3.1 Choosing the optimal fingerprinting strategy by perfect model analysis
We employed the perfect model method to analyze four possible fingerprint combination strategies: GHG + AER + NAT, ALL + GHG + NAT, ALL + AER + NAT and ALL + GHG + AER. The root mean square errors (RMSE) of the regression coefficient for the external forcings of GHG and AER in mainland China, Eastern China, and Western China under various combinations are displayed in Fig. 1. As expected, the estimated regression coefficient for the same external forcing differ across different combination strategies, even within the same climate model, resulting in several uncertainties in the prediction of SAT warming attribution. Our study found the inaccuracies in mainland China are less than those in the EC and WC, indicating that smaller region have the greater uncertainty in detection and attribution (Bindoff et al. 2014). Upon comparing the RMSE for the external forcings, it was observed that the average error of regression coefficient for the GHG forcing is smaller than that for AER, particularly in the annual and summer averages, which are both less than 0.2. Moreover, even under the same forcing, the error varies by season, with the error being substantially larger in winter than in annual average, particularly in summer.
The simulation results of the four combination strategies are elaborated upon below. As mentioned earlier, a lower the RMSE indicates a more accurate best estimate of the fingerprint regression coefficient. Since the inception of the DAMIP simulation, GHG + AER + NAT (the yellow bars) has been the most commonly used strategy combination for detection and attribution. However, as shown in Fig. 1, this combination may produce errors to some extent, particularly when estimating the regression coefficients for external forcings in winter, where the RMSE is greater than the other three combinations except AER forcing in WC. The errors could be caused by the fact that the GHG + AER + NAT combination does not account for stratospheric ozone depletion and land use or cover change (Li.C et al. 2021). In contrast, the other three combination strategies containing ALL forcing have smaller errors, with the ability of the ALL + GHG + NAT and ALL + AER + NAT combination (the orange and red bars in Fig. 1) having comparable ability to estimate the regression coefficients of GHG and AER in winter, with RMSE smaller than ALL + GHG + AER (the brown bars). However, the ALL + GHG + NAT combination would result in the highest error in predicting regression coefficients for AER in annual and summer average, which was the largest of the four combinations.
In general, the combination strategy of ALL + AER + NAT performs best in perfect model analysis, which is consistent with the Ribes’s conclusion in idealized Monte Carlo simulation (Ribes et al. 2015), as well as the results of Li's perfect model experiments using DAMIP for detection the attribution of global SAT (Li.C et al. 2021). Thus, we apply the ALL + AER + NAT combination strategy to the SAT observation to estimate the detection and attribution of anthropogenic factors to warming, and investigate the regression variables and warming contribution of GHG, AER, and NAT, respectively.
3.2 Spatial variation of annual and seasonal LSAT trends over China
Figure 2 depicts the trend of annual average LSAT in mainland China from 1951 to 2020. The observation (Fig. 2a) shows that from the middle of the twentieth century, the annual average SAT in mainland China has displayed an obvious warming tendency, with western China being greater than eastern China and northern China being greater than nouthern China. The northwest and northeast of China have the most noticeable warming, with a warming trend of almost 0.4°C/10a, while the southeast coast has a lesser warming trend, less than 0.2°C/10a. The warming trend of mainland China may be more accurately simulated by using the historical forcing of CMIP6 (Fig. 2b). This warming trend demonstrates the distribution features of China's west and northeast being bigger than those of the southeast. Nonetheless, the total warming amplitude is marginally less than the observation by roughly 0.1 ℃/10a; in contrast, the GHG forcing simulation amplitude (Fig. 2c) is superior to the historical forcing. It is observed that an abnormal large value warming trend develops in southwest China in both the historical simulation and the greenhouse gas forcing simulation. The errors in the simulations could be attributed to the altitude and the dearth of stations on the Tibetan Plateau. The cooling effect is mostly depicted in the AER forcing simulation (Fig. 2d), with the cooling trend in southeast China being the most noticeable at -0.1 to -0.2°C/10a. Furthermore, the temperature trend under the NAT forcing is negligible, essentially 0.
The observation (OBS) indicates that the warming trend in winter is the most pronounced and widespread, with warming reaching 0.3 ~ 0.4°C/10a in most of China's west, northeast, and north; the strongest warming trend in spring occurs in the region north of 40°N, with most of it exceeding 0.4°C/10a; and the warming trend in eastern China is greater than that in western China, based on the spatial distribution of the warming change on the seasonal scale (Fig. S2 in Supporting Information S1). The warming trend is smaller in the summer and autumn, with the summer warming trend being the lowest, with most of China rising by less than 0.2°C/10a; the autumn warming trend is marginally greater than the summer warming trend, and the spatial distribution indicates that the northwest of China is larger than the east. The warming trend in mainland China can be more accurately depicted by the ALL simulation, but the trend's intensity is different since the model understates the warming trend in the winter and spring, particularly in the spring, and it is unable to replicate the notable warming trend in northern China. The simulation overstates the warming trend slightly throughout the summer and fall seasons, with notable overestimations in the fall on the Tibet Plateau and northeastern China. There is a clearer warming trend from GHG forcing than from ALL forcing, which can better simulate the spatial distribution of winter warming, according to the three types of only-forcing simulations. AER forcing, on the other hand, has a negative warming trend for all seasons, reflecting its cooling effect, with little variation between seasons and the largest cooling trend in the southeast of China. Similarly, there is no clear trend change under NAT forcing.
3.3 Temporal change of annual and seasonal LSAT over China
Figure 3 displays the non-overlapping 5-year moving average of the CMIP6 simulation and SAT anomalies from 1951 to 2020. Based on the observed annual average SAT anomalies (Fig. 3, the black dotted line of the first row), the temperature anomalies in mainland China, as well as in eastern and western China, changed relatively steadily from 1951 to 1980. From 1990 to 2000, there was a clear upward trend, but after the turn of the twenty-first century, the growth slowed down and there was a warming break around 2010. These findings are in line with those of Li.Q et al. (2015) and Xie et al. (2017). When comparing the eastern and western areas of China, the western region's warming tendency is slightly greater than that of the eastern region. Winter and spring (the black dotted line in the second and third rows of Fig. 3) have seen the most warming in the last 20 years. The winter SAT has the following characteristics: First, the inter-annual variation of winter temperature anomalies is large and fluctuating; second, there was a sharp rise in temperature in the middle and late 1980s, and then the warming trend slowed; third, there was a significant interruption of warming, with temperature anomalies decreasing significantly around 2010. The valley value is more prominent in eastern China than in western China, which may be related to inter-annual changes in internal variability such as a weakening of solar radiation, PDO, ENSO, and a weakening of the westerly belt, which leads to an increase in the blocking situation (Chen et al. 2013, Ding et al. 2014, Wang and Yang 2014, Folini and Wild 2015, Xie et al. 2017). The temperature in spring is comparable to that in winter. The most obvious period of regional warming in China is from the late 1990s to the early 21st century, lagging behind the period of abrupt warming in winter, and it also showed a low value around 2010, but the temperature decrease is not as large as that in winter, and there are differences between east and west of China. The image illustrates that temperature anomalies in eastern China peaked around 2000, then varied and fell before beginning to rebound after 2010. However, this feature is not visible in western China, which has reached double peaks in 2005 and 2016. Similarly, temperature anomalies in summer and autumn (the black dotted line in the fourth and fifth rows of Fig. 3) increased in the late twentieth century but slowed after entering the twenty-first century. The warming rate in summer was the slowest of the four seasons, and it fell significantly after 2010; temperature anomalies in autumn peaked about 2008 and thereafter fluctuated.
We explore the TAS simulation of CMIP6 in further detail. The chart shows that the ALL-SSP5-8.5 simulation (orange solid line) underestimates the warming in mainland and eastern China between 1990 and 2010, yet it can accurately reproduce the annual mean temperature anomalies in western China. Summer and autumn simulations are clearly superior to winter and spring simulations, particularly in summer. The warming trend prior to 2010 is essentially in line with observation, but the warming slowdown and downward trend following 2010 are poorly represented; winter simulations have the largest error, significantly underestimating the warming trend from the 1990s to the early 2000s. The obvious warming of observation started in the mid-1980s, while the simulated warming began to reflect this warming in the late 1990s. ALL-SSP5-8.5 also underestimated the warming trend in spring, especially from the late 1990s to the early 21st century, when observed warming in eastern China exceeded the upper bound of the model ensemble (orange shadow), but the simulations in mainland China and western China were basically within the shade of orange, indicating that the model, although somewhat flawed, is still within the margin of error. The AER simulation (blue solid line and shadow) showed a fluctuating cooling from 1951s to 2000s, after which the aerosol cooling became insignificant; while the NAT simulation (green solid line and shadow) has no obvious trend in the whole simulation period, and the temperature anomalies is basically 0 ℃.
3.4 Detection and attribution of SAT warming in China
The results of the detection and attribution of the three signals (GHG, AER, and NAT) of SAT change in China are displayed in the left half of Fig. 4, which assists in identifying the external forcing responsible for SAT warming. When the external forcing's regression coefficient β is larger than 0 within a confidence interval of 5–95%, we regard it as detected; if its interval of 5–95% is larger than 0 and includes 1, the observed SAT change can be ascribed to this external forcing. As the image illustrates, GHG (red lines) is detected in both the annual and seasonal averages, with the exception of winter in eastern China. In addition, the regression coefficient in western China of summer does not include 1. This suggests that more warming may be caused by the forcing of greenhouse gases. In the annual average of the three regions, as well as in mainland China and western China during the winter, the best estimate of the GHG regression coefficient is close to 1, indicating that the model can accurately simulate the contribution of GHG forcing to SAT warming in China; however, in the spring, the best estimate of the GHG regression coefficient is greater than 1, indicating that the model underestimates the warming of LSAT; in contrast, the best estimate of the GHG regression coefficient is less than 1, indicating that the model overestimates the warming contribution of GHG simulation to SAT. With regard to AER forcing, the signal is only detected in China and western China during the summer, and the regression coefficient's best estimate is less than 1, suggesting that the model exaggerates the influence of AER external forcing on SAT change. Since the NAT regression coefficient is 0 in all seasons and regions, the NAT external forcing has little effect on the warming of SAT. The attribution has passed the residual consistency test, indicating that the CMIP6 climate model can reasonably reproduce the observed SAT internal variability, and the model fitting effect is good.
According to the analysis above, China's SAT has warmed significantly since the 21st century compared with the middle of the 20th century. We calculated the annual average and four seasonal average SAT warming trends from 1951 to 2020, the results are shown in the right half of Fig. 4. The SAT warming trend (black bar) shows that China's annual average warming trend is 0.24°C/10a, with 0.23°C/10a and 0.26°C/10a for eastern and western China, respectively. This suggests that during the past 70 years, the warming trend in western China has been slightly higher than that in the eastern region. The warming trend in winter and spring is the greatest of the four seasons, with a warming trend of 0.28°C/10a, but there are variances between eastern and western China. The warming trend in wintertime is higher in western China, reaching 0.31°C/10a, while in eastern China, it is 0.27°C/10a. Eastern China is experiencing a greater warming trend in the springtime than western China, with temperatures rising by 0.31°C/10a and 0.25°C/10a, respectively. Between 0.15°C/10a and 0.18°C/10a, the summer warming trend is the least in the last 70 years, with a little stronger trend in western China than in eastern China. The autumn warming trend in China is 0.21°C/10a, with values in western and eastern China being 0.24°C/10a and 0.18°C/10a, respectively, in accordance with the findings in Fig. 3.
The contribution of different external forcing to the SAT warming trend can be obtained by multiplying the best estimation of each external forcing regression coefficient by the corresponding linear trend. The color bars in the right side of Fig. 4 show that the GHG external forcing (red bars) is the main factor responsible of SAT warming. With the exception of China's winter average SAT and the annual and winter average SAT in eastern China, the warming trends of greenhouse gases to the SAT in other seasons and regions are greater than the warming trends that are observed; these warming trends contribute to the respective annual and winter average warming trends of 0.21 ~ 0.33°C/10a and 0.21 ~ 0.35°C/10a (see Table 1 for details). According to the previous discussion, the average SAT shows the strongest warming tendency in the spring, with the exception of the winter, and the most noticeable contribution to this warming trend comes from greenhouse gas forcing, which reaches 0.32 ~ 0.36°C/10a. The warming contribution of greenhouse gases is nearly the same in summer and autumn, at 0.24 ~ 0.26°C/10a. AER external forcing (blue bars) primarily generates a cooling impact, which offsets some of the warming effect generated by greenhouse gases. In instance, aerosol signals are identified robustly in mainland China and western China in summer, with a cooling trend of − 0.01°C/10a, which explains to some extent why the rising trend of LSAT in summer is the least of the four seasons, but the cooling effect is limited. The contribution of the warming trend of NAT external forcing (green bars) is almost zero, implying that the warming of SAT due to natural forcing is insignificant.
Table 1
Contributions of GHG, AER and NAT to the warming trend of SAT and their 5–95% uncertainty ranges
| | China | East China | West China |
| GHG | 0.27(0.07 ~ 0.47) | 0.21(-0.02 ~ 0.44) | 0.33(0.13 ~ 0.52) |
Annual | AER | 0.01(-0.08 ~ 0.10) | 0.01(-0.06 ~ 0.09) | 0.00(-0.11 ~ 0.12) |
| NAT | 0.00(-0.02 ~ 0.02) | 0.00(0.00 ~ 0.00) | -0.02(-0.07 ~ 0.02) |
| GHG | 0.27(0.07 ~ 0.48) | 0.21(-0.02 ~ 0.44) | 0.35(0.14 ~ 0.55) |
DJF | AER | 0.02(-0.15 ~ 0.18) | 0.02(-0.12 ~ 0.17) | 0.00(-0.20 ~ 0.21) |
| NAT | 0.00(-0.01 ~ 0.01) | 0.00(-0.02 ~ 0.02) | -0.02(-0.07 ~ 0.03) |
| GHG | 0.35(0.22 ~ 0.48) | 0.36(0.22 ~ 0.48) | 0.32(0.22 ~ 0.48) |
MAM | AER | -0.05(-0.17 ~ 0.07) | -0.04(-0.17 ~ 0.07) | -0.06(-0.17 ~ 0.07) |
| NAT | -0.03(-0.08 ~ 0.01) | -0.01(-0.08 ~ 0.01) | -0.05(-0.08 ~ 0.01) |
| GHG | 0.26(0.19 ~ 0.33) | 0.26(0.19 ~ 0.34) | 0.24(0.15 ~ 0.33) |
JJA | AER | -0.01(-0.02 ~ 0.00) | -0.01(-0.02 ~ 0.00) | -0.01(-0.02 ~ 0.00) |
| NAT | 0.00(-0.01 ~ 0.00) | 0.00(-0.01 ~ 0.00) | 0.00(-0.02 ~ 0.02) |
| GHG | 0.24(0.15 ~ 0.33) | 0.24(0.14 ~ 0.35) | 0.24(0.14 ~ 0.33) |
SON | AER | 0.00(-0.03 ~ 0.02) | 0.00(-0.02 ~ 0.02) | -0.01(-0.05 ~ 0.04) |
| NAT | 0.00(-0.01 ~ 0.00) | 0.00(-0.00 ~ 0.00) | -0.02(-0.06 ~ 0.01) |