Performance of RegCM4 over Major River Basins in China

One of the most important impacts of climate change is the change in water resources in response to changes in precipitation and evaporation, with the latter strongly influenced by temperature. Located in East Asia, climate in China is characterized by a large variability both in space and time, due to the complex topography, land-sea contrasts, land surface conditions, and monsoon circulations in the region. During winter (December-January-February, DJF), the prevailing cold and dry air from Siberia leads to low precipitation over the country, with the exception of the southeastern coastal regions. In summer (June-July-August, JJA), the mature summer monsoon dominates China's climate, with precipitation accounting for 50%-70% of the annual total over northern China. The annual mean precipitation shows a maximum of more than 1500 mm in the southeast and decreases towards the north and west, with a minimum of less than 25 mm observed in the northwestern deserts. In addition, pronounced interannual variability exists over the region (Tao and Chen, 1987).

Previous studies have shown that current coupled atmosphere-ocean general circulation models cannot capture the precipitation patterns over East Asia well, mostly due to their coarse resolution, while nested high-resolution regional climate models (RCMs) can greatly improve the simulations (e.g., Gao et al., 2001, 2006; Jiang et al., 2005; Qian and Leung, 2007; Xu et al., 2010; Yu et al., 2010; Zou and Zhou, 2013).

Of the various RCMs used over East Asia (Fu and Yuan, 2001; Wang et al., 2003; Xu et al., 2006; Huang and Zhang, 2007; Qian and Leung, 2007; Yu et al., 2015), the RegCM series (Giorgi et al., 1993a, b; Pal et al., 2007; Giorgi et al., 2012) are among the most commonly used. Applications of RegCM in China include present day climate simulation, studies of land-use and aerosol effects, climate change projections, paleoclimate simulations, seasonal forecast experiments, as well as further developments of the models (e.g., Liu et al., 1994; Giorgi et al., 1999; Gao et al., 2001, 2013; Luo et al., 2002; Ju and Wang, 2006; Li and Zhou, 2010; Shi et al., 2011; Wu et al., 2012; Ji and Kang, 2013; Zou and Zhou, 2013; Wang et al., 2014; Zou et al., 2014a, b; Zhang et al., 2015; Zou et al., 2016). In the latest version of the model, RegCM4 (Giorgi et al., 2012), the more advanced package of the Community Land Model (CLM) (Oleson et al., 2008) was introduced to describe land surface processes. Through a series of experiments, (Gao et al., 2016) found that RegCM4 using CLM performs the "best" with the selection of the Emanuel convection scheme (Emanuel, 1991) over China. Following that work, we updated the land cover and surface emissivity input to the model to further improve its performance over the region (see section 2 for more detail). A 21-year simulation driven by the ERA-Interim reanalysis (Uppala et al., 2008) was carried out to evaluate the model's ability to reproduce the present day climatology and interannual variability of surface climate over the region. In this paper, we describe the analysis of this simulation, focusing on the performance of the model over the main river basins throughout the China territory for the future improvement and application of the model in the projection of changes in water resources in response to global warming.

Further developments and implementation of physical processes have been carried out since the first release of RegCM4 (Giorgi et al., 2012). The version employed in the present study is RegCM4.4 (http://gforge.ictp.it/gf/project/regcm/). The dynamics of RegCM4 remained the same as in previous versions (Giorgi et al., 1993a, b; Pal et al., 2007), as a hydrostatic, compressible, sigma-p vertical coordinate model. The version of the CLM land surface scheme used here is CLM3.5, and convection is represented by the Emanuel scheme. The atmospheric radiative transfer is computed using the radiation package from the NCAR Community Climate Model CCM3 (Kiehl et al., 1998), with the non-local formulation of (Holtslag et al., 1990) to describe the planetary boundary layer. Lastly, resolvable scale precipitation is represented by the SUBEX parameterization of (Pal et al., 2000).

The East Asia domain in phase II of the International Coordinated Regional Climate Downscaling Experiment (CORDEX; Giorgi et al., 2009) is used in the study (http:// www.cordex.org/index.php?option=com_content&view= article&id=88&Itemid=625). The domain encompasses continental China, Korea, Japan, Mongolia, and the adjacent areas and oceans at a grid spacing of 25 km. Note that, compared to the domain previously used in phase I of CORDEX, the southeast Asia portion of the domain is removed, since it is considered as a separate domain (Wu, 2012; Oh et al., 2014; Huang et al., 2015; Juneng et al., 2016). The model is run at its standard configuration of 18 vertical sigma layers, with the model top at 50 hPa. The ERA-Interim dataset at a resolution of 1.5^°× 1.5^° (latitude × longitude) is used to obtain the initial and time-evolving lateral boundary conditions at six-hourly intervals to drive the model.

We found that the default land cover data used in RegCM4 and CLM3.5 shows large discrepancies compared to reality over China. For example, the dominant bare soil in the default land cover over the Tibetan Plateau should be grass, and the large portions of crop in southern China should be trees or grass. Thus, the land cover data was updated based on the vegetation and vegetation regionalization maps of China (Hou, 1981; Zhang, 2007), as reported by (Han et al., 2015). The updated land cover data show more spatial detail than the original, and some preliminary sensitivity tests showed that it improved the model simulation both for temperature and precipitation. More specifically, it reduced the cold and wet biases in south and southwestern China in DJF. In JJA, the warm bias over the Tibetan Plateau and the wet bias in northern China also showed a decrease. Furthermore, the surface emissivity for bare soil and snow in CLM are set to 0.96 and 0.97, respectively, which are larger than those observed in China (Prabhakara and Dalu, 1976; Liu et al., 2014). Thus, they were changed to 0.80 and 0.92, respectively. This reduced effectively a cold bias in DJF found in previous versions of the model (e.g., Giorgi et al., 1999).

A long-term simulation for the period 1 January 1990 to 31 December 2010, a total of 21 years, was conducted, with the first year of 1990 being excluded from the analysis as model spin up. The observational dataset of CN05.1 (Wu and Gao, 2013) is employed to validate the simulation. CN05.1 is an augmentation of CN05 (Xu et al., 2009) and comprises four variables (daily mean, minimum and maximum temperature, along with precipitation) at a resolution of 0.25^°× 0.25^°. The model outputs are interpolated bilinearly to the CN05.1 grid to facilitate the comparison. The analysis is limited to the China region and its 10 main river basins; namely, Songhuajiang River Basin (SRB), Liaohe River Basin (LRB), Haihe River Basin (HaiRB), Yellow River Basin (YLB), Huaihe River Basin (HRB), Yangtze River Basin (YRB), Zhujiang River Basin (ZRB), Southeast Rivers Basin (SERB), Southwest Rivers Basin (SWRB), and interior rivers in the Northwest River Basin (NWRB), as shown in Fig. 1 (MWRC, 1981; Chen et al., 2011; Sun et al., 2014). The river basins also roughly coincide with the primary climatic zones of China.

Figure 1.  Model domain (gray shading), topography (units: m), the major rivers, and the 10 river basins in China selected for the present study: 1. Songhuajiang River Basin (SRB); 2. Liaohe River Basin (LRB); 3. Haihe River Basin (HaiRB); 4. Yellow River Basin (YLB); 5. Huaihe River Basin (HRB); 6. Yangtze River Basin (YRB); 7. Zhujiang River Basin (ZRB); 8. Southeast Rivers Basin (SERB); 9. Southwest Rivers Basin (SWRB); 10. interior rivers in the Northwest River Basin (NWRB).

3.1.1. Temperature

Figure 2 shows the observed multi-year mean temperature and bias (model minus observations, DJF and JJA) of the RegCM4 simulation, as well as the bias of the ERA-Interim. The model is in general too warm over China in DJF, except for the Tibetan Plateau and the southeastern coasts (SERB) (Fig. 2c). It is noted that at least part of the general warm bias is inherited from the driving field of the ERA-Interim (Fig. 2e).

The warm bias of the model is typically in the range of 1^°C to 2.5^°C in eastern China and greater than 5^°C in Northwest and Northeast China (Fig. 2c), as in the previous version of RegCM3 (e.g., Zhang et al., 2008). The prevailing warm bias over the high latitudes in the cold seasons can also be found in different generations of GCM simulations (e.g., Xu et al., 2010; Jiang et al., 2016), and to some extent in the ERA-Interim (Fig. 2e). A cold bias mostly in the range of 2.5^°C to 5^°C dominates in the Tibetan Plateau, which is also a common feature in GCM simulations. However, large uncertainties in the observation data exist in this largely uninhabited region, which is characterized by sparse station distribution. Analysis has shown that the temperature differences among different gridded datasets can be up to several degrees over the region (Wu and Gao, 2013; Sun et al., 2014).

For a more quantitative evaluation of the model performance, Fig. 3a provides values of the spatial correlation coefficients (CORs) and Table 1 presents biases for mean temperature over the river basins and the whole of China between observations, ERA-Interim and RegCM4 in DJF and JJA. Note the statistical significance of the spatial correlation is calculated using the equivalent sample size method following (Zwiers and von Storch, 1995) implemented in the NCAR Command Language (NCAR, 2016). The DJF CORs for both RegCM4 and ERA-Interim are high, with values mostly in the range of 0.90 to 0.99. CORs for ERA-Interim are in general slightly larger than for RegCM4. The biases for RegCM4 range from -3^°C to 3.5^°C over the basins, with positive values in 7 out of the 10 basins. Mean biases of ERA-Interim in the basins are mostly positive, in the range of 0.5^°C to 1^°C, except in SWRB (-0.5^°C). The mean biases for RegCM4 and ERA-Interim over China are 0.8^°C and 0.7^°C, respectively.

Figure 2.  Mean temperature over China during 1991-2010 (units: ^°C): (a) observation in DJF; (b) observation in JJA; (c) bias of RegCM4 in DJF; (d) bias of RegCM4 in JJA; (e) bias of ERA-Interim in DJF; (f) bias of ERA-Interim in JJA.

The model shows a better performance in JJA compared to DJF, with the bias mostly within 2.5^°C (Fig. 2d). The model is too warm over the deserts in Northwest China, whereas it is too cold over the Tibetan Plateau. In the east, a dipole bias pattern is found, with a cold bias in the north and a warm bias in the southwest. Similar to DJF, ERA-Interim shows a warm bias in eastern China and a mixture of warm and cold biases in the mountainous areas of western China (Fig. 2f).

It is of interest to note that, while the CORs for the whole China region are the same (0.97), in 8 out of the 10 basins, slightly larger CORs for RegCM4 are found with respect to ERA-Interim in JJA (Fig. 3a). While the dominant biases are positive in ERA-Interim, more negative bias values are found for RegCM4 in the different basins. The mean bias over China for ERA-Interim and RegCM4 is -0.2^°C and 0.7^°C, respectively (Table 1).

Figure 3.  Spatial correlation coefficients over different river basins and China between observation and the RegCM4 simulation and ERA-Interim in DJF and JJA: (a) temperature; (b) precipitation; (c) interannual variability of temperature (standard deviation); (d) interannual variability of precipitation (coefficient of variation). Correlations that are not statistically significant at the 95% confidence level are marked with an asterisk (*), and negative values are annotated.

3.1.2. Precipitation

For DJF precipitation (Fig. 4a), a general overestimation by the model is found in the north and over the Tibetan Plateau (Fig. 4c). The overestimation is more significant over the mountain chains of Northwest China. Again, uncertainties in the observation dataset in these areas may contribute to the model bias, in particular considering the systematic underestimation of solid precipitation by rain gauges, which may reach up to 30% (e.g., Adam and Lettenmaier, 2003). In southern China, the precipitation center is not well simulated, whereas a band extended from the southeast to the southwest is found in the model. Also, an underestimation in the southeast and prevailing overestimation in the southwest are found in the simulation. It is clear from Fig. 4 that the spatial bias patterns for precipitation are at least partially inherited by the ERA-Interim forcing (Fig. 4e), and we note that they are found also in most GCM simulations (Xu et al., 2010; Jiang et al., 2016).

Figure 4.  Mean precipitation over China during 1991-2010 (units: mm): (a) observation in DJF; (b) observation in JJA; (c) RegCM4 in DJF; (d) RegCM4 in JJA; (e) ERA-Interim in DJF; (f) ERA-Interim in JJA.

Figure 3b and Table 1 provide the CORs and bias (in percentage) of mean precipitation over the river basins and whole of China between observations, ERA-Interim and the RegCM4 simulation. CORs for ERA-Interim are all positive in DJF, with the lowest values found in the ZRB (0.37) and SWRB (0.58, statistically insignificant) basins, both with complex topography but with fewer station observations for the latter (Wu and Gao, 2013). CORs in other basins and China are high, in the range of 0.75 to 0.97. For the RegCM4 simulation, negative and insignificant CORs are found in HaiRB (-0.23), ZRB (-0.18), and YRB (0.32), with relatively low positive correlations in SERB (0.40), SWRB (0.50), and NWRB (0.55) and a correlation of 0.67 for the whole China area.

The relative biases for RegCM4 are large in this DJF dry season (Table 1), greater than 100% in 6 out of the 10 basins and largest biases in SWRB and NWRB. The overestimation over the entire China region is by a factor of 1.5. The biases of ERA-Interim are also in general positive but mostly lower than 50% throughout the basins, except in SWRB (factor of 2.6). The mean bias of ERA-Interim over China is 37%. We should, however, recollect that the cold season biases are likely strongly amplified by the gauge undercatch problem mentioned above.

In JJA, precipitation from the RegCM4 simulation and ERA-Interim are in good agreement with observations, both for the general spatial pattern and amount (Figs. 4b, d and f). An overestimation by RegCM4 is found in Southwest China and along the Taihang and Yanshan mountains in North China. The RegCM4 simulation shows finer spatial detail compared to CN05.1 and ERA-Interim, in particular over western China. Note that a large precipitation center around (29^°N, 94^°E) is evident in the RegCM4 simulation and ERA-Interim (Figs. 4c-f), but not in the observations (Figs. 4a and b), in both DJF and JJA. This is where the grand canyon of the Yarlung Zangbo River (Fig. 1) is located, along with the path of water vapor transported from the Indian Ocean and Bay of Bengal into the inner land of the Tibetan Plateau (Gao et al., 1985).

The CORs between observations, ERA-Interim and the RegCM4 simulation over the whole China are larger in JJA compared to DJF, at 0.93 and 0.85, respectively (Fig. 3b). However, the CORs for ERA-Interim are in general lower in the river basins in JJA. Statistically insignificant CORs are found in HaiRB (-0.25) and LRB (0.73) for RegCM4, and HaiRB (0.13) and SWRB (0.47) for ERA-Interim. The biases are still mostly positive but much smaller compared to DJF over the various river basins (Table 1). The biases for ERA-Interim are in general less than 20%, except in SWRB (88%), and for RegCM4 range from 0% to 123%. The mean bias over China is 46% and 21% for RegCM4 and ERA-Interim, respectively. Note that, in summer, the under catch gauge problem is less important than in winter (Adam and Lettenmaier, 2003), which may also be a reason for the lower overestimations by the models.

Figure 5 compares the annual cycle of monthly mean temperature from the observations, RegCM4 simulation and ERA-Interim over the 10 main basins. The observed temperatures show distinct seasonal variations in all the basins, with a minimum in January and maximum in July. In the northern basin, SRB, a difference of 40^°C can be found between the summer and winter seasons. The seasonal differences are smaller in the southern basins, with less than 10^°C and around 15^°C in SWRB and ZRB, respectively.

Both RegCM4 and ERA-Interim capture the annual cycle well. A warm bias in SRB and cold bias in SWRB of up to a few degrees in the winter months can be found for RegCM4. In general, ERA-Interim agrees more with the observations; however, RegCM4 sometimes performs better. For example, a warm bias of ∼2^°C for the whole year in SERB, and a warm bias of ∼ 1^°C in the summer half of the year in SWRB, are found in ERA-Interim, while the biases of RegCM4 there are small.

Figure 5.  Monthly mean temperature over the 10 major river basins in China during 1991-2010 from observation (black lines), ERA-Interim (blue lines) and the RegCM simulation (red lines) (units: ^°C).

Figure 6.  Monthly mean precipitation over the 10 major river basins in China during 1991-2010 from observation (black lines), ERA-Interim (blue lines) and the RegCM simulation (red lines) (units: mm).

For precipitation, it is generally dry in the winter months in all basins, with only a few millimeters in the northern basins and around 50 mm in the south. Precipitation is more pronounced in the summer monsoon season (Fig. 6). It reaches a maximum in July over the northern basins and in June over the southern basins, except in SERB where a secondary peak is found in August. The model and ERA-Interim capture the observed seasonal evolution and peaks in most basins. In SWRB, an earlier peak in June instead of July is found in RegCM4; while in ZRB, two peaks in June and August instead of one in June are found in ERA-Interim. Consistent with the results in Table 1, RegCM4 largely overestimates the precipitation in LRB, HaiRB, YLB, SWRB, and NWRB, and to a lesser extent in SRB, ZRB, and SERB. ERA-Interim reproduces the amount of precipitation better; however, an overestimation is also evident in several basins, e.g., in SWRB and NWRB.

Figure 7 shows the spatial distributions of the interannual variability, as measured by the interannual standard deviation, for the observation, RegCM4 simulation, and ERA-Interim in DJF and JJA. In DJF, the observed interannual variability is characterized by a maximum (>1.6^°C) in Northeast China, larger values over the northern boundary of the country and western China, and smaller values in eastern China and the northwest basins (<1.0^°C) (Fig. 7a). Both RegCM4 and ERA-Interim capture the spatial pattern and magnitudes of the interannual variability quite well (Figs. 7c and e).

The CORs and biases of temperature interannual variability biases for RegCM4 and ERA-Interim over the river basins and the whole of China in DJF are provided in Fig. 3c and Table 2. The CORs are all positive over the basins, with values for the whole of China of 0.80 and 0.91 for RegCM4 and ERA-Interim, respectively. Slight underestimations of the temperature interannual variability are found in most basins for both RegCM4 and ERA-Interim, with generally smaller biases for the latter. The biases of RegCM4 and ERA-Interim for the whole of China are 0.01^°C and -0.03^°C, respectively.

Figure 7.  Interannual variability of temperature (standard deviation) over China during 1991-2010 (units: ^°C): (a) observation in DJF; (b) observation in JJA; (c) RegCM4 simulation in DJF; (d) RegCM4 simulation in JJA; (e) ERA-Interim in DJF; (f) ERA-Interim in JJA.

In JJA, the magnitude of the observed interannual variability is smaller compared to DJF (Fig. 7b). Again, relatively larger values (>1^°C) are found in Northeast China, while the smallest values (<0.4^°C) are found along the southern coast and in southwestern China. The pattern is reproduced well in ERA-Interim, and to a lesser extent by RegCM4. The CORs for RegCM4 and ERA-Interim are 0.70 and 0.88, respectively (Fig. 3c). The biases of RegCM4 are mostly negative in the river basins, indicating a general underestimation of interannual variability by the model in JJA (Table 2). Larger biases of around or greater than -0.2^°C are found in SRB, LRB, SERB and SWRB, with the bias for the whole country being -0.14^°C. The biases of ERA-Interim are in the range of -0.04^°C to 0.03^°C over the basins, except in SWRB (-0.12^°C). The ERA-Interim bias for the whole country is 0.01^°C.

The interannual variability of precipitation is measured here by the coefficient variation (defined as the standard deviation divided by the mean), which removes the dependency of the precipitation standard deviation from the mean. This is reported for observations, RegCM4 simulation, and ERA-Interim in Fig. 8. In DJF, the largest values (>0.8) are mainly found in the northwest basins and HaiRB, while the lowest values (<0.3) are seen in Northeast China, the mountain ranges of Northwest China (Qilian and Tianshan mountains), and areas over the Tibetan Plateau (Fig. 8a). In the middle and lower reaches of the Yangtze River, the coefficient of variation is relatively low, with values in the range of 0.3-0.4.

RegCM4 and ERA-Interim capture the general pattern of the observed spatial distribution. While RegCM4 underestimates the magnitudes of the large variability centers in the north, ERA-Interim overestimates them. Different from the observations, values lower than 0.2 dominate over the Tibetan Plateau in the model, although this underestimation needs to be seen within the context of the uncertainties in the observation dataset. The CORs between ERA-Interim, RegCM4 and observations are 0.65 and 0.44, respectively, for the whole of China. For different river basins, the CORs of SRB (0.22), HaiRB (0.58) and SWRB (0.58) for RegCM4, and LRB (0.49), ZRB (0.37) and SWRB (0.45) for ERA-Interim, are statistically insignificant (Fig. 3d). As shown in Table 2, the largest biases (around -0.2) in the RegCM4 simulation of DJF precipitation interannual variability are found over LRB and HaiRB, followed by NWRB and YLB (about -0.1). ERA-Interim also shows a general underestimation over most basins, but less pronounced than for the RegCM4, except over HaiRB and HRB. For the whole of China, the bias for RegCM4 and ERA-Interim is -0.09 and -0.04, respectively.

The interannual variability of precipitation shows much lower magnitudes in JJA than DJF. In the observations, areas with values greater than 0.30 are found in the eastern China basins, western part of Northeast China, and most of Northwest China, except over the mountain ranges, with maxima in the range of 0.4 to 0.5 (Fig. 8b). Values lower than 0.20 prevail over the Tibetan Plateau and adjacent eastern areas. The variability patterns and magnitudes are generally captured both by RegCM4 and ERA-Interim, with a slight underestimation in the east and overestimation in the northwest basins in RegCM4, and a large overestimation over the northwest basins and underestimation in southern China for ERA-Interim (Figs. 8d and f).

Figure 8.  Interannual variability of precipitation (coefficient variation) over China during 1991-2010: (a) observation in DJF; (b) observation in JJA; (c) RegCM4 simulation in DJF; (d) RegCM4 simulation in JJA; (e) ERA-Interim in DJF; (f) ERA-Interim in JJA.

The CORs in half of the river basins for RegCM4, and 3 out of the 10 for ERA-Interim, are statistically insignificant, indicating the difficulties in reproducing the monsoon precipitation interannual variability at river basin scales. For the whole of China, the CORs are 0.55 and 0.70 for RegCM4 and ERA-Interim, respectively (Fig. 3d). The biases in the river basins are all negative, in the range of -0.01 to -0.07, for RegCM4, while for ERA-Interim both positive and negative biases are found. The bias for the whole of China in RegCM4 and ERA-Interim is -0.02 and 0.01, respectively (Table 2). It should be pointed out that precipitation variability tends to increase at finer spatial scales (Giorgi, 2002), and this might help to explain some of the variability biases found in models, such as the underestimations by RegCM4.

Two indices——the simple daily intensity index [SDII; defined as the annual total divided by the number of rain days (days with precipitation ≥ 1 mm)] and the maximum number of consecutive dry days [CDD (days with precipitation <1 mm)], based on (Frich et al., 2002) and (Zhang et al., 2011), respectively——are selected to illustrate the model performance in simulating precipitation-related extremes. These indices, from the observation, RegCM4 simulation, and ERA-Interim, are presented in Fig. 9, while the bias of RegCM4 and ERA-Interim over the river basins and China are provided in Table 3.

The observed SDII spatial patterns show some agreement with the precipitation mean, with decreasing values towards Northwest and Northeast China (Fig. 9a). However, different from the latitudinal distribution of mean precipitation (Fig. 4), over eastern China the area with values greater than 10 mm d^-1 extends from south of the Yellow River down to the southern coasts. The SDII values exceed 12 mm d^-1 in the middle reaches of the Yangtze River and the southern coast. RegCM4 and ERA-Interim show agreement with observations over Northeast and western China, where the SDII is less than 7 mm d^-1. In eastern China, a significant underestimation by around 2 mm d^-1 exists in RegCM4, and to a lesser extent in ERA-Interim. The CORs between RegCM4 and ERA-Interim with observations are 0.80 and 0.64, respectively.

Figure 9.  The simple daily intensity index (SDII; units: mm d^-1) and maximum number of consecutive dry days (CDD; units: d) over China during 1991-2010: (a) observation of SDII; (b) observation of CDD; (c) RegCM4 simulation of SDII; (d) RegCM4 simulation of CDD; (e) SDII of ERA-Interim; (f) CDD of ERA-Interim.

The biases of RegCM4 over the river basins are in line with ERA-Interim, with the largest underestimation ranging from -1.6 mm d^-1 to -2.1 mm d^-1 in HRB and ZRB. RegCM4 and ERA-Interim overestimate SDII in SWRB, NWRB and YLB. However, it is possible that the overestimations are also due to the observation uncertainties in the grand canyon of the Yarlung Zangbo River for SWRB, and over the mountain ranges for NWRB and YLB, as discussed in section 3.1. The biases for the whole of China are 0.4 and 0.2 mm d^-1, for RegCM4 and ERA-Interim, respectively (Table 3).

The magnitude of the observed CDD is largest, in excess of 100 days, in the northwest basins. CDDs of less than 30 days are found in the middle and lower YRB in southern China (Fig. 9b). RegCM4 systematically underestimates the CDD in the north and over the Tibetan Plateau (Fig. 9d), in correspondence with the excessive precipitation simulated over the regions during the winter dry season (Fig. 4c). For example, the CDD observed in most of Northeast China is in the range of 50 to 70 days, while in the simulation it ranges from 20 to 30 days. The COR between the model simulation and observation is 0.40. ERA-Interim reproduces the CDD magnitudes better, but also with an underestimation in the northern part of China and in the region east of the Tibetan Plateau. The COR value of ERA-Interim is 0.77.

As shown in Table 3, the biases of RegCM4 are around 30 days in the northern basins of SRB, LRB, HaiRB, and YLB, while the biases of ERA-Interim are also relatively large, with values around 10 days over the basins. In SWRB, the bias for RegCM4 and ERA-Interim is -38 and -20 days, respectively. The largest model bias of -56 days is found in NWRB, which encompasses most of the Tibetan Plateau and Northwest China. The bias for the whole of China is -35 and -7 days for RegCM4 and ERA-Interim, respectively.

Note that the underestimation of the large precipitation extremes in RegCM4 and ERA-Interim, and the negative CDD found in RegCM4, are also typical of CMIP5 models (e.g., Chen and Sun, 2015).

In this paper, we report on the performance of RegCM4, using CLM3.5 as the land surface scheme, over the major river basins in China. The land cover of the model was updated based on the vegetation map of China and, in addition, the surface emissivity parameters based on observation. The model was driven by ERA-Interim data, and a 21-year simulation was conducted using a horizontal resolution of 25 km.

Overall, the model shows reasonable performance in reproducing the present day temperature climatology over the river basins and the whole of China, with high spatial correlation coefficients. The annual cycles of temperature are also well simulated. The model biases are smaller in JJA compared to DJF, when a warm bias in the northern basins and cold bias over the Tibetan Plateau are found (consistent to previous versions of the model).

The general mean precipitation pattern and annual cycles of precipitation are also reproduced by the model, with better performance in JJA compared to DJF. In DJF, the model tends to overestimate precipitation during the dry conditions in the north, and to underestimate it in the wetter south. This is again in line with previous RegCM versions, as well as many other climate model simulations.

RegCM4 also shows reasonably good performance in reproducing the interannual variability of temperature and precipitation over the region, albeit with a tendency towards underestimation. Similar to other climate models, RegCM4 tends to underestimate the large precipitation extremes in eastern China, while shorter than observed CDD values are simulated in northern China and over the Tibetan Plateau. Overall, ERA-Interim shows biases of temperature and precipitation mean, variability and extremes similar to those of RegCM4, although mostly lower in magnitude. This is despite the assimilation of observations in the reanalysis, further emphasizing the difficulties for climate models in reproducing the East Asian climate.

Overall, although the version of RegCM4 analyzed here generally reproduces the mean and variability of climate over China, some persistent biases found in previous model versions, as well as many other climate models, remain. More efforts are needed in the future to better understand and reduce the model biases identified in this work. As mentioned, RegCM4 is undergoing further developments and some new model physics options are available, such as an updated version of CLM (CLM4.5) and new convection parameterization options (Kain-Fristch and Tiedtke schemes). We are currently testing these new schemes to assess whether they improve these long-standing biases. Towards the goal of improving climate simulation over East Asia, it is also important that multiple models are used and assessed over the region, as envisioned, for example, in the CORDEX program.