1. Introduction
It has been widely established that the parameterization of land surface processes, in particular the way energy is partitioned on the earth’s surface, appears to significantly affect the performance of regional weather and climate models (e.g., Viterbo and Beljaars 1995, and references therein). During the last few decades various parameterizations have been continuously developed and improved. A major advance for further development and evaluation has been achieved by the Project for the Intercomparison of Land Surface Processes (PILPS; Pitman et al. 1999; Henderson-Sellers et al. 1993, 1995). Furthermore, long-term field experiments [e.g., FLUXNET (Baldocchi et al. 2001) and Northern Hemisphere Climate Processes Land Surface Experiment (NOPEX; Halldin et al. 1999)] helped to provide deeper insights into potential model limitations and weaknesses.
During recent years there has been a growing interest in the study of semiarid ecosystems to enhance the overall model performance in these areas. In semiarid ecosystems, strong feedback mechanisms exist between vegetation patterns and soil water availability (Niyogi et al. 2002). These mechanisms make semiarid ecosystems a challenging task for further parameterization evaluation.
At the present time a few long-term evaluations for semiarid regions exist (see, e.g., Unland et al. 1996; Emmerich 2003; Hogue et al. 2005). A first attempt to gain a deeper understanding of semiarid regions in West Africa was made in the context of the Sahelian Energy Balance Experiment (SEBEX; Wallace et al. 1992), which focused on the annual cycle in the northern Sahel region. The Hydrological Atmospheric Pilot Experiment (HAPEX)-Sahel (Goutorbe et al. 1994) followed in 1992 with a 3-month intensive observation period to obtain information of surface-flux estimation on the scale of a typical general circulation model (GCM) grid box. Verhoef (1995) studied various land surface schemes in this context and provided the first ideas of needed input parameters and their possible temporal variability. For the more southern region including the Volta basin there is still a lack of data and therefore a lack of model evaluation. Because of the large regional and temporal rainfall variation and its impact on soil moisture, this region is an ideal test area for model evaluation of seasonal time scales. In addition, Viterbo and Beljaars (1995) and Chen et al. (1997) stressed the need for model evaluation of seasonal time scales.
The main objective of this study is to accomplish a performance test of two operational land surface models (LSMs) for a seasonal cycle in the Volta basin by focusing on the correct representation of surface fluxes and soil moisture at two different test sites. To account for the large variability in time and space the performance test is carried out at two test sites, located in the humid tropical southern region and in the drier northern region.
The first land surface model employed in this study is the present European Centre for Medium-Range Weather Forecasts (ECMWF) land surface scheme [Tiled ECMWF Surface Scheme for Exchange Processes over Land (TESSEL)], which is based on the work of Viterbo and Beljaars (1995) and Van den Hurk et al. (2000). The second LSM is the Noah model from the National Centers for Environmental Prediction (NCEP; Ek et al. 2003); version 2.7.1 is used in this analysis. Both models were involved in a number of intercomparisons of land surface schemes (Betts et al. 1997; Van den Hurk et al. 2000; Schlosser et al. 2000; Boone et al. 2004). These two models were selected because they are updated regularly and differ in their physical aspects and the degree of complexity with which land surface processes are parameterized. In the present study an additional approach within Noah is tested to calculate canopy conductance based on plant physiology (A-gs method; Jacobs 1994; Ronda et al. 2001).
The first focus of the performance test is on the correct representation of seasonal dynamics of the different components of the energy balance and soil moisture for all parameterizations. The second focus is on adjustments of the different roughness length ratios in all models so as to obtain enhanced model performance. Based on the findings, the third focus is on obtaining a better knowledge of the changes in surface conductance within the different parameterizations, such as the critical variable for partitioning the available energy into sensible and latent heat flux during the chosen time period.
For the purpose of surface-flux evaluation a large aperture scintillometer (LAS) is utilized (see, e.g., De Bruin et al. 1995) because this robust method yields area-averaged fluxes over complex terrain, which are required when evaluating models with different subgrid surface fractions composed of high and low vegetation and bare soil. Moreover, the use of LAS for semiarid vegetation types is new.
2. Material and methods
a. Study area
The Volta basin is located in the intermediate zone and the southern part of West Africa (Fig. 1). It comprises an area of about 400 000 km2. The climate system in the Volta basin is very much controlled by the meridional movement of the intertropical convergence zone (ITCZ), the African easterly jet (AEJ), and pressure disturbances that traverse Africa from east to west (Burpee 1972). They all react to the influence of the Hadley and Walker cell circulations. At the northern and southern extremes of the Hadley cells, the climates tend to be monomodal arid to semiarid. Toward the equator the climates shift to bimodal humid and per-humid.
The data in this study are based on measurements from meteorological stations in Ejura (7°20′N, 1°16′W) and Tamale (9°29′N, 0°55′W), Ghana, from the year 2002 (Fig. 1). The mean annual rainfall at the sites is 1432 and 1082 mm (FAO 1984; including data from 1930 to 1980). For Ejura (Tamale) the total annual rainfall in 2002 was 1420 mm (1065 mm). Daily levels of precipitation are shown in Fig. 2. Forty years of rainfall at different stations in the Volta basin were analyzed regarding the beginning and end of the rainy period using the method of Kasei (1988). It was found that the start and the end of the rainy season for 2002 was within the range of one standard deviation when compared to the climatology. This demonstrated that the analyzed period was not exceptional in terms of varying vegetation cover and evapotranspiration.
The two sites show major differences concerning the vegetation, soils, land use, slopes, and also climate. Each of these sites was considered representative of larger areas and their choice took into account the different geographical and climatological parameters influencing the hydrological cycle. All measurements were gathered in the context of long-term observations concerning the water and energy balance in the Volta basin within the Global Change in the Hydrological Cycle (GLOWA)-Volta project (Van de Giesen et al. 2002).
The site in Ejura is the more tropical site. The landscape is hilly. It is a heterogeneous terrain. Here the transmitter of the LAS (further description below) and the automatic weather station (AWS) were located in a cashew orchard. The receiver of the LAS was located at the edge of a forest. The area between the transmitter and the receiver can be roughly divided into two parts. On the transmitter side the vegetation consisted of cashew trees with maize in between. On the receiver side there were bushes and trees and small swamps, but almost no agriculture.
The research site in Tamale is mainly characterized by natural grassland with scattered trees with a maximum height of 6–8 m. The landscape is slightly hilly. The automatic weather station was installed next to the receiver of the LAS.
Data availability exceeded more than 90% for most of the measurements at the two sites during the season studied. Further details can be found in Schüttemeyer et al. (2006).
b. Forcing data
c. Validation data
Net radiation was measured directly at a height of 8 m and soil heat flux was measured at a depth of 0.035 m. The soil heat flux at the surface was calculated by applying the method of Heusinkveld et al. (2004), whereby another conventional soil heat flux sensor was moved to a location just below the surface and compared to the analysis of the temperature wave propagation in the soil. On average this added 30%–40% of soil heat flux at a depth of 0.035 m. Since soil temperature measurements were only available in Tamale, the method was only applied there. Because of the denser vegetation cover, the soil heat flux was smaller in Ejura than in Tamale. Consequently, the correction of storage above the soil heat flux plate was also less.
Sensible heat flux was validated against the LAS measurements. The length of the path in between was about 2030 m in Ejura (2420 m in Tamale). The weighted effective height of the LAS was estimated as being 30.1 m in Ejura (19.5 m in Tamale). The emitted radiation is scattered by the turbulent medium in the path. The variance of intensity of received radiation is proportional to the structure parameter of the refractive index of air (C2n). At the wavelength used (940 nm), the refractive index mainly depends on temperature, so C2n is mostly determined by temperature fluctuations (C2T). The influence of humidity on the refractive index was considered in accordance with the work of Wesely (1976) and Moene (2003).
Sensible heat flux was calculated from C2T using the Monin–Obukhov similarity theory (MOST). Stability functions proposed by Wyngaard (1973) were applied for daytime values. For nighttime values we followed the formulation of De Bruin et al. (1993). For a more detailed description of the LAS theory and its applications see, for example, De Bruin et al. (1995) or Meijninger et al. (2002).
Latent heat flux was calculated as a residual from the energy balance, which showed good correspondence with eddy covariance data obtained during an intensive observation period (IOP) during the drying up of the region in 2002 (Schüttemeyer et al. 2006). One has to remember that this approach forces the measured energy balance to close. The error in daily measured energy balance closure was smaller than 10% for the IOP in 2002 in Tamale and for large parts of the season in Ejura (Schüttemeyer et al. 2006).
d. Model descriptions
A brief description of all models is given here, with a focus on the main differences, which are summarized in Table 2. The models were run in offline mode to concentrate on the evaluation of the land surface processes for the specific region and on the surface-flux formulation, without dealing with mismatches in the upper boundary condition between a three-dimensional model and the observations. Instead the models were driven by the prescribed atmospheric forcings described above. The time step for integration was set at 600 s according to the measured quantities.
Both models contain a multilayer submodel for the soil. The water and heat budget in the soil is based on the Richards equation coupled to the Fourier law of diffusion. The soil is discretized into four layers in both models (0.08, 0.24, 0.72, and 2.16 m in thickness). The lower levels differ from the original setup of both models. The changes were applied to account for the deeper rooting systems. A deep groundwater table is considered with little capillary rise and bottom boundary conditions of zero heat flux and free drainage. This is justified by the fact that minimum well depth was estimated to 8 m in the surroundings around the two test sites.
For both models an aerodynamic conductance for H and LE and a canopy conductance for LE are utilized. The turbulent exchange is based on an iterative transformation of Richardson’s number into Monin–Obukhov stability parameters (Beljaars and Holtslag 1991) for TESSEL. For Noah the formulation of Paulson (1970) is used to calculate the turbulent exchange coefficient. In both models evapotranspiration is the sum of evaporation from the bare soil (Edir) and the interception reservoir (Ec) and transpiration from the vegetation (Et), which TESSEL divides into high and low vegetation. The exact formulation of the different evapotranspiration components can be found in the appendix.
For Noah the A-gs method is utilized as well, where the photosynthetic assimilation is coupled to the leaf stomatal conductance. It has the advantage of being more physically based and fewer parameters are needed compared to the more standard Jarvis–Stewart approach. One additional advantage is the potential ability to calculate carbon dioxide fluxes based on common atmospheric variables.
Soil moisture stress is the important factor for evapotranspiration during drying-up periods. All models include a formulation for the preference of the plants to extract water from those layers where the liquid water content exceeds the wilting point. For TESSEL soil moisture stress is calculated, utilizing the root density, whereas for Noah the soil depth is used. The exact formulation for the soil moisture stress function can be found in the appendix.
e. Model parameters and initial values
The model runs for Tamale were conducted for the period from day of year (DOY) 238 until DOY 21 and for Ejura from DOY 269 until DOY 52. Both periods cover the transition time from the wet to the dry season in the studied region. For Noah the vegetation type was set to savannah for Tamale and to broadleaf–evergreen trees (tropical forest) for Ejura. For TESSEL the vegetation types of low and high vegetation were a mixture of grass and interrupted trees in the case of Tamale, and for Ejura a mixture of low (grass) and high vegetation (broadleaf–evergreen trees).
Since strong feedback mechanisms exist between vegetation patterns and soil water availability, one would expect a strong dependence of model performance on vegetation fraction. Therefore database values for the vegetation fraction within Noah were discarded and the vegetation fraction derived from the enhanced vegetation index (EVI) was applied instead. EVI data are routinely produced from the data of the Moderate Resolution Imaging Spectroradiometer (MODIS) on board the Terra and Aqua satellites. In this study 16-day composites of EVI were utilized and gridded at a spatial resolution of 1000 m. Level 3 data (version 4) from the Aqua satellite were used (see Schüttemeyer et al. 2007). Since TESSEL makes use of a fixed vegetation fraction, the highest actual EVI values were applied for both sites. The percentage of high and low vegetation for TESSEL was adopted, based on local observations. The chosen values for various parameters for the test sites are given in Tables 3 and 4. In a detailed survey of soil properties (Agyare 2004) it was found that the soil types of both sites corresponded, both being a sandy loam. For all models the needed parameters were set accordingly. A uniform root distribution was considered for all models. This setup deviates from the original setup of TESSEL, where globally averaged soil values and vegetation-dependent root distributions are utilized. Eventual impacts are discussed in section 3e. Other model parameters were set to values generally used within the operational 3D version of the ECMWF model corresponding to the values at the closest grid point (in case of geographically dependent parameters) or data are taken from the U.S. Geological Survey (USGS). Since there were no measurements for soil temperature profiles available at either site those initial values were obtained by calculating the daily, weekly, monthly, and yearly air temperature and by using those values for the four soil layers. In the case of initial soil moisture, the values for Ejura were interpolated from the closest measurements taken on an irregular basis varying from 3 to 7 days. For Tamale the routinely measured values were taken. For initial soil moisture for the lowest level the values from the level above were taken.
3. Results
In this section the comparison of the simulated and measured components of the energy balance are presented. For each location the linear Pearson correlation coefficient, a linear regression, and the mean bias error (mbe) were calculated on a 3-day basis. The linear regression was carried out using a least absolute deviation technique (Birkes and Dodge 1993). This method was chosen because of its lower sensitivity to outlying data. The mbe is the degree of agreement between model and measurement and gives an indication of the accumulated quality. Lower values indicate a better performance and negative (positive) values indicate underprediction (overprediction). To deduce conclusions regarding the quality of daytime and nighttime predictions, scatterplots are utilized. The evolution in times of roughness length ratio, soil moisture, and the role of the canopy conductance are analyzed. The main findings are interpreted in section 4.
a. Available energy
There is a clear decline in measured net radiation related to changes in the surface cover at both sites (Fig. 3). The decline starts as soon as the amount of rainfall decreases. The seasonal dynamics give a correlation coefficient that equals 0.99 for all parameterizations throughout the studied season. The mbe shows a clear trend from small negative values in the beginning to larger positive values toward the end. For Ejura the mbe is largest in the dry period between the rains, which started on DOY 343 and ended again on DOY 23. During the dry period the mbe in net radiation increased up to 34 W m−2 for Ejura (60 W m−2 for Tamale). This suggests that the process of drying up influences the model performance, since the decrease in net radiation started earlier (Fig. 3), but the mbe remained small. During the wet part of the season TESSEL showed slightly higher negative values for the mbe compared to both parameterizations of Noah. For the drier part TESSEL showed similar or lower values for the mbe. During that time Noah A-gs showed the largest mbe. Figure 4 shows that there is no phase shift in time between models and measurements, which indicates that the diurnal cycle was modeled correctly. The overestimation at the end of the period is also detectable in the group of points to the right of the 1:1 line. Noah JS and TESSEL slightly underestimated net radiation for higher values, whereas Noah A-gs showed smaller errors.
The measured soil heat fluxes also increased as soon as the rain stopped at the two sites (Fig. 5). The mbe was negative for all parameterizations during large parts of the analyzed periods. Only during days with a high rainfall did the mbe change sign. This might be related to thermal conductivity changes due to varying soil water content and might impact the quality of the dataset. The correlation for soil heat flux is small in the beginning with a minimum of 0.6 and increases toward the end with a maximum of 0.97. TESSEL shows larger errors for most of the time at both sites. The two versions within Noah give comparable results for the mbe. Again, there is no phase shift in time between models and measurements (Fig. 6). For Ejura, Noah JS overestimated soil heat flux. For TESSEL, larger values of soil heat flux and nighttime values were not modeled correctly. This clearly leads to a larger negative mbe of daily values. For Tamale nighttime values were reproduced fairly well by all models, whereas all models underestimated daytime values.
The evaluation at both sites shows that the measured decrease of net radiation and the measured increase of soil heat flux are not reproduced correctly by the LSMs and that the errors are related to the process of drying up. This suggests that parameters that influence skin temperature need to be explored in this context. In general, available energy during the wet part of the season is modeled with small errors mainly due to small errors in net radiation. Therefore, only the partitioning between sensible and latent heat flux was perhaps faulty and needs to be studied. For the dry part of the season there was too much available energy at the surface, which could be translated into overestimations of H and/or LE in the different models.
b. Sensible and latent heat flux
Because there are no measurements for sensible and latent heat flux available for the test site in Tamale until DOY 287 and for the last 3 to 4 weeks of the studied time for both sites, the validation period for H and LE will be shorter than that for available energy.
Measured sensible heat fluxes remained almost constant throughout the season at both sites (Fig. 7). The correlation for H is always higher than 0.77 with a maximum of 0.94 at the end of the period. The mbe is positive for all formulations during large parts of the season at both sites. For Ejura the magnitude of the mbe is comparable for all models at the beginning of the analyzed period. The mbe grows during periods with higher rainfalls. Furthermore, it grows for Noah A-gs and TESSEL at the Ejura test site as soon as the process of drying up is enhanced, whereas for Noah JS no real trend is detectable. For Tamale the mbe grows in both parameterizations within Noah when the drying up is enhanced, whereas TESSEL shows a smaller mbe during that time. Figure 8 shows that for both sites H has been overestimated, with large errors for stable situations occurring during the night. Newer versions of Noah include a different turbulent exchange coefficient to tackle this issue. For Ejura it was observed that Noah JS gives erroneous results for nighttime and higher values of H during daytime, whereas Noah A-gs and TESSEL modeled H more correctly with a slightly better performance of Noah A-gs. The erroneous results of Noah JS for larger values of H led to a smaller mbe compared to TESSEL and Noah A-gs. For Tamale TESSEL performs better, as already indicated by the calculated mbe.
The measured latent heat fluxes show a clear decrease at both sites (Fig. 9). The correlation shows a clear trend of high values (0.97) in the first 8 weeks. After that period it slowly decreases to 0.75 for all parameterizations. The mbe is negative with comparable magnitude for all formulations during the wet part of the season. Changes in sign occurred at the beginning of the dry season in all models, with smaller magnitudes for Noah A-gs. For Ejura the diurnal cycle is reproduced more realistically by all parameterizations (Fig. 8) compared to Tamale. For Tamale both parameterizations of Noah give similar results with underestimation of LE for large values. Here TESSEL shows a large amount of scatter.
In general, the results for sensible and latent heat flux indicate that sensible heat flux is overestimated by all models with different magnitudes. Latent heat flux is modeled with small errors during the wet part of the season and increasingly larger errors appear when the surface becomes dry. Furthermore, the largest errors occur during the period in which net radiation is overestimated, where the partitioning of the error into H and LE varies for the different parameterizations. Noah A-gs performs best for latent heat flux and shows the largest error of sensible heat flux, whereas Noah JS and TESSEL perform better for sensible heat flux.
A possible explanation for the errors in sensible heat flux could be the exchange coefficient for heat. Since Noah JS and Noah A-gs utilize the same parameterization and have different results this should not be the case. Furthermore, Chen et al. (1997) compared different atmospheric surface layer parameterization schemes and concluded that the differences in the schemes and the resulting surface exchange coefficients did not, in general, lead to significant differences in model-simulated surface fluxes and skin temperature. In general, the models were more sensitive to changes of ratio of roughness length for momentum and heat. Therefore the roughness length ratio, or kB−1 with the Zilitinkevich constant (C) as a proxy, is the critical variable for model performance.
For latent heat flux the seasonal development of soil moisture and the different canopy conductance formulations need to be explored in this context.
c. Roughness length ratio
The increasing mbe for net radiation together with larger errors in sensible heat flux during drying-up periods suggests that the surface temperature was not calculated correctly for all models. For Ejura this cannot be proven, since there are no surface temperature measurements available. For Tamale there are measurements available, but only during the IOP, which took place in November and December 2002. The direct comparison of measured and modeled surface temperatures from Noah (one universal) and TESSEL (separate surface temperatures for bare soil and low vegetation) shows that at the beginning of the IOP during daytime the differences between measured (sensor placed above a mixture of bare soil and grass) and modeled surface temperatures were small, but the differences increased toward the end of the year (Fig. 10). To improve the model results it is crucial to analyze the impact on sensible heat flux of changed roughness length ratio or kB−1 with the Zilitinkevich constant (C) as a proxy.
Therefore those ratios are varied for all parameterizations (see also Holtslag and Ek 1996). Huntingford et al. (1995) estimated a ratio of 100 for savanna-type vegetation in contrast to the ratio of 10 applied by TESSEL. To test the influence of a changed ratio of roughness length for momentum and heat, a ratio of 100 and 1000 are employed. For both parameterizations within Noah, C is set to 0.4 and 0.6 instead of 0.2. For all models the surface temperature is reduced by 2 K (C = 0.4) to 4 K (C = 0.6) and is more consistent with the measurements during the IOP. For the test site in Tamale changes in mbe for net radiation are shown in Fig. 11. The mbe is reduced toward the end of the season, but rises during the wet part with larger changes for both parameterizations within Noah.
The effect of changed roughness length ratio on soil heat flux also led to improved results; the overestimation during daytime was especially reduced. For the soil heat flux a simple one-at-once sensitivity analysis based on local observations for all important input parameters was performed. The largest impact on soil heat flux was related to quartz content and b parameter. In general, the effects were smaller than the effect of changed roughness length ratio.
Regarding sensible heat flux, the model runs with C = 0.4 (roughness length ratio of 100) reduced the mbe, whereas C = 0.6 (roughness length ratio of 1000) limited sensible heat flux excessively.
Regarding latent heat flux, the effect of changed roughness length ratio was small during the dry part of the season but it led to further underestimation of latent heat flux during the wet part of the season.
d. Soil moisture
Since measured latent heat flux decreased drastically at the test sites, it can be said that the vegetation suffers from soil moisture stress. To test the influence of the initial model settings on soil moisture stress the initial soil moisture content of the three models at both sites were varied. The volumetric soil moisture content in each layer was varied by an increase and decrease of 0.1 m3 m−3. To analyze the resulting differences, the 3-day average differences for the components of the energy balance were calculated. In a second step the seasonal course of soil moisture was compared to the measurements.
The influence on net radiation at both sites for both parameterizations within Noah is zero for higher soil moisture and negligible (<0.5 W m−2) for lower soil moisture content. For TESSEL the influence of changing initial soil moisture content is detectable throughout the season for both sites. The average difference in net radiation with the lowered initial soil moisture is about 10 W m−2 during the wet part of the season and increases to 20 W m−2 toward the end. For higher initial soil moisture the net radiation also changes, but the effect is smaller, showing maximum values of about 10 W m−2. For sensible and latent heat flux modeled by Noah the differences are small (maximum 2 W m−2) for the first 4 to 5 weeks and zero afterward. For TESSEL the differences in surface fluxes with lower initial soil moisture are about 10 to 15 W m−2 for the wet part of the season. After that the differences reach values up to 20 W m−2 in sensible heat flux and lower values of about 5 W m−2 for latent heat flux. The largest influence of changed initial soil moisture content is observed when the rain stops and the drying-up process starts.
In addition, the average soil moisture in the first two model levels was compared to the measurements. All models show slightly higher daily values during the wet part of the season (Fig. 12) and reduced soil moisture during the process of drying up. For the dry part of the season Noah JS and TESSEL overestimate soil moisture content in the upper two layers. The A-gs parameterization performs best, especially when the rapid drying starts. For Ejura a similar development was observed. The seasonal course of soil moisture in the upper two layers confirms the different model performance of latent heat flux. However, it has to be noted that a complete evaluation of soil moisture is not feasible.
e. The role of canopy conductance
Since the surface dries rapidly as soon as the rain stops, one would expect only a small amount of bare-soil evaporation. Therefore the canopy conductance and the vegetation fraction are the critical variables for partitioning the available energy into sensible and latent heat flux during the drying-up process. For both sites the vegetation characteristics show a mixture of vegetation and bare soil in both models, with a seasonally dependent vegetation fraction for Noah.
The canopy conductance is calculated from the observations using the rearranged Penman–Monteith equation (e.g., Monteith and Unsworth 1990) following Harris et al. (2003). For the calculation of the aerodynamic conductance, the method of Verma (1989) with stability corrections by Paulson (1970) was applied. The roughness length for momentum was estimated by using the approach of Martano (2000), and roughness length for heat was estimated by applying a fixed ratio of roughness length for momentum and heat of 100, which was estimated for savanna conditions by Huntingford et al. (1995). Under conditions of low available energy, the computation of gs can result in nonrealistic, large conductances. Therefore, the calculation of gs was restricted to daytime situations between 9:00 a.m. and 5:00 p.m.
Concerning the mbe of latent heat flux, the applied parameterizations gave different results (Fig. 7). For the canopy conductance all models show larger values at both sites when compared to the conductance calculated by the rearranged Penman–Monteith equation. For Ejura this is depicted in Fig. 13. Noah JS especially overestimated the canopy conductance and therefore did not react to the drying process in the same way as Noah A-gs did. TESSEL performed best concerning the canopy conductance, but nevertheless latent heat flux was not correctly reproduced. These results suggest that the vegetation fraction must be a controlling mechanism in determining evapotranspiration. For TESSEL it does not vary on a seasonal basis and together with the uniform root distribution it might explain the larger errors in latent heat flux during the drying-up period, whereas for Noah JS the vegetation fraction limits the transpiration to a certain extent. The strong dependence of Noah on vegetation fraction and its potential errors have been observed before (Kurkowski et al. 2003) and might lead to larger model errors if database values are utilized.
The important factors required to model the surface conductance correctly are the different functions that control the response on different time scales (appendix). Soil moisture dependence and vapor pressure deficit (VPD) dependence are necessary to adjust canopy conductance during the seasonal cycle, whereas radiation and temperature dependence are more important for the diurnal cycle. The response of the different functions in the framework of the total canopy conductance is shown in Fig. 14. The influence of temperature is only shown for Noah JS, since TESSEL does not make use of this dependence. It can be seen that the value of this function is close to 1, which means there is only small influence on the canopy conductance on a seasonal basis. Concerning the radiation-dependent function, both models show a similar behavior on a seasonal basis. For the VPD-dependent functions, both models start with similar values for this response function but the decrease is more pronounced in Noah JS. The soil moisture response function shows saturation in both models at the beginning but goes down for TESSEL as soon as the time of transition starts. From that point onward the biggest differences appear in this function. TESSEL shows a larger decrease of this value compared to Noah.
Since the VPD and soil moisture changed significantly at both test sites, a one-at-once sensitivity analysis was performed for hs (Noah JS), gD (TESSEL), and Θref and Θw to test the sensitivity of the stomatal conductance to the mentioned parameters.
Larger changes in the seasonal development of VPD and soil moisture response could only be achieved by applying changes higher than 20% and was therefore only conditionally suitable for the above-mentioned soil types and vegetation. The outcome indicates that the difference between the different parameterizations on a seasonal basis is not in the varied parameters but in the shape of the functions utilized in the framework of the Jarvis–Stewart approach or in the root distribution.
Therefore, the influence of changes in root density were tested by applying a variable root distribution for Noah JS and TESSEL based on the parameters of the original setup of TESSEL (Van den Hurk et al. 2000). The soil moisture response function showed a larger decrease in both models, and stomatal conductance was more consistent with the measurements, especially after the rain stopped at both sites. Furthermore, it was observed that soil moisture content in the upper two levels decreased too fast during the drying-up process.
4. Discussion and conclusions
This study is meant to serve as a test for two operational land surface models in a remote region with large contrasts between wet and dry conditions. Those contrasts are prominent in the studied regions compared to midlatitudinal or tropical regions. In addition, a different parameterization within Noah based on plant physiology was tested.
The evaluation of the modeled components of the energy balance showed rising values in time for the mbe in net radiation and soil heat flux, whereas the evolution in time for sensible heat flux and latent heat flux was tackled in different ways by the utilized parameterizations. For the wet part of the season latent heat flux is underestimated by all models. This might be related to the fact that the different models utilize a linear function for reacting to soil moisture stress [cf. Eqs. (A10) and (A18) in the appendix]. However, if soil moisture is close to field capacity, plants generally do not react linearly to soil moisture stress (Feddes et al. 1978). Using a linear function would then lead to an underestimation of the photosynthetic rate and LE is underestimated (Wang and Leuning 1998).
At the Tamale test site the model performances were worse at the end of the season compared to Ejura. This can be related to an enhanced process of drying up in the northern region. Because of this process, larger changes in surface cover occurred during the studied period compared to Ejura.
TESSEL and both formulations within Noah show small errors during the wet part of the season but fail to spot the dynamics of surface temperature and/or of the albedo caused by the drying up. One reason for this failure could be the use of database values for albedo, which has a seasonal cycle, although the seasonal dependence might be too small. This results in the overestimation of the net shortwave flux at the surface. Since the model was run in offline mode there should be no uncertainties related to cloud schemes and clear-sky absorption. Modelers consistently address the seasonal dependence of albedo, which demonstrates that there is a need for real-time parameters under the changing conditions in West Africa. The rapid changes in surface cover lead to the question of whether monthly mean values are suitable or if a higher resolution in time should be taken into account.
Tuning the Zilitinkevich constant in the case of Noah and applying a different roughness length ratio in the case of TESSEL could improve the results of both test sites. The upcoming issue of applying this is that the results were of poorer quality during the wet period. Apparently the originally applied ratios are suitable for the wet part of the season but seasonal adjustments will have to be made. The different results might be explained by the changes in surface cover, which appear due to the process of drying up. This may lead to more homogenous conditions at the beginning, especially in Tamale. For those situations, the use of one single surface temperature is justified. For more heterogeneous conditions toward the end, the concept of one surface temperature can only be applied if the roughness length ratio is adjusted. This would also explain the smaller impact of changed roughness length ratio for TESSEL, since TESSEL utilizes different skin temperatures for each tile. Furthermore, the lower model bias error for Noah when applying a different Zilitinkevich constant might be related to the more dynamic formulation of the ratio of roughness lengths, since the daily variability of wind speed increases toward the end of the season. These dynamics cannot be tackled by a fixed ratio of roughness lengths as applied to TESSEL. In future schemes, a different formulation of the ratio of roughness length for heat and momentum like the one proposed by Kubota and Sugita (1994) might help reduce the errors. A seasonal evolution of vegetation-related parameters as already mentioned by Van den Hurk et al. (2000) might also be necessary to improve the erroneous surface fluxes of both models. In this context one upcoming issue is that differentiating only between grassland and forest is not a suitable solution for savanna conditions.
It was shown that the canopy conductance parameterization, combined with a fixed vegetation fraction in TESSEL, reduces the amount of latent heat flux during the time of transition. The daily median of canopy conductance showed more consistency between the observations and TESSEL. For Noah JS one driving force in reducing evapotranspiration during the drying-up process is the vegetation fraction, since the reduction in canopy conductance did not show a comparable response as seen in TESSEL. In addition, the larger positive mbe in LE for Noah JS during the drier part of the season shows the need for further improvement of the JS formulation. Here the soil moisture response in particular needs to be optimized. The use of the A-gs parameterization clearly showed its robustness in this context. In future tests a detailed sensitivity analysis should be performed to gain further knowledge about the strengths and weaknesses of this approach. The implementation of the A-gs approach together with a formulation of soil respiration would make it possible to calculate carbon dioxide exchange based on common atmospheric variables.
The calculation of canopy conductance of TESSEL is based on soil moisture and radiation in the case of low vegetation, and water vapor pressure only being included when high vegetation is involved. Since the conductance formulation shows a large response to soil moisture content in the different layers, it is crucial to initialize the model runs with correct soil moisture content. When the initial soil moisture content was varied, it could be shown that this had a small but observable effect on the calculation of surface fluxes in TESSEL throughout the studied period. This behavior could not be observed with Noah due to the different calculation of canopy conductance and a lower influence of soil moisture. Furthermore, the changes in soil moisture showed a nonlinear response (lower values lead to larger changes) in the components of the energy balance, which was already observed by Niyogi et al. (2002). To improve the overall model performance in the region, in the first instance a better monitoring of certain input parameters and the application of seasonally dependent parameters might help to obtain improved modeling results. To obtain more robust simulations under these extreme conditions, new ideas need to be developed to capture, for example, the effect of deep rooting systems, which was beyond the scope of this study.
Acknowledgments
This research was sponsored by the Federal Ministry of Education and Research, Germany. The authors thank Dirk Malda and Ivar van der Velde for a detailed analysis of rainfall data and three anonymous reviewers for their constructive comments on the first version of this study.
REFERENCES
Agyare, W., 2004: Soil characterization and modeling of spatial distribution of saturated hydraulic conductivity at two sites in the Volta Basin of Ghana. Ecol. Dev. Ser., 17 , 194.
Baldocchi, D., and Coauthors, 2001: FLUXNET: A new tool to study the temporal and spatial variability of ecosystem-scale carbon dioxide, water vapor, and energy flux densities. Bull. Amer. Meteor. Soc., 82 , 2415–2434.
Beljaars, A. C. M., and Holtslag A. A. M. , 1991: Flux parameterization over land surfaces for atmospheric models. J. Appl. Meteor., 30 , 327–341.
Betts, A. K., Chen F. , Mitchell K. E. , and Janjić Z. I. , 1997: Assessment of the land surface and boundary layer models in two operational versions of the NCEP Eta Model using FIFE data. Mon. Wea. Rev., 125 , 2896–2916.
Birkes, D., and Dodge Y. , 1993: Alternative Methods of Regression. Wiley, 228 pp.
Boone, A., and Coauthors, 2004: The Rhône-Aggregation Land Surface Scheme Intercomparison Project: An overview. J. Climate, 17 , 187–208.
Burpee, R. W., 1972: The origin and structure of easterly waves in the lower troposphere of North Africa. J. Atmos. Sci., 29 , 77–90.
Chen, F., Janjić Z. , and Mitchell K. , 1997: Impact of atmospheric surface-layer parameterizations in the new land-surface scheme of the NCEP mesoscale Eta Model. Bound.-Layer Meteor., 85 , 391–421.
De Bruin, H. A. R., Kohsiek W. , and Van den Hurk B. J. M. , 1993: A verification of some methods to determine the fluxes of momentum, sensible heat, and water vapour using standard deviation and structure parameter of scalar meteorological quantities. Bound.-Layer Meteor., 63 , 231–257.
De Bruin, H. A. R., Kohsiek W. , and Van den Hurk B. J. M. , 1995: The scintillation method tested over a dry vineyard area. Bound.-Layer Meteor., 93 , 453–468.
Ek, M. B., Mitchell K. E. , Lin Y. , Grunmann P. , Rogers E. , Gayno G. , Tarpley J. D. , and Koren V. , 2003: Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta. J. Geophys. Res., 108 .8851, doi:10.1029/2002JD003296.
Emmerich, W. E., 2003: Carbon dioxide fluxes in a semiarid environment with high carbonate soils. Agric. For. Meteor., 116 , 91–102.
FAO, 1984: Countries North of the Equator. Vol. 1, Agroclimatological Data for Africa, FAO Plant Production and Protection Series 22, Food and Agriculture Organization of the United Nations, 36 pp.
Feddes, R. A., Kowalik P. J. , and Zaradny H. , 1978: Simulation of Field Water Use and Crop Yield. Simulation Monogr., No. 9, PUDOC, 189 pp.
Goutorbe, J. P., and Coauthors, 1994: HAPEX-Sahel: A large-scale study of land–atmosphere interactions in the semiarid tropics. Ann. Geophys., 12 , 53–64.
Halldin, S., Gryning S-E. , Gottschalk L. , Jochum A. , Lundin L-C. , and Van de Griend A. A. , 1999: Energy, water and carbon exchange in a boreal forest landscape—NOPEX experiences. Agric. For. Meteor., 98-99 , 5–29.
Harris, P. P., Huntingford C. , Cox P. M. , Gash J. H. C. , and Malhi Y. , 2003: Effect of soil moisture on canopy conductance of Amazonian rainforest. Agric. For. Meteor., 122 , 215–227.
Henderson-Sellers, A., Yang Z-L. , and Dickinson R. E. , 1993: The Project for Intercomparison of Land-surface Parameterization Schemes. Bull. Amer. Meteor. Soc., 74 , 1335–1349.
Henderson-Sellers, A., Pitman A. J. , Love P. K. , Irannejad P. , and Chen T. H. , 1995: The Project for Intercomparison of Land Surface Parameterization Schemes (PILPS): Phases 2 and 3. Bull. Amer. Meteor. Soc., 76 , 489–503.
Heusinkveld, B. G., Jacobs A. F. G. , Holtslag A. A. M. , and Berkowicz S. M. , 2004: Surface energy balance closure in an arid region: Role of soil heat flux. Agric. For. Meteor., 122 , 21–37.
Hogue, T. S., Bastidas L. , Gupta H. , Sorooshian S. , Mitchell K. , and Emmerich W. , 2005: Evaluation and transferability of the Noah land surface model in semiarid environments. J. Hydrometeor., 6 , 68–84.
Holtslag, A. A. M., and Ek M. , 1996: Simulation of surface fluxes and boundary layer development over the pine forest in HAPEX-MOBILHY. J. Appl. Meteor., 35 , 202–213.
Huntingford, C., Allen S. J. , and Harding R. J. , 1995: An intercomparison of single- and dual-source vegetation-atmosphere transfer models applied to transpiration from Sahelian savannah. Bound.-Layer Meteor., 74 , 397–418.
Idso, S. B., and Jackson R. B. , 1969: Thermal radiation from the atmosphere. J. Geophys. Res., 74 , 5397–5403.
Jacobs, C. M. J., 1994: Direct impact of atmospheric CO2 enrichment on regional transpiration. Ph.D. thesis, Wageningen Agricultural University, 179 pp.
Jacquemin, B., and Noilhan J. , 1990: Sensitivity study and validation of land-surface parameterization using the HAPEX-MOBHILY data set. Bound.-Layer Meteor., 52 , 93–134.
Jarvis, P., 1976: The interpretation of the variations in leafwater potentials and stomatal conductances found in canopies in the field. Philos. Trans. Roy. Soc. London, 273B , 593–610.
Kasei, C. N., 1988: The physical environment of semiarid Ghana. Challenges in Dryland Agriculture—A Global Perspective: Proceedings of the International Conference on Dryland Farming, P. W. Unger et al., Eds., Texas Agricultural Experiment Station, 350–354.
Kubota, A., and Sugita M. , 1994: Radiometrically determined skin temperature and scalar roughness to estimate surface heat flux. Part I: Parameterization of radiometric scalar roughness. Bound.-Layer Meteor., 69 , 397–416.
Kurkowski, N. P., Stensrud D. J. , and Baldwin M. E. , 2003: Assessment of implementing satellite-derived land cover data in the Eta Model. Wea. Forecasting, 18 , 404–416.
Mahrt, L., and Ek M. , 1984: The influence of atmospheric stability on potential evaporation. J. Climate Appl. Meteor., 23 , 222–234.
Martano, P., 2000: Estimation of surface roughness length and displacement height from single-level sonic anemometer data. J. Appl. Meteor., 39 , 708–715.
Meijninger, W. M. L., Green A. E. , Hartogensis O. K. , Kohsiek W. , Hoedjes J. C. B. , Zuurbier R. M. , and De Bruin H. A. R. , 2002: Determination of area-averaged water vapour fluxes with large aperture and radio wave scintillometers over a heterogeneous surface—Flevoland field experiment. Bound.-Layer Meteor., 105 , 63–83.
Moene, A. F., 2003: Effects of water vapour on the structure parameter of the refractive index for near-infrared radiation. Bound.-Layer Meteor., 107 , 635–653.
Monteith, J. L., and Unsworth M. , 1990: Principles of Environmental Physics. 2nd ed. Arnold, 291 pp.
Niyogi, D. S., Xue Y. , and Raman S. , 2002: Hydrological land surface response in a tropical regime and a midlatitudinal regime. J. Hydrometeor., 3 , 39–56.
Noilhan, J., and Planton S. , 1989: A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 117 , 536–549.
Paulson, C., 1970: The mathematical representation of wind speed and temperature profiles in the unstable atmospheric surface layer. J. Appl. Meteor., 9 , 857–861.
Pitman, A. J., and Coauthors, 1999: Key results and implications from phase 1(c) of the Project for Intercomparison of Land-surface Parametrization Schemes. Climate Dyn., 15 , 673–684.
Ronda, R. J., De Bruin H. A. R. , and Holtslag A. A. M. , 2001: Representation of the canopy conductance in modeling the surface energy budget for low vegetation. J. Appl. Meteor., 40 , 1431–1444.
Schlosser, C., and Coauthors, 2000: Simulations of a boreal grassland hydrology at Valdai, Russia: PILPS phase 2(d). Mon. Wea. Rev., 128 , 301–321.
Schüttemeyer, D., Moene A. F. , Holtslag A. A. M. , de Bruin H. A. R. , and van de Giesen N. , 2006: Surface fluxes and characteristics of drying semi-arid terrain in West Africa. Bound.-Layer Meteor., 118 , 583–612.
Schüttemeyer, D., Schillings C. , Moene A. F. , and de Bruin H. A. R. , 2007: Satellite-based actual evapotranspiration over drying semiarid terrain in West Africa. J. Appl. Meteor. Climatol., 46 , 97–111.
Stewart, J. B., 1988: Modelling surface conductance of pine forest. Agric. For. Meteor., 43 , 19–35.
Unland, H. E., Houser P. R. , Shuttleworth W. J. , and Yang Z-L. , 1996: Surface flux measurement and modeling at a semi-arid Sonoran desert site. Agric. For. Meteor., 82 , 119–153.
Van de Giesen, N., Kunstmann H. , Jung G. , Liebe J. , Andreini M. , and Vlek P. L. G. , 2002: The GLOWA Volta project: Integrated assessment of feedback mechanisms between climate, land use, and hydrology. Adv. Global Change Res., 10 , 151–170.
Van den Hurk, B. J. J. M., Viterbo P. , Beljaars A. C. M. , and Betts A. K. , 2000: Offline validation of the ERA40 surface scheme. ECMWF Tech. Memo. 295, European Centre for Medium-Range Weather Forecasts, 42 pp.
Verhoef, A., 1995: Surface energy balance of shrub vegetation in the Sahel. Ph.D. dissertation, University of Wageningen, 247 pp.
Verma, S., 1989: Aerodynamic resistances to transfers of heat, mass and momentum. Estimation of Areal Evapotranspiration, T. A. Black et al., Eds., IAHS Publication 177, 13–20.
Viterbo, P., and Beljaars A. C. M. , 1995: An improved land surface parameterization scheme in the ECMWF model and its validation. J. Climate, 8 , 2716–2748.
Wallace, J. S., and Coauthors, 1992: SEBEX: The Sahelian Energy Balance Experiment. ODA Rep. 92/9, 51 pp.
Wang, Y-P., and Leuning R. , 1998: A two-leaf model for canopy conductance, photosynthesis and partitioning of available energy. I: Model description and comparison with a multi-layered model. Agric. For. Meteor., 91 , 89–111.
Wesely, M. L., 1976: The combined effect of temperature and humidity fluctuations on refractive index. J. Appl. Meteor., 15 , 43–49.
Wyngaard, J. C., 1973: On surface-layer turbulence. Workshop on Micrometeorology, D. A. Haugen, Ed., Amer. Meteor. Soc., 101–149.
Zeng, X., Dai Y-J. , Dickinson R. E. , and Shaikh M. , 1998: The role of root distibution for climate simulation over land. Geophys. Res. Lett., 25 , 4533–4536.
Zhang, H., and Nobel P. S. , 1996: Dependency of cl/ca and leaf transpiration efficiency on the vapour pressure deficit. Aust. J. Plant Physiol., 232 , 561–568.
Zilitinkevich, S. S., 1995: Non-local turbulent transport: Pollution dispersion aspects of coherent structure of convective flows. Air Pollution III, H. Power, N. Moussiopoulos, and C. A. Brebbia, Eds., Vol. I, Air Pollution Theory and Simulation, Computational Mechanics Publications, 53–60.
APPENDIX
The Evapotranspiration Parameterization for Noah JS, Noah A-gs, and TESSEL
For both models a canopy conductance parameterization is utilized to calculate evapotranspiration. In this study evapotranspiration is the sum of evaporation from bare soil (Edir), the interception reservoir (Ec), and dry vegetation (Et).
Parameterization for Noah JS
Parameterization for Noah A-gs
Parameterization for TESSEL
Different locations of experimental sites in Ghana, West Africa, within the GLOWA-Volta project.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Daily sums of precipitation for the two test sites in (top) Ejura and (bottom) Tamale in 2002.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Mean measured net radiation (lines) and mbe of net radiation in W m−2 (bars) during the studied season at the two test sites: (top) Ejura and (bottom) Tamale. Net radiation shows a clear decrease with rising mbe during the dry period at both sites.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Modeled vs measured net radiation for all parameterizations. Scatterplot for the analyzed periods at the two test sites: (top) Ejura and (bottom) Tamale.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Mean measured soil heat flux (lines) and mbe of soil heat flux in W m−2 (bars) during the studied season at the two test sites: (top) Ejura and (bottom) Tamale.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Modeled vs measured soil heat flux for all parameterizations. Scatterplot for the analyzed periods at the two test sites: (top) Ejura and (bottom) Tamale.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Mean measured sensible heat flux (lines) and mbe of sensible heat flux in W m−2 (bars) during the studied season at the two test sites: (top) Ejura and (bottom) Tamale.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Modeled vs measured latent and sensible heat flux for all parameterizations: (a) Ejura and (b) Tamale.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Mean measured latent heat flux (lines) and mbe of latent heat flux in W m−2 (bars) during the studied season at the two test sites: (top) Ejura and (bottom) Tamale. Latent heat fluxes decreases, whereas the mbe rises for Noah JS and TESSEL.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Comparison of measured and modeled skin temperature (averages between 0900 and 1700 local time) for an intensive observation period in November and December 2002 at the Tamale test site.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Mean measured net radiation (lines) and mbe for net radiation (bars) with C = 0.4 for both versions of Noah and roughness length ratio of 100 for TESSEL.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Tamale daily mean values of averaged measured and modeled volumetric soil moisture content in the upper two soil layers.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Measured and modeled surface conductance at Ejura for the main period of drying up during the studied season. Noah JS overestimates surface conductance. Noah A-gs and TESSEL are closer to the measured values.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Daily averages at Ejura of the different functions fi of Jarvis–Stewart formulation for Noah JS and TESSEL used in the context of transpiration calculation. VPD and soil moisture response are important for the seasonal cycle.
Citation: Journal of Hydrometeorology 9, 2; 10.1175/2007JHM797.1
Instrument within the GLOWA-Volta project.
Major differences in concept between the two land surface models Noah and TESSEL.
Table 3a. Ejura: main model parameter values for TESSEL and Noah.
Table 3b. Tamale: main model parameter values for TESSEL and Noah.
Seasonally dependent vegetation fraction (VF) for Noah at both test sites, based on the estimated EVI.