1. Introduction
Over the past 100 years since the industrial revolution, climate change has become a growing concern and has had a powerful impact on human daily life [
1,
2]. Agricultural production, personal health, and environmental protection are affected by global warming, which can be attributed to extreme temperature and precipitation events caused by climate change. In the context of the global warming scenario, increasing temperature trends are observed in each season, especially in winter [
3,
4,
5,
6].
Indices defined by the Expert Team on Climate Change Detection and Indices (ETCCDI) have been utilised to evaluate regional- and national-scale temperature and precipitation extremes in many studies [
7]. The Coupled Model Intercomparison Project Phase 5 (CMIP5) projected future extreme weather events for the Chinese subregion based on six ETCCDI, and the results showed that cold nights (TN10P) and frost days (FD0) decreased significantly in the 21st century, with a faster decline in the late 21st century than the middle of the 21st century. In contrast, warm days (TX90p), tropical days (TD30), and daily maximum temperature (TXx) increased under both representative concentration pathway (RCP) 4.5 and RCP 8.5 scenarios [
8]. Changes in winter extreme low-temperature events in northern China under different future warming scenarios were predicted based on a simple minimum complexity earth simulator model [
9]. This showed that the intensity of low-temperature events will decrease in the future because of higher velocity of TNn, especially in Northeast China (NE), where TNx increased by 0.1 °C/0.6 °C compared to 1.5 °C/2.0 °C for TXx. The temperature prediction results based on CMIP5 by You et al. showed that under the RCP8.5 scenario, the mean temperature, maximum and minimum temperature had highest warming rate compared with the RCP4.5 and RCP2.6 scenarios. Northeast China and the Qinghai-Tibet Plateau are more vulnerable to climate change in future emission scenarios [
10]. A statistical model of extreme climates events based on generalised extreme value distribution was applied to CMIP5 predictions. The extreme high temperature in China was expected to increase by about 1.66–4.92 °C. The variation of the warm extreme value was more sensitive to the radiative forcing concentration than the cold extreme value [
11]. Chen et al. proposed that the credibility of the model can be improved by using a multi-model ensemble. The overall performance is better than a single model when the mean and median of the multiple models are considered. Model assessments vary widely across indices based on relative errors in climatology [
12]. Validation with observations and reanalysis showed that compared with CMIP5, the CMIP6 model had obvious advantages in terms of spatial patterns and temporal changes in extreme temperature indices [
1]. Previous studies have shown that winter temperature is a vital limiting factor for agricultural production and vegetation growth [
13]. China is one of the world’s largest agricultural countries. Chen et al. [
14] investigated the effects of climate change and extreme climate on maize and rice growth based on climate variable outputs from 17 general environment models (GCMs) in the CMIP5 dataset in the Yangtze River Basin. Their results showed that the extreme climate index was closely correlated with maize and rice yields, particularly on days above the temperature threshold. Thus, studying the spatiotemporal variation of climate in winter is beneficial for agricultural pattern modification and phenological adaptation. However, little research has been conducted on extreme winter temperatures and the overall warm bias.
Future climate can be predicted using global climate models made up of equations describing the interaction of energy and matter between different parts of the ocean, atmosphere, and land [
15,
16]. The Coupled Model Intercomparison Project Phase 6 (CMIP6) announced in 2016 is the latest GCM dataset, which is reliable for projecting the future climate. Compared with previous versions, CMIP6 has a higher resolution and a higher correlation with high-resolution daily observations of the Chinese mainland from 1961 to 2005 [
17]. However, the resolution of CMIP6 at a scale of hundreds of kilometres is too coarse for studying extreme climate events and climate change in small regions [
18,
19]. There are three commonly used downscaling methods for converting crudely generated data to site scale depending on research needs: scaling, dynamical downscaling, and statistical downscaling. Compared to the statistical downscaling method, the scaling method cannot explain multiple climate variabilities, whereas the dynamic downscaling method cannot efficiently and rapidly calculate large areas [
20,
21,
22]. Because of its high accuracy and low cost, the statistical downscaling method is often utilised to predict future climate with high accuracy and resolution. NWAI-WG, developed by Liu and Zuo [
23] is a widely applied statistical downscaling method that uses daily observational climate data series to improve accuracy instead of circulation data. In terms of application, Wang et al. [
24] analysed the extreme weather event indices of the wheat growing belt in southeast Australia under different RCPs through multi-model ensemble data derived from the NWAI-WG downscaled CMIP5 data. They discovered an upward trend for warm days, warm nights, and hot days and a downward trend for frost days and cold nights. Xiao et al. [
25] calculated future temperature extremes in the region of the Han River basin using the NWAI-WG downscaled CMIP5 multi-model ensembles. The results showed that the variation range of the minimum temperature extreme value was greater than that of the maximum temperature extreme value, and that the mean value was asymmetrical to the variation in daytime and night-time extreme values. Tang and Liu [
26] assessed the historical and future climate stability of summer maize using the NWAI-WG downscaled CMIP5 data from the period 1996–2100. Li et al. [
27] analysed historical extreme precipitation events based on NWAI-WG statistical downscaling data and estimated extreme precipitation events in future scenarios.
Although research on site-scale extreme temperature events during winter on the Chinese mainland could be helpful in coping with winter climate change, few studies have been conducted on this. This study provides an initial overview of a warm winter on the Chinese mainland through multi-model ensembles under SSP245 and SSP585 scenarios that represent moderate greenhouse gas emissions with strict regulation and high greenhouse gas emissions with low binding management, respectively. Monthly CMIP6 grid data obtained from WRCP were downscaled to daily site-scale data using the NWAI-WG method combined with site observation data. Future winter extreme weather events and overall seasonal warm/cold biases were predicted using the methods announced by the ETCCDI and NCC based on the multi-model ensemble mean from CMIP6. In this study, the temporal and spatial trends in extreme weather event frequency and intensity under different SSP scenarios were calculated.
4. Discussion
In summary, the accuracies of 26 GCMs were evaluated. The result showed that the GCM historical data have high goodness of fit with the observation data, with the Pearson product-moment correlation coefficient and RMSE ranging from 0.984 to 0.989 and 4.581 to 5.509 for maximum temperature and 0.993 to 0.995 and 10.352 to 11.906 for minimum temperature, respectively. The uncertainty of CMIP6 data is due to three factors: internal variation, model uncertainty, and scenario uncertainty. Internal uncertainties of maximum and minimum temperatures showed a consistent decreasing trend from approximately 70% in 2010 to 10% in 2100. Scenario uncertainty, which is the dominant uncertainty after the 2060s, showed a rapidly increasing trend during the study period. Model uncertainty decreased slightly but was observed to be greatest until the 2060s (
Figure 13).
It was found that maximum and minimum temperatures increased at different rates under a warming scenario. Average minimum temperature of the entire Chinese mainland increased by 0.298 °C/10a slower than the maximum temperature under the SSP245 scenario compared to 0.493 °C/10a under the SSP585 scenario. According to this study, NE, IM, and NW showed higher minimum temperature increasing rates, but the maximum temperature increased faster in SC and NE than in other subregions (
Figure 14). Furthermore, the variation coefficient reveals that the temperature difference between the south and north is smaller. We calculated the relationship between the warming rates for maximum and minimum temperatures and the altitudes of the observation stations under different conditions. As shown in
Table 7, there was a significant, positive correlation between the increase rate of maximum temperature and elevation in SW, a certain correlation in NC and SC, a negative correlation in IM, and a weak correlation in other regions. The increasing rate of minimum temperature is positively correlated with elevation in NE and negatively correlated with elevation in NW, which may be due to the increase in short-wave radiation sources in NW because of the distribution of deserts in basins in this region [
40]. Due to the lack of high-altitude meteorological stations over the QT Plateau, elevation-dependent warming cannot be observed.
Frequent and intense ETIs show a significant response to future warming. The intensity and frequency of the extreme cold index decreased more in NW, NC, and the QT Plateau. The extreme hot index in winter increased significantly under both warming scenarios. TN10p decreased by −0.651 and −0.761 days, TNn increased by 0.149 °C and 0.215 °C, TX90p increased by 0.153 and 0.297 days, and TXx increased by 0.259 °C and 0.489 °C under the SSP245 and SSP585 warming scenarios, respectively. A rapid and consistent decreasing trend of CSDI is shown in all subregions on the Chinese mainland under both SSP245 and SSP585 scenarios. The WSDI shows a consistently increasing trend under the SSP245 scenario but shows saltation acceleration after the 2050s under the SSP585 scenario. Maximum temperature is more sensitive to global warming, and the reduction in radiative forcing helps mitigate the warming trend of maximum temperature.
Winter average temperature bias was estimated using the warm winter index announced by the CMA. During the period 2000–2030, warm winter frequency shows a rapid increase under both SSP245 and SSP585 scenarios, but a decrease in warm winter frequency is shown after the 2030s under the SSP245 scenarios. Cold winter frequency decreases very rapidly and will remain at a low level in the future. These results reveal that warm winter events can be prevented under the SSP245 pathway. SW, QT, and IM are areas less affected by warm winters under the SSP245 scenario in the future. Warmer winters are more frequent in the eastern seaboard and arid north-western regions, which may be primarily attributed to the response of the East Asian winter monsoon to global warming [
41,
42]. Population exposure shows a downward trend in both scenarios. While the population declined faster in the ssp585 scenario, the population exposure declined more in the SSP245 scenario. This is mainly because in the SSP585 scenario, although the population shows a decreasing trend, the overall warm winter frequency and impact range are wider, while in the SSP245 scenario, the population number decreases slightly, but the overall number and scope of warm winters are significantly weakened, so the population exposure to warm winter events has a relatively large reduction. Therefore, we should strictly control carbon emissions and promote green production and lifestyle, so that the greenhouse gas emission path is controlled in the SSP245 scenario, which can avoid 25.87% of the population exposure risk in warm winters.
Extreme winter temperatures have many negative and positive impacts on agricultural production. SC is the main rice-growing regions in China. The majority of regions with maximum temperature increase are located in these regions, with the warming trend anticipated to lead to longer growing seasons and more accumulated temperature in the future. However, NE is the main wheat and rice farming region, which shows a significant increase in minimum temperature that will have a negative influence on spring phenology in future [
43,
44,
45]. In future studies, we will pay more attention to the influence of winter atmospheric circulation on warming as well as the influence of winter warming on crop production and human health.
5. Conclusions
In this study, we used 26 GCM datasets from the entire Chinese mainland derived from the CMIP6 datasets, which include maximum and minimum temperatures during the period of 1961–2100 under the SSP245 and SSP585 warming scenarios. The NWAI-WG method, which is a statistical downscaling method combining observation data, was utilised to downscale the gridded data to the station scale to enhance precision. Nine ETIs were estimated from 2010 to 2100 based on the multi-model ensemble-predicted data after checking for robustness and uncertainty. The temporal and spatial variations were calculated using the MK method and the variable coefficient, respectively. The frequency of future warm winter events was calculated by the warm winter index proposed by CMA. The population exposure to warm winter events was calculated by combining the SSP population forecast data.
Under the background of global warming, the national winter temperature shows a significant growth trend. Compared with other seasons and the whole, winter also showed asymmetric temperature rise, but the heating rate of the maximum temperature in winter was slightly higher than that of the minimum temperature. The coefficients of variation calculated based on the maximum temperature, minimum temperature, and average temperature all showed a significant decreasing trend, indicating that the region with low temperature had a faster heating rate and the north–south temperature difference decreased.
The extreme climate indices calculated according to the ETCCDI indicator showed different trends, with a significant decrease in cold-related extreme climate events and a significant increase in warm-related extreme climate events. The warm winter conditions showed a large difference between the SSP245 and SSP585 conditions. Under the SSP245 scenario, by the end of this century, the frequency of warm winters will decrease and remain at a low level. The Southwest, Qinghai-Tibet and Inner Mongolia regions will decrease from 7–9 times/10a in the early 21st century to less than 4 times/10a, while in the SSP585 scenario, the frequency of warm winter events will reach the highest, reaching 8–10 times/10a.
The overall population of the Chinese mainland shows a downward trend. The overall population change trends in the country are −26.32 million and −61.42 million/10a, respectively. The population decrease trend in NE, NC, SC, and SW is relatively high. In the future scenario, the population exposure to warm winters on the Chinese mainland shows a decline. By the 2090s, the population exposure to warm winter events in the SSP245 scenario will be 6.937 billion person-times, while that in the SSP585 scenario will be 9.358 billion person-times. The SSP245 scenario has a much lower exposure risk than the SSP585 scenario due to the decreasing frequency of both population and warm winter events.
The intensity and frequency of future extreme weather events show an increasing trend under the two warming scenarios. Warm winter events decreased under SSP245 and increased under SSP585. Population exposure showed a downward trend. A consistent warming trend was observed on the Chinese mainland under all SSPs in the 21st century, and stricter emission reduction policies can help reduce the negative impact of global warming on winter temperatures.