Introduction

Heatwaves and extreme rainfalls are the two extreme events with significant impacts on human health, food, and the environment1,2,3. The heatwaves refer to a prolonged period of much-warmer-than-average weather4. Prolonged heating causes the atmosphere to store energy, increasing the likelihood of extreme rainfall shortly after a heatwave ends5. Under climate change, the frequency of both heatwaves and extreme rainfall has increased6,7,8,9,10,11,12. During the summer of 2023, China experienced 14 extreme heat events, with about 70% of national weather stations recording temperatures above 40 °C13. Simultaneously, 37 extreme precipitation events occurred, with 55 stations recording unprecedented rainfall amounts14. High-frequency, widespread extreme weather events increase the interdependence between heatwaves and rainfall, causing regions to be consecutively affected over short periods15,16,17. For example, in July 2023, North China faced persistent heat followed by rare extreme rainfall, with cumulative precipitation reaching 1003 mm, resulting in significant losses18. In 2023, extreme weather events in China led to direct economic losses of 330.6 billion RMB and 537 people dead or missing14. The frequency and intensity of heatwaves and heavy rainfalls are projected to keep increasing through the end of the century19,20,21,22. Along with rapid population growth, extreme events are likely to have even stronger societal impacts in the future23,24,25,26,27.

Heatwaves and heavy rainfall have been widely studied as independent extreme events. However, recent research has confirmed the theoretical possibility of their consecutive occurrence28,29,30,31,32,33. During heatwaves, the atmosphere can store more moisture as it heats up. This moisture condenses into cloud droplets due to cooling at the end of heatwaves34,35. Prolonged thermal accumulation also provides energy for the formation of heavy rainfall, especially thunderstorms and convective precipitation, and increases instability in the upper troposphere36,37. When the heatwave ends, moisture in the lower troposphere converges and lifts above the unstable upper troposphere, resulting in heavy rainfall. These studies show a strong thermodynamic link between heatwaves and heavy rainfall28,32. At the same time, synoptic weather systems, such as fronts and low-pressure cyclones, can also lead to a spatial and temporal consecutive occurrence between heatwaves and heavy rainfall38,39,40. These systems might even terminate a heatwave, leading to heavy rainfall shortly thereafter41,42. Based on the theoretical possibility of consecutive heatwave and heavy rainfall (HW-HR) events, existing studies have explored patterns and dynamics of probability, frequency, duration, intensity, and future projections, offering a comprehensive understanding of these compound events5,15,43,44,45,46. Additionally, previous studies have examined the mechanisms and factors influencing these events by considering the characteristics of the independent HW-HR events, climate anomalies during the compound event, and human activities30,47,48,49,50,51,52. Furthermore, previous studies have also explored the spatial compounding of these two events. For example, Li et al.53 explored the connection between the heatwave in Pakistan and heavy rainfall in the middle and lower reaches of the Yangtze River as a spatially compound event, based on atmospheric circulation and tele-connection.

Compound or consecutive extreme events can have a greater social impact than single extreme events in terms of direct economic losses, energy, agricultural yields, and human health54,55,56. This is because the initial event often alters the environmental preconditions, reducing the region’s ability to resist the subsequent events31,57. In this context, the time interval between consecutive extreme events, such as abrupt transitions between dry and wet periods, has raised concerns and proven to be particularly critical58,59. If the time interval is too short, the affected region may not have sufficient time to recover from the effects of the previous event, which may exacerbate the impact of following events60,61. For instance, during July 2021 in Western Europe, a persistent and extreme heatwave was followed by intense rainfall, causing catastrophic flooding and heat stress risks for residents62,63. Although the time interval between consecutive HW-HR events is critically important due to their destructive impacts and superimposed losses, few studies have focused on this aspect49,64, as well as the changes in this interval and underlying mechanisms remain unknown. Understanding these changes as well as their societal impacts is essential for enhancing disaster warning systems, improving emergency management, and reducing climate change-related losses.

The objective of this study is to examine the changes in the time interval between consecutive HW-HR events in China from 1970 to 2019, using station observations. Additionally, we analyzed potential physical mechanisms affecting these time intervals with meteorological variables from the ERA5 reanalysis dataset. Finally, we examined changes in population exposure to consecutive HW-HR events with short time intervals over the past five decades using population data. This study provides valuable insights into compound extreme events and their societal impacts, aiding in targeted disaster response and climate change mitigation, and adaptation strategies.

Results

Characteristics and dynamics of the time interval between consecutive heatwave and heavy rainfall (HW-HR) events

Figure 1 illustrates the pattern and evolution of time intervals between consecutive HW-HR events from 1970 to 2019. The average time interval across all stations during this period is 3.32 days. Notably, 98.7% of stations have an average time interval of less than 5 days, indicating that in most consecutive HW-HR events, heavy rainfall typically occurs within 5 days after the heatwave ends (Fig. 1b). In terms of spatial distribution, there is no clear hot or cold spot. However, inland areas in Southwest China tend to have shorter time intervals. Similarly, stations in northern regions, including North and Northwest China, also exhibit shorter time intervals, generally below the national average. In contrast, Northeast and East China tend to have relatively longer time intervals (Fig. 1a and Supplementary Fig. S1). Regions with shorter intervals align with regions identified in previous studies as having a higher probability of consecutive HW-HR events47,49. This suggests that these areas may experience more frequent and rapid shifts from heatwaves to heavy rainfall. From 1970 to 2019, more than half of the stations show a decreasing trend in time intervals, with 22.5% of stations exhibiting a significant decrease. This suggests that at many stations, the interval between the end of a heatwave and the onset of heavy rainfall is becoming shorter. Conversely, 18.2% of the stations show a significant increasing trend (Fig. 1c, d). Since the number of stations with significant increases is roughly equal to those with significant decreases, there is no overall significant trend in the average time interval across all stations. However, in hotspot regions such as Northeast China, the time intervals show significant decreasing trends (Supplementary Fig. S2).

Fig. 1: Spatial distributions and trends of average time interval.
Fig. 1: Spatial distributions and trends of average time interval.The alternative text for this image may have been generated using AI.
Full size image

a Average time interval at each station during the study period from 1970 to 2019. b Statistics of station numbers based on time interval. c Trend of time interval of each station during the study period from 1970 to 2019. Large dots indicate trends that are significant at the 95% confidence level, while small dots represent trends that are not significant. d Statistics summary of station proportion based on trend significance. The abbreviations of sub-regions can be found in the section “Methods”.

We further examine the most urgent type of event: the short-time events (STEs). Figure 2b shows that the STEs account for a significantly higher proportion of events than the other two types, which are middle-time events (MTEs) and long-time events (LTEs), at most stations. At 32.8% of the stations, STEs constitute more than half of all consecutive HW-HR events. Stations with a high proportion of STEs are primarily located in Southwest China, consistent with the spatial pattern of shorter average time interval (Fig. 2a). In terms of temporal changes, 24.1% of the stations show a significant increase in the proportion of STEs from 1970 to 2019 (Fig. 2c). In Northeast, East and South China, stations with significant increases in the proportion of STEs are relatively concentrated and exhibit a higher rate of change compared to other regions (Supplementary Fig. S1). The average proportion of STEs across all stations also shows a significant increasing trend (p < 0.05), with a rate of 1.4% per decade (Fig. 2d). There is a strong negative correlation (R = −0.77, p < 0.01) between shortened time intervals and the increased proportion of STEs. Furthermore, 79.8% of stations show a consistent relationship between the trend of the average time interval and the trend of STE proportion (blue dots in Fig. 2e). Specifically, these stations are located in the second and fourth quadrants of the scatter plot, indicating that a decrease in the time interval is generally associated with an increase in the STE proportion, and vice versa. This suggests that the rise in the proportion of STEs is a key driver of the shortening time intervals.

Fig. 2: Spatial distributions and trends of short‐time events (STEs) proportion.
Fig. 2: Spatial distributions and trends of short‐time events (STEs) proportion.The alternative text for this image may have been generated using AI.
Full size image

a Proportion of STEs at each station during the study period from 1970 to 2019. b Density plot of station distributions by the proportion of STEs, medium‐time events (MTEs), and long‐time events (LTEs). The density is estimated using kernel density estimation with a Gaussian kernel function and a bandwidth of 1000 for each event type. The colored dashed lines corresponding to each event type indicate the median proportion of each type. The asterisk (*) denotes significant differences between the STEs and MTEs/LTEs, calculated using Student’s t‐test at the 95% confidence level. c Trend of STEs proportion per decade for each station during the study period from 1970 to 2019. Large dots indicate trends that are significant at the 95% confidence level, while small dots represent trends that are not significant. The inset provides a statistical summary of station proportions based on trend significance, where dark red represents a significant increase, while dark blue indicates a significant decrease. d Trend of average STEs proportion across all stations. The dotted line represents annual values, and the solid line represents the linear trend. e Correlation between the trend of STEs proportion and the average time interval. Blue dots represent stations where the two variables exhibit opposite trends. Correlation analysis is based on the Pearson correlation test. The abbreviations of sub-regions can be found in the section “Methods”.

We also examine the consecutive HR-HW events (Supplementary Fig. S3). The proportion of STEs within HR-HW events remains relatively low. At the same time, the number of stations with a high proportion of STEs and a significantly increasing trend in STEs is also relatively low compared to the HW-HR events. The relatively low frequency of consecutive HR-HW events is consistent with previous studies15,65. However, the proportion of STEs within HR-HW events still shows a significant increasing trend, further supporting our point that the connection between heatwaves and heavy rainfall has intensified under the changing climate.

It is important to note that there are four possible explanations for the increase in the proportion of STEs. First, both STEs and consecutive HW-HR events may be increasing, but the frequency of STEs increases at a faster rate than the frequency of consecutive HW-HR events. Second, it is possible that while the frequency of STEs increases, the total frequency of consecutive HW-HR events remains unchanged, leading to a higher proportion of STEs. Third, STEs could increase even if the total frequency of consecutive HW-HR events decreases. In this case, the proportion of STEs rises not due to an increase in STE frequency, but because of a reduction in consecutive HW-HR events. Finally, both STEs and consecutive HW-HR events may decrease, but if STEs decline at a slower rate than total consecutive HW-HR events, the proportion of STEs would still increase. Supplementary Fig. S4 shows the frequencies of STEs, MTEs, LTEs, and all consecutive HW-HR events. Over the past five decades, all four event types have exhibited a significant increasing trend, suggesting that the increase in the proportion of STEs is primarily due to their faster rate of increase compared to all consecutive HW-HR events.

Attribution of the increasing proportion of short-time events (STEs)

Heavy rainfall events have a strong temperature dependency66. Under global warming, the frequency and intensity of independent HW-HR events have been shown to increase67. The significance test proves that the frequency of consecutive HW-HR events is significantly higher than the 95% confidence interval estimates derived from moving-blocks bootstrap resampling65,68, which indicate that the occurrence of consecutive HW-HR events is triggered by underlying mechanisms rather than random chance (Supplementary Fig. S5). Heatwaves significantly modify atmospheric thermodynamic conditions, thereby promoting the occurrence of extreme precipitation events49. Elevated levels of atmospheric energy and moisture increase instability36,47,69,70. During heatwaves, high temperatures and reduced cloud cover result in strong thermal forcing, leading to substantial energy accumulation in the atmosphere71,72,73. As the heatwave ends, this accumulated energy is released, creating conditions more conducive to convection and even thunderstorms32. According to the Clausius–Clapeyron relationship, the moisture-holding capacity of the atmosphere increases by about 7% for every 1 °C rise in temperature29,35. Therefore, moisture from enhanced evaporation and convergence during heatwaves is retained in the atmosphere. Once the heatwave ends and temperatures decline, the stored moisture is lifted to the condensation level, triggering deep convection and intense precipitation30,35,70, thereby enhancing the likelihood of extreme precipitation events in the period following the end of the heatwave45,74. All these factors may have shortened the time interval between heatwaves and heavy rains, increasing the proportion of STEs. To assess the contribution of multiple factors to the increase in the STEs proportion, we employed an XGBoost (eXtreme Gradient Boosting) classifier model combined with SHAP (SHapley Additive explanation) to investigate the impact of atmospheric variables and dominant weather types on the occurrence of STEs. (Supplementary Text S1 and Fig. S6).

Existing studies have demonstrated the impact of convective available potential energy (CAPE) anomalies on consecutive HW-HR events69. During heatwaves, rising temperatures lead to energy accumulation in the atmosphere, causing an increase in CAPE75. After the heatwave ends, CAPE generally decreases, often resulting in negative ΔCAPE values. Our study finds that a higher value of ΔCAPE also accelerates the transition from heatwaves to heavy rains. The 10-year moving average of ΔCAPE anomaly showed a significant increasing trend at most stations from 1970 to 2019 (Fig. 3a). The yearly average of ΔCAPE also indicates a significant increasing trend over the past five decades, with STEs exhibiting notably higher ΔCAPE values than LTEs (Fig. 3b). The larger ΔCAPE in STEs suggests that CAPE remains relatively high even after the heatwave ends, leading to a more unstable atmosphere and increasing the likelihood of heavy rainfall occurring within a short interval76. Similarly, the spatial pattern of 2 m temperature (T2M) anomalies change in Fig. 3c shows a significant increase in T2M anomalies during the interval periods. T2M anomalies exhibited a notable upward trend, with STEs having significantly higher T2M anomalies than LTEs (Fig. 3d). Higher T2M anomalies in STEs indicate that even after the heatwave ends, the atmosphere retains a high temperature and stored energy, which enhances atmospheric instability and provides favorable conditions for heavy rainfall formation77. These findings suggest that under the changing climate, atmospheric energy accumulation after heatwaves has been increasing, leading to greater atmospheric instability during the interval in STEs compared to LTEs.

Fig. 3: Trend patterns and yearly trend of anomalies for four atmospheric variables.
Fig. 3: Trend patterns and yearly trend of anomalies for four atmospheric variables.The alternative text for this image may have been generated using AI.
Full size image

The panels a, c, e, g show the trend patterns and the panels b, d, f, h show the yearly trends. a, b Convective Available Potential Energy (CAPE, J/kg), representing the change of CAPE between interval average and the end of heatwave, c, d 2 m temperature (T2M, °C), e, f Relative Humidity (RH, %), and g, h Evaporation (ET, mm) during consecutive heatwave and heavy rainfall events. Only stations with significant changes at the 95% confidence level are shown in the trend patterns. The abbreviations of sub-regions can be found in the section “Methods”.

Besides energy and heat, the atmospheric moisture conditions also play a crucial role in the development of heavy rainfall. Changes in relative humidity (RH) and evaporation (ET) anomalies reflect these moisture conditions (Fig. 3e–h). High temperatures during a heatwave remove a large amount of soil moisture. After the heatwave ends, less soil moisture is available for evaporation. As the duration and magnitude of heatwaves increased, both STEs and LTEs experienced a significant decline in ET anomalies during the interval period78. However, RH anomalies during the interval do not exhibit a significant trend (Fig. 3f). Notably, RH anomalies in STEs intervals are significantly higher than in LTEs intervals, indicating that despite reduced soil evaporation, the atmosphere retains substantial moisture during STEs intervals, which provide conditions for heavy rainfall formation79. This may be due to the higher ET during the STEs intervals compared to LTEs, which contributes additional moisture to the atmosphere (Fig. 3h). Moreover, convective conditions may further enhance atmospheric moisture content through moisture convergence80, as the average moisture convergence on the horizontal level during STEs intervals shows a significantly increasing trend (Supplementary Fig. S7). Spatially, in the hotspots where the STEs proportion has increased significantly, such as South China, the RH and ET anomalies even show a pattern of localized aggregated increase (Fig. 3e, g). This suggests that in STEs-prone regions, more moisture is retained in the atmosphere after a heatwave, maintaining evaporation intensity and providing moisture for extreme precipitation within short intervals.

Assessment of population exposure to short-time events (STEs)

The increasing proportion of STEs raises risks to human health and society. On one hand, people are exposed to rapid shifts in temperature and humidity, which can predispose them to neurological and psychiatric diseases81,82. On the other hand, heatwaves often coincide with droughts, leading to increased water repellency in soil and reduced infiltration27,83,84. Extreme precipitation following a heatwave can trigger flooding, and the shortened interval between extreme events presents a greater challenge to society’s ability to respond to floods85.

In the past five decades, the population exposed to STEs in China has increased more than fivefold, rising from 87.7 million persons · events in 1975 to 585.5 million persons · events in 2020. The exposure proportion also surged from 9.5% to 40.8% (Fig. 4c). A total of 84.7% of stations show an increasing trend, covering 75.7% of the entire country’s area (Fig. 4b). The stations with rising exposure are widely distributed, with the highest rates of change observed in the South (852.9%) and East China (478.0%) (Supplementary Fig. S8). Notably, eight urban agglomerations have experienced above-average increases in exposure (Fig. 4d), including the Yangtze River Delta (YRD) and Pearl River Delta (PRD), China’s two largest urban agglomerations, where both population and STE frequency have risen sharply.

Fig. 4: Dynamics of short‐time events (STEs) exposure from 1975 to 2020.
Fig. 4: Dynamics of short‐time events (STEs) exposure from 1975 to 2020.The alternative text for this image may have been generated using AI.
Full size image

a Spatial pattern of changes in STEs exposure. b Area proportion of STEs exposure changes. c Dynamics of total population exposed to STEs (bars) and proportion of population exposed to STEs (dotted line), and d Changes in STEs exposure in the 20 major urban agglomerations of China (see Supplementary Table S1 for details). The abbreviations of sub-regions can be found in the section “Methods”.

Increased exposure to extreme weather events can be influenced by both climate and population factors86. To assess their respective contributions, we further examine the effects of population growth and the rise in STEs (Supplementary Text S2 and Fig. S9). The results indicate that the increase in STEs is the dominant factor, contributing 65.3% to the rise in exposure. Despite rapid population growth over the past five decades, its direct contribution remains relatively low (8.2%) due to the rarity of STEs in the early years of the study period. However, the interaction between population growth and climate change plays a significant role, contributing 26.5% to the overall increase. In the agglomerations experiencing both rapid population growth and intensified climate change, such as PRD and YRD, the interaction effects can be particularly high, reaching 66.1% and 44.7%, respectively.

Discussion

While the characteristics of consecutive HW-HR events have been extensively studied, less attention has been given to the dynamic evolution of time intervals between these consecutive events in a changing climate. This study focuses on the changes in the time interval between consecutive HW-HR events in China over a 30-year period, using meteorological observations. By integrating meteorological variables and socio-economic data, we reveal the changing characteristics, underlying mechanisms, and social impacts of the changing time interval between consecutive HW-HR events under climate change. Understanding these changes is crucial for mitigating social risks associated with compounded extreme events.

Our study found that 24.1% of the stations in China exhibited a significant decreasing trend in the average time interval between consecutive HW-HR events from 1970 to 2019. The increase in the proportion of STEs (1–2 days) is an important reason for the shortening of the average time intervals. The proportion of STEs increased significantly at a rate of 1.4% per decade on a national scale, with hotspots increasing at rates of 5.3% per decade in Northeast China, 3.1% per decade in South China, and 2.2% per decade in East China. Shifts in anomalies of atmospheric variables lead to changes in energy and moisture during consecutive HW-HR events, thereby contributing to the increased proportion of STEs. Compared to LTEs, STEs experienced a significant increase in CAPE and T2M, as well as higher positive anomalies in RH and ET, presenting conditions that were more conducive to the formation of extreme precipitation at the end of the heatwave. With urbanization and increased population, the exposure to STEs in China has increased by more than fivefold over the past five decades, with more than three-quarters of the areas showing an increase in exposure. The rising frequency of STEs was mainly responsible for this increase in exposure, with a contribution rate of 65.3%.

Thresholds used to identify extreme events can affect their characteristics. To ensure the robustness of our results, we examined outcomes using different definitions. Heatwave thresholds were set at the 85th, 90th, and 95th percentiles, while heavy rainfall thresholds were at the 90th, 95th, and 99th percentiles. Supplementary Fig. S10 shows the results of sensitivity experiments. Although there were variations in the number of events detected, the conclusions of our study remained consistent. We define consecutive events as those with a total time interval of less than 7 days between a HW-HR, and STEs as the consecutive events with time intervals of 1–2 days. This classification may also introduce some uncertainty. Therefore, we tested different combinations of total time intervals and STEs thresholds to further assess sensitivity (Supplementary Fig. S11). Under various combinations of STEs thresholds and total time intervals, most of the proportion of STEs showed a significant increasing trend. The sensitive experiments indicate that the time interval between consecutive HW-HR events is shortening in a changing climate.

Considering only consecutive HW-HR events occurring at the same station is the ideal scenario. In reality, consecutive HW-HR events may occur at larger spatial domains87. Our study provides a preliminary exploration of spatially consecutive HW-HR events (Supplementary Text 3 and Fig. S12). We observed that within a 1° spatial buffer, the frequency of spatially consecutive HW-HR events is significantly lower than the frequency of consecutive HW-HR events occurring at the same station. Moreover, spatially consecutive events did not exhibit distinct hotspot areas, which demonstrated the significance of our study even without considering spatially consecutive HW-HR events. However, the characteristics of spatially consecutive HW-HR events deserve further exploration, as they are influenced by atmospheric motions that could display notable trends under climate change.

The shortening of the time interval between consecutive HW-HR events may result from the interaction between multiple factors. Previous studies have proved that urbanization may cause impacts on extreme events by altering local microclimates and land surface properties88,89,90,91. Atmospheric circulation and synoptic weather systems can also intensify the linkage between the HW-HR. Additionally, topography also has significant effects on the formation of convective precipitation70,92. To better separate the impact of each factor on the time interval, it is essential to use multiple sources of data, such as high-resolution satellite remote sensing data for precipitation and temperature, as applied in studies of independent HW-HR events50,93,94. In addition, heatwaves may influence atmospheric chemical compositions and processes, potentially accelerating subsequent heavy rainfall. During heatwaves, atmospheric aerosol concentrations and properties change, possibly affecting atmospheric moisture and energy transport processes95,96,97. However, the microscopic mechanisms by which aerosols influence consecutive HW-HR events and their time intervals are currently unclear. Future studies could explore the interactive feedback between meteorological variables and aerosols using regional chemistry transport models such as WRF-Chem.

Our study focuses on the dynamic evolution of the time intervals between consecutive HW-HR events and their social implications. The results have practical implications for climate change policy and urban development planning. On one hand, we observe an enhanced dependence between temperature extremes and heavy rainfall events, along with a shortening of the interval, which highlights the urgent need for policymakers to address the conditions under which heavy rainfall occurs shortly after a heatwave. This necessitates enhancing early warning systems, improving public awareness, and facilitating timely evacuation plans to minimize casualties and property damage. On the other hand, our study reveals that the combined effects of climate change and urbanization are increasing population exposure to consecutive events in large urban agglomerations, such as the YRD and PRD. This implies that more urbanized areas are facing greater risks from extreme events. To address these risks, urban planning should integrate climate change adaptation strategies, such as developing resilient infrastructure and improving drainage systems. Furthermore, rationalizing the spatial distribution of urban populations can help reduce exposure to natural disasters. Policies such as decentralizing urban centers, promoting sustainable urbanization, and encouraging the development of smaller cities can alleviate the pressure on megacities. By aligning urban development policies with climate risks, the resilience of urban agglomerations to extreme events can be significantly enhanced.

Methods

Data

The daily maximum temperature data and daily precipitation data are obtained from the Resource and Environmental Science Data Platform (https://www.resdc.cn/), covering the period from 1970 to 2019. This dataset has been widely used in climate change studies across mainland China29,98. The daily records have been processed by the China National Meteorological Information Center of the China Meteorological Administration to ensure quality and homogeneity99. The original dataset includes a total of 2474 stations. To maintain completeness and continuity in the historical records, we apply selection criteria based on WMO standard practices100. First, stations with missing data for 3 consecutive years are excluded. Second, a month is considered unavailable if there are more than 11 missing values in total or if there are 5 consecutive missing values. Stations with more than 5 unavailable months during the study period are also excluded. After applying these criteria, 2001 stations are retained for the study.

The atmospheric variables used for attribution are derived from the European Centre for Medium-Range Weather Forecasts Reanalysis 5 (ERA5) hourly data on single levels (https://cds.climate.copernicus.eu/datasets/reanalysis-era5-single-levels?tab=download). The spatial resolution is 0.25° × 0.25°, and we select the nearest grid point to each station to represent the local climate conditions. The time resolution has been aggregated from hourly to a daily scale.

The population data used for exposure calculations are obtained from the Global Human Settlement Layer dataset101. The spatial resolution of the data is 30 × 30 arcseconds (~900 m near the equator) and covers the period from 1975 to 2020 at 5-year intervals.

The administrative boundary data was obtained from the National Geomatics Center of China (http://www.ngcc.cn/ngcc/). The study area was divided into seven sub-regions based on geographical location and socio-economic conditions: Northeast China (NE), North China (NC), Northwest China (NW), East China (EC), Central China (CC), Southwest China (SW), and South China (SC). The boundaries of these sub-regions, along with station locations and population information, are presented in Supplementary Fig. S13.

Identification and characterization of the time interval between consecutive heatwave and heavy rainfall (HW-HR) events

As depicted in Fig. 5, consecutive HW-HR events refer to the consecutive occurrence of HW-HR events. Heatwave is identified when the daily maximum temperature exceeds a threshold (the 90th percentile in this study) for at least 3 consecutive days during the extended summer season (May to September). To minimize the impact of station location and seasonality, the threshold is calculated for each station and each calendar day29,102. Specifically, for each station, we select the 7 days before and after each calendar day (a 15-day window) within the reference period (1970–1989). This results in 300 daily maximum temperature samples (15 days × 20 years). The 90th percentile of these 300 samples is then used as the threshold for that station and calendar day. A heatwave is considered to end when the daily maximum temperature falls below the threshold for 2 consecutive days, in order to exclude the influence of data recording biases103. Heavy rainfall is defined as that where daily precipitation exceeds the 95th percentile of the baseline period and occurs within 7 days after the end of a heatwave. If the total time interval between HW-HR exceeds 7 days, the events are treated as independent64. In this study, we focus on the time interval between the end of a heatwave and the first occurrence of heavy rainfall (Fig. 5). Concurrent HW-HR events are not considered in this study.

Fig. 5: Schematic for identifying the time interval between consecutive heatwaves and heavy rainfall events.
Fig. 5: Schematic for identifying the time interval between consecutive heatwaves and heavy rainfall events.The alternative text for this image may have been generated using AI.
Full size image

Each bar represents a one‐day period. Tmax and P represent the maximum temperature and precipitation, respectively. Concurrent events (shown as the first blue bar above the threshold line) are not considered in this research.

After identifying all consecutive HW-HR events, we categorize them into three groups based on their time intervals. Specifically, events with an interval of 1 or 2 days are classified as STEs, events with 3 or 4 days are classified as MTEs, and events with an interval of 5–7 days are classified as LTEs. By comparing the meteorological variable anomalies across the different event types, we identify significantly higher energy and moisture levels in STEs (Fig. S14). These findings support the physical mechanism for categorizing and analyzing different types of consecutive HW-HR events. Sensitivity tests regarding the definition of these three types of events are provided in the Discussion and Supplementary Information. The time interval for each station is calculated for each 5-year time window. Trend analysis is performed using linear regression, and the significance of the trend values is tested using the Mann–Kendall test.

Attribution of the shortened time intervals

The accelerated transition from heatwave to heavy rainfall may be influenced by various atmospheric variables. To identify the most influential factors, we apply the XGBoost (eXtreme Gradient Boosting) classifier model104 combined with SHAP (SHapley Additive explanation) to rank the contribution of relevant factors (see Supplementary Text S1). Based on the results of the XGBoost model, we select four atmospheric variables: CAPE, T2M, RH, and ET. We use the yearly trend to analyze the anomaly changes during the interval of STEs and LTEs, therefore, to compare the potential contributions of these variables to trigger the more frequent STEs. The yearly trend is calculated by the Mann–Kendall test.

We also conduct regional analysis to ensure whether the most influential atmospheric variables are consistent across different regions in China. The results in Fig. S15 show that, although there are slight variations in the ranking of contributing factors across different regions, ΔCAPE, RH anomalies, and T2M anomalies consistently play dominant roles.

Calculation of the exposure to the short-time events (STEs)

We use the number of people exposed to extreme events to examine the evolution of exposure from 1975 to 2020 for every 5 years (i.e., 1975, 1980, 1985, 1990, 1995, 2000, 2005, 2010, 2015, and 2020), based on the availability of population data. Accordingly, for each available year, STEs frequency is calculated as the average of 2 years prior to and the 2 years following the target year. The number of people exposed to extreme events is determined based on the risk assessment framework105 and is defined as Eq. (1):

$${Exposure}=\mathop{\sum }\limits_{i=1}^{n}{{POP}}_{i}\times {{STEF}}_{i}$$
(1)

where \({{STEF}}_{i}\) represents the frequency of STEs in station i. \({{POP}}_{i}\) represents the number of people within the areas affected by station i. The affected area for each station is defined by the Thiessen polygon method, which is commonly used to extend the observation data to ranges99. After building the Thiessen polygons, we apply the zonal statistics methods to aggregate the population data from grid cells to polygons.