Dataset Development of Population Exposure to Extreme Heatwaves in China (2000, 2010, 2020)

NONG Kaisen^{1, 2}  WANG Minghao^{1, 2}  HUANG Yaobang^{1, 2}  SUN Wenhao³ 
CHU Mingruo^3*

1. College of Urban and Environmental Sciences, Peking University, Beijing 100871, China;

2. Center for Urban Future Research, Peking University, Beijing 100871, China;

3. Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China

Abstract: To address the spatial scale mismatch between meteorological raster products and socioeconomic data aggregated by administrative units, this study integrates MERRA-2 reanalysis data with multi-source census data to develop the Dataset of population exposure to extreme heatwaves in China (2000, 2010, 2020). Using daily maximum temperature data from 1990 to 2022 as the climatological baseline, the dataset identifies heatwave events based on a criterion combining both relative and absolute thresholds, and separately calculates 3 indicators: heatwave frequency, heatwave duration, and cumulative heat intensity. To enhance the robustness of the decadal representation, national censuses data of 2000, 2010, and 2020 are used total population constraints, while high-resolution population raster data from WorldPop are incorporated to represent spatial population distributions. Through population weighting and regional statistical aggregation, gridded heatwave indicators are mapped onto 367 administrative units across mainland China. In addition, a 5-year moving average window is applied to derive total population exposure indicators for the three representative years. Accuracy assessments demon­strate a very high level of consistency (R²>0.97) between the MERRA-2 data and observa­tions from more than 2,400 meteorological stations nationwide. Furthermore, cross-validation across 35 major cities shows that the Root Mean Square Error (RMSE) of population exposure to heatwave frequency generated by this dataset, compared with in situ station measurements, is only 0.89 occurrences. This result confirms the reliability of the dataset after administrative aggre­gation and temporal reconstruction. The dataset is archived in .xlsx and .shp formats and consisting of 9 data files with a total size of 27.1 MB (17.8 MB compressed).

Keywords: severe heatwave; population exposure; MERRA-2; climate risk

DOI: https://doi.org/10.3974/geodp.2026.02.07                   

Dataset Availability Statement:

The dataset supporting this paper was published and is accessible through the Digital Journal of Global Change Data Repository at: https://doi.org/10.3974/geodb.2025.12.08.V1.

1 Introduction

Against the macro backdrop of intensifying global climate change, the frequency, intensity, and duration of extreme heat events have shown a significant increasing trend, making them one of the primary natural hazards threatening human health and constraining sustainable socioeconomic development^[1]. In particular, during China’s rapid urbanization process, the high concentration of populations in urban agglomerations and the rapid expansion of impervious surfaces have significantly intensified the urban heat island effect. The combined impacts of climate warming and accelerated urbanization have exposed China’s urban systems to unprecedented risks of heatwave exposure^[2,3].

Although multiple high-temperature datasets based on satellite remote sensing retrievals or meteorological station interpolations have been released by the academic community, existing data products still face several challenges when applied to climate risk management and socioeconomic impact assessments. Previous datasets have primarily focused on charac­terizing temperature intensity at the physical level, often neglecting the quantification of exposure that is closely linked to population distribution. As a result, it remains difficult to accurately assess the actual impacts of extreme heat on human-social systems^[4]. Meanwhile, significant data format barriers and scale mismatches exist between the high-resolution raster formats commonly used for meteorological data and the vector-based administrative units relied upon for disaster early warning, medical resource allocation, and socioeconomic statistics (e.g., GDP, mortality rates). This spatial incompatibility not only requires complex regional statistical analyses and spatial aggregation in interdisciplinary research—thereby raising the technical threshold for data processing and increasing the likelihood of calcu­lation errors—but also fails to directly support the administrative division-based decision- making needs of government agencies. Furthermore, large-scale climate drivers such as the El Niño-Southern Oscillation (ENSO) can introduce substantial interannual variability^[5]. As a result, data from a single year often contain considerable random perturbations, making it difficult to robustly reflect long-term decadal trends in climate evolution^[6].

Addressing these critical research gaps, this study integrates multi-source meteorological monitoring data and long-term census data to develop the Dataset of population exposure to extreme heatwaves in China (2000, 2010, 2020), using 367 administrative units in the mainland of China (including direct-administered municipalities, prefecture-level cities, prefectures, and province-administered counties) as the fundamental spatial units of analysis. The dataset establishes a rigorous data production framework through dual optimization in both spatial and temporal dimensions. In the spatial dimension, an administrative unit-based approach is adopted to map continuous meteorological raster variables onto prefecture-level admini­strative polygons. This approach bridges the gap between natural science data and social science statistics, enabling the seamless integration of meteorological variables with various statistical yearbooks through unified administrative codes (City IDs). In the temporal dimension, to filter out interannual climate variability and more accurately capture long-term trends, the study moves beyond the traditional single-year snapshot method. Instead, a 5-year moving average window (1998–2002, 2008–2012, and 2018–2022, respectively) is applied to generate representative values for 3 key temporal nodes: 2000, 2010, and 2020.

Based on the methodologies described above, this dataset systematically incorporates both hazard factors and their corresponding population exposure risk indicators across 3 dimensions: Heatwave Frequency (HWF), Heatwave Duration (HWD), and Cumulative Heatwave (HeatCum). By providing a vectorized dataset characterized by clearly defined administrative units, robust temporal representation, and comprehensive multidimensional indicators, this study aims to offer scientifically grounded, accessible, and highly usable fundamental data to support regional climate adaptation planning, the formulation of urban disaster risk reduction strategies, and the optimization of public health policies.

2 Metadata of the Dataset

The metadata for the Dataset of population exposure to extreme heatwaves in China (2000, 2010, 2020)^[7] are summarized in Table 1. The metadata include the full dataset name, short name, authors, reference years, temporal resolution, data format, data size, file composition, data publisher, and data sharing policy.

Table 1  Metadata summary of the Dataset of population exposure to extreme heatwaves in China (2000, 2010, 2020)

3 Methods

3.1 Data Sources

The meteorological data used to construct this dataset are derived from the MERRA-2 (Modern-Era Retrospective analysis for Research and Applications, Version 2)[1] reanalysis dataset released by the National Aeronautics and Space Administration (NASA), from this dataset, the daily maximum temperature (T_max) variable for the period 1990–2022 is extracted. Owing to its strong spatiotemporal continuity, MERRA-2 effectively compensates for the spatial coverage limitations of ground-based meteorological observations^[9]. To ensure the reliability and applicability of the reanalysis data within China, this study incorporates concurrent observational records from more than 2,400 national-level meteorological stations operated by the China Meteorological Administration[2] as ground-truth data.

The population data processing adopts a strategy that integrates gridded population data with census statistics. The spatial distribution baseline is derived from the 100-m resolution population grids of the WorldPop dataset[3], which characterize population spatial agglo­meration patterns using a random forest algorithm^[10]. To ensure the reliability and authority of the population baseline, the total population counts are strictly calibrated using county-level statistics from the 2000 (5th), 2010 (6th), and 2020 (7th) National Population Censuses[4] published by the National Bureau of Statistics of China. The spatial basemap is based on the 2020 standard national prefecture-level administrative boundary vector dataset[5], which encompasses 367 administrative units across the mainland of China. All spatial data are projected into the Albers Equal-Area Conic coordinate system to ensure the accuracy of geometric calculations.

3.2 Algorithm

(1) Definition and identification of heatwave events and metrics

Heatwave events are identified using a method that combines both relative and absolute temperature thresholds^[11]. In this study, heatwave days and heatwave events are detected at the grid scale based on daily maximum temperature (T_max). To ensure comparability across different climate zones and to capture seasonal variations, relative thresholds are constructed by calendar day using the baseline period of 1990–2022. For each grid cell i and calendar day d, a 15-day moving window [d−7,d+7] is applied to pool multi-year T_max samples within the baseline period. The 90th percentile of this sample set, denoted as P90_i,d is then calculated as the relative threshold for that specific calendar day. Subsequently, within the target year series, a day t is classified as a candidate heatwave day when the following conditions are satisfied: T_max (i,t)>P90_i,d(t) and T_max (i,t)>32 ℃^[12], and the day falls within the warm-season months (May to September in the Northern Hemisphere). A heatwave event is defined as a sequence of at least 3 consecutive candidate heatwave day^[13], only days belonging to an identified heatwave event are included in the heatwave statistics. Isolated high-temperature occurrences with a duration of fewer than 3 days are excluded from both event and duration calculations^[14].

Based on the definitions, this study calculates heatwave characteristic metrics at the grid scale across 3 dimensions: 1) Annual Heatwave Frequency (HWF), defined as the total number of heatwave events occurring within a given year. 2) Annual Heatwave Duration (HWD), defined as the total number of days associated with all heatwave events within a year. 3) Annual Cumulative Heat (HeatCum), defined as the accumulated temperature exceedance above the relative threshold during heatwave days. The calculation of cumulative heat is expressed as follows:

                                                            (1)

Where HWdays(i,t) denotes the set of all dates belonging to heatwave events for grid cell i in

year y, and t represents a specific day within this set (t ∈ HWdays(i,t)).  represents the daily maximum temperature for grid cell  on day , and  deno­tes the 90th-percentile threshold corresponding to the calendar day .

(2) Calculation of population-weighted heatwave characteristics

To address the scale mismatch between meteorological grids and administrative units, regional zonal statistics and population-weighting algorithms are employed to perform spatial aggregation and mapping^[3]. By aggregating the gridded heatwave metrics to the prefecture-level administrative scale and using the WorldPop gridded population as weights, the population-weighted heatwave indicator for prefecture-level city c in year y is calculated as follows:

                                                                               (2)

Where  denotes the total population (in 10⁴ persons) of all grid cells within prefecture-level city c in year y;  represents the population (in 10⁴ persons) of grid cell i within that prefecture-level city; and HW_i,y∈{HWF_i,y, HWD_i,y,

HeatCum_i,y}, represents the specific heatwave metric for each grid cell, corresponding to annual heatwave frequency, total heatwave duration (in days), and cumulative heat exposure, respectively. The resulting HW_c,y characterizes the population weighted average level of heatwave frequency, duration, and cumulative heat exposure experienced by resident within administrative unit c.

(3) Calculation of population heatwave exposure for typical years

To reduce interannual variability caused by large-scale climate drivers such as ENSO and to more robustly capture decadal climate trends, this study moves beyond the conventional single-year snapshot method and instead applied a 5-year moving average window for temporal reconstruction^[5]. Specifically, the years 2000 (1998–2002), 2010 (2008–2012), and 2020 (2018–2022) are selected as central reference points. For each window period, the arithmetic mean of the three heatwaves metrics described above is calculated to generate baseline heatwave indicators representing the typical climate characteristics of the corresponding decades.

                                                                                           (3)

Where T denotes the selected reference years, namely 2000, 2010, and 2020. The smoothed heatwave indicator for the typical year  is then multiplied by the census population  of each administrative unit to derive the population heatwave exposure indicators:

                                                                           (4)

Where ,  and  represent population expo­sure to heatwave frequency (million person-events), population exposure to heatwave duration (million person-days), and population exposure to cumulative heat load (million person-℃), respectively. The overall technical workflow for the dataset construction is illustrated in Figure 1.

Figure 1  Flowchart of the dataset development

4 Data Results and Validation

4.1 Dataset Composition

This dataset comprises 3 main components: (1) a list of 367 prefecture-level admini­strative units; (2) statistical tables reporting population exposure to heatwave frequency, heatwave duration, and cumulative heat load for the three typical years of 2000, 2010, and 2020 (units: million person-events, million person-days, and million person-℃, respec­tively); and (3) spatial distribution data of the above exposure indicators at the prefecture-level city scale in vector format. The dataset is archived in .xlsx and .shp formats. The abbreviations used in the attribute table of the vector dataset are defined in Table 2.

4.2 Data Products

Based on the statistical analyses for the three reference years (2000, 2010, and 2020), population exposure to heatwaves across prefecture-level administrative units in China has exhibited a non-linear and fluctuating upward trend over the past two decades. This trend is characterized by a relatively moderate increase during the early period followed by a sharp rise in recent years. All exposure indicators reached their historical peaks in 2020 (Figure 2).

Regarding population exposure to heatwave frequency, the average exposure across prefecture-level administrative units nationwide was approximately 15.6 million person- events in 2000. In 2010, this metric experienced a slight decline, with the mean dropping to 14.8 million person-events. However, by 2020, population exposure to heatwave frequency increased substantially to 25.2 million person-events, representing an increase of approxi­mately 70% compared with 2010. The distribution pattern shown in the box plots further reveals a marked increase in data dispersion in 2020, indicating a widening disparity in the impacts of heatwave frequency across different cities.

Population exposure to heatwave duration and cumulative heat load exhibited an even more pronounced upward trend. Taking HWD as an example, the average exposures levels in 2000 and 2010 were approximately 63.5 million person-days and 55.2 million person- days, respectively, whereas the mean value of this indicator exceeded 110 million person- days in 2020. An even stronger increase is observed for cumulative heat load (HeatCum). The average population exposure to cumulative heat reached 162.8 million person-℃ in

Table 2  Abbreviations in the vector data attribute table for the Dataset of population exposure to extreme heatwaves in China

Figure 2  Temporal evolution characteristics of the population heatwave exposure risk in China

2020, far exceeding the corresponding levels in 2000 (92.5 million person-℃) and 2010 (73.6 million person-℃). These results indicate that the heatwave risks faced by China’s urban population in recent years are manifested not only in a higher frequency of heatwave events but also in longer durations and substantially intensified heat exposure.

Population heatwave exposure risk in China exhibits a clear spatial differentiation pattern, characterized by higher levels in the southeast and lower levels in the northwest. Over time, high-risk areas have evolved from a relatively widespread spatial distribution to a more concentrated pattern within major urban agglomerations (Figure 3). In 2000, areas with high heatwave exposure were primarily concentrated in the Huang-Huai-Hai Plain (including Beijing-Tianjin-Hebei, Henan, and Shandong) and the Middle and Lower Reaches of the Yangtze River. Notably, North China showed consistently high exposure levels across all three dimensions of population heatwave exposure, whereas the Yangtze River Basin exhibited particularly high risks in terms of heatwave duration (HWD) and cumulative heat

Figure 3  Maps of the spatiotemporal evolution characteristics of the risk of population heatwave exposure in China

(HeatCum). During this period, high-risk areas were largely contiguous across East China, forming a broad and continuous spatial distribution.

In 2010, heatwave exposure risk across most parts of China experienced a temporary moderation. The spatial extent of high-exposure risk areas in much of North and East China contracted significantly. Moderate levels of exposure risks were retained only in certain southern coastal cities and in the Chengdu-Chongqing (Chengyu) region, a pattern that corresponds with the numerical decline observed in the box plots for this period. By 2020, the center of heatwave exposure risk had shifted markedly southward and intensified. The Middle and Lower Yangtze Plain, the Jiangnan Hilly Region, and the coastal areas of South China emerged as the most severely affected regions. In most prefecture-level cities within provinces such as Guangdong, Guangxi, Hunan, Jiangxi, and Fujian, population exposure to heatwave duration (HWD) generally exceeded 100 million person-days (Figure 3), while the HeatCum indicator also reached extreme values. Due to the combined effects of high population density and increasingly intense heatwaves, the Middle Reaches of the Yangtze River Urban Agglomeration, the Pearl River Delta, and the Chengdu-Chongqing Urban Agglomeration emerged as the primary hotspots of population exposure risk. In contrast, although the North China Plain still exhibited some exposure risk in 2020, its relative intensity had become significantly weaker than that observed in Southern China.

In summary, from 2000 to 2020, after experiencing a brief trough around 2010, population exposure to heatwaves in China increased sharply by 2020. Moreover, the center of exposure risk showed a clear tendency to shift toward the densely populated, hot and humid southern regions and major urban agglomerations. These patterns highlight the combined driving effects of climate change and population concentration on heatwave n exposure risk.

4.3 Data Validation

The data quality assessment consists of 2 stages: source data validation and result validation. First, a comparison based on 3.63 million sample points shows that the MERRA-2 reanalysis data exhibits exceptionally high consistency with temperatures observations from ground-based meteorological stations across China (R²>0.97, RMSE=2.03 ℃). These results provide a reliable physical basis for the subsequent calculations (Figure 4).

To evaluate the accuracy of the administrative unit-level dataset, this study selected in situ observational data from 35 key cities across China for cross-validation. The results indicate that, despite the spatial smoothing effect associated with the “point-to-area” aggregation process, which resulted in a coefficient of determination (R²) of 0.62 for the annual Heatwave Frequency (HWF), the Root Mean Square Error (RMSE) is only 0.89 occurrences (Figure 5). These findings demonstrate that, even after undergoing spatiote­mporal reconstruction and administrative aggregation, the dataset maintains a high level of numerical accuracy, enabling it to reliably characterize the regional evolutionary patterns of heatwave activity. Furthermore, the spatiotemporal patterns of population heatwave exposure risk revealed by this dataset are highly consistent with the findings of previous studies conducted by Tian^[15], Wang^[16], and Shen^[17].

5 Discussion and Conclusion

The core value of the Dataset of population exposure to extreme heatwaves in China (2000, 2010, 2020) developed in this study lies in provision of a standardized data production framework for mapping natural physical fields onto socioeconomic units.

Regarding its construction logic, this dataset effectively addresses the long-standing issue

of dual spatiotemporal scale mismatch in climate change research. In the temporal dimension, traditional single-year data are highly susceptible to random perturbations caused by large-scale climate drivers such as ENSO or monsoon systems, which may introduce uncertainty into analytical conclusions. The 5-year moving window smoothing strategy adopted in this study functions essentially as a low-pass filtering process. While preserving decadal-scale climate signals, it effectively filters out interannual variability and random noise, thereby providing a more robust baseline for long-term trend analysis. In the spatial dimension, this study aggregates continuous raster-based meteorological data into discrete prefecture-level administrative units and introduces high-resolution population weights. This approach not only reduces the assessment bias associated with relying solely on physical meteorological indicators but also shifts the analytical perspective from hazard intensity to actual population exposure risk. Furthermore, this dataset fills an important gap in multidimensional heatwave risk data at the meso- and macro-scales. Unlike conventional datasets that provide only a single temperature metric, this dataset includes process-oriented indicators such as heatwave frequency, duration, and cumulative heat load. Such multidi­mensional characterization is crucial for understanding the compound hazard mechanisms of heatwaves. For example, increasing heatwave frequency challenges the responsiveness of urban emergency management systems, while prolonged heatwave duration is closely associated with excess mortality among residents and increased pressure on urban energy supply systems. As foundational data integrating both physical and human geographic attributes, this dataset aims to reduce data barriers between natural and social sciences. It provides valuable support for public health and epidemiological research, climate-resilient urban planning, and vulnerability assessments of socioeconomic systems. Nevertheless, the current dataset still has several limitations. First, the prefecture-level spatial scale may masks substantial intra-regional heterogeneity. Second, the use of a single temperature indicator does not account for humid heat effects. Third, the dataset focuses solely on historical periods. Future research will therefore focus on downscaling the spatial resolution to the county level, incorporating composite indicators such as apparent temperature, and extending the dataset to include projections under future climate scenarios.

Author Contributions

Nong, K. S. performed the overall design of the dataset development. Nong, K. S. and Chu, M. R. designed the models and algorithms of the dataset. Nong, K. S. and Wang, M. H. contributed to the data collection and processing. Nong, K. S. and Huang, Y. B. conducted the data validation. Nong, K. S. wrote the data paper as the lead author. Wang, M. H., Huang, Y. B., and Sun, W. H. made supplementary contributions, and Chu, M. R. conducted a comprehensive review of the paper.

Conflicts of Interest

The authors declare no conflicts of interest.

References

[1]        Jay, O., Capon, A., Berry, P., et al. Reducing the health effects of hot weather and heat extremes: from personal cooling strategies to green cities [J]. The Lancet, 2021, 398(10301): 709–724.

[2]        Gao, S. J., Chen, Y. H., Chen, D. L., et al. Urbanization-induced warming amplifies population exposure to compound heatwaves but narrows exposure inequality between global North and South cities [J]. npj Climate and Atmospheric Science, 2024, 7(1): 154.

[3]        Stalhandske, Z., Ruiter, M. C. D., Chambers, J., et al. Global assessment of population exposure to multiple climate-related hazards from 2003 to 2021: a retrospective analysis [J]. The Lancet Planetary Health, 2025, 9(8): 11.

[4]        Yin, J. B., Gentine, P., Slater, L., et al. Future socio-ecosystem productivity threatened by compound drought-heatwave events [J]. Nature Sustainability, 2023, 6(3): 259–272.

[5]        Wang, J., Li, M. C., Liu, Y. J., et al. Large-scale climatic drivers for warm-season compound drought and heatwave frequency over North China [J]. Atmospheric Research, 2023, 288: 106727.

[6]        Liu, J. H., Chen, J., Yin, J. B., et al. Time of emergence of record-shattering compound heatwave-extreme precipitation events and their socio-economic exposures [J]. Geophysical Research Letters, 2025, 52(16): e2025GL116884.

[7]        Nong, K. S., Wang, M. H., Huang, Y. B., et al. Dataset of population exposure to extreme heatwaves in China (2000, 2010, 2020) [J/DB/OL]. Digital Journal of Global Change Data Repository, 2025. https://doi.org/10.3974/geodb.2025.12.08.V1.

[8]        GCdataPR Editorial Office. GCdataPR data sharing policy [OL]. https://doi.org/10.3974/dp.policy.2014.05 (Updated 2017).

[9]        Urraca, R., Cappucci, F., Lanconelli, C., et al. Assessing discrepancies in global aerosol trends from satellites, models and reanalyses [J]. Remote Sensing of Environment, 2025, 328: 114827.

[10]     Martin, G. K., Rojas-Rueda, D., Fong, K. C., et al. A health impact assessment of progress towards urban nature targets in the 96 C40 cities [J]. The Lancet Planetary Health, 2025, 9(4): e284–e293.

[11]     Xie, W. X., Zhou, B. T., You, Q. L., et al. Observed changes in heat waves with different severities in China during 1961–2015 [J]. Theoretical and Applied Climatology, 2020, 141(3): 1529–1540.

[12]     Yin, C., Yang, Y. P., Chen, X. N., et al. Changes in global heat waves and its socioeconomic exposure in a warmer future [J]. Climate Risk Management, 2022, 38: 100459.

[13]     Cai, F. Y., Liu, C. H., Gerten, D., et al. Sketching the spatial disparities in heatwave trends by changing atmospheric teleconnections in the Northern Hemisphere [J]. Nature Communications, 2024, 15(1): 8012.

[14]     Li, X. F., Zhao, L., Wang, S., et al. Unstable permafrost regions experience more severe heatwaves in a warming climate [J]. npj Climate and Atmospheric Science, 2025, 8(1): 147.

[15]     Tian, P., Zhang, F. Q., Yan, Y. Y., et al. Spatial inequalities in global population exposure to extreme heats and heatwaves [J]. Applied Geography, 2025, 174: 103474.

[16]     Wang, C. C., Ren, Z. B., Guo, Y. J., et al. Assessing urban population exposure risk to extreme heat: patterns, trends, and implications for climate resilience in China (2000–2020) [J]. Sustainable Cities and Society, 2024, 103: 105260.

[17]     Shen, H. J., You, Q. L., Wang, P. L., et al. Analysis on heat waves variation features in China during 1961–2014 [J]. Journal of the Meteorological Sciences, 2018, 38(1): 28–36.