Simulation Performance and Case Study of Extreme Events in Northwest China Using the BCC-CSM2 Model

Abstract

The BCC-CSM2 model is the second generation of the Beijing Climate Center Climate System Model developed by the National Center of China Meteorological Administration. Using the outputs of two versions of the BCC-CSM2 model with different resolutions, namely: BCC-CSM2-MR and BCC-CSM2-HR, their performance in simulating the climate characteristics of Northwest China was compared. The BCC-CSM2-HR had a better ability to simulate the detailed distribution of the average temperature and precipitation in Northwest China, and could delineate the influence of the topography in detail. The extreme events in Northwest China were evaluated further using the BCC-CSM2-HR and the observation data from China Meteorological Data Center. The BCC-CSM2-HR provided a good simulation of the spatial distribution of extreme climate events in Northwest China, and the spatial distribution of TXx, TNx, TXn, and TNn in Northwest China show closer proximity to the observation than that of TX90p, TN90p, TX10p, and TN10p, even in the case of extreme heavy precipitation. This case study of the extreme weather events showed that the BCC-CSM2-HR model had the best simulation performance for extreme high temperature events in Northwest China, followed by extreme low temperature events, and had the worst simulation ability for extreme precipitation events.

1. Introduction

In the context of global warming and the frequent occurrence of extreme weather, and in order to study the new problems faced in the field of climate change nowadays, the World Climate Research Program (WCRP) launched the sixth international Coupled Model Intercomparison Program (CMIP6), which contains the largest number of participating models, the most well-designed scientific experiments, and the largest amount of simulation data provided in more than 20 years of the CMIP program [1,2]. Along with the implementation of the CMIP program, the coupled model has also developed rapidly from a physical climate system model that considers the energy and water flux exchange processes among atmospheric, oceanic, snow, rock, and biological circles to a more complex Earth climate system model that further adds biogeochemical processes such as carbon and nitrogen cycles and human processes; in the future, it will also consider solid Earth (e.g., Earth plate movements and their triggers),and a relatively complete model of the Earth system will be developed, considering the interactions between the solid Earth (e.g., the Earth’s plate movements, topographic changes, earthquakes, volcanic eruptions, etc.) and space weather [3]. Therefore, climate models are not only an effective tool for describing the interaction of the five major circles of the climate system and their impact on climate change, but are also an important tool for humans to grasp the evolution of the climate system, study the characteristics and behaviors of the current climate [4,5,6], predict future climate change [5,7,8], and provide an important basis for describing the activities of humans and wildlife, and environmental protection [4,6].

The improvements in numerical model simulation’s capability have been affected by many factors such as the resolution, parameterization schemes, and surface process characteristics. The current CMIP6 models have reliable abilities for simulating the geographical distribution of climatological temperature and precipitation over China, with better performance for temperature than for precipitation; they outperform their CMIP5 predecessors [9]. In particular, the CMIP6 models set is obviously superior to the CMIP5 models set in simulations of climate and the relative variability of extreme precipitation in China, and simulations of the influence of arid and semi-arid areas have been significantly improved [10]. However, the ability of the model to simulate the magnitude and spatial distribution of precipitation still needs to be improved.

In order to keep up with the development of climate coupled models, the National Climate Center has been working on the development of the Climate System Model (BCC-CSM) and has been pushing forward. On the basis of three sets of daily reanalysis data, Zhang et al. [11] evaluated the simulation capability of the medium-resolution BCC-CSM2-MR and the individual atmospheric model (BCC-AGM3-MR) for blocking high pressure in the Northern Hemisphere at medium and high latitudes. The frequency of blocking high pressure in winter in Eurasia, especially in the Ural Mountains, was reduced, and the frequency of blocking high pressure in spring in the North Pacific was increased, and the simulation’s deviation was reduced. Xin et al. [12] presented the basic information of the Earth System Model (BCC-ESM-1.0), the medium-resolution BCC-CSM2-MR, and the high-resolution BCC-CSM2-HR, and evaluated the results of the historical experiments of the CMIP6 program in which the BCC-CSM2-MR participated, which showed the results for multi-year temperature and precipitation in the Chinese region. The results show that the BCC-CSM2-MR had better capability for simulating the evolution of multi-year temperature and precipitation in China than the previous version. Sang et al. [13] evaluated two medium-resolution BCC-CSM models for the simulation of soil moisture on the surface of Eurasia at the annual seasonal scale and showed that the BCC-CSM-MR was better at reproducing the mean of the climate and the standard deviation of soil moisture, with better correlations and significantly less bias, and the correlation coefficient of this model exceeded that of most models in CMIP6. Jiang et al. [14] used 26 models from the CMIP6 program, including BCC-CSM-MR, to assess their ability to simulate extreme temperature changes at medium and high latitudes in Asia, and the results showed that the models were in good agreement for simulating the mean climate.

Northwest China, including the five provinces of Xinjiang, Qinghai, Gansu, Ningxia and Shaanxi, is a climate-sensitive area prone to extreme events due to its unique terrain and topography, and is also the focus of meteorologists. Based on statistical methods and numerical simulation methods, a great deal of work has been carried out on the distribution characteristics of [15,16,17], the reasons for change in [18,19], and the water resources of [20,21] the arid areas in Northwest China. In the context of global warming, the frequency and intensity of extreme weather and climate events have increased in Northwest China; the water resource system in Northwest China, which is dominated by ice and snow melt water, is fragile, and Northwest China has seen an increase in precipitation, a rise in lake levels, melting ice and snow, a decrease in the number of dust storm days [22], a significant upward trend of temperature, and a significant rise in extreme low temperatures [23]. Climate change has caused the deterioration of the ecological environment and the increased the occurrence of natural disasters in Northwest China, which have affected agriculture, the ecological environment, and people’s lives in Northwest China [24,25]. Zhang et al. [25] analyzed the spatial and temporal evolution of the warming-wetting trend in Northwest China using measured climate data, CRU data, climate scenario predictions and ecological and hydrological data and found that the changes in temperature, precipitation, and the aridity index in the western part of Northwest China showed the same patterns and did not occur in the eastern region. Zheng [26] studied the spatial and temporal evolution characteristics of summer precipitation in Northwest China over the past 55 years using daily precipitation data and reanalysis data from stations in Northwest China, and the results showed that the eastern part of Northwest China had more precipitation and the western part had less precipitation from 1976 to 1996, whereas the eastern part of Northwest China had less precipitation from 1997 to 2015, the intensity of precipitation in the western part of Northwest China increased significantly, and the number of days of continuous precipitation and extreme precipitation events increased. Qi et al. [27] pointed out that in the context of climate warming, the overall temperature in Northwest China in the past 50 years has shown an increasing trend, with more extreme high temperature events and fewer extreme low temperature events; precipitation has shown a weak increasing trend and more extreme precipitation events. Spatially, the regions with a large rate of temperature increase had an increase in the number of days with extreme high temperature and abnormally high temperature and a significant decrease in the number of days with extreme low temperature and abnormally low temperature. For extreme precipitation and abnormal precipitation, the regions with large increases in the number of days were mostly located in areas with a tendency towards high precipitation.

Extreme climate events and extreme weather events are collectively referred to as extreme events. As rare events with low probability, extreme events are characterized by strong disruption and great harm. An increase in their frequency and intensity will have a serious impact on and cause losses to society, human life, the economy, and the natural ecosystem [28]. Against the backdrop of global warming, the analysis and evaluation of simulations of extreme events in Northwest China based on the climate system model can not only test the accuracy of the model’s results and provide a reference for improving the simulation capability of the model, but also have great significance and practical value for the early warning and prevention of meteorological disasters in Northwest China. Therefore, this study used versions of the BCC-CSM2 model developed by the National Climate Center to compare with different resolutions and demonstrated the improvement in the model’s simulation capability caused by the improvements in the model’s resolution, as well as evaluating the ability to simulate extreme events in Northwest China, showing the similarity in the climate models in terms of extreme events, further improving the understanding of climate change laws in Northwest China, and improving the model’s simulation capability. This will help ensure the steady development of society and agriculture in Northwest China.

2. Materials and Methods

Figure 1 shows the geographical location and topography of the selected study area.

The selected study area is the five provinces of Northwest China, including Shaanxi, Gansu, Qinghai, Ningxia, and Xinjiang. The topography of Northwest China is dominated by plateaus, plains, and basins with large differences in their elevation (Figure 1). It is roughly bounded by the Kunlun Mountains and Qilian Mountains, with lower elevations in the northern and eastern areas of Ningxia and Shaanxi, and higher elevations overall in the south because of the Kunlun Mountains. The combined effect of its unique topography and geographical location has resulted in a predominantly temperate continental climate with cold and dry winters and high temperatures in summer, with precipitation mainly concentrated in the summer. Due to the natural climatic characteristics of Northwest China, which is mainly arid, the ecosystem is fragile, composed of mostly desert and grasslands.

2.1. Data Source

The observations of temperature and precipitation used here are the daily and monthly 0.5° × 0.5° grid point datasets (V2.0) of surface temperature and precipitation in China provided by the National Meteorological Science Data Center (NMSDC), which is based on the latest compilation of Chinese high-density surface station data (about 2400 national meteorological observation stations) and the digital elevation model DEM from the National Meteorological Information Center (NMIC) Basic Data Special Project. The spatial interpolation was performed by the Thin Plate Spline (TPS) tool of ANUSPLIN software to generate a ground-level resolution of 0.5° × 0.5° for the grid data of China from 1961 until now. The data have been checked for quality and are widely used.

The BCC-CSM2 model is a second-generation climate system model of the National Climate Center that combines the atmospheric model (BCC-AGCM3.0), the land surface model (BCC-AVIM2.0), the ocean model (MOM5), and the sea ice model (SIS). The National Climate Center has provided two versions of the BCC-CSM2 model that have different resolutions to the Sixth International Coupled Model Comparison Program (CMIP6) organized by the World Climate Research Program (WCRP). The model’s simulations of changes in global temperature over the past 100 years and the climatic distribution of annual mean precipitation in China have improved significantly relative to the earlier first-generation versions [12,29], but the simulations of atmospheric wind and temperature still need to be enhanced, especially for Antarctic Sea ice, which has large biases [30]. The models used in this study were different versions of the BCC-CSM2 model with different resolutions, namely the high-resolution version (BCC-CSM2-HR) with a horizontal resolution of 0.45° × 0.45° and the medium-resolution version (BCC-CSM2-MR) with a horizontal resolution of 1.125° × 1.125°. Here, we present the results of month-by-month simulations for both versions, and the day-by-day results for the BCC-CSM-HR version.

For comparison, the monthly values of both observations and the model results were selected for 54 years from January 1961 to December 2014, and daily values for 11 years from 2004 to 2014. In addition, correlation analysis and mean squared error calculations were carried out by interpolating the model information to 0.5° × 0.5°grid points for a comparison with the measurements.

2.2. Methodology

The percentile method, commonly used in China and elsewhere, was used to determine the extreme precipitation threshold by ranking the precipitation at one station for a certain time period with the 90th, 95th or 99th percentile; and then determining the extreme precipitation events based on the extreme precipitation threshold. For regions with a small spatial scale, it is reasonable to use the absolute critical value method to define the threshold [31,32], while for regions with a large spatial scale, it is reasonable to use the percentile method to define the threshold [33,34]. The specific approach was as follows: the month-by-month precipitation of all grids in the study area from 1961 to 2014 was arranged in ascending order, and the value of the 95th percentile was used as the extreme precipitation threshold for the grid, and precipitation greater than this threshold was considered to be extreme and was identified as an extreme precipitation event.

Correlation analysis is a method of analyzing the correlation between two variables, and is an important method often used in analyses of the causes of weather and climate change processes [35]. Anand et al. [36] used an analysis applying Pearson’s correlation matrix to eliminate multi-collinearity among the variables and improve the model’s performance. The correlation coefficient is a quantity that measures the closeness of the relationship between two variables, and is calculated as follows:

$$r=\frac{{{\displaystyle \sum }}_{i=1}^n\left(x_i-\overline{x}\right)\left(y_i-\overline{y}\right)}{\sqrt{{{\displaystyle \sum }}_{i-1}^n{\left(x_i-\overline{x}\right)}^2}\sqrt{{{\displaystyle \sum }}_{i-1}^n{\left(y_i-\overline{y}\right)}^2}}$$

(1)

When the value of r is larger, this means that the model’s simulation is better.

The root mean square error standard deviation, also known as the standard deviation, shows the degree of deviation of the variable from the mean within the overall variation, and is a quantitative form reflecting the degree of dispersion among individuals in the group [35]. It is calculated as follows.

$$RSME=\sqrt{\frac{{{\displaystyle \sum }}_{k=1}^n{\left(x_{i,j}-y_{i,j}\right)}^2}{n}}$$

(2)

Extreme climate indices, which can directly reflect climate change, are common tools for studying regional climate change [37,38] and comparing the results of climate models [29,39]. In order to reflect the changes in extreme temperature, eight extreme climate indices defined by ETCCDI [39] were used, including the annual maximum of daily maximum temperature, the annual maximum of daily minimum temperature, the annual minimum of daily maximum temperature, the annual minimum of daily minimum temperature, cold nights, cold days, warm nights and warm days. The first four indices are intensity indices and the last four are frequency indices, as defined in detail in

Table 1. Core set of the selected extreme temperature indices recommended by the ETCCDI.

Label	Index Name	Definition	Unit
TXx	Max TX	Annual maximum of the daily maximum temperature	°C
TNx	Max TN	Annual maximum of the daily minimum temperature	°C
TXn	Min TX	Annual minimum of the daily maximum temperature	°C
TNn	Min TN	Annual minimum of the daily minimum temperature	°C
TN10p	Cold nights	Percentage of days with daily minimum temperatures < 10th percentile	%
TX10p	Cold days	Percentage of days with daily maximum temperatures < 10th percentile	%
TN90p	Warm nights	Percentage of days with daily minimum temperatures > 90th percentile	%
TX90p	Warm days	Percentage of days with daily maximum temperatures > 90th percentile	%

.

2.3. Selection of Individual Extreme Events

Because extreme weather events have mostly occurred in recent years and the daily data from BCC-CSM2-HR are only from 2004 to 2014, the actual observation data (daily maximum temperature, daily minimum temperature and daily precipitation) for the last 11 years (2004–2014) in Northwest China are averaged and then arranged in descending order, and three typical extreme weather events larger than the 95th percentile of their corresponding elements were selected. The resolution of the output data by BCC-CSM2-HR mode was interpolated to be the same as the observed value, and the corresponding events were found for comparison, so as to evaluate the performance of the BCC-CSM2-HR model’s ability to simulate extreme weather events in Northwest China.

Extreme weather events are mainly divided into three types: extreme high temperatures, extreme low temperatures and extreme precipitation. If we take the extreme high-temperature events as an example, first, the daily observation data of the daily maximum temperature in Northwest China from 2004 to 2014 were averaged across the region, then individual dates lying between the 95th and 100th percentiles were combined in ascending order according to the daily maximum temperature; finally, three typical extreme high-temperature events were selected. Extreme low temperature and extreme precipitation events were also selected in a similar manner. The final list of extreme events is shown in

Table 2. The nine selected extreme events in Northwest China, 2004–2014.

Extreme Events	Event 1	Event 2	Event 3
Extreme high-temperature events	31 July–4 August 2006	25 July 2011	2–3 August 2013
Extreme low-temperature events	28 January–3 February 2008	22 January 2012	11 January 2011
Extreme precipitation events	8–9 July 2014	2–3 July 2011	19–21 June 2013

. Through a literature search, it was found that the extreme weather events selected in this way were in line with reality. For example, during Event 1 of the extreme high-temperature events, which occurred in the summer of 2006, the average temperature in China was the highest in history for the same period since 1951, and it was also the 10th consecutive year since 1997 that the temperature was higher than normal. The regional average temperature of Shaanxi, Gansu, Qinghai, Ningxia, and other provinces reached the highest in the same period in history [40]. In particular, there was a high temperature of 35~38 °C in Xinjiang and the eastern part of Northwest China on 29–31 July 2006, during which the highest temperature in Turpan, Xinjiang, reached 45 °C for two consecutive days [41].

3. Results and Discussion

3.1. Comparison of the Results of the Models with Different Resolutions

To compare the effects of the models with different resolutions on the simulated temperature in Northwest China, the observed values of the average temperature in Northwest China from 1961 to 2014 and the spatial distribution of the difference between the BCC-CSM2-MR and BCC-CSM2-HR models’ simulation results and the observations are given in Figure 2. For the whole year, except for the Qinghai region and the mountain range areas with a higher elevation, the daily mean temperature of all other regions in Northwest China is above zero and is even above 4 °C in the southern border and southern Shaanxi. In winter, the daily mean temperature over all of Northwest China is below zero, but still the temperatures in Qinghai and the mountain range areas are lower. In spring and autumn, the distribution of daily mean temperature in Northwest China is similar to the annual. In spring and autumn, the distribution of the daily mean temperature in Northwest China is similar to that of the annual mean temperature, but in spring, the daily mean temperature in South Xinjiang is higher, reaching over 14 °C. Summer is the season with the highest daily mean temperature in Northwest China. In summer, the daily mean temperature in North and South Xinjiang, Gansu, Ningxia, and Shaanxi is above 20 °C, and the maximum value in Turpan Basin is above 26 °C, whereas the lowest value in Qinghai is above 4 °C. The BCC-CSM2-HR model simulated the characteristics of the variations in temperature with topography more clearly than the BCC-CSM2-MR model, and the average difference between the BCC-CSM2-HR‘s results and the observations is ±2 °C, which is lower than the difference between the BCC-CSM2-MR’s results and the observations. Meanwhile, the correlation coefficient of the daily mean temperature in Northwest China between the BCC-CSM2-HR’s results and the observations is above 0.85, both annually and for all four seasons, passing the statistical significance test at the 99.9% confidence level, and is higher than those between the BCC-CSM2-MR’s results and the observations.

The ability to reproduce the spatial distribution of summer precipitation in Northwest China at different time scales is an important criterion for judging a model’s simulation capability. The observed spatial distribution of precipitation in Northwest China and the differences between the simulated results of both the BCC-CSM2-MR and the BCC-CSM2-HR and the observations in each season during 1961~2014 are given in Figure 3.

In Figure 3, the areas with large values for observed annual precipitation in Northwest China are mainly in eastern Qinghai, southeastern Gansu, southern Shaanxi, and the Tianshan region, with values above 400 mm, while the Tarim and Qaidam basins are the areas with the least annual precipitation (less than 80 mm). However, the deviation of the BCC-CSM2-HR model is slightly smaller, and the results are closer to the observed values, effectively improving for the annual precipitation in the Kunlun Mountains and Qilian Mountains, and reducing the deviation of the simulated values for these areas to less than 120 mm. However, the BCC-CSM2-HR’s simulations has large deviations for southern Qinghai and southern Gansu, indicating that the model needs further improvement. The precipitation in Northwest China is mainly concentrated in summer. Except for the precipitation in Xinjiang, the summer precipitation in the rest of the regions simulated by both resolution models is high; the precipitation simulated by the BCC-CSM2-HR is low with a wider range, and its deviation is not large. The correlation coefficient of precipitation between the simulation and the observations is higher in summer (0.66 for the BCC-CSM2-MR and 0.75 for the BCC-CSM2-HR) and autumn (0.58 and 0.68, respectively) than in winter (0.28 and 0.32, respectively) and spring (0.37 and 0.52, respectively), and the BCC-CSM2-HR is still better than the BCC-CSM2-MR. Overall, the distribution of precipitation in Northwest China for each season simulated by the BCC-CSM2-HR is better and closer to the observations than the results of the BCC-CSM2-MR.

In summary, the BCC-CSM2-HR has a better ability to simulate the climate in Norwest China than the BCC-CSM2-MR. In view of its good simulation ability, the BCC-CSM2-HR model is used to evaluate further the extreme climate indices and typical extreme weather events in Northwest China.

3.2. Analysis of Extreme Climate Indices

The eight extreme climate indices in Table 1 were used to assess the ability of the BCC-CSM2-HR to simulate the extreme climate events in Northwest China. Figure 4 shows the spatial distribution of the TXx and TNx in Northwest China from 1961 to 2014, respectively. It can be seen that the areas where high values of TXx are observed are mainly located in the Tarim Basin and Junggar Basin on both sides of the Tianshan Mountains and the eastern part of Northwest China, and the daily maximum temperature is above 28 °C. The BCC-CSM2-HR provides good simulations of the spatial distribution of TXx, which is very consistent with the observed values, but the large difference can be seen for the boundary between the Kunlun Mountain and the Qilian Mountain (Figure 4). The simulated TNx values of the BCC-CSM2-HR in the Tarim Basin, the Junggar Basin, and the Qaidam Basin are larger than the observation, with a band of large values along the Tianshan Mountains. However, the simulated TNx values for the western part of the Kunlun Mountains are smaller than the observations. The BCC-CSM2-HR performs better for most of Qinghai, south-central Gansu, Ningxia, and Shaanxi, with the difference in TNx between the simulated and the observed values ranging from −2 °C to 2 °C.

The values of TXn in the Tianshan Mountains and the Junggar Basin simulated by the BCC-CSM2-HR are relatively large, and the maximum value reaches 8 °C, as shown in Figure 5. However, the simulated values of TXn in the southwest of Tarim Basin is smaller than the observed values by ±2 °C. According to the distribution of the observed TNn in Northwest China, the high-value areas are mainly located in the Tarim Basin, the Qaidam Basin, southern Gansu, Ningxia, and Shaanxi, especially in the southwest of Tarim Basin and southern Shaanxi; the low-value areas are mainly located in the Tianshan Mountains, the Kunlun Mountains, and most areas of Qinghai except for the Qaidam Basin. The BCC-CSM2-HR can provide good simulations of the spatial distribution of TNn, but there is some deviation in the order of magnitude for areas, such as the Qaidam Basin and the Tianshan Mountains. Zhao et al. [42] pointed that the performance of the CMIP6 models for simulating the indices of TXn and TNn in Eurasia was good. The spatial pattern of extreme temperatures in Northwest China simulated by the BCC-CSM2-HR is close to the observed values, and the correlation coefficients of TXx, TNx, TXn, and TNn are above 0.77, with the maximum being 0.90 for TNx. This indicates that there is a significant spatial difference in the performance of the climate system models for simulating extreme temperature indices.

As shown in Figure 6, the regions with more frequent cold days (TX10p) and nights (TN10p) correspond to the areas with high average temperatures in Northwest China, which are mainly located in the three major basins and most regions of central and southern Gansu, Ningxia, and Shaanxi, while the regions with a lower frequency are mainly the Kunlun Mountains and southern Qinghai. The values simulated by the BCC-CSM2-HR are generally large, but the simulated TX10p values has a negative deviation in the Qaidam Basin and southern Qinghai. Compared with the observations, an area with a large positive deviation appeared in the southwest of Qinghai, and an area with a banded negative deviation appeared in the Kunlun Mountains.

The observed spatial distributions of warm days (TX90p) and nights (TN90p) are similar, as shown in Figure 7, and the overall frequency is about 4%. The values of TX90p and TN90p simulated by the BCC-CSM2-HR are relatively large in southern Xinjiang, Eastern Gansu and central Shaanxi, and the maximum value reaches more than 4.8%. The simulated values of TX90p in southern Qinghai and TN90p in southwestern Qinghai are relatively small, leading to a positive deviation for most areas of Xinjiang and a negative deviation in the south of Qinghai.

For warm days and nights and for cold days and nights, the simulated results of the BCC-CSM2-HR model for most of Northwest China is larger than the observations. The deviation between the simulations of the BCC-CSM model and the observations for TX90p in Northwest China is negative and that of TN10p is positive [43]. Compared with the CMIP5 models and the reanalysis data, the deviation of the simulated TX90p and TN90p values from the global averages over all land is positive and that of the simulated TX10p and TN10p values is negative [39]. In addition, the correlation coefficients of TX10p, TN10p, TX90p, and TN90p between the BCC-CSM2-HR’s results and the observations are around 0.01. In short, for the intensity indices (TXx, TNx, TXn, and TNn), the performance of the BCC-CSM2-HR model is better than the performance for the frequency indices (TX10p, TN10p, TX90p, and TN90p).

Because of the vast area of Northwest China and its unique topography, the precipitation level of each place cannot be determined by a single criterion, so the percentile method is used to determine the threshold of extreme heavy precipitation at different points in Northwest China. The spatial distribution of extreme heavy precipitation exceeding the 95th percentile threshold in Northwest China from 1961 to 2014 is shown in Figure 8. The simulated extreme heavy precipitation in Xinjiang deviates from the observations, and the simulated extreme heavy precipitation in the Kunlun Mountains and the Tianshan Mountains is also smaller than the observations. However, the extreme heavy precipitation in the north of the Tianshan Mountains and the west of the Tarim Basin is not simulated, and the simulated maximum value in southern Shaanxi is about 440 mm, which deviates greatly from the observed value. From this analysis, it can be seen that the areas with large differences in the extreme heavy precipitation between the observations and the simulation are mainly in the areas with mountainous terrain, indicating that the parameterization process of the model for areas with high-altitude terrain needs to be improved [44].

3.3. Analysis of Typical Extreme Weather Events

In the context of global warming in the future, the number of extreme high temperature events in China will increase significantly, and their area will increase; the number of extreme low-temperature events will decrease significantly, and their area will decrease; and the number of heavy precipitation events will increase, and their area will expand [45]. This indicates that the probability of extreme weather events occurring will increase to a certain extent, and the consequent impact on human productivity and life will also increase. Therefore, typical extreme events during study period (from 2004 to 2014) including three extreme high-temperature events, three extreme low-temperature events and three extreme precipitation events were selected (Table 2).

From the spatial distribution of the three extreme high-temperature events according to the observations and the values simulated by the BCC-CSM2-HR (Figure 9), it can be seen that the BCC-CSM2-HR can simulate the basic distribution of the extreme high-temperature events. However, there are still some deviations in the values, among which the deviation for Event 2 is the largest, and the simulated values for the Tarim Basin and the Junggar Basin area are larger than the observed values, showing a larger range for areas with high values, and the value is higher than 45 °C. Meanwhile, for Event 1, the BCC-CSM2-HR model does not simulate a large area of extremely high temperatures in the central and western Tarim Basin, but the overall spatial distribution is consistent with the observation. The area with largest deviation in Event 3 is still in the Tarim Basin, but the deviation is smaller than that of Event 2.

As can be seen in Figure 10, the extremely low temperatures simulated by the BCC-CSM2-HR are generally larger than the observed values. Among them, the distribution of the simulated extremely low temperatures in Event 1 is closer to that of the observations, simulating an area with high values in the southeastern Tarim Basin and the southern side of the Qaidam Basin. The most obvious difference for Event 2 is that the BCC-CSM2-HR simulates a low-value area in the western part of the Tarim Basin, which is inconsistent with the observed high values in the Tarim Basin.

From the distribution of the three events of extreme precipitation (Figure 11), it can be seen that the BCC-CSM2-HR is less effective at simulating the amount of precipitation for the extreme precipitation events, which is consistent with the previous simulation results for extreme precipitation intensity. For Event 1, the BCC-CSM2-HR reproduces the area with high precipitation in northern Shaanxi, which is consistent with the observed results, but there are some deviations in the magnitude and the simulated results are large.

In order to more intuitively discern and compare the simulation performance of the BCC-CSM2-HR for the nine extreme events, the ratio of the standard deviation of the simulated results to that of the observed results and the correlation coefficient for the nine extreme events are given below using Taylor diagrams (Figure 12).

For the three extreme high-temperature events (red points in Figure 12), although the performance of the simulation for Event 2 is relatively poor, the ratio of the standard deviation for Events 1 and 3 is close to 1, i.e., the dispersion of the simulated results is small, and the spatial correlation coefficients for Event 1 and Event 3 are also larger than those for all other events, with a maximum value of about 0.86, followed by 0.8. For the three extreme low-temperature events (blue dots in Figure 12), the spatial correlation coefficient of Event 1 is the largest (0.53) and the spatial coefficients of the remaining two events are smaller, about 0.2, whereas the ratio of the standard deviations for all three extreme low- temperature events is greater than 1.25. For the three extreme precipitation events (green dots in Figure 11), the ratio of the standard deviation for all three events is less than 1, the spatial correlation coefficient of Event 1 is the largest (about 0.55), and the correlation of Event 3 is the smallest and is negative, so it does not appear in Figure 12.

In short, the overall ability of the BCC-CSM2-HR to simulate extreme events in Northwest China can be summarized as follows: extreme high-temperature events have the best performance, followed by extreme low-temperature events, but it is worst for extreme precipitation events. The ability to simulate extreme events with a long duration is better. The best simulation is found for Event 1 of the extreme high-temperature events, and the worst simulation is found for Event 3 of the extreme precipitation events.

3.4. Discussion

In line with the research purposes of this study, it can be clearly seen that the BCC-CMS2-HR model has better simulation capability than the BCC-CSM2-MR model and is closer to the actual observations. This indicates that the improvements in the model’s horizontal resolution have improved the ability to show the influence of the terrain and have greatly improved the performance for simulating air temperature and heat flux. Similar results have been found for comparing the performance of the BCC-CSM1.1 and BCC-CSM1.1m for simulating surface temperatures [29] and extreme climate events [43] over China. However, the influence of the complex terrain of the area on the simulations still needs to be improved, especially for the Tianshan Mountain area, where the simulated results for the temperature always showed a positive deviation from the observations. The BCC-CSM2-HR provide a good simulation of the distribution of extreme events in Northwest China, but there are some differences in their magnitude. The distribution of the extreme temperature indices simulated by the BCC-CSM2-HR shows larger areas with high values and smaller areas with low values than the observations, and the positive deviation of the simulated values is large in the Tianshan Mountains. The area where the BCC-CSM1.1m underestimated the average surface temperature is in the Tarim Basin [29]. This again shows that the strong influence of the terrain. The study by Tan et al. [46] showed the BCC-CSM-MR model could better simulate the spatial distribution of and variation in various land surface variables, but there were still differences in the intensity. The improvement in the horizontal resolution of BCC-CSM climate model can improve the simulation of westerly circulation on 200 hPa, the 850 hPa circulation field and surface heat fluxes, thus improving the simulation’s performance for summer temperatures and precipitation in Central Asia [30,47].

However, the capability of the BCC-CSM2-HR to simulate summer precipitation in Northwest China still needs to be improved. Our results for the extreme events indicate that the performance of the BCC-CSM2-HR for the temperature indices is better than that for extreme heavy precipitation. Huang et al. [48] have proposed two possible reasons for the large uncertainties in the simulated summer precipitation over eastern China: different partitioning between stratiform and convective precipitation, and the horizontal resolution. On the basis of the simulated precipitation, Sillmann et al. [39] and Huang et al. [44] expected that improvements in model in terms of the parametrization of unresolved physical processes, most notably convective precipitation, may also play a role, which needs to be investigated further.

4. Conclusions

Using daily and monthly gridded data including precipitation, daily mean temperature, daily maximum temperature, and daily minimum temperature obtained by spatial interpolation for five northwestern provinces from 1961 to 2014 and the output data of the related versions of the BCC-CSM2 model with different resolutions for the CMIP6 program’s historical experimental simulations, the spatial distribution of extreme events in Northwest China was analyzed via correlation analysis, the percentile method, and Taylor diagram. The spatial distribution of extreme climate events and extreme weather events in Northwest China was analyzed comparatively. The following main conclusions were drawn.

For the daily mean temperature and precipitation in Northwest China, the mean deviation of the BCC-CSM2-HR from the observations is generally smaller than that of the BCC-CSM2-MR, except for some differences between the simulated results and the observations in summer. The simulated results of the BCC-CSM2-HR can reflect the detailed distribution of temperature and precipitation in Northwest China better, especially in the regions with mountainous terrain.

The BCC-CSM2-HR has a good ability to simulate the detailed distribution of extreme climate events in Northwest China, including extreme temperature indices and extreme heavy precipitation. The distribution of TXx, TNx, TXn and TNn, which are the intensity indices of extreme temperature, show values that are closer to the observations than the frequency indices (TX10p, TN10p, TX90p, and TN90p). The spatial correlation coefficient of TXx between the BCC-CSM2-HR’s simulations and the observations is 0.92. However, the spatial distribution of simulated extreme heavy precipitation is close to the observations in terms of both range and magnitude in all regions except near the Tianshan Mountains. It again shows that the influence of the terrain on the simulated results.

For extreme weather events in Northwest China, the BCC-CSM2-HR performs well in simulations of weather events with a longer duration. The performance of simulated extreme high-temperature events is better, and the spatial correlation coefficient between the simulations and the observations reaches 0.86. This is followed by extreme low temperature events, for which the maximum spatial correlation coefficient is 0.53; the ability to simulate extreme precipitation events is the worst.

In this study, the evaluation of the BCC-CSM2-HR model’s simulation capability mainly focused on reproducing the spatial distribution of extreme temperature and precipitation events in Northwest China. If the simulation time of the BCC-CSM2-HR is extended to 2020, similar results can be obtained. This study can provide a reference for climate system model to study extreme events in arid areas. These results indicate that the climate system model can not only be used to study the climate problem with a long time series, but also to study and predict extreme events.