Coupling of a Regional Climate Model with a Crop Development Model and Evaluation of the Coupled Model across China

Cropland is an important land use type. According to the "Statistical Yearbook 2013" of the Food and Agriculture Organization, about 12% of the global land area——more than 1.5 billion hm^-2——is used for crop production (http://www.fao.org/economic/ess/ess-publications/ess-yearbook/). As a special vegetation type, the process of sowing, growing, and harvesting crops is affected by both human activity and climate forcing; dynamic changes in crop growth also give feedback about the local climate. Although managed crops physically resemble grass (i.e., compared to woody plants), studies have revealed that they play different roles in the climate system. For example, (McPherson et al., 2004) examined meteorological observations over the corn belt of the central United States and found that the difference in vegetation dynamics between cropland and natural grassland gave rise to local anomalies in the temperature and humidity of the land surface. (Twine et al., 2004) demonstrated that the conversion of grassland to winter wheat in the Mississippi River Basin increased the annual net radiation and evapotranspiration by 19% and 7%, respectively. In addition to the differences between crops and grass, various crops differ substantially in terms of phenology and physiology. For instance, the aboveground biomass of soybean is nearly half that of maize during the maturity period (Prince et al., 2001). It has been shown that more detailed parameterization on the unique properties of various crop types can improve model simulations (de Noblet-Ducoudré et al., 2004; Gervois et al., 2004; Lokupitiya et al., 2009).

Numerous studies have investigated the interactions between crops and climate change via coupling crop models with land modules of climate models (Lei et al., 2010; Maruyama and Kuwagata, 2010; Van den Hoof et al., 2011; Li et al., 2013). For example, (Tsvetsinskaya et al., 2001) incorporated a crop model into a regional climate model, RegCM2, and found that the coupled model improved the simulation of near-surface climate in the central United States. (Osborne et al., 2009) coupled a crop model, GLAM (General Large Area Model for Annual Crops), with the climate model HadAM3 (Hadley Center Atmospheric Model), and demonstrated that the growing season temperature variability increased up to 40% with the inclusion of dynamic crops. (Tsarouchi et al., 2014) dynamically coupled a land surface model JULES (Joint UK Land Environment Simulator) with a crop growth model InfoCrop, and found that the coupled model greatly reduced the mean monthly error of evapotranspiration.

Most of these studies have focused on the growth process of a single crop type or farming system. Although some studies have provided comprehensive consideration of multiple crop types and farming systems (e.g., Van den Hoof et al., 2011; Tsarouchi et al., 2014), few of them have focused on the main cereal crops in China. Similarly, few studies have investigated the interactions between crops and regional climate.

According to the 2014 annual statistical data released by the National Bureau of Statistics in China (http://data.stats.gov.cn/), maize, rice, and wheat covered 81% of the cereal crop planting area in China. In order to investigate the climate effects associated with the growth of these crops across China, Chen and Xie (2011, 2012, 2013) coupled the CERES (Crop Estimation through Resource and Environment Synthesis) model with the BATS (Biosphere-Atmosphere Transfer Scheme), which is the land module of RegCM3. The authors used this coupled model to examine differences associated with the farming systems of the main cereal crops (maize, rice, and wheat) across China. The coupled model was named RegCM3_CERES and was shown to reduce the RMSE of precipitation by 2.2%-10.7% over northern China and the temperature by 5.5%-30.9% over eastern China. However, the BATS model falls short in describing the heterogeneity within grid cells and lacks sustainable improvements in physical processes. More sophisticated land models are needed to include the growth processes of the main cereal crops in China in order to conduct further studies on crop-climate interactions.

The Community Land Model (CLM) was developed by the National Center for Atmospheric Research in the United States, and version 3.5 of CLM (CLM3.5) simulates local climate well when used as the land module of RegCM4 (Li et al., 2009; Qin et al., 2013). (Levis et al., 2012) coupled a crop model with CLM4.0 to describe the interactive crop management process. (Lu et al., 2015) coupled CLM4 including crop growth (CLM4crop) with the Weather Research and Forecasting model and evaluated its simulation ability across the United States. However, the crop model in CLM4 treats temperate cereals as summer crops. Cereals with multi cropping systems, such as winter wheat and spring wheat, are not considered. For these reasons, the crop model in CLM4 is not currently able to comprehensively represent the growth processes of maize, wheat, and rice with different cropping systems across China.

In this study, the CERES-Maize, CERES-Wheat, and CERES-Rice sub-models were coupled with CLM3.5 and the land module of RegCM4, as an extension of the work of Chen and Xie (2011, 2012, 2013). For ease of description below, the newly coupled model was named RegCM4_CERES.

The regional climate model RegCM4 was developed by the International Center for Theoretical Physics in Italy (Giorgi and Anyah, 2012). It employs three convective precipitation schemes (Kuo, Grell, and Emanuel) and one large-scale precipitation scheme (SUBEX) to simulate precipitation (Giorgi et al., 1993). The land surface physical schemes available in RegCM4 include BATS1e and CLM3.5.

CLM3.5 (Oleson et al., 2008) uses a tile approach to describe the heterogeneity within a grid cell. Each grid cell has a different number of land units, including glaciers, lakes, wetlands, vegetation, and urban land. The vegetated unit can be further divided into 17 plant functional types (PFTs) including crops. In the released version of CLM3.5, the crop types share the same optical and vegetation structure parameters with the C3 grass type. The LAI and SAI (stem area index) of the crop type are based on multi-year mean monthly surface datasets from MODIS (Moderate Resolution Imaging Spectroradiometer) products (Lawrence and Chase, 2007). The root distribution fraction of each crop type is a fixed function of depth and remains the same within the global scale. Although the DGVM (dynamic global vegetation model) is coupled into CLM3.5 to simulate dynamic vegetation growth, the growth of crops remains static for shrubs and crops are excluded in CLM-DGVM (Levis et al., 2004).

The CERES model consists of a number of sub-models, including CERES-Wheat, CERES-Barley, CERES-Maize, and so on (Jones and Kiniry, 1986; Tsuji et al., 1998). The sub-models employ a number of exogenous parameters that function to simulate the real-time growth status of crops. Daily weather data, soil surface, and profile information, as well as detailed crop management information, are needed as the inputs for these models. The time step of CERES is one day, and at the end of the day each crop's vegetative and reproductive development stage is updated. In this study, the CERES-Wheat, CERES-Rice, and CERES-Maize sub-models were chosen to be coupled with CLM3.5 to simulate the growth processes of the main cereal crops in China. Numerous studies have validated the simulation abilities of the three models worldwide and under different environmental conditions (Kiniry et al., 1997; Pang et al., 1997; Yao et al., 2007).

The CERES (-Wheat, -Rice, and -Maize) model was coupled with CLM3.5 by exchanging the environmental conditions and the crops' growth in a daily time step. The exchange frequency of the interface module that connected CERES and CLM3.5 was once per day. The weather and soil conditions provided by RegCM4/CLM3.5, i.e., precipitation, temperature, radiation, surface albedo, surface wind speed, soil moisture, etc., were summed or averaged at the daily level as input for CERES. The simulated daily status of the crops, i.e., LAI, SAI, root fraction in each soil layer, was updated as feedback from CERES in CLM3.5.

Based on the actual cropping system across China (Zhang et al., 1987), the crop type in each grid cell of CLM was further divided into eight sub-types, including spring maize, summer maize, spring wheat, winter wheat, early rice, late rice, single rice, and other crop types (Figure 1). The growth processes of each crop type were simulated by the corresponding crop model, i.e., CERES-Maize, CERES-Wheat, and CERES-Rice. The same crops with different farming schedules, such as spring maize and summer maize, were simulated by the same crop model, i.e., CERES-Maize, and driven by different datasets of planting and harvesting date. In this study, the planting and harvesting date for each crop sub-type was taken from the RegCM3_CERES data, which was derived from the "Agricultural Phenology Atlas of China" (Zhang et al., 1987) with GIS techniques by Chen and Xie (2011, 2012). Additionally, the growth information of the other crop type sub-category was taken from the simulated status of crops in CLM3.5. The growth processes of each crop sub-type were simulated independently of each other, and then summed in each grid to update the LAI, SAI, and root fraction of the crop PFT in CLM3.5.

Figure 1.  Schematic diagram of grid division in RegCM4_CERES.

The summing process employed the relative distribution fraction as the weight for each crop sub-type. Taking LAI as an example, the summing process can be described as: \begin{equation} \label{eq1} {\rm LAI}_{\rm crop}=\sum_{i=1}^8\gamma_i{\rm LAI}_i , \ \ (1)\end{equation} where i is the crop sub-type index and γ_i is the relative distribution fraction of the crop sub-type i. The relative fraction was employed because the crop distribution data used in BATS and CLM3.5 were derived from different data sources. In order to preserve the classification of land cover type in CLM3.5, the distribution data of each crop sub-type used in the previous RegCM3_CERES Chen and Xie (2011, 2012) was standardized in this study and treated as the relative fraction γ_i. It is worth noting that the early rice and late rice sub-types are collectively known as double rice and share the same parameters, except for the date of planting and harvesting. The early rice and late rice sub-types also share the same relative distribution fraction γ_i in RegCM4_CERES and the growth processes of the two sub-types never occur at the same time. The spatial distribution of γ_i for each crop sub-type at a resolution of 0.5^°× 0.5^° is shown in Fig. 2. Further, Table 1 lists the amount of grid cells with each crop sub-type, and the mean fraction of each crop sub-type in the grid cells with the respective crop across China.

Figure 2.  Spatial distribution of relative fraction γ for each crop sub-type.

A flowchart illustrating the coupling between CLM3.5 and CERES is given in Fig. 3. In the coupled model, the crop model for each sub-type was pre-activated for initialization on the planting date of each sub-type, and was stopped on the harvesting date. After harvesting, the crops degrade as bare soil, which has no LAI/SAI or root fraction, until the next planting date. The CLM3.5 coupled with the CERES model was employed to replace the original CLM3.5 module in RegCM4.

Figure 3.  Flowchart of the coupling process in RegCM4_CERES.

Continental China was chosen as the study domain (Fig. 4). The spatial resolution of RegCM4 was 50 km× 50 km and its central projection was at (36.5^°N, 114.5^°E). The Grell scheme with Arakawa-Schulbert closure was used as the convective precipitation scheme, with the Holtslag scheme as the boundary layer scheme and the Zeng scheme as the ocean flux scheme. The other parameterizations, such as cloud microphysics and radiation processes, used the default settings of RegCM4. The ERA-Interim data from 1999 to 2008 were used as the lateral boundary forcing of the two simulation tests conducted——one control test (CTL) using the unmodified RegCM4/CLM3.5 model and one crop growth test (SIM) using the coupled model RegCM4_CERES. The time steps were one day for the crop models, 30 min for the land surface module, and 100 s for the atmospheric module. The simulation of the two tests in the first year was used for spin-up, with the results from 1 January 2000 to 31 December 2008 selected for analysis.

Figure 4.  Simulation domain and three typical areas chosen for statistical analysis.

The LAI and aboveground biomass of the crops at the observation stations were collected from the Chinese Ecosystem Research Network (CERN) (http://www.cern.ac.cn/). Data from four stations——Taoyuan (28.9^°N, 111.5^°E), Yucheng (36.8^°N, 116.6^°E), Luancheng (37.9^°N, 114.7^°E), and Fengqiu (35.0^°N, 114.4^°E)——were employed to evaluate the crop development simulation obtained with RegCM4_ CERES in the nearest grids to the four station sites.

Figure 5 shows the daily LAI and aboveground biomass at Taoyuan station in the year 2000, where the main cereal crops were double rice (early rice and late rice). The simulation captured the development phase characteristics of double rice well; however, there were some differences in the peak LAI values (Fig. 5a). Compared with the observed LAI, the simulated peak LAI of early rice was higher, and that of late rice was lower, by about 2 m² m^-2. The aboveground biomass simulation was better than the LAI simulation, exhibiting a good match to the observed aboveground biomass data for early rice and a slightly higher value for late rice (Fig. 5b).

Figure 5.  Simulated series of (a) LAI and (b) aboveground biomass by RegCM4_CERES with scattered observation at Taoyuan Station.

At Yucheng station, the main cereal crops were winter wheat and summer maize; intensive observations were available from 2003 to 2004. As shown in Fig. 6a, the simulated LAIs of wheat and maize were higher than the observed LAIs, with the greatest difference in peak wheat LAI in 2003 and smallest difference in maize LAI in 2004. From Fig. 6b, the RegCM4_CERES model captured the rapid increase in biomass during the jointing and earing stage for winter wheat, from March to April, and for summer maize from July to August. However, for winter wheat in 2004, the simulated phenological phases were a bit later than the observed phases.

Figure 6.  Spatial distribution of the (a) observed and (b) simulated winter wheat LAI in the first 10 days of April and the spatial distribution of the (c) observed and (d) simulated winter wheat LAI in the last 10 days of April.

The LAI of winter wheat at Luancheng and Fengqiu stations in 2007 is provided in Fig. 7. The simulations at the two stations were similar; that is, slightly higher peak values and later phenological phases were found compared to the observed values. Overall, the crop development simulated by RegCM4_CERES matched the station observations well. However, the simulated peak LAI values tended to be higher than the observed values. Within the development period, in general, the simulated LAIs and biomasses during the earlier vegetative stage were slightly better than those during the later reproductive stage when the grain was rapidly filled.

Figure 7.  Spatial distribution of the mean soil moisture at 10 cm depth of (a) observation, (b) simulation by RegCM4_CERES, (c) simulation by original RegCM4, and (d) simulation difference (RegCM4_CERES minus RegCM4). The area shaded by stripes indicates that differences there are at or exceed the 95% confidence level.

3.2.1. Regional evaluation of LAI of winter wheat

Compared to the sparse farmland observations of CERN, the LAI dataset for winter wheat was derived from the crop yield dataset of the Meteorological Data Center of the China Meteorological Administration (CMA) (http://data.cma.cn/). This dataset contains observations collected from agricultural meteorological stations once every 10 days, spanning from 1999 to 2010. It includes LAI observations of several crop types, such as wheat, rice, peanuts, etc. However, aside from winter wheat, there are few stations where other crop types were observed. Thus, the LAI observation for winter wheat was selected to evaluate the regional simulation ability of RegCM4. In addition, because of the numerous missing values in the dataset, the stations with observations over more than two years were further selected to calculate the mean multi-year LAI of winter wheat from 2000 to 2008.

Figure 8 shows the spatial distribution of the mean multi-year LAI in the first and last 10 days of April, when the most valid data were available. In the first 10 days of April, the observed LAI (Fig. 8a) generally increased from northwest to southeast, except for several grids with high values. A high-value center of around 6-7 m² m^-2 exists in the northern area of Jiangsu Province. The simulated LAI of winter wheat in Fig. 8b corresponds with the observations. In North China, the simulated LAI decreased from northwest to southeast; a high-value center was also located in Henan and the northern area of Anhui Province. In addition, the spatial correlation coefficient (CC) and RMSE were introduced to evaluate the performance of RegCM4_CERES in the grids with available observations. For Fig. 8b, the spatial CC and RMSE are 0.373 and 2.438, respectively.

Figure 8.  Spatial distribution of the mean simulation difference (RegCM4_CERES minus RegCM4) of (a) total LAI and (b) vegetation transpiration. The area shaded by stripes indicates that differences there are at or exceed the 95% confidence level.

There were fewer available observations in the last 10 days of April compared to the first 10 days. As shown in Fig. 8c, two high-value areas, with mean LAIs of 5-7 m² m^-2, were located in the Huanghe-Huaihe Plain and Weihe Plain. The simulated LAIs in Fig. 8d successfully represent the two high-value areas, even though the high-value center in the Weihe Plain was not obvious. The spatial CC and RMSE were 0.431 and 2.324, respectively; slightly better than for Fig. 8b.

According to the simulation performance in April, RegCM4_CERES exhibited an excellent ability to simulate regional variations in wheat development. However, as referred to above in the single-station evaluation, the regional bias between the simulations and the observations in the later reproductive stage (May and June) may be larger than that in April. Owing to a lack of multi-station observed data for other crop types, the regional evaluation of crop development was only focused on the LAI of winter wheat.

3.2.2. Regional evaluation of soil moisture at 10 cm depth

The observed soil moisture at 10 cm depth was derived from the crop growth and soil moisture dataset of the Meteorological Data Center of the CMA. Like the LAI observations, the soil moisture dataset was also collected from the agricultural meteorological stations of the CMA and observations were made once every 10 days. This dataset spans from 1999 to 2008 and was averaged to monthly observations for the following evaluation. Figure 9 shows the observed and simulated mean multi-year soil moisture at 10 cm depth. The observations from 226 stations in Fig. 9a were scatterplotted because of the sparseness of stations in South China. The distribution of soil moisture corresponded with the mean precipitation, and increased progressively from northwest to southeast. The soil moisture simulated by both RegCM4_CERES (Fig. 9b) and RegCM4 (Fig. 9c) had less spatial variation compared to the observations; there was wet bias in most northern areas of China and dry bias in southern areas. This bias has been referred to in many previous studies and has been demonstrated to result from the modules of runoff generation and soil water movement (Zou and Xie, 2012; Wang et al., 2016).

The mean difference between the SIM and CTL test is shown in Fig. 9d. The magnitude of the difference was rather small; this is because the climatic difference between RegCM4 and RegCM4_CERES occurs only due to the dynamical LAI, SAI, and root fraction of crops, and these crops represented only one of the 17 PFTs in the model. As evident in Fig. 1, differences in crop types were summed with other PFTs to obtain the difference in the vegetation column. Then, the difference in the vegetation column was summed once again with the other grid units (glacier, lake, etc.) to obtain the difference in grids (Oleson et al., 2004). Therefore, in the areas where a crop type occupies a large proportion of one grid, climatic differences due to crop development would be more obvious. For this reason, the difference in climate variables between RegCM4_CERES and RegCM4 described hereafter was far less than the difference in RegCM3_CERES, where the grid heterogeneity is not considered and the grids with high crop distribution fractions are treated entirely as crop land (Chen and Xie, 2013).

As shown in Fig. 9d, one obvious difference, of -0.02 m³ m^-3, was detected in the Huanghe-Huaihe Plain, one of the main agricultural areas of China. The dry difference in the SIM test corrected the systematic wet bias of the CTL test to a certain extent. The areas shaded by strips in Figs. 9d, 10, 11f and 12f were at or exceeded the 95% confidence level based on t-tests. However, the differences in other areas were far less than that in the Huanghe-Huaihe Plain. A small wet difference was also detected in South China, but the magnitude was less than 0.01 m³ m^-3 and was not statistically significant. Overall, the simulated 10 cm soil moisture in the CTL test exhibited a wet bias in the north and a dry bias in the south when compared with the observed data. The SIM test corrected the wet bias of the CTL test slightly in the Huanghe-Huaihe Plain. However, for other areas, the improvement was negligible. The spatial CC and RMSE between the simulated and observed values were 0.105 and 0.0929 m³ m^-3 respectively for the SIM test, and 0.134 and 0.0941 m³ m^-3 respectively for the CTL test.

To investigate the soil moisture differences further, the mean multi-year differences in LAI and vegetation transpiration are given in Fig. 10. Figure 10a shows the mean difference in total LAI summed across all 17 PFTs. As described above, the total LAI was derived from the multi-year MODIS retrieval products in CLM3.5. When compared with RegCM4, RegCM4_CERES simulated a higher LAI in the North China Plain and Northeast China Plain, and lower LAIs were simulated in most southern areas of the Yangtze River. In the coastal area of South China, these lower LAI differences reached around -0.4 m² m^-2. The differences in LAIs induced the changes in vegetation transpiration. As shown in Fig. 10b, the transpiration difference between the SIM and CTL test (SIM minus CTL) was high in the Huanghe-Huaihe-Haihe Plain and the high-value center basically corresponded with the spatial distribution of the 10 cm soil moisture difference in Fig. 9d. The low LAI difference in South China also induced small negative differences in transpiration, which were not statistically significant.

Figure 10.  Spatial distribution of the mean 2 m air temperature of (a) observation, (b) simulation by RegCM4_CERES, (c) simulation by the original RegCM4, (d) RegCM4_CERES minus observation, (e) RegCM4 minus observation, and (f) RegCM4_CERES minus RegCM4. The area shaded by stripes indicates that differences there are at or exceed the 95% confidence level.

3.2.3. Regional evaluation of precipitation and 2 m air temperature

The dataset for monthly precipitation and 2 m air temperature during the simulation period was obtained from the Meteorological Data Center of the CMA and interpolated as grid data (0.5^°× 0.5^°) in order to evaluate the simulated spatial distributions and seasonal variations (Xie et al., 2016). Figure 11 shows the spatial distribution of the mean multi-year precipitation and the difference between the SIM and CTL test. The observed precipitation in Fig. 11a increased from northwest to southeast. RegCM4_CERES and RegCM4 exhibited good performance in simulating precipitation over the northwest arid area (Figs. 11b and c); however, both models underestimated the precipitation in the south and overestimated it in the north (Figs. 11d and e). Many previous studies have found that the Grell cumulus convection scheme in RegCM4 induces less precipitation in southern China, and no scheme is entirely suitable for application across mainland China (Gao et al., 2016, 2017). With respect to the difference between the SIM and CTL test in Fig. 11f, the changes in each crop's LAI and root distribution due to the coupled crop models did not induce significant regional differences in the precipitation simulation. A small increase in precipitation in the SIM test was detected in most monsoon areas, except for areas such as the Weihe Plain and the lower reaches of the Yangtze River. The spatial CC and RMSE between the SIM test and the observed data were 0.558 and 1.172 mm d^-1, respectively; similar to the indices between the CTL test and observed data: 0.557 and 1.171 mm d^-1, respectively.

Figure 11.  Mean monthly (a) 10 cm soil moisture, (b) precipitation and (c) 2 m air temperature over the typical area of Northeast China (NE); (d) 10 cm soil moisture, (e) precipitation and (f) 2 m air temperature over the typical area of North China (NC); and (g) 10 cm soil moisture, (h) precipitation and (i) 2 m air temperature over the typical area of South China (SC).

The spatial distribution of the mean multi-year observed and simulated 2 m air temperature is shown in Fig. 12. The observed temperature (Fig. 12a) generally increased from northwest to southeast, and the areas with the lowest temperatures were located in the Tibetan Plateau and northern region of Northeast China. In contrast to the precipitation simulation, RegCM4 demonstrated an excellent ability to simulate the spatial variations in 2 m air temperature (Figs. 12b and c). However, a cold bias was detected in most areas of China (Figs. 12d and e), especially in the Tibetan Plateau. With respect to the difference between the SIM and CTL test in Fig. 12f, a positive difference of about 0.2^°C in the RegCM4_CERES simulation was detected in South China, which reduced the cold bias of RegCM4 in this region. However, in the North China Plain and Northeast China Plain, small cold differences of about 0.1^°C were detected. The temperature differences corresponded with the difference in vegetation LAIs in Fig. 10a. The decrease in LAIs over South China led to less transpiration and more solar radiation absorbed by soil, inducing higher 2 m air temperatures. The differences over the North China Plain and Northeast China Plain, where the LAIs were increased, were basically in direct contrast. The spatial CC and RMSE between the SIM test and observed data were 0.819 and 5.130^°C, respectively; better than the indices between the CTL test and observed data: 0.817 and 5.142^°C, respectively.

Three typical areas——Northeast China (43^°-46^°N, 122^°-125^°E), North China (34^°-38^°N, 115^°-118^°E), and South China (22^°-24^°N, 109^°-115^°E)——were selected based on the distribution of LAI differences in Fig. 10a in order to analyze the seasonal variability in the SIM and CTL test (seen in Fig. 3). Table 2 lists the mean relative fraction of each crop sub-type in the grid cells containing the respective crops, for the three typical areas. According to Table 2, the main cereal crops in the typical areas were spring maize for Northeast China, winter wheat and summer maize for North China, and early rice and late rice for South China. Figure 13 shows the mean multi-year monthly 10 cm soil moisture, precipitation, and 2 m air temperature over the three typical areas.

For the typical area of Northeast China, soil water observations were missing during the freezing period from November to April. As described in Fig. 9, a wet bias in soil moisture was simulated in the CTL test, and the simulation of the SIM test slightly reduced this wet bias. Figure 13a shows that the largest soil moisture differences between the SIM and CTL test occurred from June to October, the main crop growth seasons; no obvious differences were detected in winter and spring. For the monthly precipitation in Fig. 13b, the simulated seasonal variability in the CTL and SIM test was not as obvious as in the observed data; specifically, less precipitation in summer and more in the other seasons was simulated in both models. RegCM4_CERES simulated more precipitation from May to June and less from July to October as compared with RegCM4. The monthly temperatures in Fig. 13c indicate that RegCM4 simulated higher temperatures in winter and lower temperatures in spring, as compared to the observed data. However, the total fraction of rice, wheat, and maize in the Northeast China domain was only 31% of all crop types, based on Table 2. The RegCM4_CERES simulation was much closer to that of RegCM4, except for the slight cold difference from May to August.

Figure 9.  Spatial distribution of the mean precipitation of (a) observation, (b) simulation by RegCM4_CERES, (c) simulation by the original RegCM4, (d) RegCM4_CERES minus observation, (e) RegCM4 minus observation, and (f) RegCM4_CERES minus RegCM4. The area shaded by stripes indicates that differences there are at or exceed the 95% confidence level.

For the typical area of North China, the simulated soil moisture of the CTL test was higher than the observed data. The SIM test using RegCM4_CERES reduced the wet bias of RegCM4 all year round; the largest reductions in bias occurred from April to June. The simulated precipitation in the CTL test was close to the observed data, except that July, when approximately 3 mm d^-1 less precipitation was detected. Slightly more precipitation in summer was simulated by RegCM4_CERES when compared with RegCM4. For temperature, the simulations by both models were close to the observed data; the cold bias of RegCM4_CERES was slightly higher than that of RegCM4 in spring.

In the South China area, a dry bias in moisture and cold bias in temperature were observed in the simulations by both models. With respect to the difference between RegCM4_CERES and RegCM4, higher soil moisture was simulated by RegCM4_CERES across the year, except for summer. The simulated precipitation by RegCM4_CERES was slightly less than that by RegCM4 all year round. In addition, RegCM4_CERES simulated higher temperatures than RegCM4, and slightly reduced the cold bias of RegCM4.

Table 3 lists the statistical indices of precipitation and 2 m air temperature for RegCM4_CERES and RegCM4 over the typical areas and over the whole of China. The three indices——mean bias (MB), temporal CC, and RMSE——were calculated using the mean monthly series from 2000 to 2008. For the simulation of precipitation, RegCM4_CERES reduced the mean bias of RegCM4 slightly, except for the small increase over South China. The RegCM4_CERES simulation exhibited better CC and RMSE values than RegCM4 over Northeast China and South China and poorer values over North China and the whole of China. For the simulation of 2 m air temperature, both models performed well; the indices provided by RegCM4_CERES were slightly better in general. Overall, the statistical indices provided by RegCM4_CERES were generally better than those by RegCM4, but this improvement was rather limited because of the minimal differences between the two models.

In this study, the crop model CERES (-Wheat, -Maize, and -Rice) was coupled with CLM3.5, which comprised the land surface module of the regional climate model RegCM4. The newly developed model, named RegCM4_CERES, divided the PFTs of crops into eight sub-types: spring maize, summer maize, spring wheat, winter wheat, early rice, late rice, single rice, and other crop types. The simulation for each crop type was performed independently according to the planting and harvesting dataset. RegCM4/CLM3.5 used weather and soil condition information as the input for the CERES model, and the simulated LAI, SAI, and root distribution fraction for each crop type obtained by CERES were summed based on each distribution fraction in order to update the corresponding values in RegCM4/CLM3.5.

Using the newly developed RegCM4_CERES model, one simulation test (SIM) was conducted over 10 years, from 1999 to 2008, across China. A control test (CTL) was also conducted with the same settings using the original RegCM4. An observed dataset consisting of crop LAI, soil moisture, precipitation, and 2 m air temperature was employed to evaluate the performance of RegCM4_CERES. The single-station evaluation indicated that the RegCM4_CERES demonstrated excellent performance in simulating the phenological changes of wheat, maize, and rice. However, some biases were still evident in the simulation of peak LAI and aboveground biomass. In general, the RegCM4_CERES simulation matched the observed data slightly better during the earlier vegetative stage (growth of leaves and roots) than during the later reproductive stage (filling of grain).

In addition to the single-station evaluation, a regional evaluation of LAI, soil moisture, precipitation, and 2 m air temperature was further conducted. Owing to data limitations, the LAI for winter wheat during the first and last 10 days of April was selected in order to evaluate the spatial distributions. The results revealed that the RegCM4_CERES simulation was a good match for the increasing gradient of the observed data from northwest to southeast. The high-value areas in Huanghe-Huaihe Plain and Weihe Plain were also simulated during the last 10 days of April.

For the soil moisture at 10 cm depth, the wet bias in the observed data over Northeast China and the dry bias over the area south of the Yangtze River was detected with the original RegCM4 model simulation. When compared with the RegCM4 simulation, the mean soil moisture simulated by RegCM4_CERES dropped by about -0.02 m³ m^-3 in the Huanghe-Huaihe Plain. This difference was statistically significant with a confidence level of 95%. Nonetheless, the differences between the two models in the other areas were rather small and insignificant. Subsequent analysis revealed that the difference in soil moisture corresponded with the differences in total LAI and vegetation transpiration. When compared with RegCM4, the decrease in soil moisture in the Huanghe-Huaihe Plain for RegCM4_CERES was mainly due to the increased transpiration; the precipitation in this region did not differ greatly.

The precipitation simulated by RegCM4 over the area south of the Yangtze River was obviously less than observed. The changes in crop LAIs for RegCM4_CERES had little effect on upper-atmospheric circulation, and thus the RegCM4_CERES simulation did not differ greatly from the RegCM4 simulation. For the simulation of 2 m air temperature, RegCM4_CERES corrected the cold bias of RegCM4 slightly, by about 0.2^°C, over South China, because of the lower crop LAI. The simulated temperature by RegCM4_CERES increased the cold bias of RegCM4 over North China in spring, but the degree of increase was rather small (<0.1^°C).

In addition to regional evaluation, mean multi-year seasonal variations in climate variables were examined for three typical areas (Northeast China, North China, and South China). The coupled model RegCM4_CERES performed slightly better than RegCM4, and the differences between the two models were greater during the crop growth seasons. For the statistical indices of the monthly series, the performance of the two models was roughly the same. The improvement in the statistics for RegCM4_CERES was rather limited because the difference between RegCM4_CERES and RegCM4 arises only from changes in LAI, SAI, and root distribution of crop type among the 17 PFTs in the model grids.