Evaluation and Projection of Extreme High Temperature Indices in Southwestern China Using NEX-GDDP-CMIP6

1.   Introduction

Global warming exacerbates extreme weather events such as heat waves, droughts, and floods, which severely impact human health, socioeconomic systems, and natural ecosystems. Since the 1950s, the number of warm days and the highest daytime temperatures have been increasing worldwide, and heat waves have become more intense, frequent, and longer lasting. With further global warming, extreme heat events are projected to increase in both frequency and intensity (IPCC, 2021; Zhou and Qian, 2021). China is significantly affected by global climate change, with a warming rate markedly higher than the global average during the same period (China Meteorological Administration Climate Change Center, 2022). The frequency and intensity of extreme events, such as heat waves, have significantly increased in China, causing great and lasting socioeconomic losses (Hu et al., 2017). Southwestern China (including Sichuan, Yunnan, Guizhou, and Chongqing), which is close to the Qinghai–Tibet Plateau, is one of the most geographically complex areas in the country (Liu et al., 2023), and one of the most sensitive and vulnerable areas to climate change in China and the world (Lan et al., 2021). Southwestern China is affected by multiple climate systems, including the South Asian, East Asian, and plateau monsoons (Li and Zhang, 2014; Li et al., 2022), and is one of the areas frequently affected by high temperatures and heat waves disasters in China (Huang et al., 2020). For example, in the summer of 2006, rare heat waves and extremely severe droughts occurred in Chongqing, Sichuan, and other areas (Zou and Gao, 2007; Li et al., 2011). In the summer of 2013, southwestern China was hit by a widespread heat wave (Tang et al., 2014). In the summer of 2022, this area experienced the longest and strongest heat wave in history, with serious adverse effects such as agricultural production reduction, river drying, energy supply shortage, and forest fires (Sun et al., 2022; Yuan et al., 2023). Therefore, accurately simulating and projecting extreme high temperature (EHT) events in southwestern China has significant scientific and practical value for disaster prevention and mitigation as well as climate change policymaking.

The Coupled Model Intercomparison Project Phase 6 (CMIP6) global climate models (GCMs) are the primary tool used to simulate historical and project future climate change (Li et al., 2019; Li C. et al., 2021). Compared with the CMIP5 models, the CMIP6 models have significantly higher resolution, more optimized parametric schemes, and more complex model physics (Eyring et al., 2016; Di Luca et al., 2020). Moreover, their simulation reults are much improved (Zhou et al., 2019; Di Luca et al., 2020) and they consider more reasonable future change scenarios (O’Neill et al., 2016; Zhang et al., 2019), namely the Shared Socioeconomic Pathway (SSP) scenarios. The CMIP6 model data have been widely used in the assessment and future projection of regional extreme climate events in China (Luo et al., 2020; Zhu et al., 2020; Wang et al., 2021; Xu et al., 2021; Zhang et al., 2021; Hu and Sun, 2022; Xie et al., 2022; Chen et al., 2023; Wei et al., 2023). For instance, the simulation of extreme temperature indices by Hu and Sun (2022) showed that China experienced continuous warming from 1951 to 2018, when the number of EHT events in most regions increased. Guo et al. (2022) estimated that the central part of southwestern China is one of the four risk centers for high temperature days in China under the SSP2-4.5 scenario. These results, along with those of many other studies, indicate a significant increase in EHT events in most regions of China, with noticeable regional differences. However, because of the relatively low output resolution, certain systematic biases, and variations in simulation performance by region (Jiang and Chen, 2021), the CMIP6 models cannot fully satisfy the requirements of regional climate change research.

On the regional scale, especially in complex terrain and climate-sensitive areas, downscaled data can provide high-resolution regional climate change information (Gao et al., 2012; Guo and Wang, 2016; Gao et al., 2017), which can reduce the simulation error of regional EHT to some extent, thereby improving the reliability of the projection results (Hawkins and Sutton, 2009; Zhou et al., 2018; Navarro-Racines et al., 2020; Park et al., 2023). The NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) is a downscaling and deviation correction effort that uses the bias-correction spatial disaggregation (BCSD) method to provide bias-corrected, high-resolution, and seamless daily climate variable simulation data from both CMIP5 and CMIP6 (Thrasher et al., 2012, 2022). Several studies have used the NEX-GDDP dataset to conduct climate change simulation studies over the Tibetan Plateau, Yangtze River valley, and the whole China. Overall, compared to CMIP5 GCMs simulations, NEX-GDDP has a better performance (Bao and Wen, 2017; Chen et al., 2017; Zhou et al., 2018; Li et al., 2019; Wu et al., 2020; Chen et al., 2021; Wang et al., 2022). As reported by Bao and Wen (2017), NEX-GDDP significantly improved simulations in areas near steep terrains with sharp slopes, such as the edges of the Tibetan Plateau. The Sichuan basin in eastern Tibet could be one of the key improved regions of the NEX-GDDP, as it is a populous area with a distinct climate that is conventionally poorly reproduced by GCMs owing to coarse terrain representation. For the NEX-GDDP-CMIP6, Park et al. (2023) analyzed the impact of a 2°C global temperature increase on land climate using NEX-GDDP-CMIP6 data and they encourage more active utilization of this dataset on the regional scale. Wu et al. (2023) evaluated the NEX-GDDP-CMIP6 in terms of its simulation performance in capturing the droughts in China. Xu et al. (2023) detected recent and future changes in concurrent temperature and precipitation extremes based on the NEX-GDDP-CMIP6 dataset in the Asian monsoon region. However, because of the short release time of the NEX-GDDP-CMIP6 data, relatively few simulation studies have used the data.

In this study, we employ the newly released high-resolution statistical downscaling dataset NEX-GDDP-CMIP6 to investigate the evolution of EHT events during historical and future periods in southwestern China. First, we calculate five EHT indices from 1961 to 2014 using the daily maximum temperature data from the CN05.1 and NEX-GDDP-CMIP6 datasets. Then, using the results of CN05.1 as a reference, we evaluate the performance of 26 models in NEX-GDDP-CMIP6 in capturing the changing trends and climatological means of the five extreme temperature indices using the Taylor diagram and Taylor skill score, after which we use the evaluation results to select the optimal models. Finally, based on the optimal models, we conduct EHT simulations for the early (2021–2040), middle (2041–2060), and late (2081–2100) periods of the 21st century for three scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) and analyzed their spatial and temporal characteristics. The primary goal of this study is to provide a scientific basis for assessing the impacts of climate change on the agriculture, water resources, and social economy in southwestern China, as well as to facilitate actionable adaptation and mitigation plans.

2.   Data and methods

2.1   Data

2.1.1   Observational data

The CN05.1 gridded dataset (Xu et al., 2009; Wu and Gao, 2013; Wu et al., 2017) in China from 1961 to 2014 with a resolution of 0.25° × 0.25° is used to assess the performance of the model simulations. This dataset was developed from the records of 2416 meteorological stations in China using the “anomaly approach” (New et al., 2000). It is one of the most accurate datasets for regional gridded near-ground meteorological fields in China (Zhou et al., 2016; Wang and Wang, 2017) and has been widely used to analyze observed climate characteristics and assess global and regional model performance (Guo and Wang, 2016; Guo et al., 2017; Zhu et al., 2020).

2.1.2   Model data

The high-resolution statistically downscaled dataset NEX-GDDP-CMIP6 (Thrasher et al., 2022) is used to evaluate and project EHT events. This dataset includes historical daily maximum temperature over 1961–2014 and future projected data under three scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) from 26 models (see https://wcrp-cmip.github.io/CMIP6_CVs/docs/CMIP6_source_id.html for model details) (Table 1). The data have a 1-day temporal resolution and a 0.25° × 0.25° spatial resolution and can be obtained from https://www.nccs.nasa.gov/services/data-collections/land-based-products/nex-gddp-cmip6.

Table  1.  List of 26 global climate models (GCMs) used in this study

Download as CSV

No.	Model name	Country/region	Institution
1	ACCESS-CM2	Australia	Commonwealth Scientific and Industrial Research Organisation, Australian Research Council of Excellence for Climate System Science
2	ACCESS-ESM1-5	Australia	Commonwealth Scientific and Industrial Research Organisation
3	BCC-CSM2-MR	China	Beijing Climate Centre, China Meteorological Administration
4	CanESM5	Canada	Canadian Centre for Climate Modelling and Analysis
5	CMCC-CM2-SR5	Italy	Fondazione Centro Euro-Mediterraneo sui Cambiamenti Climatici
6	CMCC-ESM2	Italy	Fondazione Centro Euro-Mediterraneo sui Cambiamenti Climatici
7	CNRM-CM6-1	France	Centre National de Recherches Météorologiques, Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique
8	CNRM-ESM2-1	France	Centre National de Recherches Météorologiques, Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique
9	EC-Earth3	Europe	EC-Earth Consortium
10	EC-Earth3-Veg-LR	Europe	EC-Earth Consortium
11	GFDL-CM4	USA	NOAA Geophysical Fluid Dynamics Laboratory
12	GFDL-ESM4	USA	NOAA Geophysical Fluid Dynamics Laboratory
13	GISS-E2-1-G	USA	NASA Goddard Institute for Space Studies
14	HadGEM3-GC31-LL	UK	Met Office Hadley Centre
15	INM-CM4-8	Russia	Institute for Numerical Mathematics
16	INM-CM5-0	Russia	Institute for Numerical Mathematics
17	IPSL-CM6A-LR	France	Institute Pierre-Simon Laplace
18	KACE-1-0-G	Korea	National Institute of Meteorological Sciences, Korea Meteorological Administration
19	MIROC6	Japan	Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, RIKEN Center for Computational Science
20	MIROC-ES2L	Japan	Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), National Institute for Environmental Studies, RIKEN Center for Computational Science
21	MPI-ESM1-2-LR	Germany	Max Planck Institute for Meteorology
22	MRI-ESM2-0	Japan	Meteorological Research Institute
23	NESM3	China	Nanjing University of Information Science and Technology
24	NorESM2-LM	Norway	Norwegian Climate Centre
25	TaiESM1	Taiwan, China	Research Center for Environmental Changes, Academia Sinica
26	UKESM1-0-LL	UK	Met Office Hadley Centre
Note: SSP1-2.6 scenario estimates were not available for CMCC-CM2-SR5 or GFDL-CM4.

 | Show Table

DownLoad: CSV

2.2   Methods

2.2.1   Extreme high temperature indices

In this study, the following five indices were selected to investigate the changes in EHT events in southwestern China (Table 2): annual maximum value of daily maximum temperature (TXX), high temperature days (T35), warm days (TX90P), heat wave frequency (HWF), and heat wave days (HWD). TXX is an extreme value indicator, T35 is an absolute threshold indicator, and TX90P, HWF, and HWD are relative threshold indicators. TXX and T35 mainly reflect the intensity and frequency of EHT events and are widely used in meteorological services and research in China (Wang et al., 2013; Tang et al., 2014; Zhang et al., 2021). TX90P is an indicator of EHT events defined by the Expert Team on Climate Change Detection and Indices (ETCCDI; http://etccdi.pacificclimate.org/list_27_indices.shtml). It is also a core indicator recommended by the European Climate Assessment & Dataset (ECA&D) project and has been widely used in extreme climate event assessments and future projection research (Zhang et al., 2011; Kim et al., 2020). Referring to Kong et al. (2020) and Guo et al. (2017), we defined the HWF and HWD, which represent the frequency and duration of heat wave events for the whole year, respectively. The reason for choosing the relative threshold indices TX90P, HWF, and HWD was that the number of days exceeding the percentile thresholds is more evenly distributed in space and meaningful in every region (Zhang et al., 2011).

Table  2.  Definitions of the five extreme high temperature indices used in this study

Download as CSV

Name	Index	Definition	Unit
Max Tmax	TXX	Annual maximum value of daily maximum temperature (Tmax)	°C
High temperature days	T35	Annual number of days during which the daily maximum temperature > 35°C	day
Warm days	TX90P	Annual number of days when daily maximum temperature > 90th percentile	day
Heat wave frequency	HWF	Annual number of heat waves	time
Heat wave days	HWD	Annual number of days during which heat waves occurred	day
Note: A heat wave is defined when the daily maximum temperature is higher than the relative threshold for at least three consecutive days. The relative threshold on each calendar day is calculated as the 90th percentile of daily maximum temperature based on 15-day samples centered on that day during the baseline period of 1961–1990.

 | Show Table

DownLoad: CSV

2.2.2   Model evaluation methods

The Taylor diagram and Taylor skill score (TS) (Taylor, 2001; Zhu et al., 2020) were used to evaluate the simulation results of extreme high temperature indices during the historical reference period (1995–2014). Taylor diagram can visualize both the correlation coefficient (r) and the normalized standard deviation (σ_x/σ_y) between two fields or sequences on the same plot. The correlation coefficient indicates how closely the spatial patterns of the simulation results match those of the observations, whereas the normalized standard deviation reflects the spatial uniformity of the model results and their differences from the observations. Taylor skill score (TS) can quantitatively describe the model simulation performance. The calculation formulas for the normalized standard deviation (σ_x/σ_y), correlation coefficient (r), and Taylor skill score (TS) are as follows:

where x is the model simulation result, $\overline{x}$ is the average of the simulation results, y is the observation, $\overline{y}$ is the average of the observations, n is the total number of grid points, r₀ is the maximum attainable correlation coefficient (0.999), and σ_x and σ_y are the standard deviations of the simulated and observed spatial patterns, respectively. If the model represents the observed spatial pattern well, there will be more overlap between the two distributions, and the r, σ_x/σ_y, and TS will be close to 1.

For the same index, the simulation results of different models greatly differed, along with differences in the simulation capabilities of the same model for different indices. Therefore, we optimized the models based on their comprehensive simulation capabilities. The selected optimal models in this study are the following 11 models with the highest average TS scores: BCC-CSM2-MR, CanESM5, CNRM-CM6-1, GFDL-CM4, HadGEM3-GC31-LL, INM-CM4-8, INM-CM5-0, IPSL-CM6A-LR, KACE-1-0-G, MPI-ESM1-2-LR, and NESM3. The average TS score of a certain model refers to the average value of the TS scores of the five EHT indices simulated by the model. The average TS score of the 11 preferred models was higher than the average of all 26 models and the other 15 models. In the subsequent section, the multimodel ensemble (MME) refers to the ensemble mean of all 26 models in Table 1, whereas the best-member multimodel ensemble (BMME) refers to the ensemble mean of the optimal 11 models. Statistical significance was determined by using Student’s t test.

3.   Results

3.1   Evaluation of historical EHT

3.1.1   Performance of the models

Figure 1 presents the Taylor diagrams of the simulation and observation results of extreme high temperature indices in southwestern China for the period of 1995–2014. Each individual model, MME, and BMME exhibited strong modeling capabilities and highly concentrated simulation results for TXX (Fig. 1a), with a correlation coefficient of 0.94–0.95, which is higher than that reported by Li et al. (2022) for southwestern China using nine models. The models also had a relatively strong ability to simulate T35 (Fig. 1b), with correlation coefficients ranging from 0.65 to 0.82 and normalized standard deviations ranging from 0.87 to 1.38. ACCESS-CM2, CMCC-CM2-SR5, CNRM-CM6-1, MRI-ESM2-0, TaiESM1, and UKESM1-0-LL demonstrated good ability in simulating T35, with correlation coefficients above 0.8, which is better than MME and BMME. The simulation results for the TX90P, HWF, and HWD models were unsatisfactory (Figs. 1c–e), with correlation coefficients lower than 0.69. However, the correlation coefficient of TX90P in this study was higher than those derived by Li et al. (2022) in southwestern China. The BMME generally outperformed all individual models and MME in simulating TX90P, HWF, and HWD (Figs. 1c–e). The HWF simulated by ACCESS-ESM1-5 and CMCC-CM2-SR5, and the HWD simulated by MIROC-ES2L exhibited weak negative correlations with observational data, indicating relatively poor modeling capabilities of these models. Generally, it is difficult to evaluate model performances with percentile indices because of the uncertainty in the definition of indices (Sillmann et al., 2014; Chen and Sun, 2015; Luo et al., 2020; Zhu et al., 2020), as a location could experience what would be classified as a heat wave in the middle of winter (Zhang et al., 2010). Furthermore, the estimated thresholds are subject to sampling error and the accuracy of the data can lead to errors in the indices of percentile extremes.

Download

Larger image

PowerPoint slide

Fig  1.  Taylor diagrams of the models simulating extreme high temperature indices for 1995–2014 in southwestern China: (a) TXX, (b) T35, (c) TX90P, (d) HWF, and (e) HWD. REF represents the observed data. The closer the model data are to REF, the higher the simulation performance.

Table 3 shows the Taylor skill scores and ranking of the models simulating extreme high temperature indices for the period 1995–2014 in southwestern China. Each individual model, MME, and BMME had good capabilities for simulating the spatial distribution of TXX, with TS scores exceeding 0.94, which is higher than that reported by Li et al. (2022) in southwestern China using nine models. They also had a relatively strong ability to simulate T35, with TS scores exceeding 0.68. However, they were less skilled at simulating TX90P, HWF, and HWD, with TS scores below 0.68. Previous studies have also indicated that although CMIP6 models showed some improvement in simulating TX90P compared to CMIP5, they still face certain challenges (Luo et al., 2020; Zhu et al., 2020; Li et al., 2022). ACCESS-CM2, CMCC-CM2-SR5, CNRM-CM6-1, MRI-ESM2-0, and TaiESM1 exhibited good modeling capabilities for T35, with TS scores of 0.8 or higher. ACCESS-CM2 was the best model for simulating T35, with a TS score of 0.82. KACE-1-0-G exhibited the highest TS scores for TX90P and HWF (0.68 and 0.56, respectively), which were both higher than those for BMME. In terms of HWD, BMME had the strongest modeling ability, with a TS score of 0.66. MIROC-ES2L showed low skill in simulating TX90P, HWF, and HWD, with TS scores of approximately 0.2. Generally, the models showed a smaller intermodel spread for TXX and T35 and a larger spread for TX90P, HWF, and HWD in their TS scores, indicating that the uncertainties of TX90P, HWF, and HWD in the models are larger than those of TXX and T35 (Luo et al., 2020; Zhu et al., 2020). Based on the average TS scores of each model, MME, and BMME, the highest TS scores were obtained with BMME, KACE-1-0-G, CNRM-CM6-1, HadGEM3-GC31-LL, CanESM5, MPI-ESM1-2-LR, INM-CM5-0, GFDL-CM4, INM-CM4-8, NESM3, IPSL-CM6A-LR, BCC-CSM2-MR, and MME.

Table  3.  Taylor skill scores and ranking of the models in simulating extreme high temperature indices for 1995–2014 in southwestern China

Download as CSV

Model name	TXX	T35	TX90P	HWF	HWD	Average	Ranking
BMME	0.95	0.77	0.62	0.52	0.66	0.704	1
KACE-1-0-G	0.94	0.71	0.68	0.56	0.58	0.694	2
CNRM-CM6-1	0.95	0.8	0.5	0.5	0.6	0.67	3
HadGEM3-GC31-LL	0.95	0.74	0.56	0.52	0.55	0.664	4
CanESM5	0.95	0.76	0.57	0.45	0.59	0.664	5
MPI-ESM1-2-LR	0.95	0.75	0.55	0.42	0.56	0.646	6
INM-CM5-0	0.95	0.77	0.52	0.41	0.57	0.644	7
GFDL-CM4	0.94	0.73	0.56	0.45	0.53	0.642	8
INM-CM4-8	0.95	0.76	0.51	0.44	0.53	0.638	9
NESM3	0.95	0.78	0.51	0.42	0.48	0.628	10
IPSL-CM6A-LR	0.95	0.74	0.52	0.43	0.48	0.624	11
BCC-CSM2-MR	0.95	0.76	0.4	0.46	0.5	0.614	12
MME	0.95	0.77	0.45	0.36	0.53	0.612	13
NorESM2-LM	0.95	0.76	0.42	0.39	0.48	0.6	14
EC-Earth3	0.94	0.73	0.44	0.35	0.46	0.584	15
GISS-E2-1-G	0.95	0.68	0.44	0.39	0.46	0.584	16
CNRM-ESM2-1	0.95	0.72	0.36	0.41	0.45	0.578	17
CMCC-ESM2	0.94	0.69	0.38	0.33	0.46	0.56	18
ACCESS-CM2	0.95	0.82	0.39	0.29	0.34	0.558	19
UKESM1-0-LL	0.95	0.74	0.41	0.34	0.34	0.556	20
EC-Earth3-Veg-LR	0.94	0.68	0.35	0.32	0.38	0.534	21
MIROC6	0.95	0.72	0.37	0.27	0.3	0.522	22
GFDL-ESM4	0.95	0.77	0.3	0.28	0.27	0.514	23
ACCESS-ESM1-5	0.95	0.72	0.3	0.24	0.31	0.504	24
TaiESM1	0.95	0.81	0.22	0.28	0.25	0.502	25
CMCC-CM2-SR5	0.94	0.8	0.23	0.22	0.27	0.492	26
MRI-ESM2-0	0.95	0.81	0.21	0.23	0.23	0.486	27
MIROC-ES2L	0.95	0.74	0.18	0.21	0.18	0.452	28
Note: The last column indicates the comprehensive simulation capability ranking based on the average TS scores in the seventh column.

 | Show Table

DownLoad: CSV

3.1.2   Spatial pattern

Based on the statistical analysis outlined above, Fig. 2 illustrates the spatial distributions of the observed, MME, and BMME extreme high temperature indices for 1995–2014 in southwestern China. The spatial pattern of EHT events varied with the definition of EHT indices, which was consistent with previous studies (Smith et al., 2013; Hu et al., 2017). The TXX simulated by the MME and BMME was in accordance with the observations (Figs. 2a–c). The eastern part of Sichuan and low-altitude region of Chongqing (Sichuan basin) constitute the high-value center of TXX (Li et al., 2019; Hu and Sun, 2022; Li et al., 2022), with the TXX ranging from 35 to 40°C, which may be related to the higher population density and urbanization rate. Both the MME and BMME slightly underestimated the T35 in southwestern China (Figs. 2d–f). However, their spatial structures were generally consistent with the observations. Chongqing and the adjacent areas of Sichuan are the main high-value centers for T35 (12–13 days), which was consistent with the spatial distribution of TXX and previous studies (Hu et al., 2017). Zhang et al. (2011) found a strong correlation between the extreme value indicator (TXX) and the absolute threshold indicator (T35), as well as between relative threshold indicators [TX90P, warm spell duration index (WSDI)], whereas the correlation between the extreme value indicator (TXX), absolute threshold indicator (T35), and relative threshold indicators (TX90P, WSDI) was weak. The spatial distributions of MME and BMME TX90P showed a general trend of being high in the west and low in the east (Figs. 2h, i) (Li et al., 2022), which differed from the observed spatial distribution that was low in Guizhou (35–55 days) (Fig. 2g). Both the MME and BMME underestimated the TX90P (Zhu et al., 2020), whereas the BMME was closer to the observed values in western Yunnan, which is the region with the highest TX90P (Li et al., 2019; Hu and Sun, 2022). The spatial patterns of the MME, BMME, and the observed HWF and HWD differed (Figs. 2j–o). The main center of low values in the observations was Guizhou (4–7 times) (Figs. 2j, m), whereas the low values in the models were observed in more northern and western areas and mainly located at the junction of the four southwestern provinces (Figs. 2k, l, n, o). The MME and BMME both underestimated the HWF, especially in Yunnan, whereas the BMME was closer to the observed values (Figs. 2j–l). The MME and BMME underestimated the HWD in Sichuan and Chongqing, with the BMME being closer to the observed values (Figs. 2m–o). In general, the MME and BMME simulated the spatial distributions of TXX and T35 in southwestern China reasonably well but showed poor ability to reproduce the spatial patterns of TX90P, HWF, and HWD, which was consistent with previous studies (Li et al., 2019; Fan et al., 2020; Luo et al., 2020; Zhu et al., 2020; Li et al., 2022; Wei et al., 2023). This may be related to the definition of percentile indices like what we discussed in Section 3.1.1, the complexity of the heat wave, the regional limitations of climate models, and other reasons. Southwestern China is close to the Qinghai–Tibet Plateau and presents a complex topography. The extreme high temperatures in this region could be greatly affected by the dynamic and thermodynamic effects of the Qinghai–Tibet Plateau and several monsoon circulations (Guan et al., 2015; Ma et al., 2017; Xue et al., 2020; Li et al., 2022). Heat waves are continuous extreme high temperature events; their occurrence and sustaining mechanisms are complicated (Hu et al., 2017), particularly in southwestern China. Extreme high temperatures and heat waves in the climate models are influenced by interannual to long-term temperature changes, whereas the observed heat waves are restricted by short-term climate changes (Freychet et al., 2018). North Atlantic oscillations may trigger waves spreading southeastward in northern Russia and East Asia, causing high-pressure anomalies in the Yangtze River basin of China, leading to heat waves (Deng et al., 2019). Arctic sea ice loss can lead to favorable atmospheric circulation and surface conditions for heat waves in southwestern China (Deng et al., 2020). Climate models may have difficulties in reproducing these physical processes (Shi et al., 2018), and deficiencies in simulating these indices owing to differences in model regional climate sensitivity (Seneviratne and Hauser, 2020).

Download

Larger image

PowerPoint slide

Fig  2.  Spatial distributions of the (left) observed, (middle) MME, and (right) BMME extreme high temperature indices for 1995–2014 in southwestern China: (a, b, c) TXX; (d, e, f) T35; (g, h, i) TX90P; (j, k, l) HWF; and (m, n, o) HWD.

3.1.3   Temporal changes

Figure 3 shows the temporal changes in the extreme high temperature index anomalies relative to the historical reference period (1995–2014) from 1961 to 2014 in southwestern China. The observed, MME, and BMME extreme high temperature indices exhibited significant upward trends except for TXX, with MME and BMME being highly consistent. The linear trend coefficients of the observed TXX, T35, TX90P, HWF, and HWD were −0.02, 0.02, 0.68, 0.09, and 0.47, respectively. TXX showed a slight downward trend in the observed data (black line in Fig. 3a), whereas the other indices showed clear upward trends (black lines in Figs. 3b–e). Xue et al. (2020) showed that the warming rate of observed summer maximum temperature in southwestern China in the period of 1961–2014 was about 0.01, with a decreasing trend in the period of 1975–1994 and a dramatic increasing trend in the period of 1995–2014. This decreasing trend in the period of 1975–1994 is consistent with the temporal changes of TXX observations in present study (black line in Fig. 3a). The linear trend coefficients of MME TXX, T35, TX90P, HWF, and HWD were 0.02, 0.03, 0.55, 0.07, and 0.40, respectively (red lines in Fig. 3). The linear trend coefficients of BMME TXX, T35, TX90P, HWF, and HWD were 0.02, 0.04, 0.59, 0.08, and 0.44, respectively (blue lines in Fig. 3). The temporal variability trends of all the observed and simulated indicators passed the significance test with a significance level of 0.05. Overall, the observed and simulated temporal trends of T35, TX90P, HWF, and HWD were consistent (Figs. 3b–e), but were different for TXX (Fig. 3a). Before the mid-1990s, the simulated values of TXX were significantly lower than the observed values, whereas after the mid-1990s, the values were relatively similar (Fig. 3a). After 1980, the simulated T35 was more consistent with the observed values, whereas it was lower than the observed values before this year (Fig. 3b). Both the simulations and observations indicated the occurrence of a rapid growth phase for HWF and HWD (Guo et al., 2017), as well as for TX90P (Figs. 3c–e), since the early-1990s, which could be related to the anthropogenic greenhouse gases forcing and the impacts of urbanization (Kong et al., 2020). The extent of the increase in HWD was much greater than that in HWF, indicating that the duration of individual heat waves may have increased in recent years, which may be influenced by the combined effect of the South Asian high and western North Pacific subtropical high (Hu et al., 2017; Li N. et al., 2021). Although the abilities of models to simulate spatial patterns for TX90P, HWF, and HWD were relatively weak, their ability to simulate temporal trends was good.

Download

Larger image

PowerPoint slide

Fig  3.  Temporal changes in extreme high temperature index anomalies relative to the historical reference period (1995–2014) from 1961 to 2014: (a) TXX, (b) T35, (c) TX90P, (d) HWF, and (e) HWD. The black, red, and blue lines represent the observations, MME, and BMME, respectively, and the upper and lower limits of the shaded area are the maximum and minimum values of the 26 model simulations.

3.2   Projection of future changes in EHT

3.2.1   Spatial pattern

BMME had relatively strong capabilities in simulating the spatial patterns of extreme high temperature indices in southwestern China. Therefore, this section used BMME to project the future changes of extreme high temperatures. Figure 4 shows the spatial distributions of BMME extreme high temperature index anomalies relative to the historical reference period (1995–2014) in the early (2021–2040), middle (2041–2060), and late (2081–2100) periods of the 21st century under the SSP1-2.6 scenario. For TXX, the near-term increase was in the range of 0–1.5°C (Fig. 4a), and the mid- and long-term increases were in the range of 0.5–2°C (Figs. 4b, c). The largest increases mainly occurred in the Chongqing and eastern Sichuan regions (Figs. 4b, c). For T35, the adjacent areas of Chongqing and Sichuan were centers with high growth, with an increase of 5–15 days in the near term (Fig. 4d) and 10–25 days in the mid- and long terms (Figs. 4e, f). In most of the remaining regions, the increase occurred between 0 and 5 days throughout the study period (Figs. 4d–f). TX90P showed an overall spatial distribution with higher values in the west and lower values in the east (Figs. 4g–i). Near-term growth ranged from 20 to 60 days (Fig. 4g), midterm growth ranged from 40 to 90 days (Fig. 4h), and long-term growth ranged from 50 to 100 days (Fig. 4i). The spatial distribution of HWF was basically consistent with that of TX90P, and southwestern Sichuan and northwestern Yunnan were relatively high-value areas (Figs. 4j–l). The increase was 3–7 times in the near term (Fig. 4j), 5–11 times in the midterm (Fig. 4k), and 7–11 times in the long term (Fig. 4l). Similarly, HWD exhibited an overall spatial distribution of higher values in the west and lower values in the east (Figs. 4m–o). The increase was 20–60 days in the near term (Fig. 4m), 30–100 days in the midterm (Fig. 4n), and 50–100 days in the long term (Fig. 4o). Overall, except for HWF, the increase in the historically high areas of all indicators was higher than that in the historically low areas (Figs. 2, 4). Meehl and Tebaldi (2004) also showed that areas that have already experienced serious heat waves will experience more severe heat waves in the future. Generally, the growth of the EHT indices under the SSP1-2.6 scenario tends to be moderate, which was consistent with the findings of previous studies (Jiang and Chen, 2021).

Download

Larger image

PowerPoint slide

Fig  4.  Spatial distributions of BMME extreme high temperature index anomalies relative to the historical reference period (1995–2014) in (left) 2021–2040, (middle) 2041–2060, and (right) 2081–2100 under the SSP1-2.6 scenario: (a, b, c) TXX; (d, e, f) T35; (g, h, i) TX90P; (j, k, l) HWF; and (m, n, o) HWD.

Figure 5 shows the spatial distributions of BMME extreme high temperature index anomalies relative to the historical reference period (1995–2014) in the early (2021–2040), middle (2041–2060), and late (2081–2100) periods of the 21st century under the SSP2-4.5 scenario. For TXX, the increase was 0–1°C in the near term (Fig. 5a), 0.5–2°C in the midterm (Fig. 5b), and 1.5–3.5°C in the long term (Fig. 5c). Similar to the surface air temperature, warming was projected to be more pronounced at the end of the 21st century than in the midterm (Gao et al., 2012). The largest increase in TXX mainly occurred in most areas of Chongqing and parts of eastern Sichuan (Figs. 5b, c). For T35, the adjacent areas of Chongqing and Sichuan had high growth, with an increase of 5–10 days in the near term (Fig. 5d), 10–25 days in the midterm (Fig. 5e), and 25–35 days in the long term (Fig. 5f). The largest areas of increase were also reported by Guo et al. (2022) who used 12 CMIP6 models. However, in most of the remaining regions, the increase was between 0 and 5 days in the near and midterms (Figs. 5d, e). TX90P presented an overall spatial distribution with higher values in the west and lower values in the east (Figs. 5g–i), which was consistent with the findings of Yao et al. (2012) using eight CMIP5 models in which the largest increase in TX90P corresponded to that in western Sichuan and Yunnan in 2006–2099 under the RCP4.5 scenario. Near-term growth ranged from 20 to 60 days (Fig. 5g), midterm growth ranged from 50 to 130 days (Fig. 5h), and long-term growth ranged from 90 to 170 days (Fig. 5i). The spatial distribution of HWF showed a general trend of being higher in the west and lower in the east during the early and middle periods, with an increase of 3–7 times in the former and 7–12 times in the latter (Figs. 5j, k). Relatively high-value areas were presented in southwestern Sichuan and northwestern Yunnan (Figs. 5j, k). During the late period, the spatial distribution was generally higher in the northwest and lower in the southwest, with an increase of 8–15 times (Fig. 5l). The distribution of HWD showed a west-high and east-low spatial pattern, with the high-value center mainly located in northwest Yunnan (Figs. 5m–o). Near-term growth ranged from 10 to 70 days (Fig. 5m), midterm growth ranged from 30 to 120 days (Fig. 5n), and long-term growth ranged from 90 to 190 days (Fig. 5o). Overall, except for the HWF in the late period, the spatial patterns of all indices under the SSP2-4.5 scenario were consistent with those under the SSP1-2.6 scenario (Figs. 4, 5). The growth of TX90P and HWD under the SSP2-4.5 scenario was significantly higher than that under the SSP1-2.6 scenario (Figs. 4, 5).

Download

Larger image

PowerPoint slide

Fig  5.  As in Fig. 4, but for SSP2-4.5.

Figure 6 shows the spatial distributions of BMME extreme high temperature index anomalies relative to the historical reference period (1995–2014) in the early (2021–2040), middle (2041–2060), and late (2081–2100) periods of the 21st century under the SSP5-8.5 scenario. For TXX, the increase is 0–1.5°C in the near term (Fig. 6a), 0.5–3.5°C in the midterm (Fig. 6b), and 3.5–6.5°C in the long term (Fig. 6c). Northeastern Chongqing and a small part of northeastern Sichuan experienced the most significant increases (Figs. 6b, c). For T35, the spatial difference in the near-term increase was not obvious and mainly occurred between 0 and 15 days (Fig. 6d). In the midterm, the adjacent areas of Chongqing and Sichuan were high-value centers, with an increase of 20–30 days (Fig. 6e). In the long term, the increase in the adjacent areas of Chongqing and Sichuan was mainly 60–80 days (Fig. 6f). In southern Yunnan and eastern Guizhou, the increase reached to 20–80 days (Fig. 6f), which was significantly higher than that in the near and midterms. TX90P presented an overall spatial distribution with higher values in the west and lower values in the east (Figs. 6g–i). The near-term increase was 20–70 days (Fig. 6g), midterm increase was 70–150 days (Fig. 6h), and long-term increase was 150–240 days (Fig. 6i). The increase in HWF was 3–9 times in the near term and 7–13 times in the midterm (Figs. 6j, k). In the late period, the overall spatial distribution was higher in the northeast and lower in the southwest, with an increase of 3–15 times (Fig. 6l). The high-value centers in the near and midterms were mainly located in southwestern Sichuan and northwestern Yunnan (Figs. 6j, k) and in the long term, they were mainly located in Chongqing, central and eastern Sichuan, and northern Guizhou (Fig. 6l). The spatial distribution of HWD showed higher values in the west and lower values in the east (Figs. 6m–o). The increase was 10–70 days in the near term (Fig. 6m), 50–150 days in the midterm (Fig. 6n), and 150–270 days in the long term (Fig. 6o). This phenomenon, which was rare in the past, may become normal under the high-SSP scenario in the future, with a gradually saturated trend in the number of heat wave days in a year. Consistent with the findings of Sun et al. (2023) in Southeast Asia, a very extreme increase occurred under the SSP5-8.5 scenario, that is, the HWD in the western part of southwestern China reached approximately 300 days in the late period of the 21st century. Compared to the SSP2-4.5 scenario, the SSP5-8.5 scenario showed a remarkable increase of T35 in southern Yunnan and eastern Guizhou as well as an eastward migration of HWF high-value centers in the late period (Figs. 5, 6).

Download

Larger image

PowerPoint slide

Fig  6.  As in Fig. 4, but for SSP5-8.5.

The above analysis showed that under the SSP1-2.6 (Fig. 4), SSP2-4.5 (Fig. 5), and SSP5-8.5 scenarios (Fig. 6), the indices TXX, T35, TX90P, HWF, and HWD all exhibited significant increases in the early (2021–2040), middle (2041–2060), and late (2081–2100) periods of the 21st century. These increases became more pronounced with time and radiative forcing levels, indicating higher frequency, longer duration, and stronger intensity of EHT events (Guo et al., 2017; Jiang and Chen, 2021). Li et al. (2022) also projected that from 2021 to 2100, southwestern China may experience a significant increase in TXX and TX90P under all four scenarios, especially for TX90P. Furthermore, regional differences were more pronounced under the SSP5-8.5 scenario. Zhang et al. (2021) projected that TXX will increase significantly in northern China, whereas TX90P and WSDI will increase significantly in southern China, and the warming in Tibet will exceed the global average. In the present study, TXX and T35 generally increased more in the northeastern region than in other regions (Figs. 4–6), while under SSP5-8.5, the high-value centers of TXX were located further north in the mid- and long terms (Figs. 6b, c), and the high-value areas of T35 increased in the eastern and southern edge areas in the long term (Fig. 6f). The increases in TX90P and HWD generally had a spatial distribution of being higher in the west and lower in the east (Figs. 4–6), which is in accordance with the TX90P results reported by Wei et al. (2023) and could be caused by the consistent increase in radiative-forced temperature in high-altitude regions (Ma et al., 2017). The spatial distributions of HWF were similar in three periods under SSP1-2.6 (Figs. 4j–l) and in the early and middle periods under SSP2-4.5 (Figs. 5j, k) and SSP5-8.5 (Figs. 6j, k), with the highest numbers of heat wave events occurring in the neighboring areas of Sichuan and Yunnan. However, the spatial distributions of HWF during the late periods were noticeably different, with higher number of heat wave events occurring in western Sichuan under the SSP2-4.5 scenario (Fig. 5l) and eastern Sichuan under the SSP5-8.5 scenario (Fig. 6l), respectively. Unlike in the other regions, the HWF in Yunnan and neighboring western Sichuan first increased and then decreased over time under the SSP5-8.5 scenario (Figs. 6j–l). These phenomenon of HWF may be caused by regional differences in the duration of heat wave events during different periods (Sun et al., 2023), differences in the urbanization rates under different scenarios (Li et al., 2022), or the projection uncertainty of the higher emission scenarios (Chen et al., 2022).

3.2.2   Temporal changes

Figure 7 shows the temporal changes in the extreme high temperature index anomalies relative to the historical reference period (1995–2014) from 2021 to 2100. Under the SSP1-2.6 (green lines in Fig. 7), SSP2-4.5 (blue lines in Fig. 7), and SSP5-8.5 scenarios (red lines in Fig. 7), TXX will increase by 1.8, 2.9, and 6.0°C (Fig. 7a), T35 will increase by 6.5, 16.2, and 45.0 days (Fig. 7b), TX90P will increase by 81.0, 138.4, and 225.7 days (Fig. 7c), HWF will increase by 8.5, 11.1, and 8.8 times (Fig. 7d), and HWD will increase by 76.0, 136.9, and 234.5 days (Fig. 7e), respectively, by the end of the 21st century. For the period before the 2040s, the five indices displayed almost the same increase rate under both SSP2-4.5 and SSP5-8.5. However, after the 2040s, the growth rate under SSP5-8.5 rapidly surpassed that under SSP2-4.5 (blue and red lines in Fig. 7). The increase in TXX in southwestern China under the SSP5-8.5 scenario (red line in Fig. 7a) was lower than that in the mid–high latitudes of Asia reported by Jiang and Chen (2021). Overall, higher frequency, longer duration, and greater intensity of EHT events occurred under SSP5-8.5 and in the late periods. Similarly, the projection uncertainty (shaded area in Fig. 7) also increased with radiative forcing levels (Jiang and Chen, 2021) and time. It is worth noting that unlike the HWD (red line in Fig. 7e), which showed a significant increase with time, the HWF generally increased at first and then decreased under the SSP5-8.5 scenario (red line in Fig. 7d), reaching a peak around 2060. This is different from the study by Guo et al. (2017), in which the HWF in China under the RCP8.5 scenario showed a continuous upward trend in the period of 2021–2100, which can be attributed to the definition of heat wave and the models used. However, the temporal trend of the HWF in the present study was consistent with the findings of Sun et al. (2023) showing that the HWF in Southeast Asia under the SSP5-8.5 scenario will gradually have a downward trend after the mid-21st century. What is the reason for this? Previous studies indicated that the increase in the duration of a single heat wave event may result in fewer identifiable heat wave events (Guo et al., 2017; Liu et al., 2018; Sun et al., 2023). Zhou et al. (2014) showed that the projected increase in the WSDI for China was 136 days over the period of 2081–2100 relative to the reference period of 1986–2005 under the RCP8.5 scenario. Kim et al. (2023) projected that long-lasting heat waves will occur more often in East Asia, especially under SSP5-8.5. The increase of long-lasting heat waves may be related to various factors, such as greenhouse gas warming, atmospheric circulation anomalies, and urbanization effects (Li N. et al., 2021). The decrease in HWF may also be related to the uncertainty in anthropogenic forcing of different emission paths, uncertainty related to natural variability, and uncertainty in the response of the climate system to external forcing (Guo et al., 2017).

Download

Larger image

PowerPoint slide

Fig  7.  Temporal changes in extreme high temperature index anomalies relative to the historical reference period (1995–2014) in the period from 2021 to 2100: (a) TXX, (b) T35, (c) TX90P, (d) HWF, and (e) HWD. The green, blue, and red lines represent the estimates of BMME under SSP1-2.6, SSP2-4.5, and SSP5-8.5, respectively. The upper and lower bounds of the shaded area represent the maximum and minimum values, respectively, as simulated by the 11 optimal models.

4.   Conclusions and discussion

In the present study, we utilized the newly released NEX-GDDP-CMIP6 dataset to investigate the evolution of extreme high temperature events in historical and future periods in southwestern China. The main conclusions are as follows.

NEX-GDDP-CMIP6 was highly capable of simulating the spatial patterns of TXX and T35 in southwestern China and their Taylor skill scores were above 0.94 and 0.68, respectively. Chongqing and neighboring areas in Sichuan constituted the high-value centers of TXX and T35. However, the spatial modeling ability of the models for TX90P, HWF, and HWD was relatively weak, making it difficult to accurately reproduce the spatial pattern of Guizhou as the main low-value center. Overall, the performance of the BMME was generally better than those of the MME and most individual models.

The simulated time series of T35, TX90P, HWF, and HWD in southwestern China was consistent with the observed upward trend; however, the models failed to simulate the trend of the observed TXX. The extent of the increase in TX90P and HWD was much greater than that in TXX, T35, and HWF. The estimated and observed TXX values became closer after the mid-1990s. The simulated T35 became more consistent with the observed values since 1980. After the early-1990s, a rapid increase in the HWF, HWD, and TX90P occurred. Overall, the temporal trend modeling ability of the models was relatively strong.

Generally, the projected TXX, T35, TX90P, HWF, and HWD in southwestern China are expected to increase with time and radiative forcing levels. The increases in TXX and T35 in the adjacent regions of Chongqing and Sichuan were significantly greater than those in the other areas. The increases in TX90P, HWF, and HWD generally showed a pattern of higher values in the west and lower values in the east, with relatively high-value areas located in southwestern Sichuan and northwestern Yunnan. However, in the late period, under the SSP2-4.5 and SSP5-8.5 scenarios, HWF was relatively higher in the northwest and northeast, respectively. Furthermore, HWF in Yunnan and neighboring western Sichuan first increased and then decreased over time under SSP5-8.5.

In the three different scenarios, the projected future TXX, T35, TX90P, and HWD in southwestern China show a continuous increase over time, and the projection uncertainty also increases with the scenarios. The HWF initially increased and then decreased under SSP5-8.5, while the HWD continued to increase over time, indicating that the duration of individual heat waves is also expected to increase. By the end of the 21st century, under the SSP5-8.5 scenario, TXX, T35, TX90P, HWF, and HWD were projected to increase by 6.0°C, 45.0 days, 225.7 days, 8.8 times, and 234.5 days, respectively.

Accurately projecting future changes of extreme climatic events is an important scientific problem in formulating climate change policies (Chen et al., 2020). Owing to the uncertainties associated with GCMs, SSP scenarios, and internal variability of the climate system, uncertainty in future projections has been a persistent concern (Riahi et al., 2017; Zhang and Chen, 2021). In the present study, we evaluated the simulation performance of the NEX-GDDP-CMIP6 dataset on EHT events in southwestern China and revealed possible changes in future EHT events. Our findings provide a basis for further understanding the evolution of EHT events in southwestern China as well as for developing actionable adaptation and mitigation plans. Although the NEX-GDDP-CMIP6 dataset has high accuracy, there is still some uncertainty in simulating EHT events in southwestern China. For example, the interpolation method of CN05.1 brings some uncertainty to the observations, the evaluation conclusion is affected by the selection of indices and statistical standards, and different methods of model optimization may lead to different conclusions. In addition, we did not conduct a quantitative analysis of the causes underlying the increase in EHT events. In future studies, dynamic downscaled results (Gao et al., 2017) or regional climate models (Guo et al., 2017) can be used to analyze the possible reasons for the deviation of simulations and improve the impact factor analysis of the increase in future EHT events, such as anthropogenic greenhouse gas emissions (Kong et al., 2020), atmospheric variables (Yu et al., 2020), and land use or land cover changes (Luo et al., 2020). In addition, a compari-son of multiple preferred methods and multisource observation data is required to reduce the uncertainty in climate models (Wang et al., 2022). Furthermore, a new model evaluation method, namely the emergent constraint approach, can be used in subsequent studies to better assess the responses of climate models to different emission scenarios (Eyring et al., 2019; Hall et al., 2019).