We use the outputs of two climate models developed by the BCC: BCC-CSM1.1 and BCC-CSM1.1m. The BCC-CSM1.1 is a global coupled climate system model and includes atmosphere, ocean, sea ice, and land model components and incorporates the carbon cycle. The atmospheric component is BCC_AGCM2.1, the horizontal resolution of which is 2.8° (T42), and the number of vertical layers is 26, with the model top at 2.914 hPa (Wu et al., 2008, 2010). The land model BCC_AVIM1.0 (Atmosphere–Vegetation Interaction Model) is also developed by the BCC and includes biogeophysical, eco-physiological, and soil carbon–nitrogen dynamics modules (Ji et al., 2008). The ocean component MOM4_L40 was developed by the Geophysical Fluid Dynamics Laboratory (GFDL) and uses a tripolar grid with a horizontal resolution of 1/3° × 1° (Griffies et al., 2005). The sea ice model is the sea ice simulator (SIS) developed by the GFDL and has the same horizontal and vertical resolutions as MOM4_L40 (Modular Ocean model). The differences between the BCC-CSM1.1 and BCC-CSM1.1m are the horizontal resolution of the atmospheric and land models. The horizontal resolution is increased to 110 km (T106) in BCC-CSM1.1m. A detailed introduction to the two models is given by Xin et al. (2013) and Wu (2012).
We use the historical simulations and projections performed by these two models under the RCP4.5 and RCP8.5 scenarios provided by CMIP5. According to the CMIP5 protocol, the historical experiments are carried out from 1850 to 2005 with time-varying external forcing, including greenhouse gases, ozone, aerosols, volcanoes, and solar radiation. The simulation during 1961–2005 is compared with the observational data. The climate change projections are analyzed to gain a better understanding of the variation in extreme climate events under different future scenarios of radiative forcing, which is stabilized at 4.5 and 8.5 W m−2 in 2100.
We use data from the CN05.1 gridded dataset as the observational data. This dataset is based on daily observations from 2416 stations in China starting from 1961 and has a spatial resolution of 0.25° × 0.25°. The variables used in this study include the daily mean temperature, the daily minimum and maximum temperatures, and the daily precipitation. This dataset has been widely used to evaluate the performance of climate models in China (Wu and Gao, 2013). To facilitate comparisons between the simulations and the observational data, the model outputs and observed data are interpolated into a 2.0° × 2.0° grid by the bilinear interpolation method. The climatic mean during 1961–2005 is taken as the climatology.
Twelve extreme climate indices are used in this study (Table 1). They are described in detail by Bao et al. (2015) and Zhou and Chen (2012). The baseline period for the R95T, R90N, HWDI, TX90P, and TN10P indices is 1961–90.
Number Index Definition Unit 1 R95T Percentage contribution of the sum of precipitation > 95th percentile to the total amount of precipitation % 2 R90N Number of days with daily precipitation > 90th percentile of number of wet days day 3 CDD Maximum number of consecutive dry days (RR < 1 mm) day 4 CWD Maximum number of consecutive wet days (RR ≥ 1 mm) day 5 R1 Annual number of days with RR ≥ 1 mm day 6 SDII Simple precipitation intensity index mm day–1 7 Rx5 Maximum consecutive five-day precipitation mm 8 FD Total number of days when TN < 0°C day 9 SU Total number of days when TX > 25°C day 10 HWDI Total number of days with at least five consecutive days when TX > TXnorm + 5°C day 11 TX90P Percentage of days when TX > 90th percentile % 12 TN10P Percentage of days when TN < 10th percentile %
Table 1. Definitions of extreme climate indices. RR: daily rainfall; TX: maximum temperature; TN: minimum temperature; TXnorm: base period mean
2.1. Model and experiment
2.2. Observational data
Figure 1 shows the spatial distribution of the climatological mean extreme precipitation indices over China from the observational dataset, the BCC-CSM1.1, and BCC-CSM1.1m during 1961–2005. The total annual precipitation in China increases from northwest to southeast in the observational dataset (Fig. 1a). Compared with the observations, both models overestimate the amount of precipitation in southwestern China. However, the model bias of the BCC-CSM1.1 (6%) is lower than that of the BCC-CSM1.1m (−22%) in eastern China because the BCC-CSM1.1m weakens the northeastward transport of water vapor (Kan et al., 2015). The BCC-CSM1.1m is able to simulate a more reliable spatial distribtion of the total annual precipitation (PTOT) in Xinjiang Region. Both models underestimate the amount of precipitation in southeastern China, with the BCC-CSM1.1m simulating less precipitation in this region than the BCC-CSM1 (Wang et al., 2016). The pattern correlation coefficient (PCC) between the BCC-CSM1.1m and the observations over easten China (20°–45°N, 110°–120°E) exceeds 0.8, which is higher than that (0.75) between the BCC-CSM1.1 and the observations over easten China. The BCC-CSM1.1 and BCC-CSM1.1m reproduce the spatial distribution of the percentage of extreme precipitation (R95T) well, with large amounts of extreme rainfall in northern China and little extreme rainfall in western China. Compared with the BCC-CSM1.1, the BCC-CSM1.1m improves the simulations in the southwestern, central, and eastern regions of China. The spatial pattern in eastern China simulated by the BCC-CSM1.1m shows a higher PCC (pattern correlation coefficient) of 0.91 with the observations than that by the BCC-CSM1.1. However, both models tend to underestimate R90N over the whole country by > 10 days. The PCCs between the simulations and observational dataset are −0.74 (BCC-CSM1.1) and −0.7 (BCC-CSM1.1m).
Figure 1. Spatial distributions of the annual mean extreme precipitation indices: (a1–a3) PTOT (mm), (b1–b3) R95T (%), (c1–c3) R90N (day), (d1–d3) CDD (day), (e1–e3) CWD (day), (f1–f3) R1 (day), (g1–g3) SDII (mm day–1), and (h1–h3) Rx5 (mm) during 1961–2005 over China from (a1–h1) the observational dataset (CN05.1), (a2–h2) BCC-CSM1.1 (T42), and (a3–h3) BCC-CSM1.1m (T106).
Zhou et al. (2018) used the NEX-GDDP (NASA Earth Exchange Global Daily Downscaled Projections) dataset (statistically downscaled model data) to evaluate extreme precipitation events and showed that the lowest PCC (< 0.7) of the R90N index between all models and the observational dataset occurs over the Yangtze River basin. Bao et al. (2015) evaluated the frequency of extreme precipitation (R95pF) in China simulated by GFDL-ESM2G and found a PCC of 0.67 with the observational data. Therefore, the capability of the BCC-CSM to simulate the frequency of extreme precipitation needs further improvement.
The observed CDD (consecutive dry day) has a high centered in the Tarim basin and a low centered in southern China (Figs. 1d1–d3). The two models reproduce higher values of the CDD in the Tarim basin than in other regions, but underestimate the values by about 40 days. The CDD simulated by the BCC-CSM1.1 in eastern China is reduced by about 15 days compared with the observational dataset. The PCC between the simulation by the BCC-CSM1.1 and the observational dataset is 0.94, whereas it is 0.19 for the BCC-CSM1.1m. The reduced correlation skill in BCC-CSM1.1m may be attibuted to the simulation of atmospheric circulation at a different resolution. Kan et al. (2015) reported that the BCC-CSM1.1m overestimates the intensity of the western Pacific subtropical high. By contrast, the BCC-CSM1.1m performs better than the BCC-CSM1.1 in southwestern China.
The observed CWD shows a maximum centered over southwestern China and a minimum centered over the Tarim basin (Figs. 1e1–e3). Both models overestimate the CWD over the western region, but the overestimation is reduced in the BCC-CSM1.1m. The PCCs over eastern China are 0.91 for the BCC-CSM1.1m and 0.75 for the BCC-CSM1.1.
The observed and simulated annual numbers of days with daily precipitation ≥ 1 mm (R1) are shown in Figs. 1f1–f3. Both models overestimate the frequency of light precipitation over Xinjiang Region and southwestern China. The BCC-CSM1.1m performs bettter over both these areas. The PCCs between the simulation and the observational dataset are 0.75 for the BCC-CSM1.1 and 0.91 for the BCC-CSM1.1m.
The spatial distribution of the SDII is consistent with that of the PTOT, which increases from northwest to southeast. The two models both slightly underestimate the SDII in eastern China. The BCC-CSM1.1m improves the simulation over Tibet and the PCC for the SDII between the BCC-CSM1.1m and the observational dataset is higher (0.6) than that for the BCC-CSM1.1 (0.48) in eastern China. The observed Rx5 shows a similar pattern to the PTOT and SDII. Both models overestimate Rx5 in southwestern China, with less bias in BCC-CSM1.1m, although both PCCs are > 0.8.
Based on these comparisons, both the BCC-CSM1.1 and BCC-CSM1.1m are able to reproduce the spatial distribution of extreme precipitation events over eastern China. The BCC-CSM1.1m improves the simulation of extreme precipitation indices, including R95T and CWD, especially for the complex terrain of the western region. Because the physical processes are the same for both the BCC-CSM1.1 and BCC-CSM1.1m, the finer resolution of the topography may be the main reason for the better performance of the BCC-CSM1.1m in simualating the R95T, CWD, R1 and SDII indices. Finer scale surface forcing is important in the production of realistic small-scale features. In terms of the atmospheric circulation, the BCC-CSM1.1m improves the simulation of the onset of the earlier East Asian summer monsoon (EASM) and the northward jump in the western Pacific subtropical high (WPSH) compared with the BCC-CSM1.1 (Kan et al., 2015). This may be why the BCC-CSM1.1m performs better on these indices.
Figure 2 shows the spatial distributions of the annual mean extreme temperature indices during 1961–2005 over China from the observational dataset, BCC-CSM1.1, and BCC-CSM1.1m. Both models simulate the spatial distribution of FD in China well, with PCCs between observations and simulations of 0.94 for BCC-CSM1.1 and 0.97 for BCC-CSM1.1m, except in the eastern area of southwestern China. The BCC-CSM1.1m improves the simulation of FD in Xinjiang and Tibet regions. The two models both simulate a greater number of FD over the Yangtze River valley and southern China (Figs. 2a2–a3). The simulations of these two models of the SU index are consistent with the observations. The higher resolution BCC-CSM1.1m reduces the negative biases over southwestern China relative to the BCC-CSM1.1 (Figs. 2b2–b3).
Figure 2. Spatial distributions of the annual mean extreme temperature indices: (a1–a3) FD (day), (b1–b3) SU (day), (c1–c3) HWDI (day), (d1–d3) TX90P (%), and (e1–e3) TN10P (%) during 1961–2005 over China from (a1–e1) the observational dataset (CN05.1), (a2–e2) BCC-CSM1.1 (T42), and (a3–e3) BCC-CSM1.1m (T106).
The PCC of the SU index between the BCC-CSM1.1m and the observational dataset is > 0.9 and is higher than that between the BCC-CSM1.1 and the observations (0.87). Both models simulate an artificially high center of HWDI over southwestern China. The BCC-CSM1.1m overestimates the HWDI over southwestern Xinjiang and northeastern China (Figs. 2c2–c3). The PCCs over eastern China between the simulations and observations are 0.58 (BCC-CSM1.1) and 0.51 (BCC-CSM1.1m), respectively.
Both models perform better over southeastern China than over other areas for TX90P, although both models underestimate TX90P (Figs. 2d2–d3) and overestimate TN10P (Figs. 2e2–e3) over northern China. The PCCs of the TX90P index between simulations and observations are relatively low, with values of 0.17 (BCC-CSM1.1) and 0.09 (BCC-CSM1.1m), respectively. The BCC-CSM1.1m gives a lower PCC between the simulation and observations for the TN10P index (−0.31) than the BCC-CSM1.1 (0.69). In general, most climate models are unable to simulate the distributions of TX90P and TN10P well in China (Yang et al., 2014; Chen and Sun, 2015). Zhang (2015) also found that CMIP5 models show a large deviation between simulations and observational data for TN10P in northern China. By contrast, the MIROC5 model simulates TX90P and TN10P fairly well, with correlation coefficients of 0.94 and 0.98, respectively (Yu et al., 2015).
These comparisons show that increasing the resolution of the model does not significantly improve its ability to simulate the indices of extreme temperature events (HWDI, TX90P, and TN10P).
Table 2 shows the annual mean (mean), standard deviation (SD), relative bias and PCCs between the simulations of the BCC-CSMs and the observational dataset. The extreme indices (including R90N, CDD, and HWDI) show larger deviations from the observations. Most of the PCCs between the simulated extreme indices by the BCC-CSM1.1m and observed indices are > 0.8. The largest improvement in the indices is for R95T, with the correlation coefficient increasing from 0.21 to 0.91. Both models simulate the mean of the extreme temperature indices in FD and SU well, but can only weakly simulate the other extreme temperature indices.
Mean SD Relative bias PCC OBS T42 T106 OBS T42 T106 T42 T106 T42 T106 PTOT 987.2 1047.69 768.12 179.62 185.57 174.58 6 −22 0.75 0.86 R95T 46.51 40.81 43.46 10.04 9.04 11.12 −12 −7 0.21 0.91 R90N 39.29 24.69 22.36 16.11 9.00 9.65 −37 −43 −0.74 −0.70 CDD 42.60 27.47 34.61 12.66 8.39 12.90 −36 −19 0.94 0.19 CWD 9.03 10.64 10.92 2.94 3.29 3.60 18 21 0.75 0.91 R1 100.55 126.84 108.39 11.37 13.41 13.57 26 8 0.88 0.95 SDII 8.84 7.87 6.70 1.33 1.13 1.24 −11 −24 0.48 0.60 Rx5 117.87 135.95 118.40 41.59 42.66 57.18 15 0 0.85 0.86 FD 93.51 90.63 82.39 8.96 6.14 7.44 −3 −12 0.94 0.97 SU 120.73 113.14 128.26 10.80 11.81 12.98 −6 6 0.87 0.91 HWDI 6.98 8.43 8.02 8.15 7.29 7.38 21 15 0.58 0.51 TX90P 41.07 40.81 39.41 15.86 17.22 17.48 −1 −4 0.17 0.09 TN10P 32.01 33.38 35.08 12.72 15.21 19.65 4 10 0.69 −0.31
Table 2. Annual mean and standard deviation (SD) of the 13 extreme climate indices averaged over East China (20°–45°N, 110°–120°E) calculated from the observational dataset (OBS), the BCC-CSM1.1 (T42), and BCC-CSM1.1m (T106) during 1961–2005. The relative biases of the two experiments with respect to OBS and the PCCs between the climatology over East China in the two simulations and the OBS are also given
Both models show a weak ability to simulate the R90N index. Most CMIP5 models do not capture the frequency of precipitation well, particularly extreme rainfall events (such as the R90N index) in Asia (Sooraj et al., 2016). Possible reasons for this are as follows.
The Asian summer monsoon (ASM) sub-systems and tropical cyclones have major roles in triggering extreme events (Freychet et al., 2015) in China. As a result of the coarse resolution of the CMIP5 models, current climate models, including the BCC-CSM1.1 and BCC-CSM1.1m, cannot simulate tropical cyclones well. The ASM shows many variations on intraseasonal, interannual，and multidecadal timescales (Ding and Chan, 2005). Most CMIP5 models are unable to reproduce the ASM rainfall at different timescales due to their coarse resolution and incorrect convection parameterization (Saha et al., 2014; Sabeerali et al., 2015). The descriptions of sub-seasonal signals [such as Madden–Julian oscillation (MJO) and Boreal Summer Intraseasonal Oscillation (BSISO)] in the model are not sufficiently accurate. These low-frequency oscillations also have a large impact on extreme precipitation events.
Previous studies have shown that the WPSH has a vital role in affecting summer rainfall in China. Although the BCC-CSM1.1m improves the northward jump of the WPSH, its simulation of the intensity of the WPSH is stronger than that in the BCC-CSM1.1m and it gives a location further to the northwest, leading to a weak simulation of the northeastward transport of water vapor (Wang et al., 2016).
The SST modes [such as the PDO (Pacific Decadal Oscillation) and ENSO] affect the interannual and interdecadal variability of the EASM. The CMIP5 models are unable to simulate well the temporal and spatial variation of the Indo-Pacific warm pool and North Pacific Ocean dipole (Yu et al., 2019). The SST simulated over Northwest Pacific by climate models shows a cold bias (Wang et al., 2018) and has dramatic effects on the EASM. Therefore, the fact that the two models do not perform well on the R90N index may be attributed to their poor ability to simulate large-scale precipitation, their coarse resolution, the incorrect parameterization of convection, and their weak ability to simulate SST modes over the Indo-Pacific warm pool and North and Northwest Pacific Ocean.
The poorer simulations of the CDD, HWDI, TX90, and TN10P indices in East China indicate that these indices are not sensitive to the horizontal resolution of the atmospheric model. For the TN10P index, the BCC-CSM1.1m simulates more cold nights than the BCC-CSM1.1 model. The poorer performance of the TN10P index is probably due to the larger biases in the simulated atmospheric circulation over East Asia in the BCC-CSM1.1m (Kan et al., 2015). Another possible reason is that the BCC-CSM1.1m underestimates the thermal contrast between the land and the sea (Kan et al., 2015). Previous studies have shown that ocean–atmosphere coupling and SST modes are important in reproducing the TN10P index (Peings et al., 2012) and that the increased horizontal resolution may affect air–sea interactions. In addition, regional processes (such as cloud and snow radiative feedbacks) and the large-scale dynamics in the BCC-CSM1.1m change with the resolution.
These results show that the BCC-CSM1.1m has more advantages in simulating PTOT, R95T, CWD, R1, Rx5, FD, and SU than the BCC-CSM1. The following discussion will mainly focus on indices with correlation coefficients > 0.8.
Extreme temperature (such as heat waves and droughts) and precipitation (such as rainstorms and floods) events have shown an increasing trend in China in recent decades (Liu et al., 2006). An important criterion for evaluating climate models is their simulation of long-term trends. We select the extreme indices that are reproduced well by the two climate models (PTOT, R95T, CWD, R1, Rx5, FD, and SU) to examine their linear trends.
Figure 3 shows the distribution of the long-term trend for the selected regional extreme climate indices of China from 1961 to 2005. The observed total precipitation shows an increasing tendency from Northwest to Southeast China, especially over the Yangtze River basin and southern China. The BCC-CSM1.1 simulate the increasing trend in western China and the decreasing trend in North China, but produce the opposite trend over the middle and lower reaches of the Yangtze River and in southern China (Fig. 3a2). The BCC-CSM1.1m improve the simulation for southern China, but the trend of the simulation for North China is the opposite to the observations (Fig. 3a3).
Figure 3. Spatial distributions of the long-term trends for extreme climate indices: (a1–a3) PTOT (mm), (b1–b3) R95T (%), (c1–c3) CWD (day), (d1–d3) R1 (day), (e1–e3) Rx5 (mm), (f1–f3) FD (day), and (g1–g3) SU (day) in China during 1961-2005 from (a1–g1) the observational data (CN05.1), (a2–g2) BCC-CSM1.1 (T42), and (a3–g3) BCC-CSM1.1m (T106).
Figures 3b1–b3 display the trends of R95T in the observations and the simulations during 1961–2005 and show that R95T is increasing in most areas of China. Both models are able to simulate the increasing trend in northern China. The BCC-CSM1.1m simulates the increasing trend in central and eastern China more realistically than the BCC-CSM1.1.
The observed trends in the CWD index are small. The BCC-CSM1.1 performs better over Xinjiang Region, the Hexi corridor, and northeastern China than the BCC-CSM1.1m. The BCC-CSM1.1m improves the simulation of the trend in R1 over northeastern China and Xinjiang Region, but is unable to simulate the distribution of the increasing trends over northwestern China and the decreasing trend over eastern China (Figs. 3d2–d3). The BCC-CSM1.1 shows a decreasing trend in Rx5 (Figs. 3e2–e3) over southeastern China and BCC-CSM1.1m improves the simulations over this area. Both models simulate the trends in FD (Figs. 3f2–f3) and SU (Figs. 3g2–g3) well.
Annual mean time series and regression trendlines for extreme climate index anomalies averaged over eastern China during 1961–2005 from the observational data, BCC-CSM1.1, and BCC-CSM1.1m are shown in Fig. 4. Both models are able to reproduce the evolution of the extreme indices over time, especially FD. An increasing trend of the PTOT index is found in the observational data [5.2 mm (10 yr)−1] and in the BCC-CSM1.1m [6.2 mm (10 yr) −1], whereas the trend in the BCC-CSM1.1 is the opposite of the trend in the observational dataset (Fig. 4a). The magnitude of the trends in the observational dataset and the two simulations are not statistically significant for R95T, CWD, R1, and Rx5 (Figs. 4b–4e). The correlation coefficients between the simulations and the observational datasets are low. Both models are able to reproduce the decreasing trend in the FD index [−3.5 day (10 yr)−1] and the increasing trend in the SU index [1.8 day (10 yr)−1], but they underestimate the magnitude of the trend in the FD index (Fig. 4f) and overestimate the magnitude of the trend in the SU index (Fig. 4g). The interannual variation in FD is better simulated by the BCC-CSM1.1 (0.46 day) than the BCC-CSM1.1m (0.33 day).
Figure 4. Annual mean time series of extreme climate index anomalies: (a) PTOT (mm), (b) R95T (%), (c) CWD (day), (d) R1 (day), (e) Rx5 (mm), (f) FD (day), and (g) SU (day) averaged over eastern China (20°–45°N, 110°–120°E) during 1961–2005 from the observational data (CN05.1; black solid line), BCC-CSM1.1 (T42; blue solid line), and BCC-CSM1.1m (T106; red solid line). The dotted lines denote the trend of corresponding extreme index anomaly from observations and simulations.
The simulations of the spatial distribution and the linear trend of extreme indices in China show that the BCC-CSM1.1m is better than the BCC-CSM1.1 at reproducing some of the extreme indices. This increases the credibility of the results of the model simulation. We therefore explored the future projection of extreme climate indices in different periods in China in the 21st century using the BCC-CSM1.1m.
Figure 5 shows the projected changes in future extreme climate indices under the RCP4.5 and RCP8.5 scenarios for 2080–99 relative to 1986–2005. The PTOT index (Figs. 5a1–a2) clearly increases, except for some parts of southwestern China. The increasing magnitude of PTOT is stronger under the RCP8.5 scenario than under the RCP4.5 scenario. R95T (Figs. 5b1–b2) is projected to have a large increase in amplitude; under the RCP8.5 scenario, most parts of China are projected to increase by > 10%, especially northern China. There is a clear increase in magnitude of CWD in Xinjiang and Inner Mongolia, whereas the magnitude of CWD decreases on the Qinghai–Tibetan Plateau and in central and northeastern China. R1 increases over northern China and decreases over southern China in the late 21st century under RCP4.5 (Fig. 5d1), whereas the area in which R1 decreases expands under RCP8.5. Under RCP8.5, R1 over northeastern China increases by 4–8 days, but reduces by 8–12 days over southwestern China. The increasing magnitude of Rx5 (Figs. 5e1–e2) is prominent in eastern China, with the maximum > 50 mm. Under the high-emission scenario (RCP8.5), FD (Figs. 5f1–f2) reduces in western and northern China by 40–50 days, whereas SU (Figs. 5g1–g2) increases in the eastern and southeastern parts of southwestern China by > 50 days.
Figure 5. Changes in climatology (2080–99 minus 1986–2005) of extreme climate indices: (a1–a2) PTOT (mm), (b1–b2) R95T (%), (c1–c2) CWD (day), (d1–d2) R1 (day), (e1–e2) Rx5 (mm), (f1–f2) FD (day), and (g1–g2) SU (day) under (a1-g1) the RCP4.5 and (a2-g2) RCP8.5 emission scenarios in mainland China during the late 21st century projected by the BCC-CSM1.1m.
In summary, the R95T index over northern China increases remarkably in RCP8.5. The Rx5 index increases more dramatically in eastern China in RCP8.5 than in RCP4.5, which indicates there will be more extreme precipitation in the RCP8.5 scenario than in RCP4.5. SU increases more over some parts of southern and northern China in RCP8.5 than in RCP4.5, with more summer days in RCP8.5 than in RCP4.5. FD shows a more significant reduction over western and central China in RCP8.5 than in RCP4.5, which suggests that there will be fewer frost days under the higher emission scenario.
Table 3 lists the future changes in extreme climate events in eastern China simulated by the BCC-CSM1.1m. The PTOT index clearly increases in the middle and late 21st century, although the change of PTOT is not prominent in the early 21st century. The R95T index is projected to increase in the three periods, which is consistent with the conclusion of Yang et al. (2014). In the early 21st century, the CWD index decreases slightly in eastern China under the two emission scenarios. There is no significant change in the CWD index in the middle and late 21st century. The increases in the CWD index in northern China and the decrease in southern China in the late 21st century projected by the BCC-CSM1.1m are similar to the results of Bao et al. (2015), who projected increases in the northern regions and decreases in the southern regions from the future projections of GFDL-ESM2G. R1 decreases over eastern China in the early 21st century under both the RCP4.5 and RCP8.5 scenarios, but increases in the middle of the 21st century. R1 will increase under RCP4.5 in the late 21st century and decrease under RCP8.5. The increase in R1 in northern China under both the RCP4.5 and RCP8.5 scenarios in the late 21st century is consistent with the conclusions of Sun et al. (2016), who used the future projection of the CMIP5 models to determine that northern China will become wetter in the future. The variation in Rx5 is similar to the variation in R95T and will increase over the whole country under both scenarios. The projected pattern of change in Rx5 is similar to the result presented by Li et al. (2018), who used six regional climate models to downscale the global BCC-CSM1.1/HadGEM2-AO (Hadley Centre Global Environmental Model version 2 coupled with atmosphere–ocean configuration) model. FD is projected to decrease and SU is projected to increase in the 21st century. This is similar to the result presented by Yang et al. (2014), who used the output of the CMIP5 models to project future changes in China and found that the FD index will increase under three RCP scenarios and that the SU will increase remarkably under RCP8.5.
Index 2016–35 2046–65 2080–99 RCP4.5 RCP8.5 RCP4.5 RCP8.5 RCP4.5 RCP8.5 PTOT 2.05 −1.55 70.46 86 77.27 97.76 R95T 12.48 12.02 14.87 15.64 14.56 17.3 CWD −0.6 −0.5 0.1 0.2 0.09 −0.02 R1 −3.37 −3.59 2.04 0.93 1.82 −3.25 Rx5 12.33 6.5 17.01 27.96 20.89 36.52 FD −8.91 −9.27 −13.54 −18.09 −16.53 −32.02 SU 12.66 13.72 19.59 28 23.3 46.99
Table 3. Changes in extreme climate indices in China (20°–45°N, 110°–120°E) averaged over three 20-yr periods under the RCP4.5 and RCP8.5 scenarios projected by the BCC-CSM1.1m relative to the annual mean during 1986–2005
This analysis shows that the frequency of extreme precipitation events in eastern China will increase further in the late 21st century as global warming intensifies. The SU index will clearly increase and the FD index will continue to decrease. This will have a strong influence on the social and ecological environment.