Analysis of the Variability of Canopy Resistance over a Desert Steppe Site in Inner Mongolia, China

Land surface and atmospheric interactions include the exchange of energy, momentum, and water vapor, which are affected by land-surface properties, such as the underlying soil and vegetation structures and their unique characteristics. The heterogeneity in surface conditions significantly affects the daytime and horizontal distribution of water vapor in the lower atmosphere, the development of the boundary-layer structure, and summer precipitation (e.g., Chen et al., 2001, 2003; Trier et al., 2004, 2008, 2011; Holt et al., 2006; LeMone et al., 2007, 2010a, 2010b). This heterogeneity is represented specifically by physical and physiological properties (surface albedo, roughness length, vegetation fraction, canopy resistance, etc.). Among these properties, canopy resistance plays a critical role in determining the partitioning of the available energy into sensible and latent heat fluxes (Monteith, 1965; Avissar et al., 1985; Ronda et al., 2001), and the dry deposition of gases (Charusombat et al., 2010; Wu et al., 2011, 2012).

Among the methods for estimating canopy resistance, the Jarvis scheme (Jarvis, 1976; Stewart, 1988) and the Ball-Berry scheme (Ball et al., 1987; Leuning, 1990) are two widely used parameterizations in land-surface models that may be uncoupled or coupled to atmospheric weather and climate models (Noilhan and Planton, 1989; Bosilovich and Sun, 1995; Pleim and Xiu, 1995; Viterbo and Beljaars, 1995; Chen et al., 1996; Sellers et al., 1996; Calvet et al., 1998; Cox et al., 1999a, b; Dai et al., 2003; Ek et al., 2003; Niyogi et al., 2009; Niu et al., 2011). However, the application of these schemes does not distinguish species, for instance, in grasslands such as prairie pastures, typical steppes, and desert steppes. Default or generic parameters in those schemes resolve the differences between different cover types (Jarvis and Mcnaughton, 1986; Alfieri et al., 2008). (Dickinson et al., 1991) suggested prescribing the statistical distributions of soil and canopy properties, as well as those of solar radiation and precipitation or to incorporate parameters more specific to the given land covers.

Many modifications have been made to the two aforementioned canopy-resistance formulations. By representing the increase in canopy temperature caused by additional energy to extract the water from vegetation elements, (Todorovic, 1999) developed a new method to calculate the canopy resistance of grass maintained under optimal moisture conditions. The linear specific humidity function in the Jarvis-Stewart model was adapted for a red oak forest in the Netherlands to an exponential form, and gave slightly better results (Ogink-Hendriks, 1995). Also, the effect of soil water deficit on canopy resistance was formulated by a quadratic form in the Jarvis scheme over a Caumont Soybean site and a tall C₄prairie grass site (Ronda et al., 2001), or by using a simple equation to modify the slope based on the mean soil water potential incorporated into the Ball-Berry scheme (Liu et al., 2009). These modifications demonstrated that both models and their associated parameters are empirical, which need to be adjusted for specific vegetated surfaces. Thus, it is crucial that a dataset with a large range of variables is used for parameter estimation.

In arid and semiarid environments, where water supply is often limited, the stomatal control over water loss from the ecosystem is particularly important for the growth of the steppe vegetation (Li et al., 2006). Desert steppe is the driest grassland ecosystem and is the prevalent grassland type in China, covering almost 34.7% percent of China's northern grassland. It is located in a transitional zone from typical steppe to desert, and is very sensitive to climate change.

We used the two-source model of (Shuttleworth and Wallace, 1985) to partition the observed latent heat flux into soil evaporation and plant transpiration. The average contribution of transpiration to the total latent heat flux varied from 30% to 40% during the growing season over the desert steppe site (see section 2.1) in 2008-10, and sometimes even accounted for 60%, depending on the vegetation fraction. Therefore, plant transpiration is a significant part of the total evaporation at the observation site, and so is the role of canopy resistance in regulating the amount of water transpired by vegetation (Jarvis and Mcnaughton, 1986).

While the Ball-Berry scheme, a photosynthesis-based gas exchange evapotranspiration approach, has been applied in climate models, the Jarvis scheme continues to be predominantly employed for mesoscale or numerical weather forecast models (Noilhan and Planton, 1989; Pleim and Xiu, 1995; Viterbo and Beljaars, 1995; Chen et al., 1996; Calvet et al., 1998; Niu et al., 2011). For instance, the Noah land surface model (LSM) (Chen et al., 1996; Chen and Dudhia, 2001; Ek et al., 2003) using the Jarvis scheme has been widely used in the research and operational communities and has been coupled to community weather models. While there are focused efforts to validate the Noah LSM for tall prairie grasslands (Chen et al., 1996; Sridhar et al., 2002; Chen et al., 2003; LeMone et al., 2007, 2008; Alfieri et al., 2008), little is known about the variability and the driving factors of canopy resistance over desert steppe ecosystems and about the appropriate applications of land surface models. Therefore, it is critical to carefully examine the characteristics of canopy resistance and its formulation used in models to obtain consistent conclusions (Mascart et al., 1991; Kumar et al., 2011) and to improve the global application of models.

In this study, we analyzed the seasonal pattern of canopy resistance as determined from field observations and various environmental factors affecting the variability of canopy resistance during 2008-10, by employing the principal component regression method. We then investigated the relationship between the major factors and canopy resistance over the desert steppe site and modified the Jarvis scheme to improve the calculation of evaporation in the Noah LSM. We describe the dataset, the method of estimating canopy resistance determined by the Penman-Monteith equation and the Jarvis scheme, and the analysis framework in section 2. In section 3, we present the results of the variability of canopy resistance, and the development and validation of a new exponential water-stress function for the desert steppe site that can be applied to the Jarvis scheme. Finally, the conclusions are provided in section 4.

The Desert Steppe Ecosystem Research Station (44^#cod#x000b0;05'N, 113^#cod#x000b0;34'E) is located in Sonid Zuoqi in Inner Mongolia in China. The growing season is usually from late April to October (Yang and Zhou, 2011) with a plant community consisting of the dominant grasses Stipa Klemenziiand Allium polyrrhizum. The average height of the grass clusters is 20-35 cm at the grass's peak growth stage (Fig. 1), and the grass root depth is around 30-50 cm. The soil type is brown calcic, and, given its mechanical composition, is classified as sandy loam by the international soil texture classification system.

Figure 1.  Seasonal vegetation change of 2008 over the desert steppe site in Inner Mongolia, China.

The climate of this region is classified as an arid-semiarid, temperate, continent climate. Compared with other steppe ecosystems, the annual mean air temperature is much lower (Table 1). The peak leaf-area index (LAI) is comparable to grazed prairie in North Dakota and Mongolia, but is smaller than that of other grasslands (e.g., desert grassland in New Mexico). This site has the smallest annual precipitation among the grassland sites listed in Table 1, and its evapotranspiration is similar to that of grazed prairie in Mongolia. Even with a comparable precipitation amount to that of the grazed/semiarid prairie in Mandan on dry years, this site has less evapotranspiration (Table 1).

The measurements were taken from 2008 to 2010, including surface meteorological data measured by the automatic micrometeorological observing system (Table 2) and flux data measured by the open-path eddy covariance system (Table 3). All meteorological data were recorded every two seconds, and half-hourly averaged data were logged. The data from the eddy covariance system were recorded at 10 Hz by a data logger from which the half-hourly flux data were computed by the eddy covariance method (Monteith and Unsworth, 1990). Additionally, LAI was observed using a destructive method once a month from May to September in 2008 and 2009.

The corrections to the raw flux estimates included double coordinate rotations (Wilczak et al., 2001) and the density effects of heat and water vapor transfer (Webb et al., 1980). The outliers caused by occasional spikes of unknown origin in the half-hourly flux values were removed afterward (Papale et al., 2006).

The energybalance ratio (EBR) was used to assess the performance of the eddy covariance system (Wilson et al., 2002). It was calculated using the following equation for half-hourly periods where all the data (S, H, LE, and G) were available:

where Sis the net radiation (W m^-2), H, LE and Gare sensible, latent and soil heat fluxes respectively (W m^-2)

The soil heat storage flux (G) was calculated as

where G₀is the measured soil heat flux at depth z(0.08 m), T_sis the average soil temperature (K, at a depth of 0.05 m) above the heat flux plates, #cod#916; T_sis the change of the soil temperature over one time step, #cod#916; Tis the time step (in this case #cod#916; T=30min), and C_sis the soil heat capacity (Oliphant et al., 2004).

EBR was 0.89, 0.81, and 0.84 for 2008, 2009, and 2010, respectively, over the desert steppe in Inner Mongolia. These results are consistent with previously reported results (Twine et al., 2000; Yates et al., 2001; Wilson et al., 2002; Chen et al., 2007).

2.2.1. Canopy resistance calculated from the Penman-Monteith equation

Due to the uncertainty in empirical parameters in the approach of (Shuttleworth and Wallace, 1985), we followed the method of (Alfieri et al., 2008) and (Amer and Hatfield, 2004) to calculate the canopy resistance. The surface resistance was calculated by rearranging the Penman-Monteith equation (Monteith, 1965), using measurements of latent heat flux, soil heat flux, and weather variables over the desert steppe in Inner Mongolia:

The canopy resistance was estimated by using the LAI as the scaling factor (Amer and Hatfield, 2004; Alfieri et al., 2008):

where S, G, and LE are the same as for Eq. (2); (e_s-e_a) represents the vapor pressure deficit of the air (kPa); #cod#923;is the latent heat of vaporization (2.45#cod#215; 10⁶J kg^-1); #cod#961;is the mean air density (kg m^-3); c_pis the specific heat of the air (1.013#cod#215; 10³J kg^-1K^-1); #cod#916;represents the slope of the saturation vapor pressure-temperature relationship (kPa K^-1); #cod#947;is the psychrometric constant (kPa K^-1); r_sand r_aare the (bulk) surface and aerodynamic resistances (s m^-1); and r_cis the canopy resistance (s m^-1).

These calculated r_cvalues were used as the observed canopy resistance in this study, and our analysis was limited to data from May to September (i.e., the growing season) in 2008-10 that met the following criteria:

(1) Daytime: photosynthetically active radiation (K_p) #cod#62;200#cod#956;mol m^-2s^-1(Blanken et al., 1997);

(3) No precipitation events (excluded the data when precipitation occurred because latent heat flux measurements have errors when water is present on LI-7500 sensors);

(4) Avoiding periods of very small fluxes: latent heat flux (LE) #cod#62;10W m^-2(Wilson et al., 2002);

(5) Cloud effects removal, using accurate atmospheric incoming longwave radiation together with air temperature and humidity measurements (Marty and Philipona, 2000);

(6) EBR discrepancy #cod#60; 25%(Alfieri et al., 2008);

(7) Eliminating outliers: the final data were trimmed to retain data between the 2.5th percentile and the 97.5th percentile values.

The resulting sample for each year was 345, 452, and 382 data points, respectively.

2.2.2. Canopy resistance determined by the Jarvis scheme

The same Jarvis scheme as used by the Noah LSM (Chen et al., 1996) was applied here to compare with the modified regression model developed in this study. The Jarvis scheme was originally proposed by (Jarvis, 1976) and was extended by (Noilhan and Planton, 1989). Four environmental stress functions of the Jarvis model were used to predict stomatal resistance in this scheme and then LAI was used as the scaling factor to obtain the canopy resistance, described as follows:

where r_c,minis the minimum canopy resistance (40 s m^-1for grass as used in the Noah LSM), r_c,maxis the maximum canopy resistance (5000 s m^-1), and R_glis taken as a constant of 100 W m^-2. K_pis the photosynthetically active radiation (W m^-2), VPD is the vapor-pressure deficit (kPa), T_ais the air temperature (K), #cod#952;_iis the observed soil water content for the ith layer, d_iis the thickness of the ith soil layer, #cod#952;_wis the wilting point, #cod#952;_refis the field capacity, and d_totis the total thickness of the rooting depth. Values of #cod#952;_wand #cod#952;_reffor sandy loam are taken from (Chen and Dudhia, 2001). n_rootis the number of root layers. The root depth is 30-50 cm over this site; therefore, d_totis 40 cm and n_rootis taken as 4 in light of the observations of soil water content in this study.

Principal component regression analysis is a technique used to address the problem of multicollinearity and produce stable and meaningful estimates for regression coefficients (Fekedulegn et al., 2002). It transforms the original set of linear-correlated variables into a new set of an equal number of uncorrelated principal components that are linear combinations of the original variables (Abdul-Wahab et al., 2005). It allows the relative influences of the predictor variables to be determined from the loading coefficients associated with the principal components (Alfieri et al., 2007). The relative influence for a specific variable is expressed by the ratio of its loading coefficient to the sum of all the loading coefficients. For this study, principal component regression analysis was used to determine the relative influence of the factors driving the variability of canopy resistance.

Once the influences of the environmental factors had been determined, the relationships between canopy resistance and the key controlling factors were investigated. Then, based on the Jarvis model, a single regression model appropriate for the desert steppe in Inner Mongolia was developed.

The index of agreement (IOA) was used to evaluate the performance of the modified Jarvis scheme:

where M_iand O_iare simulated and observed fluxes, respectively. #cod#x0014c; is the averaged observed fluxes for the whole period.

The peak LAI of the in situ observation over the desert steppe was around 0.4 and 0.5 m²m^-2in Inner Mongolia for 2008 and 2009, comparable to the reported values for the northern semiarid grassland and grazed Mongolia steppe (Table 1). These in situ-measured LAI values agreed well with the Moderate Resolution Imaging Spectroradiometer (MODIS) MOD15A2 1 km LAI data (Fig. 2). For 2010, we used the MODIS LAI data to compute canopy resistance because of the continuity of the MODIS data.

Figure 2.  Comparison of in situLAI to MODIS LAI in 2008, 2009 and 2010 (dots represent MODIS observations, measured every seven days from 2008 to 2010; stars represent in situ observations, measured once a month for 2008 and 2009).

The annual accumulated rainfall was 136.3, 190.8, and 141.3 mm for 2008, 2009, and 2010, respectively (Fig. 3). The mean annual precipitation was 183.9 mm over the desert steppe (40-yr average for 1965-2004 data collected at the Sonid Zuoqi weather station), with 85% (i.e., approximately 156 mm) occurring from May to September. The accumulated rainfall amounts in 2008 and 2010 were similar, slightly less than normal years, but much less than that in 2009. Although they were similar in 2008 and 2010, it should be noted that this rainy season occurred from late April in 2009 and 2010, which was one month earlier than that of 2008, and the accumulated rainfall in 2010 was always greater than that of 2008 until late August.

Figure 3.  Time series of daily total rainfall, daily averaged soil water content of four layers (left column), and half-hourly canopy resistance (right column) for 2008 (upper panel), 2009 (middle panel), and 2010 (lower panel).

There were clear patterns for the soil water content (SWC) of the upper three layers, which responded well to rain events; the surface SWC was most responsive. During dry spells, the surface SWC was always around 5%, likely approaching the soil wilting point. It increased sharply following large rainfall events. Given the water-limited nature of arid ecosystems (Schlesinger et al., 1990), vegetative growth is generally precipitation driven (Neilson, 1986; Gosz, 1993) and can be influenced by episodic events of extreme precipitation (Mielnick et al., 2005). More frequent rains tend to enhance the growth of the more shallow-rooted grasses (Gibbens and Lenz, 2001). The increased SWC after rain events enhanced the root-zone available soil water and hence reduced plant water stress (Alfieri et al., 2008). In 2009 and 2010, because of the earlier start of rain (Fig. 3) and the resultant wetter soil, grasses began growing earlier and the variations of canopy resistance for the first two months were smaller than in 2008 (Figs. 4 and 5a). Due to wetter soil and less soil-water stress in 2009, the seasonal variability of canopy resistance was less than for the other two years. As expected, the precipitation primarily controlled the seasonal and annual variation pattern of the canopy resistance.

Figure 4.  Accumulated rainfall in 2008, 2009 and 2010; the period between the vertical dashed lines denotes the growing season.

Figure 5.  Monthly mean diurnal cycle of (a) canopy resistance, (b) photosynthetically active radiation, (c) vapor pressure deficit, (d) air temperature, and (e) minimum, median, and maximum of four layers averaged soil water content for May, June, July, August, and September, in 2008, 2009, and 2010.

After applying the quality controls, the retained data for the canopy resistance showed that it was most concentrated in the afternoon (Fig. 5a). The monthly mean canopy resistance was generally low around noon and higher in the late afternoon after a relatively stable stage. The seasonal variability of canopy resistance calculated from Eq. (4) for 2008 and 2010 was greater than for 2009, due to the relatively dry year. The seasonal patterns of canopy resistance in the three years were opposite to those of the range of the SWC (Fig. 5e). Because the root depth of grass over this site was around 30-50 cm, we used the average SWC from the four layers in this study for the principal component regression analysis.

The maximum peak K_poccurred in June and was approximately 380 W m^-2. There was no significant annual variability in the K_pwithin the three years, indicating a lack of solar radiation variations. As shown in Fig. 5c, there was seasonal and interannual variability in the VPD, with patterns closely tracking air temperature (Fig. 5d). Values of VPD and air temperature peaked in July in 2008 and 2010 and in August in 2009.

Figure 6.  Relationship between canopy resistance and soil water content in (a) 2008, (b) 2009, and (c) 2010. The lines represent Eqs. (11), (12), and (13), respectively.

To understand the sensitivity of canopy resistance to environmental conditions, five environmental factors [root-zone soil-water content (SWC), LAI, photosynthetically active radiation (K_p), VPD, and air temperature], based on the Jarvis scheme, were selected to compute the relative influences on canopy resistance through the principal component regression analysis (Fig. 6). The SWC contributed most to the variability of canopy resistance, and its relative influence ranged from 35% (dry years) to 47% (wet year). (Li et al., 2006) demonstrated that canopy resistance was sensitive to incoming short-wave radiation when the soil moisture availability was low at a grazing steppe. But this study found that there was a smaller sensitivity to the K_p(less than 10%) and larger sensitivity to changes in the LAI (#cod#62;20%) for 2008 and 2010 when the soil conditions were relatively drier. This did not necessarily mean that the effect of the K_pon canopy resistance was less important. The small sensitivity to the K_pwas perhaps because only the afternoon canopy resistance was examined and the K_placked interannual variability across the three years (Fig. 5) at this particular site. Under wetter soil conditions in 2009, the variability of canopy resistance became less sensitive to changes in the LAI but more sensitive to changes in the SWC. Regardless of the soil conditions, the relative influences of the VPD were similar for the three years, consistently about 20%. Air humidity controls whether the stomata were open or closed. This was found to be true in spite of an improving water status in the leaf at stomatal closure and an increasing water stress at stomatal opening (Schulze et al., 1974).

Figure 7.  The relative influences of five environmental factors driving canopy resistance in 2008, 2009, and 2010. T_a: air temperature; VPD: vapor pressure deficit; K_p: photosynthetically active radiation; SWC: soil water content; LAI: leaf area index.

Figure 8.  Comparison of observed canopy conductance (r_c^-1, m s^-1) and simulated canopy conductance (r_c^-1, m s^-1) for (a) modeling data and (b) the validation data. Blue circles represent the simulated values of the original Jarvis scheme; red stars represent the simulated values from the modified model; the lines indicate a 1:1relationship.

Because the canopy resistance was most sensitive to the SWC over this site, the following analysis focused on the relationship between the canopy resistance and the SWC. The canopy resistance decreased exponentially with the SWC (Fig. 7), consistent with the assumption of (Stewart and Verma, 1992) and (Cox et al., 1999a) that the canopy resistance was an exponential function of the SWC over grazed/ungrazed grass. (Alfieri et al., 2007) also found that the soil resistance decreased as an exponential decay function of the SWC for the range of SWC from 10% to 28% over a grazed rangeland during drought conditions. The regression relationships over the desert steppe were as follows:

2008 : r_c= e^{5.253+26.418/SWC}, R² = 0.712, P < 0.01, (11)

2009 : r_c= e^{6.223+18.647/SWC}, R²= 0.461, P < 0.01, (12)

2010 : r_c= e^{5.514+29.787/SWC}, R²= 0.778, P < 0.01. (13)

The coefficients were similar between the two relationships for 2008 and 2010, but were quite different for 2009. However, the differences were reasonable given that there were differences in the patterns of SWC between 2009 and the other two years (Figs. 4 and 7). If all the data points from the three years were put together, this exponential relationship could explain the approximately 67% variations of canopy resistance (not shown), and explain well the variations of canopy resistance during wet soils and dry soils.

Based on the analysis discussed in sections 3.2 and 3.3, we noticed that the patterns of SWC and the coefficients in Eqs. (11)-(13) were different during the dry years and wet years. If the regressed relationships were developed for each specific year, they could not have explained the variability of canopy resistance under different soil conditions. Therefore, we randomly split all the data from these three years including the SWC, VPD, T_a, K_p, and LAI into two halves. We then used one half to develop a regression model based on the Jarvis scheme and the other half to verify it.

We modified the linear SWC expression in the Jarvis scheme in Eq. (7) to an exponential expression for steppe grass. As a result, we obtained the regression model below:

where a=-0.7857; b=-16.3506; and adjusted R²=0.811, P#cod#60;0.01. It can explain approximately 81% of the variation of canopy resistance. In the original Jarvis scheme, the values of the four weighting functions were between 0 and 1. But in the proposed scheme [Eq. (2)], the function of the soil water content is not restricted. As shown in Fig. 8, the new modified scheme slightly underestimated canopy conductance, but the original Jarvis scheme always produced too high canopy conductance (i.e., smaller canopy resistance). Table 4 provides the statistical metrics used to evaluate the modified scheme. Although the coefficient of determination in the modified model was not much improved compared with the Jarvis scheme, the slope and the index of agreement (IOA) were much closer to 1 between the observed canopy conductance and the simulated values with the modified regression model. The modified regression model was better at simulating canopy resistance than the Jarvis scheme. However, we should note that the regression model was still an empirical relationship, and it depended mainly on the accuracy of the SWC.

The effect of our modified canopy-resistance scheme on latent heat flux was further quantified for different soil conditions. As shown in Table 5, the latent heat flux increased with soil moisture, as expected. The modified regression model produced substantially better latent heat flux for moist and wet soils, and effectively reduced the high bias estimated by the original Jarvis scheme. Neither scheme simulated the latent heat flux well when soil moisture was less than 5 m³m^-3, but the latent heat fluxes were low and the differences between simulations and observations were generally smaller. Over this desert steppe site, the extremely dry and wet soil conditions did not occur often during the growing season in 2008-10. The fact that different canopy-resistance schemes could produce largely different latent heat fluxes demonstrates that canopy resistance has an important control on latent heat fluxes over the desert steppe.

Given the significant contribution of plant transpiration to the total evaporation and the important control of canopy resistance at the investigated desert steppe site in Inner Mongolia, the seasonal and interannual variability of canopy resistance and the driving environmental factors were analyzed using field observations collected in 2008, 2009, and 2010. Compared with other grassland sites, the desert steppe had a low LAI with a seasonal maximum around 0.4 to 0.5 m²m^-2, low annual total precipitation, and low annual air temperature. Its evapotranspiration was comparable to that of grazed prairie in Mongolia, but smaller than that of grazed/semiarid prairie in North Dakota in a dry year. Grass growth was closely linked to the SWC and was essentially driven by precipitation in this water-limited ecosystem (Schlesinger et al., 1990).

Canopy resistance had distinct seasonal and interannual variabilities. Using the principal component regression analysis method, we analyzed the relative influences of five main environmental factors as identified in the Jarvis scheme, and found that the SWC contributed most to the changes of canopy resistance with relative contributions greater than 35%. The amplitude of seasonal canopy-resistance variations in those years was inversely proportional to that of the SWC, and there was a strong exponential relationship between them. This study revealed a strong sensitivity of canopy resistance to changes in the LAI, which contributed to more than 20% of the canopy resistance variations in 2008 and 2010 during dry spells with water-stressed grass. The VPD contributed to about 20% of the canopy resistance variations, and had an interannual variability similar to that of the air temperature, with a single-peak seasonal tendency centered in July, except for 2009 when the peak occurred in August. Variations of canopy resistance did not respond to the K_p(its relative contribution being less than 10%) due to the lack of interannual variability in the latter across those three years.

Based on these analyses, we proposed a modified Jarvis scheme by incorporating an exponential relationship between canopy resistance and the SWC, which substantially improved the calculation of canopy resistance. We considered the modified scheme as an empirical relationship and approximately 19% to 23% of the variations in canopy resistance still cannot be explained by this new formulation. Nevertheless, it much improved the calculation of latent heat fluxes, especially for moist and wet soils, and effectively reduced the high bias in evaporation estimated by the original Jarvis scheme. This study highlights the important control of canopy resistance on plant evaporation and growth at the investigated desert steppe site despite the relatively low LAI. Regarding future work, we will explore the use of this analysis framework for observation sites with different land-cover and further improve the widely used Jarvis scheme. Some studies have suggested that the Jarvis scheme could improve the simulations of surface energy flux by a significant tuning of the minimum stomatal resistance (r_c,min) (Alfieri et al., 2008; Niyogi et al., 2009). However, we could not derive a reliable r_c,minfrom the limited observations we had and there is no observed r_c,minin existing literature. For instance, our preliminary analysis revealed that the original Jarvis scheme with a value of 80 s m^-1is a good candidate for r_c,minover this desert steppe site rather than the 96 s m^-1that (Alfieri et al., 2008) suggested. Therefore, we also plan to calibrate r_c,minand other physical parameters systematically with optimization software——for instance, the shuffled complex evolution (SCE-UA) algorithm (Duan et al., 1992, 1993)——in order to improve flux simulations.