sitesand low-elevation shrub- and grass-dominated sites couldalso be due to the fact that the model simulates too muchbare soil evaporation The 11LAY version has an optionalbare soil evaporation resistance term that is not activated inthe default version therefore the 11LAY simulations pre-sented thus far have not included any such a resistance termTherefore we tested the inclusion of the bare soil resistanceterm at all sites Although there is no bare soil fraction atthe high-elevation forested sites (US-Fuf and US-Vcp) inthe 11LAY version E still occurs over the bare soil sub-fraction of the vegetated soil tiles The bare soil sub-fractionof the vegetated soil tiles increases at low LAI during wintermonths (see Sect 221) therefore including the bare soilresistance term caused a reduction in E during the winter(lower LAIndash Fig 8a bottom panel) The reduction in winterE at the forested sites in turn allowed for higher overall soilmoisture content (Fig S10a and b in the Supplement) andtherefore a greater T (and E) during the spring and summer(Fig 8a) As a result TET ratios were increased with theaddition of the bare soil evaporation term thus potentiallypartially resolving the issue of negatively biased TET ratiosseen in the default 11LAY simulations (see Sect 33) Theincrease in plant-available moisture with the addition of theresistance term also led to a strong increase in LAI at US-Vcpfrom a mean around 05 to a mean around 21 (Fig S10b)which is much closer to the observed LAI for the site How-ever the dramatic increase in T resulted in a simulated ET atboth forest sites that strongly overestimated the observations(Figs 8 and S10a and b) therefore overall the addition of

Figure 8 Monthly mean seasonal cycle for evapotranspiration (ET) transpiration T and bare soil evaporation E averaged across all high-elevation forest sites (a) and low-elevation monsoon grass- and shrub-dominated sites (b) for the default 11LAY simulations (blue curve)compared to a simulation that included an additional bare soil evaporation resistance term (red curve) ET is also compared to observations(black curve) Units are millimeters per month (mmmonthminus1)

the bare soil evaporation resistance term did not improve theET modelndashdata fit at these sites As discussed in Sect 32spring ET may also be overestimated at these sites due to thelack of a persistent snowpack

At all the low-elevation grass and shrub sites the addi-tion of the bare soil resistance term resulted in a strong de-crease in soil evaporation during the monsoon season anda lesser but non-negligible decrease to almost zero evapo-ration during the winter (Fig 8 ndash right column) Bare soilevaporation remained much the same during the spring andthe hot dry season months of May and June As seen for theforest sites the decline in bare soil evaporation during themonsoon period results in a slightly higher moisture storage(Fig S10cndashf) which in turn fractionally increases T through-out the year (Fig 8) The net effect is a reduction in ET dur-ing summer and winter and an increase in spring and dryseason ET (Fig 8) However as for the forested sites thisnet effect in the simulated ET produces a worse fit to thedata Therefore the addition of this term does not resolve theET issues documented in Sect 33 a further positive bias inspring ET estimates is observed at US-SRM (Fig S10c) andthe underestimate in monsoon ET at US-SRM and US-Whs(Fig S10c and d) is further exacerbated Furthermore thenear-zero evaporation in the winter months with the introduc-

tion of the bare soil resistance term results in an increase inwinter TET ratios Therefore at the low-elevation sites themonthly seasonality of TET differs quite considerably fromthe default 11LAY model runs (Fig S10cndashf) and generallydoes not follow the seasonal trajectories estimated by eitherZhou et al (2016) or Scott and Biederman (2017) (Fig 6)

In a final test we combined both the decrease in bare soilfraction with the addition of the bare soil resistance term forthe low-elevation sites The addition of the bare soil resis-tance term reduced the positive bias seen with the increasein C4 grass (decrease in the bare soil fraction) (Fig S11 inthe Supplement) However as seen in the bare soil resistancetests with the original vegetation and bare soil fractions theaddition of the resistance term increased spring T due to thehigher spring soil moisture ndash thus exacerbating the positivebias in ET It is clear that neither of these tests fully dealwith remaining ET modelndashdata biases in the 11LAY ver-sion ndash nor do they account for the issues in the model sea-sonality of TET ratios The ET seasonal temporal dynamicsremain much the same in all tests We point out however thatthe model fit to ET observations was still greatly improvedin the 11LAY version compared to 2LAY and many of theremaining modelndashdata discrepancies are less significant bycomparison It is therefore possible that some combination

of the additional bare soil evaporation resistance term de-creased bare soil fraction improved semi-arid leaf phenol-ogy schemes and further calibration of hydrology phenol-ogy stomatal conductance and water-limitation parameterswould be able to resolve most if not all of the remainingmodelndashdata discrepancies in ET and TET estimates at thesesites This is beyond the scope of this study but the optionsare discussed more in Sect 4

4 Discussion

This study showed that in comparison to a simple bucketmodel (Manabe 1969) a discretized soil hydrology schemebased on the Richards equation ndash and associated model de-velopments ndash results in considerable improvements in sim-ulated semi-arid site soil moisture temporal dynamics thatexhibit a more realistic response to rainfall events (contraryto the modelndashdata comparison of Lohou et al 2014) As aresult we see dramatic improvements in high temporal fre-quency to seasonal ET simulations Previous studies havealso demonstrated that the more mechanistic descriptions ofsoil hydrology included in the latest LSM versions have re-sulted in improvements to surface latent and sensible heatfluxes (de Rosnay et al 2002 Best et al 2015) howeverfew studies have specifically compared these two model ver-sions across a range of semi-arid ecosystems as we have at-tempted in this study However there remain a number ofmissing hydrological processes that have not yet been in-corporated into LSMs andor inadequate existing processeswhich will clearly have an impact on semi-arid hydrologicalmodelling (Boone et al 2009 Grippa et al 2017) and mayresolve some of the remaining modelndashdata discrepancies wewere not able to address in this study We highlight these inthe sections below

41 Issues with modelling vegetation dynamics insemi-arid ecosystems

Our analysis has suggested that biases in low-elevation shruband grassland site ET might be due to incorrect simulationsof seasonal vegetation dynamics therefore in order to ob-tain realistic estimates of ET and its component fluxes it isimportant that the model can accurately simulate seasonalchanges in leaf area andor grass vs bare soil fractionalcover The connection between vegetation fractional coverand LAI is a particular issue in sparsely vegetated regionswhen low LAI effectively means more bare soil is coupledwith the atmosphere and E increases To account for this inORCHIDEE the bare soil fraction is slightly increased whenLAI is low (see Sect 221) which is often the case at thesesites however there are only limited observations to sup-port this model specification Similarly there are not manyLAI measurements for grasses and shrubs in these ecosys-tems therefore we have relied on estimating the LAImax

parameter from MODIS LAI data While different satelliteLAI products often correspond well to each other in termsof temporal variability there is often a considerable spreadin their absolute LAI values (Garrigues et al 2008 Fanget al 2013) therefore the MODIS LAI peak values maynot be accurate for these ecosystems In any case the satel-lite LAI values represent a mix of different vegetation typesand unlike satellite reflectance data it is not possible to lin-early unmix the satellite LAI estimates based on fractionalcover More field LAI measurements are needed from differ-ent vegetation types (especially annual vs perennial grassesand shrubs) to verify what the likely maximum LAI is foreach PFT

As mentioned in the results it is also possible that LSMscontain an inaccurate representation of different semi-aridvegetation phenology including drought-deciduous shrubsand annual vs perennial C4 grasses The model does yet dis-cern between perennial grasses and annual C4 grasses thatonly grow during the warmest wettest periods (Smith et al1997) It is possible that LSMs need new phenology modelsthat account for annual C4 grass strategies in order to obtainaccurate simulations of semi-arid water and carbon fluxesFinally it is possible that incorrect seasonal LAI trajectoriesare also causing the issues in the TET seasonality seen atthe higher-elevation forested sites due to the lack of an ev-ergreen phenology module in ORCHIDEE Recently a newevergreen phenology module has been implemented in OR-CHIDEE (Chen et al 2020) however this scheme was de-veloped for humid tropical forests Testing it for evergreentrees in semi-arid regions is beyond the scope of this studybut will be investigated in future work Again seasonal LAImeasurements of different high-elevation semi-arid vegeta-tion types would significantly help to improve or further de-velop semi-arid phenology models

Alternatively it may be that other model parameters andprocesses involved in leaf growth ndash for example phenol-ogy root zone plant water uptake water-limitation andphotosynthesis-related parameters ndash are inaccurate and inneed of statistical calibration (eg MacBean et al 2015)Incorrect representations of how we model low tempera-ture and high VPD constraints on stomatal conductance mayalso play a role At the high-elevation sites we assumedthe ponderosa pine trees should be modelled as temperateneedleleaved evergreen PFT The default model parametersassigned to this PFT may not be appropriate for modellingthis plant functional type in water-limited semi-arid envi-ronments Another likely issue for modelling low-elevationsparsely vegetated semi-arid ecosystems with ORCHIDEEis that there is no specific shrub PFT although a recent OR-CHIDEE version includes shrub PFTs for high-latitude tun-dra ecosystems (Druel et al 2017) In future work we willadapt similar shrub parameterizations for semi-arid environ-ments

The importance of vegetation cover and seasonal changesin leaf area for modelling hydrological fluxes ndash particu-

larly T ndash is not a new observation (eg Ha et al 2015Grippa et al 2017) Baldocchi et al (2010) found that LAIwas important at five Mediterranean sites in California andEurope for determining how much carbon is assimilated andhow much water is lost Hogue et al (2005) also found theNoah LSM was not able to replicate monsoon period LE in-creases at the Walnut Gulch sites which they suggested maybe related to inaccuracies in the satellite greenness fractionestimates that are used to run the model Whitley et al (20162017) also proposed that any improvements needed for ter-restrial biosphere modelling of savanna ecosystems shouldinclude modifications to the phenology schemes and the splitbetween fractional cover of trees and grasses

42 ET partitioning (TET ratio)

In agreement with this study Lian et al (2018) also showthat CMIP5 models vastly underestimate TET ratios Theyestimated a new global TET ratio of 062plusmn 006 whichis similar to the upscaled estimate of 057plusmn 007 of Weiet al (2017) and suggest that model underestimates could becaused by misrepresentation of vegetation structure impactson canopy light use interception loss and root water up-take Their conclusions lend further weight to our suggestionthat further improvements in TET ratios may result frommore accurate simulations of seasonal phenology and frac-tional vegetation cover (see Sect 33) Alternatively Changet al (2018) have suggested that neglecting to account forlateral redistribution of moisture is responsible for model in-ability to capture TET partitioning Current LSM versionsdo not simulate extensive shallow root systems that are typ-ical of semi-arid vegetation that is more adapted to water-limited conditions However they also mention other LSMissues that might be affecting the TET ratio such as the lackof root dynamics vegetation shading topographic effectsand the representation of bare soil evaporation In order toproperly diagnose whether discrepancies in modelled TETare caused by inaccurate representation of lateral moisture re-distribution we need to perform a comparison of a spatiallydistributed model simulation with a high-density network ofhydrological observations Nevertheless in spatially hetero-geneous mixed shrubndashgrass ecosystems it seems likely thatmissing model processes will need to be accounted for beforeaccurate simulations of TET ratios can be achieved One ex-ample of this might be the need to include in the model a rep-resentation of shrub understory and below-canopy E Diag-nosing and addressing discrepancies between modelled andestimated TET is important specifically for dryland ecosys-tems where increases in vegetation productivity andor coverin response to rising atmospheric CO2 appear to be drivinghigher TET rates (Lian et al 2018)

43 Bare soil evaporation

The addition of a term that simulates bare soil evaporation re-sistance to dry soil served to alleviate discrepancies in TETratios compared to data-derived estimates however result-ing changes in modelled ET provided a worse fit to the ob-servations It is possible that the bare soil resistance is onlypart of the solution as discussed in Sect 34 Future stud-ies could also investigate the impact of uncertainty in the useof pedotransfer functions (eg Mermoud and Xu 2006) inderiving soil hydraulic parameters from soil texture informa-tion The low-elevation sites typically have a very cobblyrocky soil surface that is not accounted for in ORCHIDEEIncluding soil texture variability with different soil horizonscould further improve ORCHIDEErsquos capability to capturethe correct E ET and TET ratios Alternatively the rela-tively simple implementation of a bare soil resistance term(Eq 2 ndash Sect 223) might need to be adapted to include baresoil evaporation resistance across a litter or biocrust layerAt the sparsely vegetated grass- and shrub-dominated sites insouthern Arizona biological soil crusts (biocrusts) composedof assemblages of lichens bryophytes cyanobacteria algaeand microbes form across much of the bare soil surface (Bel-nap et al 2016) Biocrust layers may significantly alter baresoil evaporation (and other aspects of ecosystem ecology andfunctioning ndash Ferrenberg and Reed 2017) in sparsely vege-tated regions in ways that have not yet been considered in anyLSM bare soil evaporation scheme Therefore it is possiblethat in addition to a more mechanistically based formulationof resistance to bare soil evaporation due to a litter layer (asper Swenson and Lawrence 2014 or Decker et al 2017)separate formulations of evaporation through biocrustmulchlayers may need to be developed (eg Saux-Picard et al2009)

44 High-elevation model snowpack and snowmeltpredictions

The model also needs to be tested at other high-elevationsemi-arid mountainous sites (such as the Sierra Nevada inCalifornia) for which spring snowmelt is the predominant(and controlling) annual source of moisture More specifi-cally more information on snow cover depth or mass par-ticularly under closed forest canopies would be useful todiagnose potential sources of bias in the snowfall simula-tions It is crucial that LSMs accurately capture semi-aridhigh-elevation snowfall temporal dynamics if we are to haveunbiased projections in future moisture availability and pro-ductivity for these regions

45 Implications for modelling plant water stress

Similar to Whitley et al (2016) the original 2LAY ver-sion of the model underpredicted wet monsoon season ETThe peak ET fluxes were generally much better captured in

the 11LAY version However in contrast to the findings ofWhitley et al (2016) the 2LAY simulations overestimatedET during the hottest driest period between May and JuneOur results demonstrated that a modified empirical beta wa-ter stress function (used to downregulate stomatal conduc-tance during periods of limited moisture) that takes into ac-count available soil moisture and root density across the en-tire soil column (Sect 224) helped to better capture dryseason ET dynamics These results are interesting in lightof previous studies showing that LSMs employing empiri-cal beta water stress functions show considerable differencesin their simulated response to water-stressed periods (Med-lyn et al 2016 De Kauwe et al 2017) These studies arguefor more evidence-based formulations of plant response todrought De Kauwe et al (2015) also highlight the need formodels to incorporate dynamic root zone soil moisture up-take down profile as the soil dries It is therefore possiblethat while the modified beta function used in 11LAY doeshelp to capture seasonal water stress as seen across sites inthis study new mechanistic plant hydraulic schemes that cantrack transport of water through the xylem (eg Bonan et al2014 Naudts et al 2015) may be needed when simulatingplant response to prolonged drought periods However com-paring beta functions vs plant hydraulic schemes under se-vere water-stressed periods was not within the scope of thisstudy When discussing woody plant responses to drought itis also worth noting that many LSMs to date are also miss-ing any representation of groundwater (Clark et al 2015)As described in Sect 21 the water table is typically verydeep (tens to hundreds of meters) at these sites Previousmodelling studies have shown that only rather shallow wa-ter tables (sim 1 m) are likely to significantly increase ET inthe SW US (eg by ge 24 mmdminus1 in Fig 4g of Wang et al2018) However the fact that LSMs typically do not includeadequate descriptions of groundwater (and deeper tap roots)could impact their ability to simulate semi-arid ecosystemwater uptake in the dry season given that drought-deciduousshrubs are more resilient to droughts due to their ability toaccess groundwater reserves (eg Miller et al 2010) A newgroundwater module is being developed for ORCHIDEE andwill be tested in future studies

5 Conclusions

These results strongly suggest that a more complex process-based hydrology model ndash in particular one which containsfine-scale discretization of the upper soil moisture layers andassociated improvements in bare soil evaporation and plantwater stress functions ndash improves daily to seasonal predic-tions of the upper-layer root-zone soil moisture dynamics andET (as seen in de Rosnay et al 2002) In particular thereis a dramatic improvement in the modelrsquos ability to capturethe decline in ET during the hot dry MayndashJune period As-sociated changes in the calculations of runoff soil moisture

infiltration and bottom-layer drainage also appear to resultin more plausible (lower) simulations of total runoff (surfacerunoff plus drainage) at the forest sites given that across allthese semi-arid sites most precipitation is accounted for byET at the annual scale Such improvements might counterprevious work highlighting that models tend to overestimaterunoff (Grippa et al 2017)

ORCHIDEE CMIP5 simulations used the two-layer con-ceptual bucket scheme of Manabe (1969) therefore OR-CHIDEE CMIP5 predictions of semi-arid water availabilityand consequent impacts on ecosystem functioning and feed-backs to climate were likely inaccurate Despite the appeal ofsimplicity and low calculation costs two-layer simple buckethydrology models are likely unsuitable for accurate semi-arid water flux simulations (at least in the semi-arid SW US)The forthcoming ORCHIDEE CMIP6 simulations will likelyprovide more accurate and reliable results of semi-arid soilmoisture availability and evapotranspiration

Remaining discrepancies in both overestimated and under-estimated winter and spring soil moisture at high-elevationsemi-arid forested sites might be respectively related to is-sues with soil moisture data during periods of soil freez-ing andor underestimated snowfall forcing data causing alimited duration snowpack with consequent implicationsfor predictions of water availability in regions that rely onspringtime snowmelt However biases in soil moisture atboth the forested sites do not translate into the same biases inmodelled ET suggesting other factors such as issues in ever-green phenology or the lack of resistance to bare soil evapo-ration may also play a role

The addition of an empirical bare soil evaporation resis-tance term by itself did not improve estimates of ET in theseecosystems although TET ratios were increased potentiallyreducing the negative biases in the monsoon season whencomparing to data-derived TET estimates The increase intranspiring leaf area (from a reduction in bare soil fraction) atthe low-elevation forest sites also could account for the samemonsoon season TET bias However issues in the timingof the simulated transition from low to high TET ratios atthe start of the monsoon remain Our analysis shows that re-maining discrepancies in semi-arid site ET simulations (andits constituent fluxes) might therefore be related to a combi-nation of factors impacting both the amount and timing oftranspiring leaf area and resistance to bare soil evaporationWe recommend that future work on improving LSM semi-arid hydrological predictions focuses not only on issues high-lighted in previous studies such as dynamic root zone mois-ture uptake inclusion of ground water lateral and verticalredistribution of moisture (eg Whitley et al 2016 2017Grippa et al 2017) but also on (i) multi-variable calibra-tion of vegetation and hydrology-related parameters acrossall sites (ii) more data to better evaluate the seasonal trajec-tory of LAI across all sites as well as the vegetation frac-tional cover and peak LAI magnitude at low-elevation sites(iii) more data to test modelled snow mass or depth at high-

elevation sites and (iv) testing of a more mechanistic de-scription of resistance to bare soil evaporation

Code availability The ORCHIDEE v20 model code and docu-mentation are publicly available via the ORCHIDEE wiki page(httpforgeipsljussieufrorchideebrowsertagsORCHIDEE_2_0ORCHIDEE Peylin et al 2020) under the CeCILL license(httpwwwcecillinfoindexenhtml CeCILL 2020) The OR-CHIDEE model code is written in Fortran 90 and is maintainedand developed under an SVN version control system at the InstitutePierre Simon Laplace (IPSL) in France Simulation post-processingand plotting scripts were performed in Python and are pro-vided on NMrsquos GitHub repository httpsgithubcomnmacbeanSW-US-Hydro-Model-Eval-HESS (last access 2 November 2020httpsdoiorg105281zenodo4198088 MacBean 2020)

Data availability Meteorological forcing and evapotranspirationdata for each site can be downloaded via the Ameriflux site httpsamerifluxlblgovdatadownload-data (US-SRM Scott 2004andashPresent US-SRG Scott 2008ndashPresent US-Whs Scott 2007ndashPresent US-Wkg Scott 2004bndashPresent US-Fuf Dore and Kolb2006ndash2010 US-Vcp Litvak 2007ndashPresent) Soil moisture wasobtained directly from site PIs Vegetation and soil texture char-acteristics were derived from the published literature as speci-fied in Table 1 and from site PIs Model simulations are pro-vided on NMrsquos GitHub repository httpsgithubcomnmacbeanSW-US-Hydro-Model-Eval-HESS

Supplement The supplement related to this article is available on-line at httpsdoiorg105194hess-24-5203-2020-supplement

Author contributions NM RLS JAB and DJPM designed theoverall study NM carried out the model simulations post-simulation analysis and figure plotting CO NV and AD provideddetailed inputs on model descriptioncode and recommendationsfor further tests to diagnose modelndashdata deficiencies NV providedscripts to gap-fill the meteorological data JAB gap-filled the ETdata RLS JAB TK and ML provided gap-filled soil moisturedata and information on site characteristics and typical behaviourof seasonal vegetation cover LAI and snowfall NM wrote themanuscript All the co-authors provided detailed comments sug-gestions and edits on the first and second drafts of the manuscript

Competing interests The authors declare that they have no conflictof interest

Acknowledgements We would like to thank the ORCHIDEE teamfor developing and maintaining the ORCHIDEE code and for pro-viding the ORCHIDEE version used in this study Finally we thankthe two anonymous referees for their comprehensive and useful re-views

Financial support Funding for AmeriFlux data resources and datacollection at US-SRM US-SRG US-Wkg and US-Whs was pro-vided by the US Department of Energyrsquos Office of Science andthe USDA Data collection at US-Fuf was supported by grantsfrom the North American Carbon ProgramUSDA CREES NRI(grant no 2004-3511115057) the US National Science Founda-tion MRI Program Science Foundation Arizona (grant no CAA 0-203-08) the Arizona Water Institute and the Mission Research Pro-gram School of Forestry Northern Arizona University (McIntire-StennisArizona Bureau of Forestry) The US-Vcp site was fundedby the US DOE Office of Science through the AmeriFlux Man-agement Project (AMP) at Lawrence Berkeley National Labora-tory (award no 7074628) and the Catalina-Jemez Critical Zone Ob-servatory (grant no NSF EAR 1331408) NM was funded by USNational Science Foundation Award Numbers 1065790 (EmergingFrontiers Program) and 1754430 (Division of Environmental Biol-ogy Ecosystems Program)

Review statement This paper was edited by Anke Hildebrandt andreviewed by two anonymous referees

References

Abramowitz G Leuning R Clark M and Pitman A Evaluatingthe performance of land surface models J Climate 21 5468ndash5481 2008

Allen C D Chapter 4 ndash Forest ecosystem reorganization underwayin the Southwestern US A preview of widespread forest changesin the Anthropocene in Forest Conservation and Managementin the Anthropocene Adaptation of Science Policy and Practiceedited by Bixler R P and Miller C University Press of Col-orado Boulder Colorado 57ndash79 2016

Anderson-Teixeira K J Delong J P Fox A M Brese D Aand Litvak M E Differential responses of production and res-piration to temperature and moisture drive the carbon balanceacross a climatic gradient in New Mexico Glob Change Biol17 410ndash424 httpsdoiorg101111j1365-2486201002269x2011

Archer S R and Predick K I Climate Change and Ecosystemsof the Southwestern United States Rangelands 30 23ndash28httpsdoiorg1021111551-501x(2008)30[23ccaeot]20co22008

Ault T R Cole J E Overpeck J T Pederson G T and MekoD M Assessing the Risk of Persistent Drought Using ClimateModel Simulations and Paleoclimate Data J Climate 27 7529ndash7549 httpsdoiorg101175jcli-d-12-002821 2014

Ault T R Mankin J S Cook B I and Smerdon J E Relativeimpacts of mitigation temperature and precipitation on 21st-century megadrought risk in the American Southwest ScienceAdvances 2 e1600873 httpsdoiorg101126sciadv16008732016

Baldocchi D D Ma S Rambal S Misson L Ourcival J-MLimousin J-M Pereira J and Papale D On the differentialadvantages of evergreenness and deciduousness in mediterraneanoak woodlands a flux perspective Ecol Appl 20 1583ndash1597httpsdoiorg10189008-20471 2010

Belnap J Weber B and Buumldel B Biological Soil Crustsan Organizing Principle in Drylands in Biological soil crustsan organizing principle in drylands edited by Weber BBuumldel B and Belnap J Springer Cham Switzerland 3ndash13httpsdoiorg101007978-3-319-30214-0 2016

Berg A Findell K Lintner B Giannini A Seneviratne Svan den Hurk B Lorenz R Pitman A Hagemann S MeierA Cheruy F Ducharne A Malyshev S and Milly P CD Land-atmosphere feedbacks amplify aridity increase overland under global warning Nat Clim Change 6 869ndash874httpsdoiorg101038nclimate3029 2016

Best M J Abramowitz G Johnson H R Pitman A J Bal-samo G Boone A Cuntz M Decharme B Dirmeyer PA Dong J Ek M Guo Z Haverd V Van Den Hurk BJ J Nearing G S Pak B Peters-Lidard C Santanello J AStevens L and Vuichard N The Plumbing of Land SurfaceModels Benchmarking Model Performance J Hydrometeorol16 1425ndash1442 httpsdoiorg101175jhm-d-14-01581 2015

Biederman J A Scott R L Goulden M L Vargas RLitvak M E Kolb T E Yepez E A Oechel W CBlanken P D Bell T W Garatuza-Payan J Maurer GE Dore S and Burns S P Terrestrial carbon balancein a drier world the effects of water availability in south-western North America Global Change Biol 22 1867ndash1879httpsdoiorg101111gcb13222 2016

Biederman J A Scott R L Bell T W Bowling D R DoreS Garatuza-Payan J Kolb T E Krishnan P Krofcheck DJ Litvak M E Maurer G E Meyers T P Oechel W CPapuga S A Ponce-Campos G E Rodriguez J C Smith WK Vargas R Watts C J Yepez E A and Goulden M LCO2 exchange and evapotranspiration across dryland ecosystemsof southwestern North America Glob Change Biol 23 4204ndash4221 httpsdoiorg101111gcb13686 2017

Bierkens M F P Global hydrology 2015 State trendsand directions Water Resour Res 51 4923ndash4947httpsdoiorg1010022015wr017173 2015

Bonan G B Williams M Fisher R A and Oleson K WModeling stomatal conductance in the earth system linkingleaf water-use efficiency and water transport along the soilndashplantndashatmosphere continuum Geosci Model Dev 7 2193ndash2222 httpsdoiorg105194gmd-7-2193-2014 2014

Boone A de Rosnay P D Balsamo G Beljaars A ChopinF Decharme B Delire C Ducharne A Gascoin S GrippaM Guichard F Gusev Y Harris P Jarlan L Kergoat LMougin E Nasonova O Norgaard A Orgeval T Ottleacute CPoccard-Leclercq I Polcher J Sandholt I Saux-Picart STaylor C and Xue Y The AMMA Land Surface Model Inter-comparison Project (ALMIP) B Am Meteorol Soc 90 1865ndash1880 httpsdoiorg1011752009bams27861 2009

Botta A Viovy N Ciais P Friedlingstein P and Monfray PA global prognostic scheme of leaf onset using satellite dataGlobal Change Biol 6 709ndash725 httpsdoiorg101046j1365-2486200000362x 2000

Carsel R F and Parrish R S Developing joint probability dis-tributions of soil water retention characteristics Water ResourRes 24 755ndash769 httpsdoiorg101029wr024i005p007551988

CeCILL httpscecillinfoindexenhtml last access 2 Novem-ber 2020

Chang L-L Dwivedi R Knowles J F Fang Y-H NiuG-Y Pelletier J D Rasmussen C Durcik M Barron-Gafford G A and Meixner T Why Do Large-Scale LandSurface Models Produce a Low Ratio of Transpiration toEvapotranspiration J Geophys Res-Atmos 123 9109ndash9130httpsdoiorg1010292018jd029159 2018

Chen X Maignan F Viovy N Bastos A Goll D Wu J LiuL Yue C Peng S Yuan W Conceiccedilatildeo A C OrsquoSullivanM and Ciais P Novel Representation of Leaf Phenology Im-proves Simulation of Amazonian Evergreen Forest Photosyn-thesis in a Land Surface Model J Adv Model Earth Sy 12e2018MS001565 httpsdoiorg1010292018ms001565 2020

Choisnel E M Jourdain S V and Jacquart C J Climato-logical evaluation of some fluxes of the surface energy andsoil water balances over France Ann Geophys 13 666ndash674httpsdoiorg101007s00585-995-0666-y 1995

Chubb T Manton M J Siems S T Peace A D and Bil-ish S P Estimation of Wind-Induced Losses from a Precip-itation Gauge Network in the Australian Snowy Mountains JHydrometeorol 16 2619ndash2638 httpsdoiorg101175jhm-d-14-02161 2015

Clark M P Fan Y Lawrence D M Adam J C BolsterD Gochis D J Hooper R P Kumar M Leung L RMackay D S Maxwell R M Shen C Swenson S C andZeng X Improving the representation of hydrologic processesin Earth System Models Water Resour Res 51 5929ndash5956httpsdoiorg1010022015wr017096 2015

Cook B I Ault T R and Smerdon J E Unprece-dented 21st century drought risk in the American South-west and Central Plains Science Advances 1 e1400082httpsdoiorg101126sciadv1400082 2015

De Kauwe M G Taylor C M Harris P P Weedon G P andEllis R J Quantifying Land Surface Temperature Variabilityfor Two Sahelian Mesoscale Regions during the Wet Season JHydrometeorol 14 1605ndash1619 httpsdoiorg101175jhm-d-12-01411 2013

De Kauwe M G Zhou S-X Medlyn B E Pitman A JWang Y-P Duursma R A and Prentice I C Do landsurface models need to include differential plant species re-sponses to drought Examining model predictions across amesic-xeric gradient in Europe Biogeosciences 12 7503ndash7518httpsdoiorg105194bg-12-7503-2015 2015

De Kauwe M G Medlyn B E Walker A P Zaehle S AsaoS Guenet B Harper A B Hickler T Jain A K Luo YLu X Luus K Parton W J Shu S Wang Y P Werner CXia J Pendall E Morgan J A Ryan E M Carrillo Y Di-jkstra F A Zelikova T J and Norby R J Challenging terres-trial biosphere models with data from the long-term multifactorPrairie Heating and CO2 Enrichment experiment Global ChangeBiol 23 3623ndash3645 httpsdoiorg101111gcb13643 2017

de Rosnay P and Polcher J Modelling root water uptake in a com-plex land surface scheme coupled to a GCM Hydrol Earth SystSci 2 239ndash255 httpsdoiorg105194hess-2-239-1998 1998

de Rosnay P Bruen M and Polcher J Sensitivity of sur-face fluxes to the number of layers in the soil modelused in GCMs Geophys Res Lett 27 3329ndash3332httpsdoiorg1010292000gl011574 2000

de Rosnay P Polcher J Bruen M and Laval K Im-pact of a physically based soil water flow and soil-

plant interaction representation for modeling large-scale landsurface processes J Geophys Res-Atmos 107 4118httpsdoiorg1010292001jd000634 2002

Decker M Or D Pitman A and Ukkola A New turbulent resis-tance parameterization for soil evaporation based on a pore-scalemodel Impact on surface fluxes in CABLE J Adv Model EarthSy 9 220ndash238 httpsdoiorg1010022016ms000832 2017

Diffenbaugh N S Giorgi F and Pal J S Climate changehotspots in the United States Geophys Res Lett 35 L16709httpsdoiorg1010292008gl035075 2008

Dirmeyer P A A History and Review of the Global SoilWetness Project (GSWP) J Hydrometeorol 12 729ndash749httpsdoiorg101175JHM-D-10-050101 2011

Donat M G Lowry A L Alexander L V OrsquoGormanP A and Maher N More extreme precipitation in theworldrsquos dry and wet regions Nat Clim Change 6 508ndash513httpsdoiorg101038nclimate2941 2016

Dore S and Kolb T AmeriFlux US-Fuf Flagstaff ndash UnmanagedForest Dataset httpsdoiorg1017190AMF1246051 2006ndash2010

Dore S Kolb T E Montes-Helu M Eckert S E Sullivan BW Hungate B A Kaye J P Hart S C Koch G W andFinkral A Carbon and water fluxes from ponderosa pine forestsdisturbed by wildfire and thinning Ecol Appl 20 663ndash683httpsdoiorg10189009-09341 2010

Dore S Montes-Helu M Hart S C Hungate B A Koch GW Moon J B Finkral A J and Kolb T E Recovery ofponderosa pine ecosystem carbon and water fluxes from thinningand stand-replacing fire Global Change Biol 18 3171ndash3185httpsdoiorg101111j1365-2486201202775x 2012

Druel A Peylin P Krinner G Ciais P Viovy N Pere-gon A Bastrikov V Kosykh N and Mironycheva-TokarevaN Towards a more detailed representation of high-latitudevegetation in the global land surface model ORCHIDEE(ORC-HL-VEGv10) Geosci Model Dev 10 4693ndash4722httpsdoiorg105194gmd-10-4693-2017 2017

Ducharne A Laval K and Polcher J Sensitivity ofthe hydrological cycle to the parametrization of soilhydrology in a GCM Clim Dynam 14 307ndash327httpsdoiorg101007s003820050226 1998

Ducharne A Ghattas J Maignan F Ottleacute C Vuichard NGuimberteau M Krinner G Polcher J Tafasca S BastrikovV Cheruy F Gueacutenet B Mizuochi H Peylin P Tootchi Aand Wang F Soil water processes in the ORCHIDEE-20 landsurface model state of the art for CMIP6 in preparation GeosciModel Dev 2020

Ducoudreacute N I Laval K and Perrier A SECHIBAa New Set of Parameterizations of the HydrologicExchanges at the Land-Atmosphere Interface withinthe LMD Atmospheric General Circulation ModelJ Climate 6 248ndash273 httpsdoiorg1011751520-0442(1993)006lt0248sansopgt20co2 1993

Dufresne J-L Foujols M-A Denvil S Caubel A Marti OAumont O Balkanski Y Bekki S Bellenger H BenshilaR Bony S Bopp L Braconnot P Brockmann P CaduleP Cheruy F Codron F Cozic A Cugnet D Noblet ND Duvel J-P Etheacute C Fairhead L Fichefet T FlavoniS Friedlingstein P Grandpeix J-Y Guez L Guilyardi EHauglustaine D Hourdin F Idelkadi A Ghattas J Jous-

saume S Kageyama M Krinner G Labetoulle S Lahel-lec A Lefebvre M-P Lefevre F Levy C Li Z X LloydJ Lott F Madec G Mancip M Marchand M Masson SMeurdesoif Y Mignot J Musat I Parouty S Polcher J RioC Schulz M Swingedouw D Szopa S Talandier C TerrayP Viovy N and Vuichard N Climate change projections us-ing the IPSL-CM5 Earth System Model from CMIP3 to CMIP5Clim Dynam 40 2123ndash2165 httpsdoiorg101007s00382-012-1636-1 2013

drsquoOrgeval T Polcher J and de Rosnay P Sensitivity ofthe West African hydrological cycle in ORCHIDEE to in-filtration processes Hydrol Earth Syst Sci 12 1387ndash1401httpsdoiorg105194hess-12-1387-2008 2008

Falge E Baldocchi D Olson R Anthoni P Aubinet MBernhofer C Burba G Ceulemans R Clement R Dol-man H Granier A Gross P Gruumlnwald T Hollinger DJensen N-O Katul G Keronen P Kowalski A Lai CT Law B E Meyers T Moncrieff J Moors E WilliamMunger J Pilegaard K Rannik Uuml Rebmann C SuykerA Tenhunen J Tu K Verma S Vesala T Wilson Kand Wofsy S Gap filling strategies for defensible annual sumsof net ecosystem exchange Agr Forest Meteorol 107 43ndash69httpsdoiorg101016s0168-1923(00)00225-2 2001

Fang H Jiang C Li W Wei S Baret F Chen JM Garcia-Haro J Liang S Liu R Myneni RB and Pinty B Charac-terization and intercomparison of global moderate resolution leafarea index (LAI) products Analysis of climatologies and the-oretical uncertainties J Geophys Res-Biogeo 118 529ndash5482013

Ferrenberg S and Reed S C Biocrust ecology unifying micro-and macro-scales to confront global change New Phytol 216643ndash646 httpsdoiorg101111nph14826 2017

Garrigues S Lacaze R Baret F J T M Morisette J T WeissM Nickeson J E Fernandes R Plummer S Shabanov NV Myneni R B and Knyazikhin Y Validation and inter-comparison of global Leaf Area Index products derived fromremote sensing data J Geophys Res-Biogeo 113 G02028httpsdoiorg1010292007JG000635 2008

Green J K Seneviratne S I Berg A M Findell K L Hage-mann S Lawrence D M and Gentine P Large influence ofsoil moisture on long-term terrestrial carbon uptake Nature 565476ndash479 httpsdoiorg101038s41586-018-0848-x 2019

Gremer J R Bradford J B Munson S M and Duni-way M C Desert grassland responses to climate and soilmoisture suggest divergent vulnerabilities across the south-western United States Global Change Biol 21 4049ndash4062httpsdoiorg101111gcb13043 2015

Grippa M Kergoat L Frappart F Araud Q Boone Ade Rosnay P D Lemoine J-M Gascoin S BalsamoG Ottleacute C Decharme B Saux-Picart S and RamillienG Land water storage variability over West Africa esti-mated by Gravity Recovery and Climate Experiment (GRACE)and land surface models Water Resour Res 47 W05549httpsdoiorg1010292009wr008856 2011

Grippa M Kergoat L Boone A Peugeot C Demarty JCappelaere B Gal L Hiernaux P Mougin E DucharneA Dutra E Anderson M and Hain C Modeling SurfaceRunoff and Water Fluxes over Contrasted Soils in the PastoralSahel Evaluation of the ALMIP2 Land Surface Models over

the Gourma Region in Mali J Hydrometeorol 18 1847ndash1866httpsdoiorg101175jhm-d-16-01701 2017

Guimberteau M Drapeau G Ronchail J Sultan B Polcher JMartinez J-M Prigent C Guyot J-L Cochonneau G Es-pinoza J C Filizola N Fraizy P Lavado W De OliveiraE Pombosa R Noriega L and Vauchel P Discharge sim-ulation in the sub-basins of the Amazon using ORCHIDEEforced by new datasets Hydrol Earth Syst Sci 16 911ndash935httpsdoiorg105194hess-16-911-2012 2012a

Guimberteau M Perrier A Laval K and Polcher J A compre-hensive approach to analyze discrepancies between land surfacemodels and in-situ measurements a case study over the US andIllinois with SECHIBA forced by NLDAS Hydrol Earth SystSci 16 3973ndash3988 httpsdoiorg105194hess-16-3973-20122012b

Guimberteau M Ronchail J Espinoza J C Lengaigne M Sul-tan B Polcher J Drapeau G Guyot J-L Ducharne A andCiais P Future changes in precipitation and impacts on extremestreamflow over Amazonian sub-basins Environ Res Lett 8014035 httpsdoiorg1010881748-932681014035 2013

Guimberteau M Ducharne A Ciais P Boisier J P PengS De Weirdt M and Verbeeck H Testing conceptual andphysically based soil hydrology schemes against observationsfor the Amazon Basin Geosci Model Dev 7 1115ndash1136httpsdoiorg105194gmd-7-1115-2014 2014

Gupta H V Kling H Yilmaz K K and Martinez G F Decom-position of the mean squared error and NSE performance criteriaImplications for improving hydrological modelling J Hydrol377 80ndash91 httpsdoiorg101016jjhydrol200908003 2009

Harris C R Millman K J van der Walt S J Gommers R Vir-tanen P Cournapeau D Wieser E Taylor J Berg S SmithN J and Kern R Array programming with NumPy Nature585 357ndash362 2020

Haverd V Ahlstroumlm A Smith B and Canadell J GCarbon cycle responses of semi-arid ecosystems to positiveasymmetry in rainfall Global Change Biol 23 793ndash800httpsdoiorg101111gcb13412 2016

Hogue T S Bastidas L Gupta H Sorooshian S Mitchell Kand Emmerich W Evaluation and Transferability of the NoahLand Surface Model in Semiarid Environments J Hydrometeo-rol 6 68ndash84 httpsdoiorg101175jhm-4021 2005

Huang J Yu H Dai A Wei Y and Kang L Drylands facepotential threat under 2 C global warming target Nat ClimChange 7 417ndash422 httpsdoiorg101038nclimate32752017

Humphrey V Zscheischler J Ciais P Gudmundsson L SitchS and Seneviratne S I Sensitivity of atmospheric CO2 growthrate to observed changes in terrestrial water storage Nature 560628ndash631 httpsdoiorg101038s41586-018-0424-4 2018

Hunter J D Matplotlib A 2D graphics environment Comput SciEng 9 90ndash95 2007

IPCC Climate Change 2013 The Physical Science Basis Con-tribution of Working Group I to the Fifth Assessment Reportof the Intergovernmental Panel on Climate Change edited byStocker T F Qin D Plattner G-K Tignor M Allen S KBoschung J Nauels A Xia Y Bex V and Midgley P MCambridge University Press Cambridge UK and New York NYUSA 1535 pp 2013

Keenan T Sabate S and Gracia C The importance of mes-ophyll conductance in regulating forest ecosystem productivityduring drought periods Global Change Biol 16 1019ndash1034httpsdoiorg101111j1365-2486200902017x 2010

Koster R D Dirmeyer P A Guo Z Bonan G Chan E CoxP Gordon C T Kanae S Kowalczyk E Lawrence D LiuP Lu C-H Malyshev S McAvaney B Mitchell K MockoD Oki T Oleson K Pitman A Sud Y C Taylor C MVerseghy D Vasic R Xue Y Yamada T and GLACE TeamRegions of strong coupling between soil moisture and precipita-tion Science 305 1138ndash1140 2004

Krinner G Viovy N Noblet-Ducoudreacute N D Ogeacutee J PolcherJ Friedlingstein P Ciais P Sitch S and Prentice I CA dynamic global vegetation model for studies of the cou-pled atmosphere-biosphere system Global Biogeochem Cy 19GB1015 httpsdoiorg1010292003gb002199 2005

Lian X Piao S Huntingford C Li Y Zeng Z Wang XCiais P Mcvicar T R Peng S Ottleacute C Yang H YangY Zhang Y and Wang T Partitioning global land evapotran-spiration using CMIP5 models constrained by observations NatClim Change 8 640ndash646 httpsdoiorg101038s41558-018-0207-9 2018

Litvak M AmeriFlux US-Vcp Valles Caldera Ponderosa PineDataset httpsdoiorg1017190AMF1246122 2007ndashPresent

Lohou F Kergoat L Guichard F Boone A Cappelaere BCohard J-M Demarty J Galle S Grippa M Peugeot CRamier D Taylor C M and Timouk F Surface responseto rain events throughout the West African monsoon AtmosChem Phys 14 3883ndash3898 httpsdoiorg105194acp-14-3883-2014 2014

MacBean N Maignan F Peylin P Bacour C Breacuteon F-M andCiais P Using satellite data to improve the leaf phenology ofa global terrestrial biosphere model Biogeosciences 12 7185ndash7208 httpsdoiorg105194bg-12-7185-2015 2015

Maestre F T Salguero-Gomez R and Quero J L It is gettinghotter in here determining and projecting the impacts of globalenvironmental change on drylands Philos T R Soc B 3673062ndash3075 httpsdoiorg101098rstb20110323 2012

Manabe S Climate And The Ocean Circulation1 MonWeather Rev 97 739ndash774 httpsdoiorg1011751520-0493(1969)097lt0739catocgt23co2 1969

MacBean N nmacbeanSW-US-Hydro-Model-Eval-HESS v1Zenodo httpsdoiorg105281zenodo4198088 2020

Medlyn B E Kauwe M G D Zaehle S Walker A P Du-ursma R A Luus K Mishurov M Pak B Smith B WangY-P Yang X Crous K Y Drake J E Gimeno T E Mac-donald C A Norby R J Power S A Tjoelker M G andEllsworth D S Using models to guide field experimentsa pri-oripredictions for the CO2 response of a nutrient- and water-limited native Eucalypt woodland Global Change Biol 22(8)2834ndash2851 httpsdoiorg101111gcb13268 2016

Mermoud A and Xu D Comparative analysis of three methodsto generate soil hydraulic functions Soil Till Res 87 89ndash100httpsdoiorg101016jstill200502034 2006

Miller G R Chen X Rubin Y Ma S and Baldoc-chi D D Groundwater uptake by woody vegetation ina semiarid oak savanna Water Resour Res 46 W10503httpsdoiorg1010292009wr008902 2010

Mualem Y A new model for predicting the hydraulic conductivityof unsaturated porous media Water Resour Res 12 513ndash522httpsdoiorg101029wr012i003p00513 1976

Mueller B and Seneviratne S I Systematic land climate andevapotranspiration biases in CMIP5 simulations Geophys ResLett 41 128ndash134 httpsdoiorg1010022013gl058055 2014

Naudts K Ryder J McGrath M J Otto J Chen Y ValadeA Bellasen V Berhongaray G Boumlnisch G Campioli MGhattas J De Groote T Haverd V Kattge J MacBeanN Maignan F Merilauml P Penuelas J Peylin P Pinty BPretzsch H Schulze E D Solyga D Vuichard N Yan Yand Luyssaert S A vertically discretised canopy description forORCHIDEE (SVN r2290) and the modifications to the energywater and carbon fluxes Geosci Model Dev 8 2035ndash2065httpsdoiorg105194gmd-8-2035-2015 2015

Niu G Y and Yang Z L An observation-based formulationof snow cover fraction and its evaluation over large NorthAmerican river basins J Geophys Res-Atmos 112 D21101httpsdoiorg1010292007JD008674 2007

Peylin P Ghattas J Cadule P Cheruy F Ducharne AGuenet B Lathiegravere J Luyssaert S Maignan F MaugisP Ottle C Polcher J Viovy N Vuichard N BastrikovV Guimberteau M Lanso A-S MacBean N McgrathM Tafasca S and Wang F The global land surface modelORCHIDEE ndash Tag20 available at httpforgeipsljussieufrorchideebrowsertagsORCHIDEE_2_0ORCHIDEE last ac-cess 19 October 2020

Raoult N Delorme B Ottleacute C Peylin P Bastrikov V MaugisP and Polcher J Confronting Soil Moisture Dynamics fromthe ORCHIDEE Land Surface Model With the ESA-CCI Prod-uct Perspectives for Data Assimilation Remote Sens-Basel 101786 httpsdoiorg103390rs10111786 2018

Rasmussen R Baker B Kochendorfer J Meyers T LandoltS Fischer A P Black J Theacuteriault J M Kucera P GochisD Smith C Nitu R Hall M Ikeda K and Gutmann EHow Well Are We Measuring Snow The NOAAFAANCARWinter Precipitation Test Bed B Am Meteorol Soc 93 811ndash829 httpsdoiorg101175bams-d-11-000521 2012

Reynolds C A Jackson T J and Rawls W J Estimating soilwater-holding capacities by linking the Food and Agriculture Or-ganization Soil map of the world with global pedon databasesand continuous pedotransfer functions Water Resour Res 363653ndash3662 httpsdoiorg1010292000wr900130 2000

Richards L A Capillary Conduction Of LiquidsThrough Porous Mediums Physics 1 318ndash333httpsdoiorg10106311745010 1931

Saux-Picart S Ottleacute C Perrier A Decharme B CoudertB Zribi M Boulain N Cappelaere B and Ramier DSEtHyS_Savannah A multiple source land surface model ap-plied to Sahelian landscapes Agr Forest Meteorol 149 1421ndash1432 2009

Scanlon B R Zhang Z Save H Sun A Y Schmied H MBeek L P H V Wiese D N Wada Y Long D Reedy R CLonguevergne L Doumlll P and Bierkens M F P Global modelsunderestimate large decadal declining and rising water storagetrends relative to GRACE satellite data P Natl Acad Sci USA115 E1080ndashE1089 httpsdoiorg101073pnas17046651152018

Scanlon B R Zhang Z Rateb A Sun A Wiese D Save HBeaudoing H Lo M H Muumlller-Schmied H Doumlll P BeekR Swenson S Lawrence D Croteau M and Reedy R CTracking Seasonal Fluctuations in Land Water Storage UsingGlobal Models and GRACE Satellites Geophys Res Lett 465254ndash5264 httpsdoiorg1010292018gl081836 2019

Scott R AmeriFlux US-SRM Santa Rita Mesquite Datasethttpsdoiorg1017190AMF1246104 2004andashPresent

Scott R AmeriFlux US-Wkg Walnut Gulch Kendall Grass-lands Dataset httpsdoiorg1017190AMF1246112 2004bndashPresent

Scott R AmeriFlux US-Whs Walnut Gulch Lucky Hills ShrubDataset httpsdoiorg1017190AMF1246113 2007ndashPresent

Scott R AmeriFlux US-SRG Santa Rita Grassland Datasethttpsdoiorg1017190AMF1246154 2008ndashPresent

Scott R L and Biederman J A Partitioning evapo-transpiration using long-term carbon dioxide and wa-ter vapor fluxes Geophys Res Lett 44 6833ndash6840httpsdoiorg1010022017gl074324 2017

Scott R L and Biederman J A Critical Zone Water Balance Over13 Years in a Semiarid Savanna Water Resour Res 55 574ndash588 httpsdoiorg1010292018wr023477 2019

Scott R L Biederman J A Hamerlynck E P and Barron-Gafford G A The carbon balance pivot point of south-western US semiarid ecosystems Insights from the 21stcentury drought J Geophys Res-Biogeo 120 2612ndash2624httpsdoiorg1010022015jg003181 2015

Seager R and Vecchi G A Greenhouse warming andthe 21st century hydroclimate of southwestern NorthAmerica P Natl Acad Sci USA 107 21277ndash21282httpsdoiorg101073pnas0910856107 2010

Seager R Ting M Held I Kushnir Y Lu J Vecchi G HuangH-P Harnik N Leetmaa A Lau N-C Li C Velez Jand Naik N Model Projections of an Imminent Transition toa More Arid Climate in Southwestern North America Science316 1181ndash1184 httpsdoiorg101126science1139601 2007

Sellers P J Heiser M D and Hall F G Relations between sur-face conductance and spectral vegetation indices at intermediate(100 m2 to 15 km2) length scales J Geophys Res 97 19033httpsdoiorg10102992jd01096 1992

Seneviratne S I Wilhelm M Stanelle T Hurk B Hage-mann S Berg A Cheruy F Higgins M E Meier ABrovkin V Claussen M Ducharne A Dufresne J LFindell K L Ghattas J Lawrence D M Malyshev SRummukainen M and Smith B Impact of soil moisture-climate feedbacks on CMIP5 projections First results from theGLACE-CMIP5 experiment Geophys Res Lett 40 5212ndash5217 httpsdoiorg101002grl50956 2013

Sippel S Zscheischler J Heimann M Lange H MahechaM D van Oldenborgh G J Otto F E L and ReichsteinM Have precipitation extremes and annual totals been increas-

ing in the worldrsquos dry regions over the last 60 years HydrolEarth Syst Sci 21 441ndash458 httpsdoiorg105194hess-21-441-2017 2017

Smith S D Monson R K and Anderson J E Physiologi-cal Ecology of North American Desert Plants Springer-VerlagBerlin Heidelberg 1997

Su Z Schmugge T Kustas W P and Massman W J An Evalu-ation of Two Models for Estimation of the Roughness Height forHeat Transfer between the Land Surface and the Atmosphere JAppl Meteorol 40 1933ndash1951 httpsdoiorg1011751520-0450(2001)040lt1933aeotmfgt20co2 2001

Swenson S C and Lawrence D M Assessing a drysurface layer-based soil resistance parameterization for theCommunity Land Model using GRACE and FLUXNET-MTE data J Geophys Res-Atmos 119 10299ndash10312httpsdoiorg1010022014jd022314 2014

Tietjen B Jeltsch F Zehe E Classen N Groengroeft A Schif-fers K and Oldeland J Effects of climate change on the cou-pled dynamics of water and vegetation in drylands Ecohydrol-ogy 3 226ndash237 httpsdoiorg101002eco70 2010

Ukkola A M Kauwe M G D Pitman A J Best M JAbramowitz G Haverd V Decker M and Haughton NLand surface models systematically overestimate the intensityduration and magnitude of seasonal-scale evaporative droughtsEnviron Res Lett 11 104012 httpsdoiorg1010881748-93261110104012 2016a

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 httpsdoiorg105194hess-20-2403-2016 2016b

van Genuchten M T A Closed-form Equation forPredicting the Hydraulic Conductivity of Unsatu-rated Soils Soil Sci Soc Am J 44 892ndash898httpsdoiorg102136sssaj198003615995004400050002x1980

Vuichard N and Papale D Filling the gaps in meteoro-logical continuous data measured at FLUXNET sites withERA-Interim reanalysis Earth Syst Sci Data 7 157ndash171httpsdoiorg105194essd-7-157-2015 2015

Wang T Ottleacute C Boone A Ciais P Brun E Morin SKrinner G Piao S and Peng S Evaluation of an im-proved intermediate complexity snow scheme in the ORCHIDEEland surface model J Geophys Res-Atmos 118 6064ndash6079httpsdoiorg101002jgrd50395 2013

Wang F Ducharne A Cheruy F Lo M-H and Grandpeix J-Y Impact of a shallow groundwater table on the global watercycle in the IPSL landndashatmosphere coupled model Clim Dy-nam 50 3505ndash3522 httpsdoiorg101007s00382-017-3820-9 2018

Waskom M Botvinnik O OrsquoKane D Hobson P LukauskasS Gemperline D C Augspurger T HalchenkoY Cole J B Warmenhoven J and de Ruiter Jmwaskomseaborn v0 81 (September 2017) Zenodohttpsdoiorg105281zenodo883859 2017

Wei Z Yoshimura K Wang L Miralles D G Jasechko S andLee X Revisiting the contribution of transpiration to global ter-restrial evapotranspiration Geophys Res Lett 44 2792ndash2801httpsdoiorg1010022016gl072235 2017

Whitley R Beringer J Hutley L B Abramowitz G De KauweM G Duursma R Evans B Haverd V Li L Ryu YSmith B Wang Y-P Williams M and Yu Q A modelinter-comparison study to examine limiting factors in modellingAustralian tropical savannas Biogeosciences 13 3245ndash3265httpsdoiorg105194bg-13-3245-2016 2016

Whitley R Beringer J Hutley L B Abramowitz G De KauweM G Evans B Haverd V Li L Moore C Ryu Y ScheiterS Schymanski S J Smith B Wang Y-P Williams Mand Yu Q Challenges and opportunities in land surface mod-elling of savanna ecosystems Biogeosciences 14 4711ndash4732httpsdoiorg105194bg-14-4711-2017 2017

Zhou S Duursma R A Medlyn B E Kelly J W andPrentice I C How should we model plant responses todrought An analysis of stomatal and non-stomatal responsesto water stress Agr Forest Meteorol 182ndash183 204ndash214httpsdoiorg101016jagrformet201305009 2013

Zhou S Medlyn B Sabateacute S Sperlich D PrenticeI C and Whitehead D Short-term water stress im-pacts on stomatal mesophyll and biochemical limitationsto photosynthesis differ consistently among tree speciesfrom contrasting climates Tree Physiol 34 1035ndash1046httpsdoiorg101093treephystpu072 2014

Zhou S Yu B Zhang Y Huang Y and Wang G Parti-tioning evapotranspiration based on the concept of underly-ing water use efficiency Water Resour Res 52 1160ndash1175httpsdoiorg1010022015wr017766 2016

Abstract

Introduction

Methods and data

South-western US study sites

ORCHIDEE land surface model

General model description

Soil hydrology

Bare soil evaporation and additional resistance term

Empirical plant water stress function

Snow scheme

Data

Site-level meteorological and eddy covariance data and processing

Soil moisture data and processing

Simulation set-up and post-processing

Results

Differences between the 2LAY and 11LAY model versions for main hydrological stores and fluxes

Comparison of 11LAY soil moisture against observations at different depths

Remaining discrepancies in ET and its component fluxes

Testing decreased bare soil cover and the addition of the 11LAY bare soil resistance term

Discussion

Issues with modelling vegetation dynamics in semi-arid ecosystems

ET partitioning (TET ratio)

Bare soil evaporation

High-elevation model snowpack and snowmelt predictions

Implications for modelling plant water stress

Conclusions

Code availability

Data availability

Supplement

Author contributions

Competing interests

Acknowledgements

Financial support

Review statement

References

were generally lower than estimated values across all sitesparticularly during the monsoon season Adding a soil re-sistance term generally decreased simulated bare soil evapo-ration E and increased soil moisture content thus increas-ing transpiration T and reducing the negative bias betweenmodelled and estimated monsoon TET ratios This negativebias could also be accounted for at the low-elevation sites bydecreasing the model bare soil fraction thus increasing theamount of transpiring leaf area However adding the baresoil resistance term and decreasing the bare soil fraction bothdegraded the model fit to ET observations Furthermore re-maining discrepancies in the timing of the transition fromminimum TET ratios during the hot dry MayndashJune periodto high values at the start of the monsoon in JulyndashAugustmay also point towards incorrect modelling of leaf phenol-ogy and vegetation growth in response to monsoon rains Weconclude that a discretized soil hydrology scheme and associ-ated developments improve estimates of ET by allowing themodelled upper-layer soil moisture to more closely match thepulse precipitation dynamics of these semi-arid ecosystemshowever the partitioning of T from E is not solved by thismodification alone

1 Introduction

Semi-arid ecosystems ndash which cover sim 40 of the Earthrsquosterrestrial surface and which include rangelands shrublandsgrasslands savannas and seasonally dry forests ndash are inzones of transition between humid and arid climates and arecharacterized by sparse patchy vegetation cover and lim-ited water availability Moisture availability in these ecosys-tems is therefore a major control on the complex interac-tions between vegetation dynamics and surface energy wa-ter and carbon exchange (Biederman et al 2017 Haverdet al 2016) Given the sensitivity to water availability semi-arid ecosystem functioning may be particularly vulnerable toprojected changes in climate (Tietjen et al 2010 Maestreet al 2012 Gremer et al 2015) IPCC Earth system model(ESM) projections and observation-based datasets indicatethese regions will likely experience more intense warmingand droughts increases in extreme rainfall events and agreater contrast between wet and dry seasons in the future(IPCC 2013 Donat et al 2016 Sippel et al 2017 Huanget al 2017)

To simulate the impact of climate change on semi-aridecosystem functioning it is essential that the land sur-face model (LSM) component of ESMs accurately repre-sent semi-arid water flux and storage budgets (and all asso-ciated processes) In the last 2 to 3 decades LSM groupshave progressively updated their hydrology schemes fromthe more simplistic ldquobucketrdquo-type models included in earlierversions (Manabe 1969) The resulting schemes typically in-clude more physically based representations of vertical diffu-

sion of water in unsaturated soils (Clark et al 2015) In ad-dition to increasing the complexity of soil hydrology severalstudies have attempted to address the issue that models tendto miscalculate partitioning of evapotranspiration (ET) intotranspiration (T ) and bare soil evaporation (E) with modelssystematically underestimating TET ratios (Wei et al 2017Chang et al 2018) One such mechanism that models haveintroduced is an evaporation resistance term that reduces therate of water evaporation from bare soil surfaces (Swensonand Lawrence 2014 Decker et al 2017) The developmentof these more mechanistic soil hydrology schemes shouldmean that LSMs better capture high temporal frequency toseasonal and long-term temporal variability of water storesand fluxes However it is not always apparent that increasingmodel complexity provides more accurate representations ofreality (as encapsulated by observations of different vari-ables at multiple spatio-temporal scales) Further increasingmodel complexity comes at a cost of increased computationalresources and unknown parameters Therefore it is imper-ative that we test models of increasing complexity againstmultiple types of observations at a variety of sites represent-ing different ecosystem types

New-generation LSM water flux and storage estimateshave been extensively tested at multiple scales from the sitelevel to the globe (Abramowitz et al 2008 Dirmeyer 2011Guimberteau et al 2014 Mueller and Senerviratne 2014Best et al 2015 Ukkola et al 2016b Raoult et al 2018Scanlon et al 2018 2019) Modelndashdata biases are observedacross all biomes however a key finding common to thesestudies is that models do not capture seasonal to inter-annualwater stores and fluxes well during dry periods andor atdrier sites (Mueller and Senerviratne 2014 Swenson andLawrence 2014 De Kauwe et al 2015 Best et al 2015Ukkola et al 2016a Humphrey et al 2018 Scanlon et al2019) Mueller and Senerviratne (2014) showed that CMIP5models overestimated multiyear mean daily ET in many re-gions with the strongest bias in dryland regions (particu-larly western North America) Likewise Grippa et al (2011)and Scanlon et al (2019) demonstrated that LSMs underes-timate seasonal amplitude of total water storage in semi-arid(and tropical) regions However compared to more mesicecosystems semi-arid ecosystem LSM water flux and stor-age simulations have rarely been tested extensively againstin situ observations apart from a few exceptions (Hogueet al 2005 Abramowitz et al 2008 Whitley et al 2016Grippa et al 2017) Whitley et al (2016) compared carbonand water flux simulations from six LSMs at five OzFlux sa-vanna sites Their study highlighted two key deficiencies inmodelling water fluxes (i) modelled C4 grass T is too lowand (ii) models with shallow rooting depths typically under-estimate woody plant dry season ET As part of a modelinter-comparison for western Africa (the AMMA LSM In-tercomparison Project ndash ALMIP) LSM water storage fluxesrunoff and land surface temperature were evaluated againstin situ and remote sensing data in the Malian Gourma region

2 Methods and data

ndashse

ndash201

08ndash2

ndash201

07ndash2

2LAY 11LAY

222 Soil hydrology

sumnroot(i)= 1

β = exp(minuscj h

dt) (3)

β(j)=

11sumi=2

nroot(i)

middotmax

) ))) (4)

225 Snow scheme

23 Data

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

σθMod+ θObs (6)

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

2 Methods and data

ndashse

ndash201

08ndash2

ndash201

07ndash2

2LAY 11LAY

222 Soil hydrology

sumnroot(i)= 1

β = exp(minuscj h

dt) (3)

β(j)=

11sumi=2

nroot(i)

middotmax

) ))) (4)

225 Snow scheme

23 Data

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

σθMod+ θObs (6)

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

2 Methods and data

ndashse

ndash201

08ndash2

ndash201

07ndash2

2LAY 11LAY

222 Soil hydrology

sumnroot(i)= 1

β = exp(minuscj h

dt) (3)

β(j)=

11sumi=2

nroot(i)

middotmax

) ))) (4)

225 Snow scheme

23 Data

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

σθMod+ θObs (6)

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

ndashse

ndash201

08ndash2

ndash201

07ndash2

2LAY 11LAY

222 Soil hydrology

sumnroot(i)= 1

β = exp(minuscj h

dt) (3)

β(j)=

11sumi=2

nroot(i)

middotmax

) ))) (4)

225 Snow scheme

23 Data

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

σθMod+ θObs (6)

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

2LAY 11LAY

222 Soil hydrology

sumnroot(i)= 1

β = exp(minuscj h

dt) (3)

β(j)=

11sumi=2

nroot(i)

middotmax

) ))) (4)

225 Snow scheme

23 Data

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

σθMod+ θObs (6)

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

222 Soil hydrology

sumnroot(i)= 1

β = exp(minuscj h

dt) (3)

β(j)=

11sumi=2

nroot(i)

middotmax

) ))) (4)

225 Snow scheme

23 Data

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

σθMod+ θObs (6)

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

sumnroot(i)= 1

β = exp(minuscj h

dt) (3)

β(j)=

11sumi=2

nroot(i)

middotmax

) ))) (4)

225 Snow scheme

23 Data

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

σθMod+ θObs (6)

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

23 Data

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

σθMod+ θObs (6)

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

5 cm (6)15 cm (7)30 cm (8)

2 cm (4)20 cm (8)50 cm (9)

5 cm (6)20 cm (7)50 cm (9)

σθMod+ θObs (6)

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

3 Results

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

Site Modelversion

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

4 Discussion

5 Conclusions

References

Abstract

Introduction

Methods and data




Soil hydrology



Snow scheme

Data




Results





Discussion






Conclusions

Code availability

Data availability

Supplement


Competing interests

Acknowledgements

Financial support

Review statement

References

Documents

Testing water ﬂuxes and storage from two hydrology ......to miscalculate partitioning of evapotranspiration (ET) into transpiration (T) and bare soil evaporation (E), with models