Asymmetric intensification of heatwave-ozone compound extremes across China at 1 km resolution from 2000 to 2023

Abstract

Compound heatwave and ozone pollution events threaten public health as warming and photochemistry intensify. Yet the internal architecture of these events—how heatwaves and ozone separately influence frequency versus intensity—remains poorly resolved at national scale. Here we show, using 1 km resolution grids across China from 2000 to 2023, that heatwaves dominate the frequency of compound extremes while ozone dominates their intensity, and that this asymmetry strengthened after 2016. Population exposure rose primarily because the events themselves intensified, rather than from population growth. City-level influencing factors proved highly heterogeneous, reflecting local combinations of meteorology, pollution and socioeconomic conditions. These findings uncover an asymmetric architecture of compound risk and identify distinct levers—heat versus ozone—for targeted mitigation under continued warming.

Graphical abstract

Highlights

• •Heatwaves dominate frequency while ozone dominates intensity of compound extremes.
• •This frequency–intensity asymmetry strengthened after 2016.
• •National 1 km records from 2000 to 2023 reveal the asymmetric pattern across China.
• •Population exposure rose mainly from event intensification, not population growth.
• •Findings identify distinct levers—heat versus ozone—for targeted risk reduction.

1. Introduction

The escalating trend of global warming has led to a surge in heatwaves and ozone (O₃) pollution, raising urgent concerns about their pronounced adverse effects on human health [1]. In recent decades, the frequency and intensity of heatwaves have increased dramatically. The Intergovernmental Panel on Climate Change reported that the occurrence of once-in-a-decade heatwave events has risen to 1.8–3.2 times that of the 1850–1900 period [2]. Surface O₃ concentrations have climbed 3 × 10⁻⁴ μg m⁻³ per year over the past two decades, driven by intensified photochemical reactions under rising temperatures and anthropogenic emissions [3]. These dual threats often co-occur and reinforce each other. Elevated temperatures not only enhance the release of O₃ precursors, including biogenic volatile organic compounds (VOCs) and soil-released nitrogen oxides (NO_x), but also accelerate photochemical reactions and disrupt atmospheric equilibrium, which promote O₃ formation and accumulation [4]. As a major greenhouse gas, O₃ was specifically selected among common air pollutants because it in turn amplifies heatwaves through longwave radiative trapping and the deepening of isopycnals [5], exhibits the strongest thermally driven response, and frequently peaks concurrently with heatwave events. This tight coupling provides a clear mechanistic basis for defining compound heatwave and O₃ pollution events (CHOEs), enabling a focused investigation of their spatiotemporal distribution and population exposure as a scientifically coherent environmental stressor. In addition, the co-occurrence of heatwaves and O₃ pollution has been widely reported to pose serious threats to both public health and ecological stability [6,7]. Identifying the spatiotemporal distribution and population exposure to CHOEs is essential for mitigating ecosystem stress and reducing disease burdens.

CHOEs have been increasing annually worldwide, with a significant rise detected in mid-latitude regions, including China [8,9]. To date, only a limited number of studies have investigated the occurrence of CHOEs at the city level in China, which are generally defined as the co-occurrence of a heatwave and an ozone pollution event on the same day, with minor variations in thresholds and duration across studies [7,10,11]. CHOEs were more prevalent in cities with a large population (≥10 million), arid climate, and high solar irradiance [10]. For 1028 Chinese counties, the annual occurrence of CHOEs was reported to be approximately 3500–3700 days during 2017–2019 [7]. Although studies that relied on city-level aggregated data or specific regions have provided valuable insights, they did not fully capture fine-scale spatial patterns. The limited grid-scale research that does exist is often constrained to discrete study years or narrow geographic scopes, restricting the generalizability of the findings [[12], [13], [14]]. Moreover, these studies have understandably prioritized the occurrence of CHOEs in the absence of corresponding intensity metrics [10,11]. However, frequency alone may not adequately reflect the actual burden of CHOEs. Some regions may experience frequent, yet mild, events, whereas others face infrequent but highly intense episodes with potentially greater health implications. Incorporating intensity metrics is therefore essential for a more comprehensive understanding of the severity and public health relevance of CHOEs. In addition, only one study incorporated population data to assess future exposure levels to CHOEs [15]. Although such a projection offered important perspectives on potential health risks, understanding the spatiotemporal patterns of historical population exposure remains essential yet largely overlooked.

To address these gaps, we conducted a nationwide assessment of CHOEs in China from 2000 to 2023 at high spatial resolution (1 km × 1 km), providing fine-scale insight into their spatiotemporal dynamics. Drawing inspiration from quantitative approaches used to assess heatwave intensity [9], our study introduced a new dimension by evaluating the intensity of CHOEs, thus providing fresh insight into their severity. We quantified the spatiotemporal distribution and population exposure levels of CHOEs in frequency, duration, and intensity for each year. To better understand the spatiotemporal changes, we elucidated the proportion of heatwaves and O₃ pollution to CHOEs in frequency and intensity and decomposed the impacts of demographic dynamics and event intensification on population exposure trends. We additionally applied a spatiotemporal random forest model to identify factors influencing CHOEs, spanning population density, urban infrastructure, green space coverage, economic indicators, healthcare resources, particulate matter smaller than 2.5 μm (PM_2.5), and key meteorological variables including radiation, pressure, precipitation, and wind. This integrated approach, combining high-resolution spatial data, long-term trends, refined exposure metrics, and analysis of influencing factors, offers a comprehensive perspective on the patterns, determinants, and public health implications of CHOEs across China. We aim to systematically characterize the spatiotemporal dynamics of CHOEs, quantify population exposure, and identify key environmental and socioeconomic influencing factors, with the goal of informing mitigation strategies and public health planning.

2. Methods

2.1. Study area

China is located in East Asia and covers approximately 9.6 million km², extending from 73° to 135° E and from 3° to 54° N [16]. China is officially divided into 34 province-level administrative divisions, which are typically categorized into seven distinct geographical regions: Northeast China, Northwest China, Southwest China, North China, Central China, South China, and East China. We also considered five major urban clusters that have been widely examined in environmental studies: the Yangtze River Delta, the Guangdong–Hong Kong–Macao Greater Bay Area (hereafter Greater Bay Area), the Beijing–Tianjin–Hebei Region, the Chengdu–Chongqing Region and the Triangle of Central China [17,18]. These regions differ substantially in climate, socioeconomic development and population distribution, which may influence the occurrence and impacts of CHOEs. Details of these subregion classifications in China are provided in Supplementary Text S1.

2.2. Data sources

Meteorological data were obtained from ERA5-Land, the land component of the fifth-generation reanalysis product ERA5 developed by the European Centre for Medium-Range Weather Forecasts through its Integrated Forecasting System (https://www.ecmwf.int/en/era5-land), with a spatial resolution of 0.1° × 0.1° (approximately 9 km × 9 km). This dataset incorporated both direct observations from ground-based meteorological stations and indirect observations from satellites. We used daily mean 2 m air temperature and daily mean 2 m dewpoint temperature, which were calculated by interpolating between the lowest model level and the Earth's surface while accounting for atmospheric conditions [19].

Surface O₃ and PM_2.5 data were obtained from our previously validated ChinaHighAirPollutants (CHAP) dataset (https://geodata.nnu.edu.cn/featured_data.html), which provides long-term, spatially continuous, high-resolution estimates of ground-level air pollutants across China. For O₃, CHAP employed an advanced four-dimensional spatiotemporal deep forest model, integrating national surface O₃ observations, atmospheric reanalysis data, satellite remote sensing products, and an emission inventory, with input predictors including surface downward shortwave radiation, air temperature, boundary-layer height, relative humidity, surface pressure, normalized difference vegetation index, and surface elevation. Based on these variables, the model estimated surface daily maximum 8-h average O₃ (MDA8 O₃) concentrations at a spatial resolution of 1 km × 1 km, achieving a cross-validated coefficient of determination (R²) of 0.9 and root mean square error (RMSE) of 15.8 μg m⁻³ for the period after 2013. Independent validation against Tropospheric Ozone Assessment Report measurements for the pre-2013 period further confirmed reliability, with daily, monthly, and annual correlation coefficients of 0.8, 0.9, and 0.9, respectively [20,21].

The Landscan Global Population Dataset (https://landscan.ornl.gov) provided population data at a spatial resolution of 1 km × 1 km. Developed by the Oak Ridge National Laboratory of the United States Department of Energy, LandScan integrates remote sensing satellite imagery, population census data, road network information, and urban construction data. It applies image analysis, multivariate dasymetric modeling, and machine-learning techniques to generate high-resolution estimates of population distribution and has been widely used in demographic research [21].

Socioeconomic indicators, including city-level population density, per capita road area, green coverage of built-up areas, urban construction land area, per capita gross domestic product (GDP), number of large-scale industrial enterprises, total retail sales of consumer goods, and hospital bed capacity, were obtained from the China Urban Statistical Yearbook and the China Urban Construction Statistical Yearbook. These datasets integrate official city-level survey data, administrative records, and urban infrastructure statistics, providing comprehensive coverage of socioeconomic and urban development characteristics across Chinese cities. The data are systematically compiled and quality-controlled, and have been widely used in previous studies [[22], [23], [24]].

2.3. Definition of CHOEs

A unified definition of a heatwave has yet to be established [25]. Extensive epidemiological studies have demonstrated that human adaptability to heat varies across locations with different climate conditions [26]. To better capture local meteorological and health-related characteristics, we defined heatwaves at the grid level, considering both the relative temperature threshold and duration. We used the heat index (HI), which incorporates air temperature and relative humidity, as the temperature indicator, providing a more accurate representation of human thermal stress than air temperature alone [27]. Details of the HI calculation are provided in Supplementary Text S2.

At each grid cell, we defined the heat threshold as the 90th percentile of daily mean HI during 1961–1990 (Supplementary Fig. S1) [28]. A heatwave was identified when the daily mean HI reached or exceeded this threshold for three or more consecutive days [29]. We selected 1961–1990 as a fixed reference period for long-term climate comparisons, consistent with the guidance from the World Meteorological Organization [28].

For O₃, we used the World Health Organization air quality guideline value (100 μg m⁻³) as the threshold for the MDA8 O₃ concentration, which also served as the Grade I limit standard in China's Ambient Air Quality Standards (GB 3095-2026) [30]. Accordingly, O₃ pollution was identified at a given grid cell on days when MDA8 O₃ concentration ≥100 μg m⁻³ [9]. CHOEs were then identified when both heatwave and O₃ pollution criteria were met at the same grid cell on the same day [8].

We then masked and resampled the multisource data to construct a grid dataset on daily CHOEs across China during 2000–2023 at a spatial resolution of 1 km × 1 km. We first identified heatwave events at the original ERA5-Land resolution of approximately 9 km and then resampled the binary identification results to 1 km resolution to match the O₃ and population grids.

2.4. Characteristics of CHOEs

We quantified the annual frequency, duration, and intensity of CHOEs using six metrics: event number (N), total duration (D_t), longest duration (D_l), average duration (D_a), cumulative intensity (I_c), and average intensity (I_a). We also derived population-exposure metrics for D_t, D_a, or I_c by multiplying each metric by the population of the corresponding grid cell [31].

For each grid cell and year, D_t is the total number of CHOE days and D_l is the longest event duration. The average duration is:

$$\begin{array}{c}{D}_{\mathrm{a}}=\frac{{\sum }_{i=1}^N{D}_i}{N}\end{array}$$ (1)

where $D_i$ is the duration of event i and N is the number of CHOEs.

We combined heatwave and O₃ pollution intensities after independently normalizing their threshold exceedances across all CHOE days within each grid cell and year. The normalized heatwave intensity is:

$$\begin{array}{c}{T}_{ij}^{\prime }=\frac{\left(T_{ij}-T_{\mathrm{R}}\right)-\underset{1\le x\le D_{\mathrm{t}}}{\min }\left(T_x-T_{\mathrm{R}}\right)}{\underset{1\le x\le D_{\mathrm{t}}}{\max }\left(T_x-T_{\mathrm{R}}\right)-\underset{1\le x\le D_{\mathrm{t}}}{\min }\left(T_x-T_{\mathrm{R}}\right)}\end{array}$$ (2)

where $T_{ij}$ is the mean HI on day j of event i, $T_{\mathrm{R}}$ is the relative threshold at the 90th percentile of daily mean HI during 1961–1990, and $T_x$ is the mean HI on the xth CHOE day among all CHOE days within the corresponding grid cell and year. $T_{ij}^{\prime }$ is the min–max-normalized heatwave intensity on day j of event i.

For O₃ pollution, the normalized O₃ intensity is:

$$\begin{array}{c}{O}_{ij}^{\prime }=\frac{\left(O_{ij}-{\mathrm{O}}_{\mathrm{R}}\right)-\underset{1\le x\le D_{\mathrm{t}}}{\min }\left(O_x-{\mathrm{O}}_{\mathrm{R}}\right)}{\underset{1\le x\le D_{\mathrm{t}}}{\max }\left(O_x-{\mathrm{O}}_{\mathrm{R}}\right)-\underset{1\le x\le D_{\mathrm{t}}}{\min }\left(O_x-{\mathrm{O}}_{\mathrm{R}}\right)}\end{array}$$ (3)

where $O_{ij}$ is the MDA8 O₃ concentration on day j of event i, $O_{\mathrm{R}}$ is the fixed threshold of 100 μg m⁻³, and $O_x$ is the MDA8 O₃ concentration on the xth CHOE day among all CHOE days within the corresponding grid cell and year. $O_{ij}^{\prime }$ is the min–max-normalized O₃ pollution intensity on day j of event i.

We defined cumulative intensity as the sum of the normalized heatwave and O₃ intensities across all CHOE days:

$$\begin{array}{c}{I}_{\mathrm{c}}={\sum }_i^N{\sum }_j^{D_i}{T}_{ij}^{\prime }+{\sum }_i^N{\sum }_j^{D_i}{O}_{ij}^{\prime }\end{array}$$ (4)

Average intensity was obtained by dividing I_c by the total number of CHOE days:

$$\begin{array}{c}{I}_{\mathrm{a}}=\frac{{\sum }_i^N{\sum }_j^{D_i}{T}_{ij}^{\prime }+{\sum }_i^N{\sum }_j^{D_i}{O}_{ij}^{\prime }}{{\sum }_{i=1}^N{D}_i}\end{array}$$ (5)

2.5. Statistical analysis

2.5.1. Spatiotemporal analysis

Spatial autocorrelation is a foundational concept in geography and statistics that provides insights into how values cluster across regions [32]. We used Moran's I index to identify the spatial autocorrelation of CHOEs in China. The global Moran's I index (I_G) was applied to quantify the degree of spatial correlation among units as a whole [33]. I_G ranges from −1 to 1, where values > 0 suggest positive spatial autocorrelation, values < 0 indicate negative spatial autocorrelation, and values = 0 reflect a random distribution of variables. I_G captures overall trends but fails to reveal the degree of spatial clustering among neighboring units. To overcome this limitation, local Moran's I index (I_L) offers a more nuanced perspective, allowing us to gain insights into spatial clustering and differentiation characteristics of variables [34]. We used the Z-score (Z_I) to quantify the statistical significance of spatial autocorrelation for each unit. Four types of spatial clustering were identified based on the range of I_L and Z_I. The details of the spatial clustering classification are provided in Supplementary Text S3.

We further conducted nonparametric Mann–Kendall tests, with and without prewhitening, along with Sen's slope, to assess long-term trends in CHOEs in China from 2000 to 2023, which are recognized for their robustness in trend detection [35,36].

2.5.2. Proportion of heatwave and O₃ pollution to CHOEs

We examined the associations between heatwaves and O₃ pollution and CHOE total duration and cumulative intensity. For total duration, the heatwave co-occurrence proportion was calculated as the number of CHOE days divided by the number of heatwave days. The O₃ co-occurrence proportion was calculated analogously as the number of CHOE days divided by the number of O₃ pollution days.

We then decomposed CHOE cumulative intensity into its heatwave and O₃ components. $P_{\mathrm{H}}$ denotes the percentage contribution of normalized heatwave intensity to CHOE cumulative intensity and was calculated as:

$$\begin{array}{c}{P}_{\mathrm{H}}=\frac{{\sum }_i^N{\sum }_j^{D_i}{T}_{ij}^{\prime }}{{\sum }_i^N{\sum }_j^{D_i}{T}_{ij}^{\prime }+{\sum }_i^N{\sum }_j^{D_i}{O}_{ij}^{\prime }}\times 100\%\end{array}$$ (6)

$P_{\mathrm{O}}$ denotes the percentage contribution of normalized O₃ pollution intensity to CHOE cumulative intensity and was calculated as:

$$\begin{array}{c}{P}_{\mathrm{O}}=\frac{{\sum }_i^N{\sum }_j^{D_i}{O}_{ij}^{\prime }}{{\sum }_i^N{\sum }_j^{D_i}{T}_{ij}^{\prime }+{\sum }_i^N{\sum }_j^{D_i}{O}_{ij}^{\prime }}\times 100\%\end{array}$$ (7)

By definition, $P_{\mathrm{H}}$ and $P_{\mathrm{O}}$ sum to 100%.

2.5.3. Decomposition of population exposure variation

Variation in population exposure to CHOE total duration, average duration, and cumulative intensity arises from changes in both event indicators and population distribution. We decomposed this variation into an event effect, representing changes in the event indicator with population held constant; a population effect, representing changes in population with the event indicator held constant; and a combined effect, representing simultaneous changes in both the event indicator and population [37]:

$$\begin{array}{c}\Delta X=E_{\mathrm{b}}\times \Delta P+P_{\mathrm{b}}\times \Delta E+\Delta E\times \Delta P\end{array}$$ (8)

where $\Delta X$ is the total change in population exposure between the base year (2000) and each subsequent year during 2001–2023, $E_{\mathrm{b}}$ and $P_{\mathrm{b}}$ denote the event indicator and population in 2000, respectively, whereas $\Delta E$ and $\Delta P$ denote their respective changes relative to the base year. The population effect ($E_{\mathrm{b}}\times \Delta P$), event effect ($P_{\mathrm{b}}\times \Delta E$), and combined effect ($\Delta E\times \Delta P$) together constitute the total change in population exposure.

2.5.4. Spatiotemporal random forest analysis

We applied a spatiotemporal random forest (STRF) model to systematically quantify the relative importance of multiple influencing factors to CHOEs. Influencing factors comprised three categories: meteorological variables (surface pressure, solar radiation, total precipitation, and wind speed), atmospheric pollution (PM_2.5), and socioeconomic indicators (population density, per capita road area, green coverage of built-up areas, urban construction land area, per capita GDP, number of large-scale industrial enterprises, total retail sales of consumer goods, and hospital bed capacity). Outcome variables comprised CHOE event number, total duration, longest duration, average duration, cumulative intensity, and average intensity.

The STRF model, an extension of conventional random forests, accounts for spatial and temporal dependencies, enabling exploration of complex, non-linear relationships in geospatially structured data [38]. Multicollinearity among influencing factors was assessed using the variance inflation factor (VIF), and a permutation-based significance test evaluated the statistical significance of each factor's importance. The STRF model was evaluated using spatially stratified 10-fold cross-validation at the city level and temporally stratified cross-validation to assess model robustness.

Several sensitivity analyses tested the robustness of our results, including adjusting the referent period for the heatwave temperature threshold from 1961 to 1990 to 2000–2023 (Supplementary Fig. S2), and changing the O₃ pollution threshold to the 90th percentile of MDA8 O₃ concentration over the study period (2000–2023) (Supplementary Fig. S3). All analyses were performed using R (version 4.4.3) and Python (version 3.12.3); a two-sided P < 0.05 was considered statistically significant.

3. Results

3.1. Situation of CHOEs in 2023

In 2023, China experienced an average of 7.1 CHOE event number and 42.3 CHOE total duration, with the longest and average duration reaching 13.9 and 6.2 days, respectively (Table 1). CHOE cumulative intensity increased 28.3-fold relative to baseline, while CHOE average intensity increased 0.6-fold. Population exposure to CHOE total duration, average duration, and cumulative intensity reached 4238.2 person-days, 564.8 person-days, and 2530.5 person-fold, respectively. Regionally, South China exhibited the highest mean CHOE event number among the seven geographical regions (Table 1), although substantial spatial heterogeneity was observed within the region, with the highest values concentrated in the Greater Bay Area (Supplementary Fig. S4). Northwest China and North China (including the Beijing–Tianjin–Hebei Region) recorded the highest CHOE total duration, average duration, and cumulative intensity. East China (including the Yangtze River Delta) had the highest population exposure levels for CHOE total duration, average duration, and cumulative intensity (Table 1 and Supplementary Fig. S4).

Table 1.

Compound heatwave and ozone pollution events (CHOEs) in China and its subregions in 2023.

Location	Raw indicators						Population exposure indicators
	Event number	Total duration	Longest duration	Average duration	Cumulative intensity	Average intensity	Population exposure to total duration	Population exposure to average duration	Population exposure to cumulative intensity
China	7.14	42.34	13.94	6.24	28.34	0.64	4238.16	564.81	2530.52
Geographical regions
Northeast China	7.19	27.86	7.59	3.90	15.31	0.54	3132.72	367.56	1727.17
North China	7.23	48.99	15.78	7.02	34.60	0.68	4085.76	594.28	2692.01
Northwest China	7.04	52.09	17.27	7.76	37.00	0.70	1177.83	173.03	876.94
East China	6.59	31.98	9.98	5.11	17.45	0.53	14148.87	1915.80	7631.57
Central China	7.06	34.58	11.63	5.12	20.02	0.56	11967.62	1737.08	6850.64
Southwest China	7.28	37.36	13.48	5.59	24.48	0.64	2938.53	356.11	1850.82
South China	7.91	27.86	7.59	3.49	16.42	0.61	9457.38	980.50	4959.96
Urban clusters
Beijing–Tianjin–Hebei Region	8.34	59.17	14.79	7.52	42.09	0.71	4238.16	2535.23	12794.46
Greater Bay Area	11.53	39.05	7.90	3.40	19.11	0.50	20107.08	1968.12	18041.14
Yangtze River Delta	6.56	29.90	9.95	4.95	16.61	0.56	38208.29	2965.97	8111.55
Chengdu–Chongqing Region	7.62	42.60	14.76	5.72	25.85	0.60	15404.04	1300.55	8830.18
Triangle of Central China	6.68	28.96	10.25	4.54	14.93	0.50	15151.84	2198.25	4327.25

3.2. Spatiotemporal distribution of CHOEs in China from 2000 to 2023

CHOEs in China exhibited pronounced spatial heterogeneity (Supplementary Table S1 and Figs. S5–S22). Xizang, Xinjiang, Gansu, and Hebei exhibited severe CHOE event number, total duration, longest duration, average duration, cumulative intensity, and average intensity, while high population exposure was concentrated in Beijing, Tianjin, Shanghai, Guangdong, Sichuan, and Chongqing. The I_L index similarly identified these regions as CHOE hotspots (Supplementary Figs. S23–S46). CHOEs in China and most of its subregions exhibited increasing trends from 2000 to 2023 (Table 2 and Supplementary Tables S2–S4), with increasing trends after 2016 generally higher than those observed prior to 2016 (Supplementary Tables S3–S4). China experienced a slight decline in CHOEs around 2020, followed by continued growth (Supplementary Fig. S47). South China (including the Greater Bay Area), Southwest China (including the Chengdu–Chongqing Region), and Central China (including the Triangle of Central China) experienced greater increases in CHOE event number, total duration, longest duration, average duration, cumulative intensity, and average intensity (Supplementary Figs. S48–S49). Urban areas with high population density (Beijing, Tianjin, Guangzhou, Shenzhen, Shanghai, and Chengdu) exhibited faster growth in CHOE exposure levels (Supplementary Figs. S48–S49).

Table 2.

Trend analysis of compound heatwave and ozone pollution events (CHOEs) in China and its subregions by Mann-Kendall test and Sen's slope with pre-whitening from 2000 to 2023.

Location	Raw indicators						Population exposure indicators
	Event number	Total duration	Longest duration	Average duration	Cumulative intensity	Average intensity	Population exposure to total duration	Population exposure to average duration	Population exposure to cumulative intensity
China	0.13∗	0.93∗	0.35∗	0.10∗	0.73∗	0.02∗	87.69∗	8.03∗∗	53.54∗
Geographical regions
Northeast China	−0.01	0.23	0.11∗	0.03	0.26∗	<0.01	41.72	2.52	36.22∗
North China	0.01	0.43	0.27∗	0.08∗	0.54∗	0.01∗	53.43∗	7.38∗	58.36∗
Northwest China	0.07	0.84	0.39∗	0.16∗	0.73	0.02∗	19.74	3.59∗	16.13
East China	0.08∗	0.76∗	0.29∗	0.11∗	0.43∗	0.01∗	293.76∗	48.73∗∗	187.55∗
Central China	0.01	0.82∗	0.31∗	0.10∗	0.60∗	0.01∗	256.68∗	32.59∗	213.38∗
Southwest China	0.29∗	1.52∗	0.36∗	0.17∗	0.5	0.02∗	50.55∗	3.46∗	−9.33
South China	0.15∗	0.66∗	0.17∗	0.04	0.40∗	0.01∗	273.75∗∗	17.10∗∗	165.92∗∗
Urban clusters
Beijing–Tianjin–Hebei Region	0.03	0.7	0.32	0.08	0.61	0.01∗	272.56∗	38.77∗	255.38∗
Greater Bay Area	0.23∗	1.00∗∗	0.15∗	0.01	0.61∗∗	0.01∗	1315.50∗∗	0.26	760.29∗∗
Yangtze River Delta	0.08	0.50∗	0.25∗	0.11∗	0.39	0.01∗	253.54∗	73.97∗∗	141.52
Chengdu–Chongqing Region	0.04	0.6	0.28∗	0.05	0.47∗	0.01∗	206.33	16.53∗	135.29
Triangle of Central China	0.06	0.93∗	0.31∗	0.07∗	0.46∗	0.01∗	218.86∗	62.35∗∗	121.11∗

∗P < 0.05.

∗∗P < 0.001.

3.3. Proportions of heatwave and O₃ pollution to CHOEs

The frequency and intensity of individual heatwave events, O₃ pollution events, and CHOEs all increased to varying degrees between 2000 and 2023 (Fig. 1). The co-occurrence proportion of heatwave to CHOE total duration (47.9–80.2%) was higher than the corresponding proportion of O₃ pollution (17.9–32.5%). However, the relative proportion of O₃ pollution to CHOE cumulative intensity (43.9–67.2%) exceeded that of heatwave (32.9–56.1%) during most years of 2000–2023 (Fig. 1). These proportions of heatwave and O₃ pollution to CHOEs remained consistent from 2000 to 2016; thereafter, the co-occurrence proportion of heatwave to CHOE total duration rose while its relative proportion to CHOE cumulative intensity declined. By contrast, O₃ pollution's co-occurrence proportion to CHOE total duration remained stable, while its relative proportion to CHOE cumulative intensity surged after 2016 (Supplementary Table S5). These proportions showed greater variability in South China, Southwest China, and Northwest China (Supplementary Table S5 and Figs. S50–S53).

Fig. 1.

Proportions of heatwaves and O₃ pollution to compound heatwave and ozone pollution events (CHOEs) in China from 2000 to 2023. a,b, Annual occurrence days (a) and cumulative intensity (b) of heatwaves, O₃ pollution events, and CHOEs, together with the corresponding proportions of heatwaves and O₃ pollution to CHOEs. Red, blue, and gray bars represent heatwaves, O₃ pollution events, and CHOEs, respectively. Red and blue lines indicate the proportions attributable to heatwaves and O₃ pollution, respectively. Gray dashed line represents the year 2016 as the breakpoint. Trend lines and associated 95% confidence intervals, along with regression statistics are shown for each component. Values are aggregated at the national scale for each year. The figure summarizes the temporal variations in the co-occurrence proportion of heatwaves and O₃ pollution to CHOE total duration and the relative proportion of heatwaves and O₃ pollution to CHOE cumulative intensity. The co-occurrence proportions to CHOE total duration were calculated as the ratios of CHOE days to heatwave days and to O₃ pollution days, respectively. The relative proportions to CHOE cumulative intensity were derived by decomposing the CHOE cumulative intensity equation into its heatwave and O₃ components.

3.4. Decomposition of population exposure to CHOEs

The decompositions of population exposure to CHOE total duration, average duration, and cumulative intensity in China showed accelerating upward trends from 2000 to 2023 (Fig. 2), with growth after 2016 higher than in 2000–2016, except for the population effect. Relative to 2000, the annual change in population exposure to CHOE total duration, average duration, and cumulative intensity increased from 214.2 to 2214.3, from 51.5 to 242.9, and from 29.7 to 1942.0 (in person-days, person-days, and person-fold, respectively). The contribution of event effects to total exposure change exceeded that of population effects and combined effects (Supplementary Table S6). The distribution and change patterns of population exposure decomposition effects across China's subregions were consistent with national trends. South China (including the Greater Bay Area), East China, and the Beijing–Tianjin–Hebei Region in North China exhibited the highest growth rates among these decomposition effects (Supplementary Table S6).

Fig. 2.

Decomposition of changes in population exposure to compound heatwave and ozone pollution events (CHOEs) in China from 2000 to 2023. Population exposure to CHOE total duration, CHOE average duration, and CHOE cumulative intensity is decomposed into exposure change, event effect, population effect, and combined effect. The first row shows population exposure, followed by the contributions of exposure change, event effect, population effect, and combined effect. Red lines represent annual changes in total population exposure, whereas yellow, green, and blue bars and lines indicate the event effect, population effect, and combined effect, respectively. Gray dashed line represents the year 2016 as the breakpoint. Population exposure changes are decomposed into contributions from changes in CHOE characteristics (event effect), population changes (population effect), and their interaction (combined effect). This decomposition illustrates the relative contributions of these factors to variations in population exposure over time.

3.5. Influencing factors of CHOEs

The STRF analysis quantified the relative importance of meteorological variables, atmospheric pollution, and socioeconomic indicators to multiple CHOE dimensions (Table 3). Multicollinearity diagnostics confirmed VIF values below five for all influencing factors, indicating no significant collinearity (Supplementary Table S7). Spatially stratified cross-validation yielded R² values ranging from 0.5 to 0.8, whereas temporally stratified cross-validation yielded R² values ranging from 0.7 to 0.8 across all influencing factors (Supplementary Table S8). Across six raw CHOE indicators, the importance percentages were broadly comparable at the national level, with the most influencing factors accounting for approximately 7–8% of total explained importance, primarily due to the averaging of city-level variability.

Table 3.

Relative importance (%) of influencing factors to compound heatwave and ozone pollution events (CHOEs) derived from the spatiotemporal random forest model.

Influencing factor	Event number	Total duration	Longest duration	Average duration	Cumulative intensity	Average intensity
Population density	8.19	7.58	7.74	7.53	7.23	7.72
Road area	6.96	7.56	7.96	7.44	7.64	7.70
Green coverage	7.80	7.96	7.70	7.55	7.72	7.92
Urban built-up area	7.80	7.68	7.92	7.14	8.18	7.84
GDP	7.73	7.56	7.35	8.20	7.60	8.11
Industrial enterprises	7.83	7.47	7.93	8.04	8.09	7.49
Retail sales	7.75	7.45	8.05	7.83	7.40	7.55
Hospital beds	7.27	7.77	7.18	7.24	7.65	8.03
PM2.5	7.93	7.68	7.62	7.99	7.89	7.64
Solar radiation	7.66	8.03	7.75	8.12	7.41	7.75
Surface pressure	7.65	7.66	7.59	7.69	7.62	7.11
Total precipitation	7.53	7.87	7.46	7.67	7.83	7.93
Wind speed	7.91	7.48	7.75	7.56	7.74	7.21

Permutation-based significance (P < 0.05) tests further identified solar radiation and total precipitation as significant influencing factors for CHOE event number, total duration, longest duration, and cumulative intensity, and identified total precipitation as a significant influencing factors for CHOE average duration (Supplementary Tables S9–S14). The spatial distribution of the leading influencing factor (Supplementary Fig. S54) and relative importance for individual factors (Supplementary Figs. S55–60) displayed pronounced regional variability. No single factor uniformly dominated nationwide, and cities instead exhibited distinct factor profiles shaped by local atmospheric conditions, pollution mixtures, and urban development characteristics.

Sensitivity analyses adjusting the referent period for the heatwave temperature threshold from 1961 to 1990 to 2000–2023 and changing the O₃ pollution threshold from 100 μg m⁻³ to the 90th percentile of MDA8 O₃ concentration over 2000–2023 yielded similar results (Supplementary Tables S15–S18 and Figs. S61–S90).

4. Discussion

In this national spatiotemporal analysis of CHOEs using high-resolution datasets, CHOE event number, total duration, longest duration, average duration, cumulative intensity, average intensity, and population exposure to CHOE total duration, average duration, and cumulative intensity in China exhibited pronounced spatial heterogeneity and significant growth from 2000 to 2023, accelerating after 2016. South China, East China, and the Beijing–Tianjin–Hebei Region in North China experienced the most severe CHOEs, concentrated mainly in highly populated areas. CHOE total duration was more closely associated with heatwave conditions, while CHOE cumulative intensity was more strongly linked to O₃ pollution. The co-occurrence proportion and relative proportion of heatwave and O₃ pollution to CHOEs also showed distinct temporal trends before and after 2016. Changes in population exposure to CHOE total duration, average duration, and cumulative intensity were dominated by event effects rather than population effects or combined effects. Our STRF results further indicate that CHOEs are not shaped by a uniform set of factors nationwide but instead reflect region-specific combinations of meteorological, pollution, and socioeconomic conditions.

Consistent with our findings, previous studies have demonstrated that the frequency of CHOEs has increased in some cities of China since the 21st century, except that two studies have reported no significant upward trends [7,11,15,39,40]. Our study, focusing on a nationwide scale, further examined the durations, intensities, and population exposure, and observed marked accelerated growth trends after 2016. The disparities might be attributable to inconsistencies in the study areas or periods of CHOEs. Global warming, atmospheric circulation anomalies, rapid urban expansion, and industrial development in China were likely to contribute to the exacerbation of CHOEs after 2016 [41,42]. Although studies have reported the spatial heterogeneity of CHOEs at the city level, our research uncovered such patterns at the national scale [10,39,40]. CHOEs were observed to be more severe in densely populated urban areas, likely due to a combination of regional climatic conditions, pollutant emissions, urbanization processes, and population density [[43], [44], [45]].

This study provides a comprehensive assessment of CHOEs in Northwest China and reveals increasing CHOE severity in western regions over the study period. The observed patterns may be associated with a combination of natural and anthropogenic factors. Natural factors, such as high elevation, intense solar radiation, specific dry deposition conditions, and stratosphere–troposphere exchange, may contribute to elevated O₃ levels in western China [46,47]. Additionally, as part of national strategies such as the “China's Western Development Strategy” and the “West–East Gas Transmission Project”, the establishment of large chemical enterprises may have influenced precursor emission patterns in the region. O₃ pollution in remote areas can sometimes exceed that in densely populated regions [48]. Although direct, high-resolution data linking CHOEs with anthropogenic emissions in western China are currently unavailable, previous studies of heatwaves and air pollution support the broader plausibility of this mechanism [49,50]. Temperature-dependent emissions of biogenic VOCs, shifts in NO_x-VOC sensitivity, and enhanced photochemical reaction rates have been identified as key mechanisms driving the co-occurrence of heatwaves and O₃ pollution [47,51,52]. Elevated temperatures during heatwaves enhance the emissions of biogenic and anthropogenic VOCs, which in turn accelerate photochemical O₃ formation. Furthermore, the O₃ formation regime may shift under high temperatures, altering the sensitivity of O₃ to its precursors [52]. These mechanisms collectively help explain the observed spatial and temporal patterns of CHOEs, although a full mechanistic understanding of these processes would require integrated assessment modeling with detailed emission inventories and chemical transport simulations, which represents an important direction for future research.

Our results revealed that heatwave exhibited a stronger association with CHOE total duration compared with O₃ pollution, aligning with previous studies [7,15]. Heatwave days were more often accompanied by high O₃ concentrations, leading to a rise in the co-occurrence rate of heatwave and CHOEs. Unexpectedly, we observed that CHOE cumulative intensity was more linked to O₃ pollution. One potential cause was the selected O₃ thresholds, and another may be that severe O₃ pollution across China has been further exacerbated by the combined effects of stratosphere–troposphere O₃ exchange, climate change, and rapid industrialization [21]. Through decomposition analysis of population exposure levels, we revealed that both event effects and population effects exhibited persistent growth. Although changes within the CHOEs themselves served as a primary positive driver of population exposure change, this regional disparity suggests latent changes in population structure with migration from less urbanized areas to more highly urbanized regions [53]. According to World Urbanization Prospects, by 2050, 70% of the global population would be concentrated in urban clusters that occupied only 3% of the Earth's land area [54]. This trend underscored the urgency of urban environmental governance. It was imperative to prioritize the environmental challenges of highly populated urban areas within the sustainable development agenda.

Our analysis of the influencing factors of CHOEs highlights a highly heterogeneous system shaped by region-specific atmospheric and urban development contexts. Although the average importance of individual influencing factors were broadly comparable, several variables showed slightly elevated importance, including surface pressure, PM_2.5, solar radiation, industrial enterprises, and wind speed. This spatially differentiated structure aligns with evidence from international studies of heatwaves and O₃ pollution, which similarly report that influencing factor dominance varies across climatic regimes and urbanization gradients. Studies in the United States have highlighted the dominant influence of meteorological conditions on summertime O₃ in coastal and monsoon-affected regions, whereas socioeconomic and emission-related factors exert greater control in densely urbanized or industrialized areas [55]. Similar patterns have been reported in China for O₃ formation [56], and global analyses of heatwave risks show that both climate and human-activity factors jointly shape extreme heat exposure [57]. This regional diversification of influencing factors suggests that the effective mitigation of CHOEs will require context-specific strategies rather than uniform national measures.

CHOEs are an urgent issue that may grow more severe globally [9]. Although our study focuses on China, inequitable exposure to CHOEs is increasingly recognized worldwide. Global analyses indicate that heatwaves can exacerbate O₃ pollution, disproportionately affecting socioeconomically vulnerable populations, whereas in East Asia, urbanization and synoptic weather patterns further intensify these compound risks [58,59]. Epidemiological studies have demonstrated that compound exposure to elevated temperatures and high O₃ concentrations exacerbated the incidence of asthma, chronic obstructive pulmonary disease, and cardiovascular events [7,40]. This dual exposure not only increased the frequency of acute health episodes but also contributed to long-term health deterioration, placing additional strain on healthcare systems. To address these challenges, future investigations on CHOEs require broader scientific engagement. Given that CHOEs occurred more frequently during heatwaves, the development and implementation of heat-health action plans should be prioritized, including real-time early warning systems, enhanced urban cooling infrastructure, and targeted public health outreach [60]. In parallel, as O₃ pollution is more strongly linked to CHOE cumulative intensity, targeted emission control strategies focusing on coordinated NO_x and VOC reductions are essential, especially in densely populated and industrialized regions [61]. National-level contingency frameworks should be complemented by community-based interventions tailored to local vulnerabilities, fostering both resilience and adaptive capacity [54]. Efforts should also be intensified to promote green, climate-resilient, and health-supportive urban environments.

This study improves the characterization of CHOEs in China by integrating three complementary aspects. First, high-spatial-resolution data enabled a detailed assessment of regional patterns, allowing more accurate identification of vulnerable areas. Second, the long study period (2000–2023) allows robust evaluation of temporal variations in CHOEs and provides a basis for understanding their recent evolution. Third, we identified CHOEs using the HI, which accounts for humidity in addition to air temperature and more accurately reflects human thermal stress, potentially supporting more targeted assessments of population health risks.

The study has several limitations. Although input datasets were harmonized to a common spatial resolution through resampling, minor inconsistencies stemming from their native resolutions might persist, although these are unlikely to materially affect the overall findings. The study period was limited to 2000–2023, as O₃ data for 2024 were unavailable at the time of analysis. In addition, China's nationwide ground-level O₃ monitoring network was not fully established until 2013, which may increase uncertainty in O3 estimates before then. Finally, the lack of age- and sex-specific population data prevented us from assessing differences in population exposure among demographic groups; this may mask heterogeneity in exposure levels or vulnerability. Future studies incorporating demographic-specific population distributions and health outcome data could further refine exposure assessments and evaluate population vulnerability.

5. Conclusion

This study provides evidence on the spatial heterogeneity and increasing trends in CHOE event number, total duration, longest duration, average duration, cumulative intensity, average intensity, and population exposure to CHOE total duration, average duration, and cumulative intensity in China from 2000 to 2023. Densely populated regions, including the Greater Bay Area, the Beijing–Tianjin–Hebei Region, and the Yangtze River Delta, were more severely affected by CHOEs and should be prioritized as hotspots for environmental governance. Heatwaves had a stronger influence on CHOE total duration, while O₃ pollution was more closely associated with CHOE cumulative intensity. Rising population exposure levels stem largely from intensified CHOEs themselves rather than population change. The influencing factors of CHOEs varied among cities, reflecting the interplay of meteorological, pollution, and socioeconomic conditions. These findings improve understanding of the spatiotemporal patterns, population exposure, and influencing factors of CHOEs across China, providing scientific evidence to support future research and risk assessment.

CRediT authorship contribution statement

Yuxin Bi: Writing – original draft, Visualization, Methodology, Investigation, Formal analysis. Ruijun Xu: Writing – original draft, Visualization, Investigation, Formal analysis. Yi Zheng: Writing – review & editing, Investigation. Nongping Feng: Writing – review & editing. Jing Wei: Resources, Funding acquisition. Sirong Wang: Writing – review & editing. Lu Luo: Writing – review & editing. Jinying Huang: Writing – review & editing. Kaiwen Zuo: Writing – review & editing. Qiling Wang: Writing – review & editing. Xinzong Zhang: Writing – review & editing. Ziquan Lv: Writing – review & editing. Suli Huang: Writing – review & editing. Xue-Yan Zheng: Writing – review & editing. Hong Sun: Writing – review & editing. Ronghai Deng: Writing – review & editing, Supervision, Conceptualization. Yuewei Liu: Writing – review & editing, Supervision, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article: