The regional climate model, RegCM3, is used to simulate the 2004 summer surface air temperature (SAT) and precipitation at different horizontal (i.e., 30, 60, and 90 km) and vertical resolutions (i.e., 14, 18, and 23 layers). Results showed that increasing resolution evidently changes simulated SATs with regional characteristics. For example, simulated SATs are apparently better produced when horizontal resolution increases from 60 to 30 km under the 23 layers. Meanwhile, the SATs over the entire area are more sensitive to vertical resolution than horizontal resolution. The subareas present higher sensitivities than the total area, with larger horizontal resolution effects than those of vertical resolution. For precipitation, increasing resolution shows higher impact compared to SAT, with higher sensitivity induced by vertical resolution than by horizontal resolution, especially in rainy South China. The best SAT/precipitation can be produced only when the horizontal and vertical resolutions are reasonably configured. This indicates that different resolutions lead to different atmospheric thermodynamic states. Because of the dry climate and low soil heat capacity in Northern China, resolution changes easily modify surface energy fluxes, hence the SAT; due to the rainy and humid climate in South China, resolution changes likely strongly influence grid-scale structure of clouds and therefore precipitation.
Using dynamical climate models to study climate change can help people understand climate variations. High-resolution regional climate models (RCMs) offer more advantages than general circulation models (GCMs) when used for regional climate simulations [
RCM research and applications include several works regarding the RegCM RCM model [
Few studies have considered the effects of vertical resolution in RCM simulations. For example, Ruti et al. [
The above studies on the effects of resolution in RCM simulations have two features.
Therefore, to study the effects of the different horizontal and vertical resolutions of RegCM3 on SAT and precipitation and the correlations between them over China, we simulated the 2004 summer climate using three different horizontal and vertical resolutions. Note that we used version 3 of RegCM, while the newest version is RegCM4.5; the findings in this paper are suggested to be version-independent. Hence, we have arranged the paper as follows. First, the RegCM3 model is briefly described and the experiments are designed in Section
The regional climate model (RegCM3) used in this work was developed by the Abdus Salam International Centre for Theoretical Physics (ICTP) [
Due to many factors, such as mathematical calculation errors in numerical models and the applicability of physical parameterization schemes that are relevant to spatial extent and scale, different spatial resolutions are expected to result in different simulation results. To test differences in resolution, we used the same physical parameterization schemes for each experiment (Tables
Designed experimental schemes based on horizontal resolutions.
Scheme | Horizontal resolution | Simulation area (buffer) grid |
---|---|---|
H3 | 30 km |
|
H6 | 60 km |
|
H9 | 90 km |
|
Designed experimental schemes based on different vertical layers with different sigma (
Scheme | Layer |
|
|
|
---|---|---|---|---|
V14 | 14 layers | 1, 0.99, 0.97, 0.93, 0.86 | 0.77, 0.67, 0.56, 0.46, 0.35, 0.25 | 0.17, 0.1, 0.04, |
|
||||
V18 | 18 layers | 1, 0.99, 0.98, 0.96, 0.93, 0.89, 0.84 | 0.78, 0.71, 0.63, 0.55, 0.47, 0.39, 0.31, 0.23 | 0.16, 0.1, 0.05, |
|
||||
V23 | 23 layers | 1, 0.99, 0.98, 0.96, 0.93, 0.89, 0.85 | 0.8, 0.75, 0.7, 0.65, 0.6, 0.55, 0.5, 0.45, 0.4, 0.35, 0.3, 0.25, 0.2 | 0.15, 0.1, 0.05, |
Model domain and distribution of the studied subareas, where subareas 1–6 correspond to South China, Southwest China, East China, Northwest China, North China, and Northeast China, respectively.
According to nearly 30 years of variations in summer SAT and precipitation (Figure
Variations of summer SAT and precipitation over the past 30 years in mainland China.
SAT change
Precipitation change
We used NCEP/NCAR 2.5° × 2.5° reanalysis data [
In this paper, only the average seasonal results are discussed. To compare the simulated results and observations that were obtained from the Climate Research Center in the Department of Geography at the University of Delaware and to study the differences between various experiments, we define the seasonal mean model bias (BIAS), root mean square error (RMSE), correlation coefficient (CORR), and standard deviation (STD;
Figure
BIAS (a) and RMSE (b) of summer-mean SAT for the studied subareas (South China (SC), Southwest China (SWC), East China (EC), Northwest China (NWC), North China (NC), Northeast China (NEC), and total area (ALL)).
SAT bias
SAT root mean square errors
Summer-mean distribution of SAT from observations and simulations using Scheme V23 for (a) observations, (b) H3, (c) H6, and (d) H9 in °C.
Observation
V23H3
V23H6
V23H9
Figure
Meanwhile, the RMSEs for the total area (Figure
Spatial correlation coefficients of SAT for the V23 and H3 Schemes at different resolutions for the subareas.
Test | SC | SWC | EC | NWC | NC | NEC | ALL |
---|---|---|---|---|---|---|---|
V23H9 | 0.781 | 0.911 | 0.669 | 0.926 | 0.896 | 0.772 | 0.964 |
V23H6 | 0.813 | 0.923 | 0.719 | 0.938 | 0.917 | 0.859 | 0.966 |
V23H3 | 0.822 | 0.931 | 0.783 | 0.938 | 0.927 | 0.912 | 0.968 |
V18H3 | 0.829 | 0.932 | 0.772 | 0.923 | 0.928 | 0.913 | 0.964 |
V14H3 | 0.812 | 0.931 | 0.745 | 0.927 | 0.921 | 0.914 | 0.959 |
The standard deviation of SAT and precipitation at different resolutions in the studied subareas.
SAT standard deviation
Precipitation standard deviation
For each subarea, the CORR values (Table
For Scheme V23, the changes in SAT with horizontal resolution can be substantial in different areas, while the changes in the total area are slight (Figure
For Scheme V14, the H9, H6, and H3 BIAS values in the total area are −1.10, −1.28, and −1.31°C, respectively. The effects of horizontal resolution are greater in the subareas than in the total area, for example; the H9-H3 difference amounts to 0.5°C in South China. Moreover, when the horizontal resolution increases for Scheme V14, the RMSE only continuously decreases over North China. Meanwhile, for the total area, the H3 RMSE is larger than the RMSEs of H6 and H9.
As shown in Figure
The spatial correlation coefficients for the total area and the subareas are slightly different in the SAT distributions among V14, V18, and V23. In addition, these distributions are very similar to the observed SATs (Table
The SATs for the H6, H9, and H3 resolutions can differ greatly. For the total area, V23 produces slightly better SATs than V14 and V18 at the H3 resolution. For resolutions H6 and H9, V23 provides results that deviate more from the observations than V14 and V18. In addition, the largest differences in the BIAS values for the V14 and V18 simulations are 0.25 and 0.35°C, respectively, which correspond to the largest RMSE differences of 0.1 and 0.13°C, respectively. The SAT sensitivity to vertical resolution is higher than its sensitivity to horizontal resolution. In the subareas, the sensitivities of the simulated SATs to vertical resolution are weak for different horizontal resolutions. Simultaneously, the simulations of SAT are more sensitive to horizontal resolution than vertical resolution.
Figure
Bias (a) and RMSE (b) of summer-mean precipitation for the studied areas.
Precipitation bias
Precipitation root mean square errors
Summer-mean precipitation distribution from observations and Scheme V23 for (a) observations, (b) H3, (c) H6, and (d) H9 in mm d−1.
Observation
V23H3
V23H6
V23H9
Precipitation BIAS values (Figure
The RMSE values for the total area decrease as the horizontal resolution increases (Figure
Spatial correlation coefficients of precipitation from the L23 and X3 Schemes for the studied areas.
Test | SC | SWC | EC | NWC | NC | NEC | ALL |
---|---|---|---|---|---|---|---|
V23H9 | 0.1409 | −0.1086 | 0.2525 | 0.5853 | 0.5776 | 0.4614 | 0.5136 |
V23H6 | 0.2861 | −0.0461 | 0.109 | 0.7994 | 0.527 | 0.4634 | 0.5978 |
V23H3 | 0.1865 | 0.1953 | 0.2872 | 0.8296 | 0.5536 | 0.4351 | 0.6713 |
V18H3 | 0.3269 | −0.104 | 0.1712 | 0.8604 | 0.5295 | 0.4681 | 0.5968 |
V14H3 | 0.1867 | 0.2704 | −0.0382 | 0.8688 | 0.5829 | 0.4502 | 0.6407 |
Considering the BIAS for the total area, Scheme V14 shows that simulated precipitation is not very sensitive to horizontal resolution. However, the subareas show high sensitivity; for example, in North China, the H9-H3 BIAS difference reaches 1.6 mm d−1, which is approximately 20% of the mean summer value for mainland China. In addition, with increasing horizontal resolution, all the RMSE values in North, Northeast, and Southwest China consistently decrease. In the total area, the H9 RMSE value is higher than those of H3 and H6.
In general, Schemes V14 and V23 appear to perform best at a resolution of 60 km. For Scheme V18, higher horizontal resolutions correspond with closer agreement between the simulated and observed amounts of precipitation. For the total area, the largest BIAS (RMSE) difference for V23 between H3 and H6 (H3 and H9) is 0.52 mm d−1 (0.22 mm d−1). Similar results can be observed for Schemes V18 and V14. The influences of horizontal resolution on simulated precipitation are nonlinear at different vertical resolutions. Additionally, the sensitivity of simulated precipitation to horizontal resolution is highest for Scheme V23 and lowest for Scheme V18. The sensitivity in the subareas is stronger than in the total area.
With higher vertical resolution, the spatial variability of precipitation is closer to the observation of the total area (Figure
V23 CORR is the largest (Table
Additionally, V18 outperforms V14 in simulated precipitation for the total area. At resolutions of both H6 and H9, the V18-V23 precipitation difference is relatively small and the V18 and V23 precipitations are closer to the observations than V14. The largest V18 and V23 decreases in precipitation BIAS (RMSE) are 1.49 and 1.81 mm d−1 (0.84 and 0.94 mm d−1), respectively. Therefore, the precipitation sensitivity to vertical resolution is stronger than its sensitivity to horizontal resolution. In the subareas, the simulated precipitation is also sensitive to vertical resolution at different horizontal resolutions, especially for the rainy region of South China.
The simulation results also show differences between the effects of resolution on the simulated SATs and precipitation. For Scheme V23, the H6 SAT shows no improvement over the H9 SAT but the H6 precipitation is substantially improved. The H3 SAT improves compared to the H6 SAT, but the improvement in precipitation at H3 is insignificant. For Scheme V18, the simulated SATs deviate more from the observations as the horizontal resolution increases while the simulated precipitation is closer to the observations. For Scheme V14, the changes in the SAT with horizontal resolution are similar to those of Scheme V18. However, the changes in precipitation with horizontal resolution are similar to those of Scheme V23. Therefore, the SAT and precipitation are well simulated at each resolution and better results are not obtained at certain resolutions. This result clearly shows that different resolutions result in different atmospheric thermodynamic states and suggests that the coordination of the thermodynamic equilibrium of the land-atmosphere system is not treated well in the model. Thus, the simulated thermodynamic equilibrium can be modified by changing the resolution and can deviate from the real state. In addition, when comprehensively comparing the effects of resolution on the simulations, the simulations of SAT and precipitation are (or the coordination of the atmospheric thermal field is) the best at a horizontal resolution of 30 km with 23 vertical layers. Thus, a suitable configuration for horizontal and vertical resolutions exists that can improve the simulation capabilities of the models [
Meanwhile, the sensitivity of simulated SAT to resolution is different from that of precipitation. Of the different vertical resolutions, the highest sensitivity of the simulated SAT to horizontal resolution exists in Scheme V18 and the lowest sensitivity exists in Scheme V23. The simulated precipitation results contrast the SAT results. Therefore, at higher vertical resolution, the sensitivity of simulated SAT to horizontal resolution is opposite to that of the simulated precipitation. The SAT and precipitation demonstrate a stronger sensitivity to vertical resolution than horizontal resolution, indicating that the model is sensitive to vertical resolution. The CORRs of simulated precipitation are generally lower than those of SAT, and the change in the CORR of the precipitation with resolution is larger than the changes in the SAT CORR, which clearly demonstrates that simulated precipitation is affected more by resolution than SAT.
Considering the effects of resolution on the subareas, the distributions of simulated SATs become similar to the observations with increasing horizontal resolution. However, the spatial variability is not affected much by the vertical resolution. The differences between the distributions of the simulated precipitation and observations are much greater than those for the SAT, and the CORR variations are very large with changes in the horizontal and vertical resolution. Furthermore, the sensitivities of SAT and precipitation to resolution are generally stronger in the subareas than for the total area. The SAT sensitivity to resolution is the largest in Northwest and North China (corresponding to the largest SAT BIAS differences in the different experiments at 1.96 and 0.98°C, resp.), and the precipitation sensitivity is the largest in Southwest and South China (corresponding to the largest BIAS differences in precipitation of 4.78 and 2.59 mm/d, resp.). These characteristics are closely related to the dry climates of Northwest and North China and the humid climates of Southwest and South China. Because of the dry climate and low soil heat capacity in the north, the change in resolution can easily cause a change in surface energy flux and land-atmosphere exchange, which would significantly change the SAT. Due to the rainy and humid climate and high soil heat capacity in the south, the change in resolution has a large influence on the grid-scale precipitation and cloud structure. Therefore, different resolutions lead to large precipitation differences in the south and small changes in the simulated SAT.
In this paper, the regional climate model RegCM3 was used to simulate the 2004 summer surface air temperature and precipitation at three horizontal (i.e., 30, 60, and 90 km) and vertical resolutions (i.e., 14, 18, and 23 layers) over the same geographic area (i.e., the area of the model domain excluding the buffer zone) in mainland China. In addition, the effects of resolution on the SAT and precipitation simulations were evaluated using various measures.
The effects of increasing resolution on simulated SAT and precipitation for mainland China are evident. For instance, compared with the increasing summer SAT (0.15°C/10a) in the last 54 years over China [
The effects of changing the horizontal resolution on the SATs differ from the effect obtained by changing the vertical resolution. Changing the vertical resolution has a stronger influence on the simulated results than changing the horizontal resolution, particularly for precipitation simulations. The distribution of simulated SAT becomes closer to the observations with increasing horizontal resolution; however, this phenomenon is less evident with increasing vertical resolution. This result occurs because higher horizontal resolution (as opposed to higher vertical resolution) increases the more realistic local forcings at the land surface, which greatly affect the SAT [
The effects of changing the resolution on SATs differ from the effects of changing the resolution on precipitation, and the resolution-induced sensitivities show some regional characteristics. For example, the simulated SATs in the north and the simulated precipitation in the south were generally the most sensitive to changes in resolution. The difference between the simulated precipitation and observed precipitation distributions is much larger than that of the simulated SATs and changes greatly with horizontal and vertical resolution. In addition, because of the dry climate and low soil heat capacity in the north, the change in resolution tends to easily cause a change in surface energy flux; therefore, the SAT changes significantly. In contrast, due to the rainy and humid climate and high soil heat capacity in the south, the changes in resolution have a large influence on the grid-scale structure of clouds and therefore precipitation. Consequently, the different resolutions result in little change in simulated temperature but large differences in simulated precipitation in the south.
Relative to the simulated SATs with precipitation, the improvements in the simulated SATs do not correspond with the improvements in precipitation due to the change in resolution, indicating that the different resolutions result in greater modifications in the atmospheric thermodynamic state. For example, the simulations of SAT and precipitation for the total area are best at a resolution of 30 km with 23 vertical layers. The other configurations result in poorer representations of the thermodynamic state. Therefore, the coordination of the atmospheric thermodynamic state fields can be improved at a suitable horizontal resolution and at a matching suitable vertical resolution.
Theoretically, higher resolutions correspond with smaller truncation errors due to the numerical discretization of the model dynamic framework, resulting in simulation results that are closer to the observations. However, due to a deficiency of RCMs, the relationship between the simulated SAT and the precipitation using RegCM3 could depend on the selection of physical parameterization schemes, the simulation area, the buffer zone, or even the initial and lateral conditions that are used [
The authors declare that there is no conflict of interests regarding the publication of this paper.
This work was financially funded by the National Natural Science Foundation of China (Grant nos. 41275012 and 41205073).