SEtHyS_Savannah: A multiple source land surface model applied to Sahelian landscapes

Agricultural and Forest Meteorology 149 (2009) 1421–1432

Contents lists available at ScienceDirect

Agricultural and Forest Meteorology journal homepage: www.elsevier.com/locate/agrformet

SEtHyS_Savannah: A multiple source land surface model applied to Sahelian landscapes S. Saux-Picart a,3, C. Ottle´ b,*, A. Perrier c, B. Decharme b,1, B. Coudert a,2, M. Zribi a, N. Boulain d, B. Cappelaere d, D. Ramier d a

CETP/IPSL, 10 Avenue de l’Europe, 78140 Ve´lizy, France LSCE/IPSL, Centre d’Etudes de Saclay, Orme des Merisiers, 91191 Gif-sur-Yvette, France INAPG, 16 rue Claude Bernard, 75231 Paris Ce´dex 05, France d HSM/IRD, 911 avenue Agropolis, BP 64501, 34394 Montpellier Ce´dex 05, France b c

A R T I C L E I N F O

A B S T R A C T

Article history: Received 19 December 2007 Received in revised form 24 March 2009 Accepted 26 March 2009

The existing SEtHyS SVAT model was developed further to model heat and water fluxes over savannah landscapes with the final objective to use remote sensing for regionalization and monitoring at larger scale. This new development incorporates two vegetation layers (low and high covers) above the soil. The transfer of soil water was revised in order to improve the simulated hydrology over Sahelian and semi-arid regions, by incorporating a mulch representation in a three-layer soil scheme. The two versions of the model, original and modified, have been compared at local scale over two instrumented local sites of the AMMANiger supersite: a fallow and a millet field equipped with surface flux, soil moisture and vegetation measurements. After calibration of the model parameters using the Multiobjective Calibration Iterative Process (MCIP) methodology on the 2005 dataset, the simulations of the two versions of SEtHyS were compared to observations over 2006. Significant differences were found between the simulations, and a better agreement with in situ measurements was observed for the new model. These differences are discussed in relation to the parameterizations of the hydrological and vegetation processes. ß 2009 Elsevier B.V. All rights reserved.

Keywords: Plant canopy modelling Surface fluxes Soil surface mulch SVAT Savannah Sahel

1. Introduction The understanding of land–atmosphere processes is crucial in many environmental sciences such as hydrology, meteorology and ecology, because these interactions impact atmospheric dynamics and water resources at different spatial and temporal scales. These links are particularly strong in semi-arid regions because of the climate characteristics, i.e. short rainy season and strong atmospheric demand. In order to study the feedbacks between land and atmosphere in semi-arid zones, various experimental programs have been designed. One can mention as an example the HAPEXSahel experiment (Goutorbe et al., 1997; Monteny et al., 1997), the Semi-Arid Land Surface Atmosphere (SALSA) program (Goodrich et al., 2000), the Sustainability of Semi-Arid Hydrology and Riparian Areas (SAHRA) program (Sorooshian et al., 2002) and more recently the African Monsoon Multidisciplinary Analysis

* Corresponding author. Tel.: +33 1 69 08 62 68; fax: +33 1 69 08 77 16. E-mail address: [email protected] (C. Ottle´). 1 Present address: CNRM, Me´te´oFrance, 42 Avenue G. Coriolis, 31057 Toulouse, France. 2 Present address: CESBIO, IUT-A, 24 rue d’Embaque`s, 32000 Auch, France. 3 Present address: PML, Prospect Place, The Hoe, Plymouth, PL1 3DH, United Kingdom. 0168-1923/$ – see front matter ß 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.agrformet.2009.03.013

(AMMA, http://www.amma-eu.org) program (Redelsperger et al., 2006) which addresses the West African Monsoon (WAM). The WAM is a coupled land–ocean–atmosphere system characterized by summer rainfall over the continent and winter drought. Apart from these seasonal changes, this complex system also shows a large interannual variability with extreme changes between humid years (pre-1970) and dried years (post-1970) leading to devastating environmental and socio-economic consequences. In order to understand and quantify the land surface–atmosphere exchanges, a modelling strategy was implemented, supported by observations at different scales: field campaigns, aircraft measurements and the use of a remote sensing database. Land–atmosphere feedbacks in the WAM have been shown to be of critical importance for atmospheric prediction at various scales. Koster et al. (2004) demonstrated the strong coupling between soil moisture and rainfall over the northern hemisphere, particularly the persistence of soil moisture anomalies across West Africa. At a smaller scale, Taylor and Lebel (1998) noted persistent effects in the spatial organization of precipitation in successive convective systems. These findings show that it is necessary to develop land–atmosphere-coupled models for a better understanding of the role of land and vegetation in WAM dynamics in order to better predict its variability. However, land surface modelling in semi-arid environments is a great challenge for

1422

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

various reasons. The difficulties are linked to the extreme climatic conditions, the large land cover heterogeneity (vegetation and soil), the poor amount of data available for process calibration and finally, the particular properties developed by the vegetation to withstand the arid conditions (Porporato et al., 2001). The semi-arid regions of West Africa such as the Sahel, are covered by savannah landscapes characterized by heterogeneous ecosystems presenting various components with distinct properties. Two main layers of vegetation may be differentiated: the upper layer generally consists of bushes and trees, and the understorey is composed of grasses or bare soil. Such an ecosystem is difficult to simulate with a general-purpose land surface model (LSM) that considers only one layer of vegetation to simulate hydrological and energy transfers. The reason is related to the strong spatial heterogeneity of this ecosystem, presenting various physiological and physical properties, and different temporal dynamics between the high vegetation layer and the understorey. For all these reasons there is a necessity to develop an LSM able to simulate such multi-source environments, and also able to be validated by remote sensing data in order to supplement the lack of observations. Space observations are indeed a valuable source of information in these regions. Surface temperature, soil moisture, land cover and vegetation phenology are important parameters which can be assessed by remote sensing instruments and can be used to control a distributed LSM at a regional scale. In this context the objective of our work is to develop an LSM which can effectively be spatialized over Sahelian savannah landscapes and monitored using remote sensing data. For that purpose, the first step is the development of the model and its validation at the local scale against in situ surface measurements. These developments are conducted in the framework of the AMMA program, taking advantage of the experimental surface network installed to measure the energy and water transfers inside the biosphere (Lebel et al., 2009). For this study, the SetHyS LSM (Coudert et al., 2006) was modified in order to simulate savannah landscapes. Thus a second vegetation layer was added and the soil representation was modified in order to better simulate the water and energy budgets. The new SEtHyS model, called SEtHyS_Savannah, is briefly described in Section 2. These two models are then evaluated at the local scale against the measurements available within the AMMA supersite in Niger (Cappelaere et al., 2009), for the two main vegetation types consisting of millet crop and fallow bush. The experimental design is explained in Section 3. The results from the two models are given in Section 4, while a short discussion and the main conclusions are provided in Section 5. 2. The SEtHyS_Savannah model The SEtHyS_Savannah model is derived from the SEtHyS model (Coudert et al., 2006), which is a one-dimensional soil–vegetation– atmosphere–transfer (SVAT) model. SEtHyS has a two-layer soil hydrology, with a 0.1 m thick top layer included in the total root zone (Deardorff, 1978). The land surface is represented by two components: the soil and the overlying vegetation. The transfers between these two sources and the atmosphere are based on the resistance concept, whereby the fluxes are determined by integration of the diffusion equations for heat and energy. Various parameterizations are included to represent the resistance network (Coudert et al., 2006). Because of the heterogeneous characteristics of savannah compared to agricultural areas, two sources are not sufficient to represent such complex vegetation with differing photosynthetic functioning (C3 and C4 species) as well as differing heights and phenology. Previous works carried out over such semiarid vegetation showed the importance of differentiating between at least the grass and the woody components (Verhoef and Allen, 2000; Tuzet et al., 1997). When considering the soil, several studies (Jalota, 1993; Braud et al., 1997) have shown the importance of having a dry

soil mulch description – especially under semi-arid conditions – to better describe evaporation and runoff/infiltration mechanisms. The dry soil mulch layer is a totally dry surface layer generated by the loss of water due to soil evaporation. Consequently, various modifications have been brought to the SEtHyS model in order to represent savannah landscapes. A three source description was chosen to represent the soil and the two vegetation layers (grass and trees), respectively. It is assumed that there is a grass layer under a top vegetation layer which could be crops, such as millet, or fallow bushes. Indeed, in the study area, weeding between millet pockets is only very partial, both in time and space, leading to the presence of grass most of the time during the growing season. Regarding the soil representation, a deliberately conceptual scheme was adopted as a compromise between the complexity needed to correctly represent soil water content and soil evaporation, and the necessarily limited number of parameters required to allow for future calibration with remote sensing data. Consequently, in the new SEtHyS_Savannah model, the soil is still represented in two fixed layers which are now the grass root zone about 0.6 m thick, and a deeper layer down to 1 m reached only by tree roots, but a dynamic mulch description was added. Then, the functioning of the upper ground layer becomes as follows: when a wet soil is subject to strong evaporative demand (which is generally the case under arid climates), a dry mulch layer (at the residual water content value, wresid ) forms at the soil surface. This layer limits soil evaporation and thus contributes to keeping water in the soil beneath. No water exchange by capillary action between the dry soil mulch layer and the wet layer beneath is assumed to occur. Under semi-arid conditions, this mulch layer has a significant impact on the soil water budget and it is therefore important to incorporate it into the model. Moreover, when the mulch surface is wetted by rain, a wet layer at field capacity, wcc , is represented. Hence, for a soil at initial water content w covered by a mulch, the evolution of the mulch depth (zmulch) may be written:

@zmulch Es pr ¼  @t w  wresid wcc  w

(1)

where Es and Pr are the soil evaporation and infiltration rates (mm s1). Thus, soil evaporation is converted into a variation of mulch thickness. If rain occurs, the new wet layer appearing at the surface with water content wcc forms a new mulch layer when evaporation begins. Thus, the model is able to predict the evolution of two mulch layers (with thickness zmulch1 and zmulch2 ) and of the water content in between three soil layers (z1, z2 and zrt). Fig. 1 presents the soil description in the case of only one mulch layer. The infiltration rate is computed as the difference between the through-fall, which is the net rainfall that enters the soil after passing through (and not being intercepted by) the canopy, and the surface runoff. The surface runoff takes into account the infiltration excess mechanism (Horton runoff) described by Decharme and Douville (2006). Horton runoff is the largely dominant type of runoff over the Sahelian region, characterized by high-intensity rainfall events and widespread soil surface crusting (Casenave and Valentin, 1992; Peugeot et al., 1997). In SEtHyS_Savannah, surface runoff can also occur when the soil is completely saturated. Three energy budgets are solved for the bare soil, grass and tree components, respectively. A shielding factor (Deardorff, 1978) is used to partition incoming downward shortwave and longwave radiations, which assumes a spherical distribution of the leaves for the two vegetation layers. The resistance scheme is slightly more complicated than in the original model, as shown in Fig. 1. For sensible and latent heat fluxes, aerodynamic resistances are computed between the surface, the vegetation component source heights and the reference level. The wind profile is assumed to be logarithmic above the canopy and then exponential within the

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

1423

Fig. 1. Schematic representation of the multi-source SEtHyS and SEtHyS_Savannah resistance networks.

canopy. For the latent heat flux, two resistances are added for each component: the stomatal resistance based on Sellers et al. (1992) following Ball (1988) and Collatz et al. (1991, 1992) and the mulch resistance which is linearly dependent on the mulch thickness: r mulch ¼

t Dv

zmulch

(2)

where Dv is the diffusivity of water vapor in the soil, and t is the tortuosity coefficient, which are equal to 2.5  105 m2 s1 and 1.5 respectively. This new SetHyS_Savannah model requires 28 parameters and 8 initial values which are summarized in Table 1. The specification of the model parameters has a direct influence on the system response. Thus, the model calibration consists in the minimization

Table 1 List of model parameters and variables used in SEtHyS_Savannah, and ranges of parameter values (in brackets: initial parameter intervals before calibration; bold: calibration results; others: non-calibrated parameters). Name Optical properties es, eg, et albsec albhum winf, wsup albg, albt

Description

Parameter range for millet site

Parameter range for fallow site

Unit

Emissivity of bare soil, grass and tree Bare and wet soil albedo

[0.95, 0.99] 0.96 0.99, 0.99 [0.25, 0.4] 0.38 [0.12, 0.24] 0.23 0.006, 0.2 0.32, 0.32

[0.95, 0.99] 0.98 0.99, 0.99 [0.25, 0.4] 0.33 [0.12, 0.24] 0.22 0.006, 0.2 0.32, 0.32

– –

30, 100 0.01, 0.01 0.02

30, 60 0.01, 0.01 0.02

mmol m2 s1

‘Half critic’ hydric potential Saturated water content Field capacity Residual water content Surface saturated hydraulic conductivity Deep saturated hydraulic conductivity Correction coefficient of the volumetric heat soil capacity Soil tortuosity Soil water vapour diffusivity Water potential of saturated soil Retention curve slope Depth of soil reservoirs

200, 300 0.27 0.2 0.006 [5  107, 5  105] 1  105 [5  107,5  105] 9.1  106 [1, 2] 1.5 [1, 3] 1.5 7.5  105 0.16 [3, 9] 4.2 0.6, 1.0

200, 300 0.27 0.2 0.006 [5  107, 5  105] 3  106 [5  107,5  105] 8  107 [1, 2] 1.2 [1, 3] 1.22 7.5  105 0.16 [3, 9] 8.4 0.4, 1.0

m m3 m3 m3 m3 m3 m3 m s1 m s1 J m3 K1 – m2 s1 m – m

Water content of the different soil layers

[0.005, 0.15] 0.08 [0.005, 0.15] 0.125 [0.01, 0.15] 0.125 0, 0.18, 0.05, 0.37 [300, 310] 305

[0.005, 0.15] 0.02 [0.005, 0.15] 0.02 [0.01, 0.15] 0.14 0, 0.11, 0.05, 0.24 [300, 310] 303

m3 m3

Soil moisture parameters for albedo calculation Grass and tree albedos

Vegetation characteristics V max 0g ; V max 0t Leaf photosynthetic capacity (Rubisco) lgfg, lgft Leaf width kwstr Empirical parameter for water stress calculation Ground properties phcg, phct wsat wcc wresid Ksat1 Ksat facttherm

t diffv

csat b zrg, zrt Initial variables wg10 wg20 wt0 zmulch10 ; zmulch20 , z10, z20 t20

Thickness of the mulch and inter-mulch soil layers Deep soil temperature

m3 m3 –

m –

m K

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

1424

of a cost function expressing the divergence between the modelsimulated output and the observation data. In order to achieve such calibration, the multi-objective calibration iterative process (MCIP) was applied (Gupta et al., 1999; Bastidas et al., 1999; Demarty et al., 2005). This approach is purely stochastic and based on the iterative reduction of the parameter space by optimization of multi-criteria cost functions. Coudert et al. (2006), using the original SEtHyS model, showed that this model calibration can be performed with relatively good accuracy and robustness when surface fluxes are used as optimization variables. In this study, the same methodology has been successfully adapted to calibrate SEtHyS_Savannah parameters. 3. Experimental design 3.1. AMMA-Niger site The studied area is located in southwest Niger, close to the center of the HAPEX-Sahel’s square degree (2–38E, 13–148N) at 60 km East of Niamey where the major part of that experiment was concentrated (Goutorbe et al., 1997). This area has been chosen as one of the experimental sites within AMMA and equipped in order to measure the main surface processes (Cappelaere et al., 2009). The region is representative of a Sahelian semi-arid environment and climate, where rainfall is confined to a single wet season from May–June to September–October. The mean annual temperature is around 29 8C, and the mean monthly temperature reaches its highest value in April (34 8C) and its lowest value in January (24 8C). As part of the AMMA experiment, two sites have been intensively instrumented by IRD (Ramier et al., 2009) to study basic processes such as infiltration and evapotranspiration. The first site (138380 3800 N, 28370 4700 E) is a hand-cultivated field of millet. The second site is fallow (138380 5100 N, 28380 100 E) and covered by annual grass and scattered guiera bushes (Guiera senegalensis). On each site, an eddy-covariance flux station was installed to characterize the heat, water and carbon dioxide surface fluxes. Moreover at the same sites, soil temperature and soil moisture probes were installed. Technical characteristics of all instruments are provided in Ramier et al. (2009). The

instruments were installed at the beginning of 2005 and the first measurements were acquired on day of experiment (DOE) 166 (15 June 2005). Two complete rainy and vegetation growth seasons (2005 and 2006) were available for analysis. The instruments produce standard forcing variables (air temperature and relative humidity, shortwave and longwave incoming radiations, wind speed and rainfall) for land surface schemes as well as the turbulent fluxes and outgoing shortwave and longwave radiations for verification. 3.2. Vegetation assessment During the whole 2005 and 2006 rainy seasons, leaf area index (LAI) and vegetation heights were measured around the two flux stations in six experimental plots approximately every 15 days (Fig. 2; see Boulain et al., 2009, for measurement details). In the millet plots, only millet plants have been considered for the LAI measurements, whereas in the fallow plots only grasses were considered. Since a significant grass layer usually develops in millet fields because of the local agricultural practices (little soil weeding) and since shrubs coexist with grasses in fallow sites, it has been necessary to superimpose grasses and shrubs LAI seasonal cycles, respectively, onto the measured ones. It is assumed, following the in situ observations, that the grass layer at the millet site has a vegetation cycle strongly dependent on rainfall, and that the maximum height never exceeds 0.3 m. The beginning of the growing cycle is supposed to begin after the first rainfall event and the maximum stage of development is in phase with the crop. Finally, the senescent phase begins 10 days after the last precipitation. The height of the grass layer for the fallow site is prescribed with a maximum of 0.4 m. The shrubs consist almost entirely of G. senegalensis bushes which have a very low LAI because of the small size and number of leaves. Up to 80% of the LAI of a fallow site is due to the grass layer (Van Leeuwen et al., 1997). The beginning of the vegetation cycle of the G. senegalensis is prescribed in phase with the herbaceous underlayer dynamics, with an LAI up to 0.2 and a constant height equal to 1.9 m. The resulting LAIs for the two sites are presented in Fig. 2 where the in situ data are interpolated by a spline method and a smooth function (boxcar average IDL Reference Guide, 2007).

Fig. 2. Seasonal courses of vegetation parameters (LAI, height) for the two land cover types: millet (left) and fallow (right) (DOE 167–730).

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

1425

Fig. 3. Comparison between measured and modelled (Seth_sav) fluxes (net radiation, evapotranspiration, sensible heat and ground heat fluxes) at both sites during the validation period (DOE 379–729). Millet site results are presented on the left and fallow site on the right.

1426

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

1427

Table 2 Statistics results for energy balance components (W m2) obtained for the calibration and the validation periods with model versions Seth_ini and Seth_sav (LE: latent heat flux; H: sensible heat flux; G: ground heat flux; RN: net radiation. Calibration period [DOE 168–369] Seth_ini

Validation period [DOE 379–729] Seth_sav

Seth_ini

Seth_sav

RMSE

Corr

RMSE

Corr

RMSE

Corr

RMSE

Corr

Millet site LE H G RN

46.0 78.8 59.4 18.0

0.79 0.88 0.77 0.99

36.3 48.6 45.1 20.6

0.88 0.90 0.75 0.99

44.3 50.2 59.9 14.0

0.81 0.90 0.82 0.99

31.6 36.0 46.5 24.8

0.91 0.93 0.78 0.99

Fallow site LE H G RN

51.3 87.3 45.0 19.9

0.82 0.82 0.75 0.99

45.4 71.2 34.4 25.0

0.87 0.89 0.83 0.99

65.7 87.7 49.1 17.5

0.80 0.75 0.80 0.99

53.9 78.9 35.3 36.8

0.88 0.84 0.83 0.99

3.3. Model calibration The MCIP methodology which was used to calibrate the SEtHyS_Savannah model consists in the reduction of an initial parameter space by the optimization of the output surface fluxes against observations. The method requires the generation of an ensemble of simulations (6000 in this case) by sampling an initial parameter space. The iterative Pareto ranking of the best simulations (which best fit the observed data) is then used to reduce the parameter space. The calibration is generally achieved with a few iterations (10 is a maximum value) when cost functions have reached a global minimum. Several calibration tests have been performed, as a first step, to identify the sensitive parameters in this semi-arid environment. They showed that the parameters related to the vegetation processes (transpiration, interception, radiation) were not sensitive in the simulations of the two sites. Actually, this is not surprising as the vegetation is sparse with very low LAI throughout the year. Consequently, the model calibration has been performed only on the soil parameters governing the soil water and energy processes and initial conditions. The initial ranges for the parameters (in brackets in Table 1) have been chosen according to the literature and a priori knowledge of the field conditions. The values obtained by the calibration are indicated in bold in Table 1. The non-sensitive parameters were fixed according to Sellers et al. (1992). Both the initial and Savannah versions of the SEtHyS model (called the Seth_ini and the Seth_sav versions respectively in the following) were calibrated over the period between 17 June 2005 and 4 January 2006 (DOE 168 and 369, respectively). This period was chosen so as to include both wet and dry conditions. It has to be noted that, for parameters that are common to both model versions, calibration does not converge to the same final values. Moreover, one can see differences between parameters for the millet site and for the fallow site, particularly the wet soil albedo and the slope of the retention curve which respectively converge to 0.23 and 4.2 on the millet site, and to 0.22 and 8.4 on the fallow site (see Table 1). 4. Results 4.1. Energy balance components After calibration of both models, validations were performed over the two sites during the following 1 year period (DOE 366– 730) and more precisely on the second growing season (DOE 550–

650). Statistical results in terms of root mean square errors (RMSE) and correlation coefficients for the energy balance components are presented in Table 2 for the calibration and the validation periods. Graphical comparison of the time series of the fluxes over the 2 years of the dataset are not shown because the figures are almost illegible given the lengths of the two growing seasons. Instead Fig. 3 shows the comparison of observed and modelled (Seth_sav model) terms of the energy balance as scatter plots for the validation period. Then short time series were extracted in order to compare the predictions of both versions of the model with the ground truth measurements (Figs. 4 and 5). Globally the new version of the SEtHyS model (Seth_sav) performs better over the whole period than the previous version (Seth_ini). As shown in Table 2, all fluxes are better represented at the two sites except for net radiation. For the millet site there is not much difference in the net radiation RMSE with only a 2 W m2 difference during the calibration period and 12 W m2 for the validation period. However we have almost a 20 W m2 RMSE difference in net radiation during the second year of simulation for the fallow site. For this site the RMSE regarding net radiation is more than 5 W m2 larger with the new version of the model for both periods. On the contrary, latent heat flux, sensible heat flux and ground heat flux are better represented by Seth_sav model. Enhancement of up to 15 W m2 in RMSE for the 2-year simulation can be observed. On the scatter plots presented in Fig. 3, all fluxes except ground heat fluxes, are better simulated for the millet field (left) than for the fallow site (right), with RMSE less than 47 W m2 and correlation coefficients greater than 0.78 for the millet site. For the fallow site, the plots are more scattered and the RMSE are larger (54 W m2 for evapotranspiration, 79 W m2 for sensible heat), except for the ground heat flux which seems to be in better agreement with observations (RMSE = 46.5 W m2, correlation coefficient = 0.78 for the millet and RMSE = 35 W m2, correlation coefficient = 0.83 for the fallow). These differences between the two sites may be explained by the larger spatial heterogeneity of the surface (land cover, soil moisture) for the fallow cover type, which makes the flux measurements, as well as the model calibration, more variable as already noted by Lloyd et al. (1997). It must also be noted that ground heat flux measurements have been performed at the 0.05 m depth and have not been extrapolated to the soil surface, which may have a significant impact on comparisons with simulated ground heat fluxes. Better results are expected when extrapolated fluxes become available within the AMMA database.

Fig. 4. Time series of observed (grey dots) and modelled surface fluxes for the millet site: net radiation, evapotranspiration, sensible heat and ground heat fluxes for the two model versions (Seth_ini, dashed line; Seth_sav, solid line), for two time periods: DOE 300–307 (left) and DOE 607–613 (right).

1428

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

Figs. 4 and 5 show, for the millet and the fallow site respectively, the simulated and observed energy balance components (from top to bottom: latent heat flux, sensible heat flux, ground heat flux and net radiation) over short time slices (DOE 300–307, end of the monsoon season and 607–613, beginning of the rainy season for the millet, left and right of Fig. 4, respectively; DOE 612–616 and 640–647 for the fallow, left and right of Fig. 5, respectively). Because rain event conditions are generally unsuitable for eddy flux measurements, there is a lack of latent and sensible heat flux observations during and immediately after most rain events. The differences are analyzed at each site individually during periods when observations are available with no gaps, to show the impact of the new model parameterizations. At the millet site, all over the simulated period, the Seth_sav simulations show large improvements regarding the estimation of both components of the evapotranspiration. At the end of the vegetated period (DOE 300–307, Fig. 4, top left), the soil evaporation seems to be better represented with Seth_sav compared to Seth_ini. The same improvements can be noted over vegetation and in the response to precipitation events. For example, during DOE 607, a small rain event (0.5 mm) occurred around midday. Fig. 4 (right) shows the strong difference in the response of the two models. The Seth_ini model (dashed line) is not sensitive to that event while the Seth_sav model (solid line) matches quite well latent heat flux observations (top right). Moreover, when a high rainfall occurs (DOE 608, around midday), Seth_ini overestimates evapotranspiration while Seth_sav shows better agreement with the observations. The behaviour of the evapotranspiration after a rain event is not either correctly reproduced by Seth_ini model: latent heat flux is under-estimated during day time while it is over estimated during night time (DOE 610–612). Sensible heat flux is also better simulated by the new version of the model, although it is still overestimated during day time and under-estimated during night time, with a slightly higher correlation coefficient and a much smaller RMSE value. However net radiation is not represented as well as the original model: during day time, the new model seems to under-estimate net radiation. This under-estimation is due to an overestimation of the calibrated surface albedos. Indeed, the values reached at the end of the calibration process (as already noted in Section 3.3) are large compared to literature values and previous simulation works (0.22 and 0.23 (Seth_sav) compared to 0.18 (Seth_ini) for the wet soil albedo of respectively the fallow and the millet sites). Such differences have a strong impact on net radiation because the solar radiation is the predominant component in the net radiation calculation. Finally, the ground heat flux comparison shows a time shift and an amplitude deviation between simulations and observations corresponding to the fact that measurements were acquired at 0.05 m depth and model estimations are done at the surface level. These deviations are larger (as expected) at the end of the vegetation season (DOE 300–307) when LAI is null, compared to the full vegetation period (DOE 607–613). At the fallow site, as already noted, RMSE are generally higher than at the millet site, especially for the latent and sensible heat fluxes. But the two models show similar tendencies with a better simulation of evapotranspiration using Seth_sav compared to Seth_ini. As an example (Fig. 5, left), during DOE 612 and 613 two rain events occur, and strong differences in the two model responses are observed: the Seth_ini model does not correctly simulate evapotranspiration after either small or large rain events, while Seth_sav does much better. In Sahelian regions, evapotranspiration is strongly dependent on soil surface water content and atmospheric conditions. The

1429

formalism used for the water budget and the soil surface resistance in SVAT models is thus very important. The Seth_ini model assumes a fixed 0.1 m surface layer included in a thicker root layer, following Deardorff (1978). Soil evaporation is driven by a soil surface conductance that depends on surface soil moisture and potential evaporation. Soil evaporation is then limited by an exponential function of the surface soil moisture content (Soares et al., 1988). This formulation is not suitable for semi-arid climate conditions. First, soil evaporation cannot be represented correctly with the bucket surface layer, because the water content averaged over a 0.1 m soil layer depth is not sensitive to very small rain events. Second, the exponential limitation of soil evaporation does not allow the model to reproduce the diurnal cycle properly. The proposed mulch layer representation solves these two problems. The soil resistance is directly proportional to the mulch thickness which is reinitialized to zero after a rain event (irrespective of intensity). For example, at the beginning of DOE 612 at the fallow site (Fig. 5, left) a small rain event is immediately followed by a peak in evaporation. Evaporation is not controlled by soil water content alone, but also by mulch thickness which depends on antecedent precipitation. This formulation (still conceptual) of soil evaporation leads to more realistic simulations after a rain event. It can be noted that this new parameterization offers the advantages of simulating precisely the superficial soil drying and soil evaporation without the over-parameterization of multi-layer models (here the number of parameters has not changed compared to the original model). 4.2. Transpiration versus soil evaporation Fig. 6 shows the cumulative components of evapotranspiration for both sites (left and right) and both models along the simulation period. With both models, soil evaporation is larger than transpiration. Soil evaporation is slightly higher with Seth_sav compared to Seth_ini, especially after the rainy season, as a result of the mulch parameterization. Seth_sav simulates higher vegetation transpiration, compared to the Seth_ini simulation, for the millet site and lower vegetation transpiration for the fallow site. Discrepancies can be explained by the fact that the Seth_ini model is not able to handle two vegetation layers. Total LAI was thus used as forcing and a single stomatal resistance was computed. In contrast, the Seth_sav model can take two vegetation layers into account with different stomatal conductance related to the plant type. For the same soil hydric potential, grass, millet and guiera plants do not have the same stomatal conductance thanks to the formulation of Sellers et al. (1992), following Ball (1988), that enabled the parameterization of the stomatal conductance according to plant type (C3 and C4). Millet and grass are C3 species whereas Guiera is a C4 plant. At the fallow site, vegetation transpiration almost stops before DOE 300 with Set_sav although the vegetation cycle is not over. Actually the parameter b which represents the slope of the retention curve does not have the same value (after calibration) for each site (4.2 and 8.4 respectively for the millet and fallow site). Thus the soil hydric potential is not the same for similar soil moisture. The absolute value is larger for the fallow site than for the millet site leading to a smaller root uptake capacity. Consequently, stomatal resistance is larger for the fallow site as it depends on vegetation water stress. 4.3. Soil water content The impacts of the water transfer parameterizations are significant for the soil water content predictions, presented in

Fig. 5. Time series for observed (grey dots) and modelled surface fluxes for the fallow site: net radiation, evapotranspiration, sensible heat and ground heat fluxes for the two model versions (Seth_ini, dashed line; Seth_sav, solid line), for two time periods: DOE 612–616 (left) and DOE 640–647 (right).

1430

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

Fig. 6. Cumulative seasonal courses of the two components of evapotranspiration (soil evaporation and vegetation transpiration) and rainfall, for the two model versions (millet site: on the left and fallow site: on the right).

Fig. 7. Root zone water content, at the millet site (left) and the fallow site (right), observed (grey dots) and simulated with the two model versions over the 2-year period: Wrg is the water content in the grass root zone of Seth_sav (solid line), W2 is the water content in the Seth_ini root zone (dashed line) and rainfall as vertical bars.

Figs. 7 and 8 for the full simulation and a 1-month period, respectively. The relative soil water contents (in percentages of saturation) are plotted for the root zone (aggregation of component layers) and the soil surface layer (Fig. 8 only), in order to allow for the comparison of models and with observations. Fig. 7 presents the simulated and observed root zone relative water content (or soil humidity) for both experimental sites (millet, left and fallow, right) and for the 2005–2006 simulation period (beginning at DOE 168). Strong discrepancies are observed between the two models especially for the millet site. In the Seth_ini model, when the top 0.1 m soil layer is dry (i.e., at the residual water content), soil evaporation is blocked. Thus after DOE 320 and until the next rainfall event (DOE 520) the soil moisture is constant as there is no deep drainage in the initial model. In the Seth_sav model, soil moisture continues to decrease until reaching residual value in the grass root zone because soil evaporation through the mulch layer and drainage are allowed. This behaviour looks more realistic when compared to the observations. The plots show also the large biases between the two models, larger for the millet simulation than for the fallow one. This can be explained by the differences in the size of the soil layers contributing to the evapotranspiration. In Seth_sav model, the total root zone was calibrated to 0.62 and 1.0 m for millet and fallow respectively (in Seth_ini the total soil depth for both sites is equal to 1.0 m). This difference for millet lead to a larger soil moisture dynamics in the new model compared to the initial version.

Fig. 8 shows the soil water content in the root zone (top) and in the surface layer (bottom) for a 1-month period (DOE 600–630) during the 2006 rainy season at the millet (left) and fallow (right) sites. One can see the larger root zone relative humidity simulated by the Seth_ini model compared to measurements and the obvious enhancement provided by the Seth_sav model, especially at the millet site. Regarding the soil surface relative humidity (bottom plots), it can be seen that after a rainy event the Seth_ini model often reaches saturation, whereas the Seth_sav model does not, and shows better agreement with the observations for both sites. These curves also show that soil evaporation seems to be correctly simulated after rainy events because the temporal variations in soil water content are very well reproduced by the Seth_sav model. 5. Discussion and conclusions A new version of the SEtHyS land surface model was developed to simulate energy and water budgets over savannah landscapes for future applications at larger scales using remote sensing data. The existing and new models were compared against ground measurements available from two sites instrumented in the framework of the AMMA experiment. The results show large improvements to the flux simulations provided by the new soil parameterizations. The proposed soil description, with a mulch concept, allows better simulation of the surface soil moisture. The infiltration model improves root zone and surface water content

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

1431

Fig. 8. Root zone water content (top) and surface soil water content (bottom) observed and modelled for the millet site (left) and the fallow site (right), with the two model versions during a 1-month period (DOE 600–630). Same conventions as Fig. 7.

simulations. Moreover simulation of the latent heat flux (and especially the soil evaporation component) is particularly enhanced at a daily time scale, and the diurnal cycle of evapotranspiration is in better agreement with observations due to the mulch description. The principle of the calibration process is to find the best set of parameters that minimizes the differences between observations and simulations. Calibration of the new model enables enhanced simulation of turbulent fluxes and ground heat fluxes compared to the former version of the model. However soil albedos do not converge to the same values for both models and thus comparison of simulated net radiation with observations show higher RMSE for the new model version. In order to correct this problem a new calibration should be done with inclusion of land surface temperature diurnal cycle characteristics in the cost function minimization (Coudert and Ottle´, 2007; Coudert et al., 2008). Correction of ground heat fluxes for shallow soil storage, when available, should also contribute to improve calibration. Statistical criteria show better agreement at all time scales between models and observations over the millet site, compared to the fallow site. This is consistent with Lloyd et al. (1997) who discussed the large vegetation and soil moisture heterogeneities in fallows which make more difficult the model calibration and validation compared to the millet site. Tests have been carried out (not reported in this paper) with the Seth_sav model assuming a single vegetation layer, showing that a simple modellization (two sources) seems enough to simulate the water cycle and energy balance terms. However, the enhancement provided by the second layer of vegetation lies principally in the

new capacity to simulate vegetation carbon fluxes. It is indeed important to handle two vegetation layers for a good simulation of these fluxes as the guiera, millet or grass do not have the same photosynthetic functioning (assimilation rate). Comparison of the two models’ behaviour shows strong discrepancies. It is not easy to compare results regarding soil moisture, as they are not calculated over the same soil layer depths. Nevertheless, when comparing relative values, Seth_sav clearly enhances soil moisture simulation. In particular, the new mulch representation enables water to infiltrate deeper whereas in the Seth_ini model the water was restricted to the top layer. It can be noted that the new parameterization, albeit conceptual, is able to simulate surface soil moisture and soil evaporation quite precisely, probably as well as a multi-layer model would, but with as few calibration parameters as the initial model. Another problem is the definition of the bush LAI at the fallow site, and of the grass LAI at the millet site. LAI is a key forcing variable that controls transpiration and consequently soil moisture, hence the hydrological and energy budgets. The LAI values and phenologies have been adjusted according to the literature, which had a positive effect on the flux simulations. These results validate the model at the local scale for the two main vegetation types of the region (millet crop and fallow sites). The next steps concern the spatialization of the model with remote sensing data and its validation against surface temperature and soil moisture products over the AMMA-Niger super-site (Saux-Picart et al., 2009), then over the whole meso-scale site. Estimating land– atmosphere exchanges of water and energy at these larger scales, with their time/space variability, should help improving our understanding of the West African Monsoon mechanisms and

1432

S. Saux-Picart et al. / Agricultural and Forest Meteorology 149 (2009) 1421–1432

effects. Impacts of land use management in this highly sensitive environment could also be better investigated. Acknowledgements The authors wish to acknowledge A. Tuzet for fruitful discussions and the French Education and Research Ministry for funding S. Saux-Picart Ph.D. scholarship. Field data were obtained thanks to the AMMA-Catch observatory (ORE) in Niger and the French ECCO-PNRH program (‘‘Eau et ve´ge´tation au Niger’’). Based on a French initiative, AMMA was built by an international scientific group and is currently funded by a large number of agencies, especially from France, UK, US and Africa. It is also the beneficiary of a major financial contribution from the European Community’s Sixth Framework Research Programme. Detailed information on scientific coordination and funding is available on the AMMA International web site http://www.amma-international.org/. References Ball, J.T., 1988. An analysis of stomatal conductance. Ph.D. Thesis, Stanford University, 89 pp. Bastidas, L.A., Gupta, H.V., Sorooshian, S., Shuttleworth, W.J., Yang, Z.L., 1999. Sensitivity analysis of a land surface scheme using multicriteria methods. Journal of Geophysical Research 104 (D16), 19481–19490. Boulain, N., Cappelaere, B., Ramier, D., Issoufou, H.B.A., Halilou, O., Seghieri, J., Guillemin, F., Gignoux, J., Timouk, F., 2009. Towards an understanding of coupled physical and biological processes in the cultivated Sahel—2. Vegetation and carbon dynamics. Journal of Hydrology. doi:10.1016/j.jhydrol.2008.11.045. Braud, I., Bessemoulin, P., Monteny, B., Sicot, M., Vandervaere, J.P., Vauclin, M., 1997. Unidimensional modelling of a fallow savannah during the HAPEX-Sahel experiment using the SiSPAT model. Journal of Hydrology 188–189, 912–945. Cappelaere, B., Descroix, L., Lebel, T., Boulain, N., Ramier, D., Laurent, J.-P., Le Breton, E., Mamadou, I., Boubkraoui, S., Bouzou Moussa, I., Favreau, G., Issoufou, H.B.A., Nazoumou, Y., Quantin, G., Chaffard, V., Ottle´, C., 2009. The AMMA-Catch experiment in the cultivated Sahelian area of south-west Niger—strategy, implementation, site description, main results. Journal of Hydrology AMMACatch special issue. Casenave, A., Valentin, C., 1992. A runoff capability classification system based on surface features criteria in semi-arid areas of West Africa. Journal of Hydrology 130, 231–249. Collatz, G.J., Ball, J.T., Grivet, C., Berry, J.A., 1991. Physiological and environmental regulation of stomatal conductance, photosynthesis and transpiration: a model that includes a laminar boundary layer. Agricultural and Forest Meteorology 54, 107–136. Collatz, G.J., Ribas-Carbo, M., Berry, J.A., 1992. Coupled photosynthesis-stomatal conductance model for leaves of C4 plants. Australian Journal of Plant Physiology 19, 519–538. Coudert, B., Ottle´, C., Boudevillain, B., Guillevic, P., Demarty, J., 2006. Contribution of thermal infrared remote sensing data in multiobjective calibration of a dual source SVAT model. Journal of Hydrometeorology 7 (3), 404–420. Coudert, B., Ottle´, C., 2007. An improved SVAT model calibration strategy based on the optimization of the surface temperature temporal dynamics. Geophysical Research Letters 34, L04402 doi:10.1029/2006GL028778. Coudert, B., Ottle´, C., Briottet, X., 2008. Monitoring land surface processes with thermal infrared data: calibration of SVAT parameters based on the optimisation of diurnal surface temperature cycling features. Remote Sensing of Environment 112, 872–887. Deardorff, J.W., 1978. Efficient prediction of ground surface temperature and moisture, with inclusion of a layer of vegetation. Journal of Geophysical Research 83 (C4), 1889–1903. Decharme, B., Douville, H., 2006. Introduction of a sub-grid hydrology in the ISBA land surface model. Climate Dynamics 26, 65–78. Demarty, J., Ottle´, C., Braud, I., Olioso, A., Frangi, J.P., Gupta, H.V., Bastidas, L.A., 2005. Constraining a physically based SVAT model with surface water content and

thermal infrared brightness temperature measurements using a multiobjective approach. Water Resources Research 41(W01011): doi:10.1029/ 2004WR003695. Goodrich, D.C., 68 coauthors, 2000. Preface paper to the semi-arid land–surface– atmosphere (SALSA) program special issue. Agricultural and Forest Meteorology 105 (1–3), 3–20. Goutorbe, J.P., Lebel, T., Dolman, A.J., Gash, J.H.C., Kabat, M., Kerr, Y.H., Monteny, B., Prince, S., Stricker, J.N.M., Tinga, A., Wallace, J.S., 1997. An overview of HAPEXSahel: a study in climate and desertification. Journal of Hydrology 188–189, 4– 17. Gupta, H.V., Bastidas, L.A., Sorooshian, S., Shuttleworth, W.J., Yang, Z.L., 1999. Parameter estimation of a land surface scheme using multicriteria methods. Journal of Geophysical Research 104 (D16), 19491–19503. IDL Reference Guide, Version 6.4, April 2007 edition. ITT Visual Information Solutions. Jalota, A., 1993. Evaporation through a soil mulch in relation to mulch characteristics and evaporativity. Australian Journal of Soil Research 31 (2), 131–136. Koster, R.D., 24 coauthors, 2004. Regions of strong coupling between soil moisture and precipitation. Science 305 (5687), 1138–1140. Lebel, T., Cappelaere, B., Galle, S., Hanan, N., Kergoat, L., Levis, S., Vieux, B., Descroix, L., Gosset, M., Mougin, E., Peugeot, C., Seguis, L., 2009. AMMA-CATCH studies in the Sahelian region of West-Africa: an overview. Journal of Hydrology, doi:10.1016/j.jhydrol.2009.03.020. Lloyd, C.R., Bessemoulin, P., Cropley, F.D., Culf, A.D., Dolman, A.J., Elbers, J.A., Heusinkveld, B., Moncrieff, J.B., Monteny, B., Verhoef, A., 1997. A comparison of surface fluxes at the HAPEX-Sahel fallow bush sites. Journal of Hydrology 188–189, 400–425. Monteny, B.A., Lhomme, J.P., Chebouhni, A., Troufleau, D., Amadou, M., Sicot, M., Verhoef, A., Galle, S., Said, F., Lloyd, C.R., 1997. The role of the Sahelian biosphere on the water and the CO2 cycle during the HAPEX-Sahel experiment. Journal of Hydrology 188–189, 516–535. Peugeot, C., Esteves, M., Galle, S., Rajot, J.L., Vandervaere, J.P., 1997. Runoff generation processes: results and analysis of field data collected at the East Central Supersite of the HAPEX-Sahel experiment. Journal of Hydrology 188–189, 179– 202. Porporato, A., Laio, F., Ridolfi, L., Rodriguez-Iturbe, I., 2001. Plants in water-controlled ecosystems: active role in hydrological processes and response to water stress. III. Vegetation water stress. Advances in Water Resources 24, 725–744. Ramier, D., Boulain, N., Cappelaere, B., Timouk, F., Rabanit, M., Lloyd, C.R., Boubkraoui, S., Me´tayer, F., Descroix, L., Wawrzynia, V., 2009. Towards an understanding of coupled physical and biological processes in the cultivated Sahel—1. Energy and water. Journal of Hydrology. doi:10.1016/j.jhydrol.2008.12.002. Redelsperger, J.L., Thorncroft, C., Diedhiou, A., Lebel, T.H., Parker, D., Polcher, J., 2006. African monsoon multidisciplinary analysis (AMMA): an international research project and field campaign. Bulletin of the American Meteorological Society (December), 1739–1746. Saux-Picart, S., Ottle´, C., Decharme, B., Andre´, C., Zribi, M., Perrier, A., Coudert, B., Boulain, N., Cappelaere, B., 2009. Water and energy budgets simulation over the Niger super site spatially constrained with remote sensing data. Journal of Hydrology doi:10.1016/j.jhydrol.2008.12.023. Sellers, P.J., Berry, J.A., Collatz, G.J., Field, C.B., Hall, F.G., 1992. Canopy reflectance, photosynthesis, and transpiration. III. A reanalysis using improved leaf models and a new canopy integration scheme. Remote Sensing of Environment 42, 187–216. Soares, J.V., Bernard, R., Taconnet, O., Vidal-Madjar, D., Weill, A., 1988. Estimation of bare soil evaporation from airborne measurements. Journal of Hydrology 99, 281–296. Sorooshian, S., Bales, R., Gupta, H., Woodard, G., Washburne, J., 2002. A brief history and mission of SAHRA: A National Science Foundation Science and Technology Center on ‘Sustainability of semi-arid hydrology and riparian areas’ (invited commentary). Hydrological Processes 16, 3293–3295. Taylor, C., Lebel, T., 1998. Observational evidence of persistent convective scale rainfall pattern. Monthly Weather Review 126, 1597–1607. Tuzet, A., Castell, J.F., Perrier, A., Zurfluh, O., 1997. Flux heterogeneity and evapotranspiration partitioning in a sparse canopy: the fallow savanna. Journal of Hydrology 188-189, 482–493. Van Leeuwen, W.J.D., Huete, A.R., Walthall, C.L., Prince, S.D., Begue´, A., Roujean, J.L., 1997. Deconvolution of remotely sensed spectral mixtures for retrieval of LAI, fAPAR and soil brightness. Journal of Hydrology 188–189, 697–724. Verhoef, A., Allen, S.J., 2000. A SVAT scheme describing energy and CO2 fluxes for multi-component vegetation: calibration and test for a Sahelian savannah. Ecological Modelling 127, 245–267.