Land surface processes are the link between the earth’s surface and the atmosphere, including the exchange of momentum, energy, gas, and matter. These processes not only have significant implications for global climate, but also are a key part of the physical processes used in atmospheric numerical models on various scales. In addition, land surface process models play an important role in the assimilation of land surface data. The land surface model is coupled to the assimilation of atmospheric data to produce the atmospheric forcing and provides a highly precise background field for the assimilation of land surface data. Surface energy and material exchanges directly affect the output of the land surface assimilation system and the statistical prediction of the next time step (Huang and Li, 2004). Improving the land surface process model is an important aspect of increasing the atmospheric model’s predictive capability.

Over the previous three decades, numerical land surface models have become increasingly numerous and complex. These models have developed from the original and simple “bucket” model into the soil–vegetation–atmosphere transfer models that we see today. In these latter models, the physiological processes of vegetation are also coupled. More recently, complex land surface models [CLM (Community Land Model), CoLM (Common Land Model), etc.] have been developed by involving the carbon cycle. As the models have become more complex, the results produced by the models have improved (Bonan, 1998; Shi, 2001; Zheng et al., 2014; Cai et al., 2017; Jing and An, 2017). Several land surface model schemes have been developed in China, each with a different research priority. These schemes have been summarized in previous papers (Shi, 2001; Chen and Sun, 2002; Huang and Li, 2004). However, a land surface model specifically developed for the arid and semi-arid regions of China is rarely mentioned in these summaries.

Desert and Gobi are the primary land surfaces in the arid region of northwestern China. As a result of its unique albedo, heat capacity, roughness, and surface energy exchange, which differ from those of other surfaces, Gobi plays an important role in determining local, regional, and even global climate. Understanding Gobi–air interactions and environmental controls on the surface energy exchanges over Gobi surfaces is necessary (Liu et al., 2009).

Many efforts have been made in the past, using indirect observations, to estimate or measure the mass transfer and surface energy budget in the arid region of Northwest China (Zhang et al., 2005). Wang et al. (2002) simulated the land surface characteristics of the underlying desert surface in the Hexi Corridor region of Northwest China using the LSM (Land Surface Model). Cao and Zhang (2003) used the modified land surface process model SVA (Soil Vegetation Atmosphere) developed by Zhao et al. (1995) to simulate land–air interactions in the arid region of Northwest China, especially the land surface processes over the arid Gobi Desert. Bao et al. (2005) modified the LSM of CCM (Community Climate Model) by using the Gobi surface parameters at Dunhuang station. Their simulation showed that changes in the parameterization of the underlying surface in the arid region significantly improved regional climate simulations in East Asia.

Until recently, only a few direct measurements of surface energy balance and evaporation over this kind of land surface have been made by using the eddy covariance technique. In this study, we collected eddy covariance observations of the surface energy budget over the arid region of northwestern China during summer in the Gobi landscape at Dunhuang station. Our aims were to better understand the interaction between the lower atmosphere and the Gobi surface, and to improve CoLM so that it can be used to simulate land surface processes.

In this study, we used CoLM, a relatively advanced land surface process model. CoLM consists of four parts: (1) physical processes of the biological earth—the exchange of energy, water, and momentum between the land and atmosphere; (2) the hydrological cycle—vegetation-intercepted water, throughfall (drips off the vegetation), runoff, infiltration, soil-water movement, and snow surface runoff, which directly or indirectly affect rainfall, temperature, and runoff; (3) biogeochemical processes—chemical exchanges between the land and atmosphere, commonly including biological flux, carbon, dust, dry deposition, and other mass exchanges; (4) dynamic vegetation processes—the dynamic description of the material or energy exchanges between vegetation and the environment (Dai et al., 2003, 2004).

The data analyzed in this paper were obtained during the Dunhuang experiment at the Shuangdunzi Gobi micrometeorology station. The Shuangdunzi Gobi is a typical arid region. We focused mainly on the bare-soil-related model in relation to the parameterization of surface momentum flux, sensible heat, latent heat, and their calculations. The observational site is about 5 km from the edge of an oasis, with an elevation of 1150 m above sea level, and the site is located at 40°10′N, 94°31′E. The mean annual air temperature at the site is 10.8°C (Wang et al., 2011) and annual precipitation averages 40 mm. The study site is flat, with the upper soil layer composed mainly of pebble-sized sediment, and the lower layer being sand (Wang et al., 2010).

Since May 2000, we have been conducting routine meteorological observations at Shuangdunzi Gobi micrometeorological station. At this site, enhanced/intensive observation experiments were performed in August 2008, during which additional measurements of sensible and latent heat fluxes, soil thermal flux at 2.5-cm depth, and three-dimensional wind fields were carried out. These data were not available in routine meteorological observations. Therefore, we selected a period of 22–28 August 2008 and collected all the relevant data for driving the CoLM model. Overall, the observational data include wind, air temperature and humidity gradients (measured on a tower), radiation components, soil temperature and moisture (observed at depths of 0, 5, 10, 20, 40, 80, 160, and 320 cm), and heat fluxes. For details of these field experiments, please refer to Wei et al. (2006) and Wang et al. (2011). During the observation period, turbulence measurements were taken with a supersonic anemometer/thermometer and an open-path infrared gas analyzer. The sensible and latent heat fluxes were observed by using an eddy covariance system with a krypton hygrometer. All data, including the values used in the comparison with the simulation results, have been processed by abnormal data deletion, WPL (Webb–Pearman–Leuning) correction, etc.

By using the observational data, simulations were conducted for the period 22–28 August 2008. Both the observational data and simulation data have a temporal resolution of 30 minutes (all times are in local time). The weather was sunny during the simulation period. The aerodynamic roughness of the Gobi land surface used in the simulation was obtained from the study of the Dunhuang Gobi by Zhang et al. (2004). In the model simulation, wasteland was selected as the vegetation type, and the soil texture structure was sand (82%), clay (6%), and silt (12%), and the momentum roughness in the model was 0.0019 m. By the linear interpolation method, the initial values of soil temperature and moisture were calculated, as shown in Table 1. Other boundary and initial forcing fields were also prescribed based on the observational data for inputing into the CoLM at 30-min intervals.

The following eight observed variables were used as atmospheric forcing in CoLM: incoming solar shortwave radiation, atmospheric longwave radiation, precipitation rate, atmospheric temperature, zonal wind component, meridional wind component, atmospheric pressure, and specific humidity, at a reference height. Air temperature and specific humidity were measured at 1.5 m, and wind velocity was measured at 2 m. No precipitation occurred during the simulation period. The observed variability of some of the forcing variables is shown in Fig. 1, based on the enhanced field measurements at the Shuangdunzi Gobi station during 22–28 August 2008.

Figure 1 Time series of observed atmospheric variables: (a) solar radiation (Rsd) that reached the ground surface; (b) longwave radiation (Rld) that reached the ground surface; (c) zonal (east–west) surface wind speed (Ux); (d) meridional (south–north) surface wind speed (Uy); (e) surface air temperature (Ta); and (f) surface specific humidity (q). These are used as forcing conditions for driving the CoLM simulations at 30-min intervals.

Figure options

The energy fluxes play an important role in land–atmosphere interactions. With the prescribed forcing and initial data at 30-min intervals, a simulation experiment was conducted with the unmodified CoLM, referred to as the original CoLM experiment or Exp1. Figure 2 shows the simulated values (dotted line) from the original CoLM experiment and the observed values (solid line). These include the sensible and latent heat fluxes, net radiation, and soil heat flux.

Figure 2 Original CoLM experiment showing the simulated (dotted line) and measured (solid line) values for (a) surface sensible heat flux (H), (b) surface latent heat flux (LE), (c) surface net radiation (Rn), and (d) soil heat flux (Gs).

Figure options

We first determine whether CoLM was able to adequately simulate the energy fluxes over the surface. The simulated sensible heat flux and eddy-covariance observations are compared in Fig. 2a. The variations of the simulated and observed sensible heat fluxes were consistent, and the daily trends of the simulated and observed values were also generally consistent. However, there was an overestimation of the sensible heat flux around mid day. Figure 2b shows that the model had limited capability in simulating the observed latent heat flux. This was mainly due to the dynamic ranges of soil moisture in the model being too narrow. Also, soil moisture was extremely low during daytime. The simulated net radiation (Fig. 2c) was also lower than observed. Comparison of the simulated surface heat flux (Fig. 2d) to the observed subsurface soil heat flux at the depth of 2.5 cm reveals that they exhibited similar variability, while the simulated value was larger and the maximum value was overestimated.

Figure 3 Variation in solar radiation (Rns) at the ground surface: observed (open circles), original CoLM experiment simulated (dotted line), and corrected albedo experiment simulated (solid line) values.

Figure options

The original CoLM experiment (dotted line in Fig. 3) produced solar radiation at the surface that was significantly lower than observed (open circles in Fig. 3). This apparently resulted when the model’s parameterized value was too large for the surface albedo. Furthermore, the unmodified CoLM also produced net longwave radiation at the surface (Fig. 4a; green line) that was significantly smaller than observed (Fig. 4a; black line). The simulated daytime upward longwave radiation from the ground surface was significantly small, because the simulated surface soil temperature was significantly low. Figure 4b shows the observed (black line) and unmodified CoLM simulated (green line) soil temperature variations at the surface.

Figure 4 Unmodified CoLM simulated and observed (a) net longwave radiation at the surface (Rnl) and (b) soil temperature variations at the surface (Tgs) (Obs: observed value; Exp1: unmodified CoLM experiment; Exp2: improved albedo experiment; Exp3: improved rah experiment).

Figure options

Overall, we found two main problems with the original CoLM simulation. The first was that the simulated surface albedo was too high, and the second that the simulated surface soil temperature was too low. In general, the simulated sensible heat flux should also be low, when the simulated surface soil temperature is low. However, the simulated daily maximum value for the sensible heat flux was higher than expected. The surface sensible heat flux is controlled by two different properties: (1) it is directly proportional to the temperature difference between the surface and the air, and (2) it is inversely proportional to the thermal aerodynamic impedance (r_ah). Apparently, the higher simulated sensible heat flux can be attributed to the values of r_ah prescribed in the model being too low.

In addition, when r_ah is small, the sensible heat flux at the surface will be too large and the amount of heat entering the soil will decrease, resulting in a low soil temperature. In the unmodified CoLM experiment, the significantly lower surface soil temperature was compensated by decreased surface sensible heat flux and low r_ah, while sensible heat was also dependent on the thermal aerodynamic impedance. It is expected that the model can not only better simulate sensible heat flux, but also better simulate soil temperature. In the next section, we will discuss how to improve the simulation results by modifying the surface albedo and r_ah.

The surface albedo was the first parameter that we modified. CoLM limits the maximum value of the albedo of a bare land surface to 0.32. However, the average albedo of the Dunhuang Gobi surface was calculated with the observational data collected by radiation instruments at the station, and the value was 0.26. Therefore, we changed the albedo value in the model into 0.26, and named the associated experiment as Exp2. All other parameters were the same as those used in the unmodified CoLM experiment. This simulation was also termed the corrected albedo experiment. Figure 3 shows the solar radiation absorbed by the ground surface (solid line) obtained from the corrected albedo simulation. Meanwhile, the RMSE (root-mean-square error) of the solar radiation between the observed and simulated data was calculated, showing that it decreased from 44.46 to 36.13 W m^–2. It can be seen that the albedo modifications produced values for the simulated solar radiation absorbed by the ground surface that were much closer to the observations. Some improvements occurred for the surface temperature (Fig. 4b; blue line) and the net longwave radiation (Fig. 4a; blue line), but they were still significantly lower than the observed values. The RMSE of the simulation, shown in Table 2, was decreasing in Exp2.

After modifying the albedo, the simulation of soil temperature did not meet our expectation. Next, we tried to improve r_ah. The value of r_ah calculated by the model was small; therefore, it was necessary to select a new parameterization formula for r_ah.

Wei et al. (2006) used the aerodynamic method to calculate the sensible heat exchange coefficient (C_h) of the Dunhuang Gobi surface between September 2000 and September 2001. Their study obtained an annual mean C_h value of 0.00203 ± 0.00045. Under neutral and near-neutral atmospheric conditions, the sensible heat exchange coefficient in neutral atmospheric conditions (C_hn) of the Dunhuang Gobi is constant at 0.002, where the Monin–Obukhov atmospheric stability parameter (ζ) is between –0.01 and 0.01. Under stable conditions (ζ ≥ 0.01), C_h is smaller, and increases with increasing surface wind velocity, eventually approaching C_hn. The relationship between the wind velocity (V_a) and the sensible heat exchange coefficient is: C_h = 0.0026 – 0.0002 × ln(V_a), while r_ah = 1/(C_hV_a). The thermal aerodynamic impedance r_ah can be calculated by using the empirical formulae (Wei et al., 2006) below:

For the next simulation, we substituted the above equations into CoLM, and also used the modified surface albedo of 0.26 from the corrected albedo experiment, while all other parameters were kept the same as those of the original CoLM experiment. Hereafter, this simulation will be referred to as the improved r_ah experiment or Exp3. Figure 4 shows the simulated net longwave radiation at the surface (Fig. 4a; red line) and the surface soil temperature (Fig. 4b; red line). It can be seen that the simulated and observed surface soil temperatures were fairly consistent. The RMSE of net longwave radiation and soil temperature are shown in Table 2, showing a greater decrease in Exp3 than Exp2. However, there was still a difference between the observed and simulation value. Further improvement to the calculation scheme is needed to solve this problem, and this will be investigated in a future study.

Figure 5 Variations in (a) surface sensible heat flux (H) and (b) surface net radiation (Rn), in the observation (open circles), corrected albedo experiment (dotted line), and improved rah experiment (solid line).

Figure options

The surface sensible heat flux and the net radiation, as simulated by the corrected albedo and improved r_ah experiments (Fig. 5) showed that the simulated sensible heat flux was significantly larger than the observed values in the corrected albedo experiment. This enlarged the anomaly in the sensible heat flux simulated by the unmodified CoLM. The r_ah experiment greatly improved the simulation of the sensible heat flux, producing a value that was closer to the observed than that produced by the unmodified CoLM simulation. The RMSE of the sensible heat flux decreased from 50.27 to 28.79 W m^–2, by correcting r_ah in the model.

It can be seen from Fig. 5b that, in the corrected albedo experiment, the simulated surface net radiation was closer to the observed value, but the surface temperature was relatively lower. The RMSE of surface net radiation decreased from 28.29 to 21.70 W m^–2. The improvement in the surface net radiation was due to a large improvement in the model’s ability to simulate the solar radiation absorbed by the ground.

As described above, by modifying the surface albedo and r_ah, the ability of CoLM to simulate surface soil temperature was greatly improved. Meanwhile, the simulation of surface sensible heat flux and net radiation were also improved. CoLM can be further improved by modifying the parameterization scheme of the M–O (MoninObukhov) universal function. We are currently improving the simulation capability of CoLM by modifying the flux profile relationship, and the results of this work will be described in a future paper.

The Dunhuang observation experiment provides an opportunity to research the land surface processes of the Gobi landscape, and to evaluate and reduce land surface model errors. With this information, and a wealth of surface observations, we are able to test and improve CoLM’s ability to simulate the Gobi surface. From this investigation, we have drawn the following conclusions:

(1) Despite providing CoLM with reliable observational data, the unmodified CoLM simulations exhibited large differences when compared with observations. The issues include: the soil temperature being significantly lower than observed; the simulated surface sensible heat flux being higher; and the surface net radiation being lower.

(2) CoLM’s parameterization scheme for the surface albedo is not suitable for the arid Gobi region. By replacing the surface albedo with a measured albedo of 0.26, the model produced values for solar radiation absorbed by the ground surface that were more consistent with observations.

(3) Inclusion of an empirical parameterization for r_ah, determined from a wealth of unique surface observations, dramatically improved the simulation. Furthermore, the simulated surface sensible heat flux and surface net radiation were also improved. These results indicate that improvements in the model require careful calibrations of the parameterization schemes used within CoLM.

More detailed observations allow for more robust parameterizations. However, this new empirical scheme works well solely in a local region. Further improvement in the characterizations of the land surface will allow for further upgrades to CoLM’s thermodynamic and dynamic parameterizations, and will produce more accurate simulations of near-surface atmospheric variables.