Changes in concurrent precipitation and temperature extremes over the Asian monsoon region: observation and projection

Concurrent precipitation and temperature extremes exert amplified impacts on the ecosystems and human society; however, they have not been well documented over the Asian monsoon region with dense population and agricultures. In this study, the spatiotemporal variations of four concurrent extreme modes (cold/dry, cold/wet, warm/dry, and warm/wet) are detected based on observations and model projections. From 1961 to 2014, the ‘dry’ modes manifest large values at high latitudes, while the ‘wet’ modes occur frequently in tropical regions. Based on the linear congruency, the trends of the four modes are largely determined by extreme temperature. Furthermore, the interaction between extreme precipitation and extreme temperature (IEPET) facilitates the trends of the dry modes, and inhibits the trends of the wet modes. Three modeling datasets (CMIP6, NEX-GDDP-CMIP6, and BCSD_CMIP6) are employed to project future changes in the occurrences of four concurrent modes. The BCSD_CMIP6, generated by statistical downscaling of the CMIP6 simulations, stands out in simulating the observed features of extreme precipitation and extreme temperature over the Asian monsoon region. Extreme temperature is also identified as the main driver in the future trends of the four modes, while the IEPET is not conducive to the decreasing trend of the cold/dry mode, implying that the IEPET would change under global warming. The warm/wet mode manifests the largest change among the four compound extremes from 1995 to 2014 and two projected periods (2046–2065 and 2080–2099) relative to 1961–1980. On the annual timescale, the change magnitudes over Southeast Asia, South Asia, the Tibetan Plateau, and Eastern Central Asia are relatively larger than in the other sub-regions during historical and future periods, which are quantified as the hotspots of the warm/wet mode. On the seasonal timescale, the future hotspots will change relative to the historical period. Our findings are critical for formulating adaptation strategies to cope with the adverse effects of compound extremes.


Introduction
The Asian monsoon region (60 • E-150 • E, 15 • S-55 • N) is highly vulnerable to global warming due to the dense population and complex monsoon climate. In recent decades, this region has suffered from increased frequency and intensity of natural disasters, such as heatwaves, floods, and droughts, with severe ramifications on the environmental and social systems , Ge et al 2021, Sun et al 2022, Wang et al 2023. On the other hand, climate change over the Asian monsoon region is essential for the climate variability over tropical regions and even on a global scale through the stratospheric pathway (Rosenlof et al 1997, Park et al 2007, Randel et al 2015, Zhang et al 2016. Exploring the observational characteristics of climate extremes over this region and the response to global warming could enhance our understanding of local climate variability and the potential global climate impacts. Climate extremes of precipitation and temperature in the Asian monsoon region have been generally investigated separately (Zhou et al 2014, Zhai et al 2018, Chen et al 2020, Zhang et al 2020, Gao et al 2021, Ge et al 2021, Yang et al 2021, Sun et al 2022. Overall, the temperature extremes manifested spatially coherent variations, while the changes in precipitation extremes exhibited a large regional diversity. Nevertheless, from time to time, climate extremes occur simultaneously. For example, Tencer et al (2014) found that heavy rainfall often occurred with high temperatures in winter at certain high latitudes. The occurrence of high temperature extremes was often accompanied by precipitation deficits in the warm season over most continental regions Seneviratne 2012, Tencer et al 2014).
The simultaneous occurrences of multiple extremes are commonly referred to as concurrent extremes, which have attracted increased attention due to their amplified impacts (Hao et al 2013, Zscheischler et al 2018, Ge et al 2021, Sun et al 2022, Wang et al 2022. The concurrent extremes are often described as cold/dry, cold/wet, warm/dry, and warm/wet combinations (Beniston et al 2009, Estrella and Menzel 2012, Hao et al 2013, Aihaiti et al 2021, Zhang et al 2022. In a pioneering study, Beniston (2009) investigated the changing behavior of these four concurrent modes over Europe. Based on various observations, Hao et al (2013) elucidated that the 'warm' modes (warm/dry and warm/wet) manifested substantial increasing trends, while the 'cold' modes (cold/dry and cold/wet) depicted significant decreasing trends from 1901 to 2000 across most of the global land. However, analyses of concurrent extremes over the Asian monsoon region are scarce, and the response to global warming is even more rarely considered.
The Coupled Model Intercomparison Project Phase 6 (CMIP6) output has been widely used to project future climate change. It is known that the CMIP6 models exhibit a general improvement in simulating climate extremes at a global scale compared with the CMIP5 simulations, but bias still exists at regional scales compared with observations (Chen et al 2020, Zhu et al 2020, Yang et al 2021. To enrich the studies of climate change impacts at regional scales, the NASA Earth Exchange Global Daily Downscaled Projections generated a high-resolution dataset (0.25 • × 0.25 • , NEX-GDDP-CMIP6) by statistically downscaling the daily outputs from 35 CMIP6 models, released to the public in 2022 (Wood et al 2002, 2004, Maurer and Hidalgo 2008, Thrasher et al 2012. So, which of the CMIP6 simulations and the NEX-GDDP-CMIP6 dataset could better replicate the observed statistics of the four compound extremes? This issue is critical for accurately projecting future variations of the compound extremes, and thus warrants a detailed investigation. Based on these premises, this study detects the recent and future changes in concurrent temperature and precipitation extremes based on observations and multiple modeling datasets. Three key scientific questions will be addressed: (1) What are the spatiotemporal characteristics of cold/dry, cold/wet, warm/dry, and warm/wet combinations over the Asian monsoon region from 1961 to 2014? (2) Are the CMIP6 simulations and the NEX-GDDP-CMIP6 output able to capture the observed features of the four concurrent extremes over this region? (3) How can the future variations of compound extremes be projected accurately based on the multiple modeling simulations? The remainder of this paper is organized as follows. Section 2 describes the datasets and analysis methods. In section 3, we present the results obtained. Conclusions and a further discussion are provided in section 4.

Data
The Asian Precipitation-Highly-Resolved Observational Data Integration Towards Evaluation (APH-RODITE) is employed as observational data of daily precipitation and temperature, which is a gauge-based gridded dataset for Asia and covers the period of 1951-2015 with a horizontal resolution of 0.25 • × 0.25 • (Yatagai et al 2012). It has been used as reference data in the studies of data inter-comparison, model validation, and climate extremes (Bollasina and Messori 2018, Ge et al 2019, Utsumi and Kim 2022. We also analyze the modeling simulations from the first ensemble member ('r1i1p1') of 13 models in the CMIP6 project and the NEX-GDDP-CMIP6 dataset (Maurer and Hidalgo 2008, Thrasher et al 2012, Eyring et al 2016. These models (table S1) are selected based on the availability of daily precipitation and temperature from the historical simulations  and the 21st-century projections (2015-2100) under SSP126 and SSP585 scenarios. NEX-GDDP-CMIP6, with a horizontal resolution of 0.25 • × 0.25 • , is a set of statistically downscaled climate projections from 1950 to 2100, generated by utilizing the Bias Correction Spatial Disaggregation (BCSD) algorithm and observations, to bias correct and downscale daily outputs from the 35 CMIP6 models. The BCSD is a trendpreserving statistical downscaling technique, which has been widely applied to hydrological and meteorological fields , Wood et al 2002, 2004. The Global Meteorological Forcing Dataset (GMFD) is employed as the observational validation of NEX-GDDP-CMIP6, which merges information from observation, reanalysis, and satellite remote sensing products (Sheffield et al 2006).
Previous studies have revealed that the NEX-GDDP-CMIP5 dataset is in good agreement with observations in simulating the climatologies of daily precipitation and maximum and minimum temperatures, but bias still exists in capturing the observed climate extremes at regional scales, partly due to the limited accuracy of the GMFD (Bao and Wen 2017, Chen et al 2017, Raghavan et al 2018. Since NEX-GDDP-CMIP6 is generated utilizing the same calculation principle as NEX-GDDP-CMIP5, we compare the GMFD with APHROD-ITE over the Asian monsoon region, and find that the two observations show discrepancies in depicting the extreme precipitation and extreme temperature (figure not shown). Thus, we generate a statistically downscaled dataset (hereafter referred to as 'BCSD_CMIP6') based on the daily precipitation and temperature in 13 CMIP6 models and APHRODITE utilizing the BCSD method. In this paper, we evaluate the abilities of the three modeling datasets (CMIP6, NEX-GDDP-CMIP6, and BCSD_CMIP6) from 1961 to 2014, and select the best performance to project future variations of the four concurrent extremes.

Definitions of concurrent extremes
The concurrent extremes focused are defined based on two extreme precipitation and two extreme temperature thresholds as delineated by Beniston (2009). The extreme precipitation thresholds for a given grid are separately calculated as the 75th and 25th percentiles of wet days from 1961 to 2014 (hereafter referred to as P75 and P25, respectively). To facilitate the intercomparison among different sub-regions and months, a wet day in each grid represents a day with a precipitation amount of at least 1 mm. P75 and P25 are calculated based on the daily mean precipitation for each month of 1961-2014. A similar computation process is carried out for the temperature thresholds (T75 and T25).
In each month, the extreme wet (dry) days are defined as the total number of days when the daily precipitation is above (below) the P75 (P25) threshold, and the extreme warm (cold) days are calculated as the total number of days when the daily temperature is above (below) the T75 (T25) threshold. The combinations of extreme precipitation and extreme temperature, including cold/dry, cold/wet, warm/dry, and warm/wet, are defined as the total number of days when the daily precipitation and daily temperature simultaneously meet the thresholds. As has been reported in previous studies, the statistical characteristics of concurrent extreme precipitation and extreme temperature are quite insensitive to the choice of percentile levels (Beniston 2009, Estrella and Menzel 2012, Hao et al 2013. The regimes of the four compound extremes based on the 75th/25th percentile are similar to those based on the 90th/10th percentile (figure not shown). We set the 75th/25th percentile to capture a relatively larger number of events. In addition, the criterion of the wet day exerts an impact on the number of extreme wet and dry days, but the spatiotemporal characteristics of the four modes are insensitive to the changes in the definition of the wet day.
Since the changes in extreme precipitation over the Asian monsoon region exhibit spatial heterogeneity (Peel et   , the spatial characteristics and long-term trends of concurrent extremes over the Asian monsoon region will be explored in seasonal and annual timescales based on APHRODITE. In addition, the percentage changes in concurrent extremes over each sub-region from 1995 to 2014 with respect to the reference period (1961)(1962)(1963)(1964)(1965)(1966)(1967)(1968)(1969)(1970)(1971)(1972)(1973)(1974)(1975)(1976)(1977)(1978)(1979)(1980) are calculated to determine the regions that are most sensitive to climate change. In the future period (2015-2100), the percentage changes in concurrent extremes will also be explored during the two periods of 2046-2065 and 2080-2099 under SSP126 and SSP585 scenarios with respect to the reference period. The two-tailed Student's t-test is employed to assess the statistical significances at the 95% confidence level.

Linear congruency
To reveal the trend of each concurrent mode associated with the extreme temperature and extreme precipitation, we divide the trend into linearly congruent and linearly independent components with respect to the two individual extreme events based on the method proposed by Thompson et al (2000). The linearly congruent component of the concurrent mode is estimated as follows: where TrendA cong is the fraction of linear trend in time series A that is linearly congruent with time series B, R is the regression coefficient describing the relationship between the detrended time series A and B, and TrendB is the linear trend in time series B. Taking the cold/wet mode from 1961 to 2014 as an illustrative example, time series A is the seasonal mean frequency of the cold/wet mode averaged over the Asian monsoon region. Time series B denotes the corresponding seasonal mean frequencies of extreme cold

Spatiotemporal characteristics of concurrent extremes from 1961 to 2014
The spatial patterns of multi-year averaged occurrences of cold/dry, cold/wet, warm/dry, and warm/wet combinations from 1961 to 2014 derived from APHRODITE are shown in figures 1(a)-(d). The area averaged frequencies of the cold/dry (55.1 d yr −1 ) and warm/dry (67.9 d yr −1 ) modes over the whole domain are higher than those of the cold/wet (14.7 d yr −1 ) and warm/wet (6.2 d yr −1 ) combinations. The dry modes occur frequently in the mid-high latitudes across the Asian monsoon region, while the frequency of the cold/wet mode exhibits large values over tropical regions and the warm/wet combination rarely occurs in subtropical regions.
The spatial structure of each concurrent mode on a seasonal timescale bears a large resemblance to that on the annual timescale (figure not shown). Furthermore, the seasonal mean occurrences of the four compound extremes over eight sub-regions are explored (figures 1(e)-(h)). Among these compound extremes, the frequency of the warm/wet mode is the lowest, while the cold/dry and warm/dry modes exhibit relatively higher frequencies in the four seasons over most regions. The frequency of the warm/dry mode is higher than that of the cold/dry mode in spring and summer, and these two modes display comparable frequencies in autumn and winter over almost the entire domain. The frequency of the cold/wet mode in summer is higher than that in the other three seasons. Figure 2 depicts the interannual variations of frequencies of the four combinations from 1961 to 2014 over the Asian monsoon region. Overall, the annual averaged cold modes exhibit significant downward trends, and the decreasing trend of cold/dry mode (−0.67, p = 0.05) is much sharper than that of the cold/wet mode (−0.03, p = 0.05; table 1). The annual averaged frequencies of the warm modes signify remarkable upward trends, and the warm/dry mode manifests a much larger trend (0.51, p = 0.05) than that of the warm/wet mode (0.08, p = 0.05). The 'dry' modes depict more obvious trends in autumn than in the other three seasons, while the trends of 'wet' modes in summer are the largest among the four seasons.
Temperature and precipitation have been recognized to be closely associated with each other at different timescales due to their thermodynamic relationship (Seneviratne et al 2006, 2010, Allan and Soden 2008, Mueller and Seneviratne 2012, Holmes et al 2017. Thus, the long-term trend of each mode is determined by extreme precipitation, extreme temperature, and the interaction  between extreme precipitation and extreme temperature (IEPET). We further investigate the relative contributions of each element to the long-term trends of compound extremes based on the linear congruency (table 1). In general, the trends of the cold/dry, cold/wet, warm/dry, and warm/wet combinations are largely determined by extreme temperature on the annual and seasonal timescales. This result is consistent with that of Beniston (2009), who elucidated that temperature was the key driver of the changes in the four modes over Europe. The IEPET facilitates the trends of the dry modes, while it is not conducive to the trends of the wet modes. Specifically, the trends of extreme cold and extreme dry days explain about 38.8% (−0.26/−0.67) and 3.0% (−0.26/−0.67) of the trend of the cold/dry mode, respectively. Thus, the IEPET exerts a positive impact on the occurrence of the cold/dry combination mode. The long-term trend of the cold/wet mode is largely determined by the cold days, and the IEPET exerts a negative impact on this concurrent mode. The trends of the warm/dry and warm/wet modes are mainly modulated by the warm days, while the IEPET contributes to the former but inhibits the latter. Why the IEPET manifests opposite effects on the dry and wet modes is still worthy of further investigations. Extreme climate change is characterized by obvious regional diversity, and the regions with the greatest change magnitudes are usually identified as climate change 'hotspots' (Giorgi 2006, Diffenbaugh and Giorgi 2012, Turco et al 2015. Under the context of global warming, the identification of hotspots is essential for prioritizing mitigation and multifaceted adaptation strategies. The percentage changes in the four concurrent extremes over the Asian monsoon region from 1995 to 2014 with respect to 1961 to 1980 are calculated to identify the observed hotspots of the compound extremes (figure 3). The annual mean frequencies of the warm modes depict positive changes over all the subregions, and the cold/dry mode exhibits consistent negative changes in the whole domain, while the cold/wet mode shows a large spatial heterogeneity ( figure 3(a)). Although the annual mean frequency of the warm/wet mode is lower than the other three concurrent modes (figures 1(a)-(d)), its percentage change manifests the largest positive values over most of the Asian monsoon region, especially over Southeast Asia, South Asia, the Tibetan Plateau, and Eastern Central Asia ( figure 3(a)). These sub-regions are quantified as the observed hotspots of the warm/wet mode on the annual timescale. Similar results are found in spring and autumn (figures 3(b) and (d)), while the change in the warm/wet mode displays a slight decrease over South Asia (figure 3(e)) in winter. In summer, this mode manifests large change magnitudes over all the sub-regions ( figure 3(c)).

Projections of concurrent extremes under SSP126 and SSP585 scenarios
To accurately project the future changes in the four compound extremes over the Asian monsoon region, we first assess the performances of CMIP6 simulations and the NEX-GDDP-CMIP6 and BCSD_CMIP6 datasets in capturing the observed statistical metrics of precipitation and temperature from 1961 to 2014 (figures S1 and S2). Apparently, the area averaged temperature and precipitation over the Asian monsoon region obtained from BCSD_CMIP6 are in better agreement with observations, and their inter-model spreads are largely reduced compared with the CMIP6 simulations and NEX-GDDP-CMIP6 dataset (figures S1(a) and (b)). Since the four compound extremes are defined based on T25/T75 and P25/P75, we further examine the ability of BCSD_CMIP6 in capturing the mean values and standard deviations of these thresholds. The CMIP6 multi-model ensemble mean (MME) shows wet biases over Southeast Asia and the Tibetan Plateau (figures S1(c) and (e)), and cold biases over the Tibetan Plateau (figures S1(g) and (i)). It can be clearly seen that these biases in the mean values of extreme precipitation and extreme temperature thresholds are reduced in the BCSD_CMIP6 dataset (figures S1(d), (f), (h), and (j)). Furthermore, the biases of standard deviations of these thresholds in CMIP6 MME are remarkably reduced in the BCSD_CMIP6 dataset ( figure S2). Therefore, the BCSD_CMIP6 stands out among the three modeling datasets in capturing the observed features, and is employed to assess the future changes in the four compound extremes over the Asian monsoon region. Figure 4 depicts the time series of annual mean occurrences of cold/dry, cold/wet, warm/dry, and warm/wet modes averaged over the Asian monsoon region under the SSP585 scenario from 2015 to 2100. The cold modes, which exhibit significant downward trends, may diminish almost completely by the end of 2100, while the warm modes show remarkable upward trends and have already been observed in recent decades. The relative contributions of extreme precipitation and extreme temperature to the longterm trends of the four compound extremes under SSP126 and SSP585 scenarios are also explored (table  S2). Overall, the extreme temperature is the main driver of the future trends of the four compound extremes. The IEPET is not conducive to the increasing trend of the warm/wet mode and the decreasing trends of the cold/dry and cold/wet modes, while it is beneficial to the increase of the warm/dry mode occurrences (table S2). This result is inconsistent with that obtained from the observations (table 1), implying that the IEPET changes under global warming. The climate zones over the Asian monsoon region exhibit tendencies of the shift to the arid climate zone due to global warming (Mueller and Seneviratne 2012, Son and Bae 2015, Holmes et al 2017, which may partly explain why the IEPET would change in the future. Our results show some consistencies with Zscheischler and Seneviratne (2017), who found that the interaction between drought and heatwave in the warm season would strengthen from 2001 to 2100 as compared to 1871 to 1969 based on the CMIP5 output under the RCP85 scenario. However, how the IEPET changes under global warming warrants further investigation.
The percentage changes in the occurrences of cold/dry, cold/wet, warm/dry, and warm/wet modes from 2046 to 2065 and from 2080 to 2099 under SSP126 and SSP585 scenarios over the eight Asian monsoon sub-regions with respect to the period of 1961-1980 and their model uncertainties are also investigated (figures 5 and S4). The occurrence of warm modes increases from 2046 to 2065 and from 2080 and 2099 under both SSP126 and SSP585 scenarios, while the occurrence of cold modes decreases. The percentage changes in the warm modes over the eight sub-regions and their model spreads during the two periods under the SSP585 scenario are much larger than those under the SSP126 scenario (figures 5(a), (b) and (c), (d)) while the cold modes exhibit little discrepancy between the two scenarios. In addition, the percentage changes in the four compound extremes between the two periods show much larger discrepancies under the SSP585 scenario than under the SSP126 scenario (figures 5(a), (c) and (b), (d)). In particular, the annual mean warm/wet mode depicts the largest changes and model uncertainties among the four compound extremes, especially over Southeast Asia, South Asia, the Tibetan Plateau, and Eastern Central Asia. That is, the hotspots of the warm/wet mode at the annual scale would remain unchanged. For the seasonal mean frequencies, the change in magnitude of the warm/wet mode over each sub-region is larger than that of the other three modes, and it exhibits a relatively larger seasonal diversity than the other three modes during the two future periods under the SSP126 and SSP585 scenarios ( figure S3). The warm/wet mode displays a relatively larger change in magnitude over Southeast Asia, South Asia, the Tibetan Plateau, and Eastern Central Asia in spring, autumn, and winter, whereas in summer, the warm/wet mode depicts remarkable changes over all the sub-regions from 2046 to 2065 and from 2080 to 2099 under the SSP126 and SSP585 scenarios.

Conclusions and discussion
Based on the observations and model projections, we explore the spatiotemporal characteristics and longterm trends of cold/dry, cold/wet, warm/wet, and warm/dry combinations over the Asian monsoon region. In the period of 1961-2014, the occurrences of the dry modes over the Asian monsoon region are higher than the wet modes, and the trends of the four modes are largely determined by extreme temperature. The IEPET facilitates the trends of the dry modes, and inhibits the trends of the wet modes based on the linear congruency. In the projected period, extreme temperature is also the main driver in the future trends of the four modes, but the IEPET is not conducive to the decreasing trend of the cold/dry mode, implying that the IEPET would change under global warming.
Based on the changes in the occurrences of the four modes in the periods 1995-2014, 2046-2065, and 2080-2099 with respect to 1961-1980, we find that the warm/wet mode displays a larger change in magnitude than the other three modes over the Asian monsoon region. On the annual timescale, the change magnitudes over Southeast Asia, South Asia, the Tibetan Plateau, and Eastern Central Asia are relatively larger than in the other sub-regions during both the historical and future periods. These subregions are quantified as historical and future hotspots of the warm/wet mode. On the seasonal timescale, the hotspots change with time. Our findings are critical for formulating adaptation strategies to cope with the adverse effects of compound extremes.
Several potential caveats should also be mentioned in this study. The BCSD_CMIP6 dataset shows a better performance in simulating the mean values and standard deviations of extreme precipitation and extreme temperature thresholds than the CMIP6 simulations and the NEX-GDDP-CMIP6 dataset. However, bias still exists with respect to the observations in capturing the frequencies of the four compound extremes over the Asian monsoon region (figure 6). Specifically, both the raw and statistical downscaled CMIP6 simulations can accurately capture the observed decline trend and mean value of the cold/dry mode from 1961 to 2014 ( figure S3(a)). The cold/wet mode is overestimated by the CMIP6 simulations before 1990 and underestimated after 1990 ( figure S3(b)). However, this mode is underestimated by the BCSD_CMIP6 throughout the historical period. The CMIP6 simulations underestimate (overestimate) the occurrence of the warm/dry (warm/wet) mode over the Asian monsoon region. After statistical downscaling, the model spread of the warm/dry (warm/wet) mode is reduced, but the underestimated (overestimated) bias has not been improved. Therefore, the BCSD method shows limited ability in reproducing the observed features of the four concurrent modes, and more advanced algorithms considering the relationship between precipitation and temperature need to be developed to produce more reliable datasets. Moreover, substantial studies have revealed that human influence is essential for the likelihood of a single extreme event (Min et al 2011, Fischer andKnutti 2015). However, how human influence affects the four concurrent modes over the Asian monsoon region warrants further investigations.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: www. nccs.nasa.gov/services/data-collections/land-basedproducts/nex-gddp-cmip6.