Spatial Distribution of High-temperature Risk with a Return Period of Different Years in the Yangtze River Delta Urban Agglomeration

The fifth assessment report of the Intergovernmental Panel on Climate Change (IPCC) points out that global warming is an undeniable fact (Alexander et al., 2013). The frequent occurrence of extreme high-temperature events caused by climate warming not only has an irreversible impact on the natural environment but also has a great impact on human production and life (Patz and Khaliq, 2002; Green et al., 2016; Im et al., 2017). The dry climate caused by continuous high temperature will hinder plant growth, and the reduction of oxygen saturation in water may lead to a large-scale outbreak of cyanobacteria. The quality and output of agriculture, forestry and animal husbandry-related products will be affected by high temperature, and dry and rainy weather will also increase the incidence of forest and grassland fires (Pascal et al., 2021). The urban heat island effect caused by many factors increases the high-temperature risk in urban central areas, making the city the main risk area for climate warming (Grimm et al., 2008). Assessing high-temperature risk can provide a reliable reference for the establishment of early warning mechanisms to prevent high-temperature effects (Pascal et al., 2021) and provide a scientific basis for urban planning and construction, rational social and economic development, and scientific disaster prevention and reduction decision-making on a regional scale.

Due to the differences in the social economy and geographical environment in different regions, there is no clear standard for the definition of high temperatures and heat waves in the world. The World Meteorological Organization (WMO) defines the process in which the daily maximum temperature is greater than 32°C and lasts for 3 d or more as a high-temperature heat wave; according to the definition of China Meteorological Administration, it is classified as high temperature when the daily maximum temperature reaches or exceeds 35°C, and a high-temperature weather process that lasts for more than 3 d is called a high-temperature heat wave defined by (Wu et al., 2019).

In recent years, many scholars have carried out high-temperature and heat wave research in different regions from the perspectives of monitoring indices, comparative simulations, spatial patterns and risk assessments. Typical research examples of high-temperature risk assessment are as follows. As early as 2003, Dousset and Gourmelon (2003) constructed a logistic risk assessment model to assess high-temperature risk in Paris. By selecting the daily maximum temperature dataset from 128 stations during 1960 to 2004 to determine daily extreme high temperature (EHT) thresholds by centesimal method for different stations, Yang et al. (2008) analyzed the characteristics of EHT spatio-temporal distribution in Northwest China. Kuglitsch et al. (2010) analyzed the high-temperature heat wave process by using data from 246 meteorological stations in the eastern Mediterranean from 1960 to 2006, and the results show that the mean heat wave intensity, heat wave number and heat wave length have increased by (7.6 ± 1.3)°C/decade, (6.2 ± 1.1)/decade and (7.5 ± 1.3) d/decade, respectively, since 1960. Johnson et al. (2012) selected 25 extreme health risk indicators to establish a risk model and analyzed high-temperature events in Chicago. Krüger et al. (2013) established high-temperature sensitivity indicators based on urban structure, population status and thermal characteristics and then performed a high-temperature risk assessment in Dresden City, Germany. Frazier et al. (2014) used the method of exposure-sensitivity-adaptability to evaluate and analyze the risk level of high temperatures. Mishra et al. (2015) analyzed the data of 217 urban observation stations around the world from 1973 to 2012, and the results show that the frequency of high-temperature and heat wave events at urban stations has increased significantly with the growth of urbanization, and the degree of enhancement is much higher than that at rural stations during the same period. Inostroza et al. (2016) established a risk assessment model based on sensitivity, exposure and adaptability to predict the urban high-temperature risk of Santiago, Chile. Hu et al. (2019) estimated the spatio-temporal distribution of the high-temperature risk of summer maize in the Huang-Huai-Hai Plain of China from 2003 to 2018 based on Moderate-resolution Imaging Spectroradiometer (MODIS) land surface temperature.

Under the background of global warming, the number of high-temperature days, as well as the frequency and intensity of heat waves are increasing all over the world (Meehl and Tebaldi et al., 2004; Perkins et al., 2012). In general, people usually discuss characteristics of summer heatwaves or evaluate high-temperature risk for an extreme heat event (Kuglitsch et al., 2010; Ye et al., 2013; Mishra et al., 2015; Chen et al., 2017; Zhu et al., 2018; Hu et al., 2019; Yu et al., 2021), while assessing the spatio-temporal distribution of the high-temperature risk with various return periods for different regions is very important for the prevention and control of high-temperature effects. At present, there are two kind of data that have been used to assess the spatio-temporal distribution of the high-temperature risk at the regional scale. The interpolation of air temperature data obtained from meteorological stations can be used to monitor high-temperature with a long time series, while the land surface temperature retrieved from remote sensing data usually assess high-temperature risk at a given moment. The main content of meteorological disaster risk assessment can be summed up in four aspects: hazard risk assessment, vulnerability assessment, disaster damage assessment, and comprehensive assessment (Wang et al., 2015), in which indices are usually used to perform risk assessment. Evaluation based on indices mainly focuses on the selection, optimization, and weight calculation of disaster risk indicators. Typical assessment methods include the analytic hierarchy process (AHP), fuzzy comprehensive evaluation, principal component analysis, historical comparison, and the Delphi method (Yu et al., 2021).

By comprehensively analyzing the existing research related to high-temperature risk assessment, we know that a unified system of high-temperature risk assessment has not been fully established. The existing disaster assessment technology methods can not truly reflect the uncertainty of risk disaster processes and the complexity, variability, and diversity of risks when risks emerge, or show the reality of risk transfer management activities (Pan et al., 2020; Yu et al., 2021). Moreover, the selection of relevant high-temperature assessment indicators differs and is not comprehensive. The spatial scale of the assessment results is usually large, which makes it difficult to express detailed spatiotemporal and continuous distribution characteristics of high-temperature risk, so further comparative research on different assessment methods and results is needed. The continuous application of remote sensing and geographic information system in obtaining spatial geographic big data has gradually become an important method to carry out high-temperature distribution computation and risk assessment.

Affected by the subtropical climate of the Western Pacific, the phenomenon of high-temperature weather in summer in the Yangtze River Delta urban agglomeration often occurs (Ye et al., 2013), so it is of great importance to evaluate the spatial distribution of high-temperature risk in this area. Previous studies (Fu et al., 2020; Yu et al., 2021) in this region generally have low spatial resolution, incomplete consideration for high-temperature vulnerability and prevention capacity assessment, and usually focus on individual high-temperature events. It is necessary to quantitatively assess the spatial distribution of high-temperature risk in a time-series with different recurrence periods by comprehensively considering a variety of influencing factors. Based on this analysis, combined with multisource data, including air temperature observed from automatic meteorological stations (AWSs), remote sensing image data, socioeconomic statistical data and basic geographic data, and taking the underlying land surface and socioeconomic conditions in 2018 as the background data, this paper selects the Yangtze RiverDelta urban agglomeration, which has experienced rapid economic development, to assess the spatial distribution of high-temperature risk with return periods of 5, 10, 20 and 30 yr. In this research, the high-temperature thresholds for different recurrence periods are determined by using the generalized extreme value method based on the daily maximum air temperature data with a resolution of 1 km interpolated from the local thin disk smooth spline function, which is used to calculate high-temperature danger factors. Eight high-temperature vulnerability factors and 11 high-temperature preventability factors are derived and spatialized to calculate different vulnerability and prevention capacity elements. After determining the influence of each factor on the high-temperature risk by combining the AHP and expert scoring method, a high-temperature risk assessment model is built and applied to analyze the spatial distribution characteristics of high-temperature risk with different return periods in the Yangtze River Delta urban agglomeration. The analysis of the spatio-temporal distribution characteristics of high-temperature risk in the Yangtze River Delta can provide scientific evidence for the spatial layout of urban public service facilities, reasonable public policy making, as well as decision-making for preventing high-temperature risk on a regional scale.

The Yangtze River Delta urban agglomeration is one of the most developed and urbanized regions in China. According to the development plan approved by China’s State Council (China’s State Council, 2016), the Yangtze River Delta urban agglomeration includes Shanghai; Nanjing, Wuxi, Changzhou, Suzhou, Nantong, Yancheng, Yangzhou, Zhenjiang, and Taizhou in Jiangsu Province; Hangzhou, Ningbo, Jiaxing, Huzhou, Shaoxing, Jinhua, Zhoushan, and Taizhou in Zhejiang Province; and Hefei, Wuhu, Maanshan, Tongling, Anqing, Chuzhou, Chizhou and Xuancheng in Anhui Province. The study area has a regional area of 211 700 km², accounting for approximately 2.2% of China’s land area and 11.0% of China’s population, while contributing approximately 25% of China’s Gross Domestic Product (GDP). The population of the study area reached 154 million, and the average urbanization rate was 67.38% at 2018 (Wang et al., 2021).

The study area is located in the subtropical monsoon climate area with four distinct seasons. The annual average temperature is 17°C–22°C, and the average annual precipitation is 1000–1500 mm. It is hot and rainy in summer. July has the highest temperature, with a monthly average temperature above 28°C and a maximum temperature above 38°C. As the largest urban agglomeration in China (Fu et al., 2020), the annual average, maximum and minimum air temperatures have increased significantly with the continuous expansion of urban areas and the continuous growth of the population. The temperature increase rate in winter and spring is higher, and the temperature increase rate in large cities is significantly higher than that in medium-sized cities and small towns (Yu et al., 2021). The urban heat island effect is aggravated, resulting in frequent high-temperature disasters (Hou et al., 2013). In particular, high-temperature events with a long duration (more than 8 d) often occur in this area (Liu et al., 2015).

The meteorological observation data are from the daily dataset of China’s land surface climate data released by the National Meteorological Center of China. The dataset provides daily meteorological data, including average temperature, maximum temperature, average wind speed and average relative humidity. This paper selects the daily maximum temperature data observed at 42 meteorological stations from June, July and August of 1951 to 2018, including 1 station in Shanghai, 9 stations in Anhui Province, 13 stations in Jiangsu Province and 19 stations in Zhejiang Province (Fig. 1). For the purpose of calculating high-temperature danger factors with a return period of different years, long time-series daily maximum temperature data are needed, so the sparsely distributed AWSs measurement are selected and interpolated to get spatial distributions with the resolution of 1 km, while remote sensing data is difficult to meet this requirement currently.

Figure 1.  Spatial distribution of meteorological stations in the Yangtze River Delta urban agglomeration, China

According to various factors required for high-temperature risk assessment, the following remote sensing data and basic geographic data are selected.

(1) MODIS_09A1 data

MOD09A1 (downloaded from http://reverb.echo.nasa.gov/reverb/) provides bands 1–7 at 500-m resolution in an 8-d gridded level-3 product in the sinusoidal projection, including reflectance values, quality assessment, and the day of the year for the pixel along with solar, view, and zenith angles. Each MOD09A1 pixel contains the best possible observation during an 8-d period as selected on the basis of high observation coverage, low view angle, the absence of clouds or cloud shadow, and aerosol loading. MOD09A1 data corresponding to July 16, 2018 are selected and resampled to 1 km resolution and then used to calculate the normalized difference vegetation index (NDVI), normalized difference building index (NDBI) and normalized difference water index (NDWI) by Eqs. (1), (2) and (3), respectively.

where ρ_i is the MODIS reflectance of the i-th channel, i = 1, 2, 4, 6.

(2) DEM data

Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) Global Digital Elevation Model (GDEM) data (V3 version) with a resolution of 30 m are mosaiced and resampled to a spatial resolution of 1 km. ASTER GDEM data, covering 99% of the land surface from 83° north latitude to 83° south latitude, are produced based on the ASTER sensor onboard the Terra satellite by National Aeronautics and Space Administration (NASA) Jet Propulsion Lab (JPL) and Japan’s Ministry of Economy, Trade, and Industry (METI).

(3) Traffic line data

The distribution of railways, expressways, county-level roads, rural roads and primary roads in the Yangtze River Delta urban agglomeration was collected and released by the National Geomatics Center of China, which is converted to the raster data with the resolution of 1 km.

All of the following data are collected and spatialized to produce the grid data with the resolution of 1 km.

(1) Statistical yearbook data

The statistical yearbooks of Anhui Province (Anhui Bureau of Statistics, 2019), Jiangsu Province (Jiangsu Bureau of Statistics, 2019), Zhejiang Province (Zhejiang Bureau of Statistics, 2019) and Shanghai (Shanghai Bureau of Statistics, 2019) released in 2019 and the statistical data provided by the municipal statistical bureaus are selected and sorted, which are statistics of the corresponding industries by the end of 2018. Thirteen socioeconomic statistical data used for the study mainly include population density, people over 60 yr and under 18 yr, male to female ratio, GDP, public budget financial expenditure, number of medical staff, institutions and beds, motor vehicle ownership, residential and total power consumption, and air conditioning ownership.

(2) Spatial distribution of GDP

The spatial distribution GDP raster data in 2015 released by the Resource and Environmental Science Data Center of the Chinese Academy of Sciences are selected. By comprehensively considering many factors related to human economic activities, such as land use type, night light brightness, residential density and so on, the GDP raster data with a spatial resolution of 1 km are calculated by the multifactor weight distribution method based on the GDP statistical data within each administrative region. The GDP of each pixel in 2018 is calculated according to the relationship between the GDP data of 2015 and 2018 in the corresponding statistical subregion.

(3) Spatial distribution of population

According to the population data of the statistical yearbook, the accuracy of the Landscape and WorldPOP population datasets is compared and analyzed. The comparison results reveal that WorldPOP datasets have higher accuracy and are selected to calculate the spatial distribution of population parameters used in computing vulnerability factors. The WorldPOP dataset (downloaded from www.worldpop.org) integrates the data of AfriPop, AsiaPop and AmeriPop population projects, which are stored in Geotiff format with a resolution of 1 km. The spatial distribution of population density in 2018, China by using the mapping approach of random forest-based dasymetric redistribution is downloaded. The projection is geographic coordinate system WGS84 and the units are number of people per square kilometer.

(4) Medical institution sites and traffic station data

Through point of interest (POI) data from the Baidu map (https://map.baidu.com/) searched in the year of 2019, 27 957 medical institution sites and 28 182 traffic station sites in the study area are retrieved as important factors of high-temperature prevention capabilities.

The research flowchart of this paper is shown in Fig. 2. The detailed methods are explained as follows.

Figure 2.  The research flowchart of high-temperature risk assessment. AWS is automatic weather station. GDP is gross domestic production

(1) Spatial interpolation of daily maximum temperature

Since the daily maximum temperature data are observations from discrete stations, it is necessary to interpolate the daily maximum temperature spatially to assess the distribution of high-temperature risk. The local thin disk smooth spline function is widely used in the spatial interpolation of meteorological elements (Jobst et al., 2017), which is an extension of the spline function and allows the addition of linear covariates based on the ordinary spline function. The local thin disk spline function model is as follows:

where z_i is the dependent variable of the i-th pixel and x_i is a d-dimensional spline independent variable. f is the unknown smooth function about x_i to be estimated; y_i is a p-dimensional independent covariate; b is the p-dimensional coefficient of y_i and T means matrix transpose; e_i is the random error of the independent variable with a mean value of 0 and the variance of w_iσ²; w_i is the known relative error variance; σ² is the error variance and is a constant at all data points.

The function f and coefficient b in the model can be estimated by the least square estimation method to ensure a minimum β value.

where J_m(f) is defined as the m-order partial derivative of function f, and ρ is a positive smoothing parameter and is usually determined by the minimization of generalized cross validation (GCV), which plays a balance between data fidelity and surface roughness. The GCV calculation method removes one sample point at a time, fits the surface with the remaining sample points under certain smooth parameters to obtain the estimated value of the point, and then calculates the variance of the observed value.

Taking longitude and latitude as independent variables, DEM as a covariate and daily maximum temperature as a dependent variable, the interpolation model of the local thin disk spline function is trained to obtain the spatial distribution of daily maximum temperature with a spatial resolution of 1 km. Interpolation results show that the RMSE (Root Mean Square Error) of the local thin disk spline function is 1.72°C.

(2) Calculation of high-temperature threshold values with different recurrence periods

The generalized extreme value (GEV) was developed by Jenkinson (1955), who combined the Gumbel, Fréchet and Weibull distributions. If h is the daily highest air temperature at a pixel location, the cumulative distribution function of the GEV distribution is:

where 1+ξ(h−u)/σ > 0, u, σ (σ > 0) and ξ are the location parameter, scale parameter and shape parameter, respectively.

The maximum likelihood method was used to fit the related models, and maximization was performed using a quasi-Newton iterative algorithm. Assuming the independence of the data, the likelihood function is the product of the assumed densities for the observations h₁, h₂..., h_n. For the GEV₀ model, we have:

The estimates of µ, σ and ξ are taken to be those values that maximize the likelihood function L. A standard way of determining the best fit model is the likelihood ratio test. To investigate the existence of a trend in extreme air temperature with respect to time, a four-parameter model (Eq. (8)) was applied to express GEV₀.

where t₀ denotes the year the record started, here t₀=1951; t is the number of years; a and b are the regressed coefficients.

Once the best models for the data are determined, the next step is to derive the return levels for air temperature. The T-year return period, x_T, is the level exceeded on average only once in every T years (Coles, 2001); then, the distribution function is:

For the best model GEV₀, the maximum temperature of the T-year return period can be calculated by Eqs. (10). In this research, T is set to 5, 10, 20 and 30.

(3) Calculation of high-temperature danger factors

The annual average high-temperature days and intensity are used to express high-temperature danger. For each recurrence period, the annual average number of high-temperature days is the mean value of the number of days exceeding the high-temperature threshold of the corresponding recurrence period. The average high-temperature intensity of different return periods is calculated as follows. The date when the daily maximum temperature is greater than the high-temperature threshold is determined, and then the average value of the difference between the daily maximum temperature and the high-temperature threshold is calculated for these dates.

High-temperature vulnerability factors mainly include population density, age structure, male to female ratio, GDP and underlying land surface factors. The population density data are directly read out from the WorldPOP data. The underlying land surface factors, including NDBI, NDWI and NDVI, are calculated by MODIS-related products.

According to the proportional relationship between the GDP of 2015 and 2018 within each administrative unit in the study area, a relationship model used to convert the GDP of 2015 to the GDP of 2018 is established in the subregion (Eq. (11)), and then the spatial distribution of GDP at each pixel in 2018 is calculated on the 1 km resolution scale.

where $GDP_{i j}^{2018}$and $GDP_{i j}^{2015}$are the GDP data of pixel (i, j) in 2018 and 2015, respectively; GDP₂₀₁₈ and GDP₂₀₁₅ are the statistical GDP data of the district where pixel (i, j) is located in 2018 and 2015, respectively.

Wang et al. (2021) shows that GDP and population density has a positive effect on the ecological risk index, and the impact of population exceeded that of GDP in the Yangtze River Delta Urban Agglomeration. So, the research estimates the spatial distribution of each vulnerability factor by using the population density relationship between the pixel scale and the administrative unit. The calculation Equation is:

where Data_ij and Pd_ij are the estimation results of the influencing factor and the corresponding population density at pixel (i, j), respectively; Data and Pd are the statistical value of the factor and population density of the district where pixel (i, j) is located.

It should be noted that when using Eqs. (11) and (12) to spatialize different impact factors, the value of each grid pixel uses the corresponding statistical value of the district, county or town where the grid is located, that is, the scale factor changes with the statistical ministration district.

The number of medical institutions, the number of beds and staff in medical institutions, public financial expenditure, total power consumption, residential power consumption, the ownership of air conditioners per 100 households, car ownership and transportation networks have a positive correlation with the spatial distribution of population density. The high-temperature risks are lower for the sites where there have better prevention capabilities, such as better medical conditions. The spatial distribution of each factor is estimated by Eq. (12). Furthermore, kernel density analysis is performed for medical institutions and traffic stations. The greater the kernel density is, the denser the distribution of medical or traffic stations in the region and the stronger the ability to prevent high temperatures. For different types and grades of traffic line networks, linear density analysis is used to analyze the distribution of traffic networks. The denser the network is, the stronger the ability to prevent high temperatures.

(1) Calculation of high-temperature risk index

The addition and subtraction assessment model is used to calculate the high-temperature risk index of the Yangtze River Delta urban agglomeration. The calculation model is as follows:

where R is the high-temperature risk index, E is the high-temperature danger factor, w_HE is the weight of E, S is the high-temperature vulnerability factor, w_a is the weight of S, A is the factor of high-temperature preventability, and w_b is the weight of A.

The calculation model of high-temperature danger factor (E) is:

where E₁ and w_HE1 are the annual average high-temperature days with an n-year return period and their weights, respectively; E₂ and w_HE2 are the annual average high-temperature intensity and its weight with an n-year return period, respectively.

The calculation formula of high-temperature vulnerability factor is:

where w_a1, w_a2 and w_a3 are the influence of population, economic development level and underlying land surface characteristics, respectively; c₁ and w_c1 are population density and its weight, respectively; c₂ and w_c2 are the proportion and weight of the population over 60 years old, respectively; c₃ and w_c3 are the proportion and weight of the population under 18 years old, respectively; c₄ and w_c4 are the proportion and weight of men and women, respectively; d is the spatial distribution of GDP; g₁ and w_g1 are the water index and its weight, respectively; g₂ and w_g2 are the building index and its weight, respectively; and g₃ and w_g3 are the vegetation index and its weight, respectively.

The calculation model of high-temperature prevention capability factor is:

where w_b1, w_b2, w_b3 and w_b4 are the influence of fund accessibility, medical facilities and technology, infrastructure accessibility and social development capacity, respectively; f is public budget financial expenditure; m₁ and w_m1 are the number and weight of medical staff, respectively; m₂ and w_m2 are the number and weight of medical institutions, respectively; m₃ and w_m3 are the distribution density and weight of medical institutions, respectively; m₄ and w_m4 are the number of beds and its weight in medical institutions, respectively; j₁ and w_j1 are the distribution density and weight of traffic lines, respectively; j₂ and w_j2 are the distribution density and weight of traffic stations, respectively; j₃ and w_j3 are motor vehicle ownership and its weight, respectively; k₁ and w_k1 are residential power consumption and its weight, respectively; k₂ and w_k2 are the total power consumption and its weight, respectively; and k₃ and w_k3 are air conditioning ownership per 100 households and its weight, respectively.

(2) Standardized processing of impact factors

During the process of high-temperature risk assessment, due to differences in units, quantities and trends among different high-temperature impact factors, these data need to be standardized. The min-max normalization method for dimensionless processing is used to preprocess various influencing factors, resulting in the spatial distribution of each factor ranging from 0 to 1.

(3) Determination of the weight for each influencing factor

Because the traditional AHP method tends to be subjective (Yu et al., 2021), the weight of each influencing factor is determined by the combination of AHP and expert scoring method. The AHP method establishes a hierarchical structure model and divide it into the target layer, the criterion layer and the factor layer. According to the influencing elements related to high-temperature danger, vulnerability and preventability factors, an electronic questionnaire was designed. Twenty-six experts from different fields and specialties, including meteorology, environmental and emergency management, were invited to fill in the questionnaire. The experts gave the score of each factor according to the principle of AHP (Aminbakhsh et al., 2013), which was counted and averaged to obtain the weight value of each influencing factor. Factors influencing high-temperature risk and its weight calculated based on AHP and expert scoring method for the target layer, the criterion layer and the factor layer are listed in Table 1.

Based on the spatialization of various high-temperature influencing factors, a high-temperature risk assessment model is built to compute the high-temperature risk index after calculating factors of high-temperature danger, high-temperature vulnerability and high-temperature preventability, which is used to analyze the spatial distribution of high-temperature risk during different return periods for the study area.

High-temperature danger factor is calculated by Eq. (14), and the spatial distribution of the danger factor in different return periods is given in Fig. 3.

Figure 3.  Spatial distribution of high-temperature danger factor at different return periods in the Yangtze River Delta, China

According to Fig. 3, high-temperature danger factor decreases with increasing return periods of 5, 10, 20 and 30 yr. The average high-temperature danger factor values of the whole study area are 0.1327, 0.1161, 0.0979 and 0.0899, indicating that high-temperature danger decreases gradually when the return period becomes greater. In terms of spatial distribution, the standard deviations of the high-temperature danger factor are 0.0403, 0.0396, 0.0373 and 0.0370 for return periods of 5, 10, 20 and 30 yr, respectively, which reveal that the spatial distribution difference gradually decreases while the relative high and low trends of high-temperature danger in different return periods are roughly the same. High-temperature danger generally decreases from the north to the south and from the west inland to the east coast, and the area of the high-temperature risk region decreases gradually with the increase in the recurrence period. For 27 cities in the study area, the higher high-temperature danger regions are distributed in Jinhua, Hangzhou and Ningbo in Zhejiang Province and Chuzhou, Xuancheng in Anhui Province. Low high-temperature danger areas are located in Yancheng and Nantong in Jiangsu Province, Anqing in Anhui Province and the southeast coastal area of Zhejiang Province. The risk of high-temperature danger in other cities is medium.

According to Eq. (15), based on the spatialization of each high-temperature vulnerability factor in 2018, the spatial distribution of high-temperature vulnerability factor is calculated and shown in Fig. 4. According to Fig. 4, high- and low-vulnerability distribution features are generally closely related to socioeconomic and underlying land surface factors such as urban scale, built-up area distribution and population density. In the study area, the high value of urban vulnerability factor in Shanghai is widely distributed, and the maximum value is 0.4236. Central urban areas in Nanjing, Hangzhou, Wuxi and Hefei also have high vulnerability, and the extreme vulnerability factor reaches approximately 0.36–0.40. The maximum value of high-temperature vulnerability factor in Ningbo and Changzhou, is approximately 0.35, but the distribution areas of vulnerability are relatively small. The high-temperature vulnerability factor in other cities is mostly less than 0.18. In addition to the influence of population, social and economic factors related to urban scale, terrain and underlying land surface types are also important factors affecting the distribution of high-temperature vulnerability. According to Fig. 4, areas close to water, such as Taihu Lake and the surrounding areas of the Yangtze River, have relatively low high-temperature vulnerability, and some mountainous areas have less population distribution and relatively low GDP. These areas also have relatively low high-temperature vulnerability, such as some areas of Jinhua in Zhejiang Province.

Figure 4.  Spatial distribution of high-temperature vulnerability factor in the Yangtze River Delta, China

The spatial distribution of high-temperature preventability factor is calculated by Eq. (16), as shown in Fig. 5, which is similar to the distribution of high-temperature vulnerability.

Figure 5.  Spatial distribution of high-temperature preventability factor in the Yangtze River Delta, China

Because the ability to prevent high temperatures is related to government revenue, medical conditions, level of traffic development, residential power consumption and so on, the ability to prevent high temperatures in economically developed regions is relatively high. Suzhou, Wuxi, and Changzhou in southern Jiangsu Province, Shanghai, and the provincial capital of Hefei, Nanjing, and Hangzhou have the capability to prevent high temperatures, close to 0.2. The ability to prevent high temperatures of other cities is mainly distributed in the central urban area. Except for provincial capital Hefei, high-temperature preventability in other cities of Anhui Province is relatively lower, in which Anqing and Chizhou has the lowest ability. Due to the influence of the spatial distribution characteristics of various influencing factors, high-temperature preventability gradually radiates and decreases from the central urban area to the surrounding suburban area.

According to Eq. (13), the spatial distribution of the high-temperature risk index for different return periods is calculated and shown in Fig. 6. According to Fig. 6, in different return periods, the high-temperature risk in each urban central area is relatively high and generally shows an increasing trend from northeast to southwest. The high-temperature risk in the coastal areas of middle Jiangsu Province, northern Jiangsu Province and eastern Zhejiang Province is relatively low, and it is relatively high in southern Anhui Province and western Zhejiang Province. Due to the high value of high-temperature danger and vulnerability, the high-temperature risk of each city, especially large cities such as the urban central areas of Shanghai, Nanjing, and Hangzhou in different return periods, is higher than that of the surrounding suburban areas. Chen et al. (2017) and Fu et al. (2020) also demonstrated that the downtown areas were mostly driven by the higher heat hazard index and the heat health risks were lower in the suburban and rural areas.

Figure 6.  High-temperature risk index at different return periods in the Yangtze River Delta, China

The minimum, maximum, and average values and the standard deviation of the high-temperature risk index in different return periods for the whole study area are counted, and the results are shown in Fig. 7. Due to the thresholds used to determine the high temperature gradually increase with the recurrence period, according to Fig. 6 and Fig. 7, the annual high-temperature days and intensity meeting the threshold conditions gradually decrease, the statistical indicators of the high-temperature risk index gradually decrease, and the distribution range of the high-temperature risk index gradually decreases. The average values of the high-temperature risk index in the whole study area are 0.2458, 0.2291, 0.2109 and 0.2029, indicating that the high-temperature risk gradually reduces. The standard deviations of the high-temperature risk index are 0.0449, 0.0439, 0.0428 and 0.0425, indicating that the difference in the spatial distribution of high-temperature risk in different regions gradually decreases. Although the high-temperature risk index shows a downward trend with the increase in return period, the impact and harm of high-temperature weather with a long return period on nature and society will be more serious.

Figure 7.  Statistical analysis of the high-temperature risk index at different return periods in the Yangtze River Delta, China

On the urban scale, the average value and the standard deviation of each city’s high-temperature risk index in different return periods are counted, and the results are shown in Fig. 8. According to Fig. 8a, the high-temperature risk indices of different cities vary greatly. In terms of average value of the high-temperature risk indices, Jinhua, Hangzhou and Xuancheng are the top three cities, Yangzhou, Taizhou and Yancheng are the bottom three, and the risk index of Jinhua in Zhejiang Province is almost twice that of Yancheng. Based on Fig. 8a, it can be found that high-temperature risk increases gradually with decreasing latitude. The high-temperature risk index of cities with higher latitudes in the northern study area, including Yancheng, Taizhou (Jiangsu Province), Yangzhou, Nantong and Shanghai, is very low, while the high-temperature risk of cities in the southern region, such as Jinhua, Hangzhou, Xuancheng and Shaoxing, is relatively higher. In the east-west direction, the farther away from the ocean, the higher the high-temperature risk. Although the latitudes of Ningbo and Taizhou (Zhejiang Province) in the south are low, due to their proximity to the East China Sea, the high temperature risk of these cities is less than that of Shaoxing, Hangzhou and Jinhua inland, and the high-temperature risk index in the eastern region is relatively lower. Suzhou, Wuxi, Changzhou, Nanjing, Maanshan and Hefei in the middle region are farther away from the sea, and the high-temperature risk there is gradually increasing. Due to the influence of a large area of Chaohu Lake, the high-temperature risk in Hefei is slightly lower than that in eastern Maanshan. The average high-temperature risk index of Jiaxing, Anqing, Tongling, Huzhou, Chizhou, Wuhu, and Xuancheng in the central and southern regions gradually increases, which is generally consistent with the distance from the East China Sea. Due to the relatively large water area and forestland in these cities in southern Anhui Province, the high-temperature risk index is reduced to a certain extent. The leeward slope area in some mountainous areas is vulnerable to the Brahma effect in summer, such as north Anqing, which leads to an increase in the high-temperature risk index. Compared to former research, Chen et al. (2017) selected MOD11A1 and MYD11A1 land surface temperature data obtained in Aug.7, 2013 to assess heat health risks in Yangtze River Delta region and derived that the high risks areas were mainly distributed in Shanghai, Changzhou, Hangzhou, Ningbo, Wuxi, Jiaxing and Taizhou, which is similar to the results of this research though the temperature data used are not the same. By using MOD11A1 land surface temperature and social economic data, Fu et al. (2020) assessed a typical heat wave event of 2017 in the middle-lower Yangtze River and concluded that the heat wave risks are higher for Hangzhou, Shaoxing, Jinhua of Zhejiang Province.

Figure 8.  Mean value (a) and standard deviation (b) of the high-temperature risk index in various cities at different return periods in the Yangtze River Delta, China

According to Fig. 8b, the distribution characteristics of the standard deviation of the high-temperature risk index in different cities with different return periods are basically the same. For coastal cities in the southern region, such as Taizhou (Zhejiang Province), due to the gradual reduction of high temperature risk from coastal to inland areas, the standard deviation of the high-temperature risk index is relatively large, and there are great differences in different locations within the city. Due to the relatively larger high-temperature danger in northern Anqing, southern Chuzhou in Anhui Province and southwestern Jinhua in Zhejiang Province, the high-temperature risk index in these areas is also larger, which is quite different from that in other locations of the same city. Therefore, the standard deviation of the high-temperature risk index in these cities is also relatively larger.

Among provincial capitals and municipalities, according to the high-temperature risk distribution and statistical results in different return periods, Hangzhou has the largest high-temperature risk, Shanghai has the smallest risk, and Nanjing and Hefei are at the intermediate risk level. The main reason for the highest high-temperature risk in Hangzhou is that the annual average high-temperature days and intensity in different return periods are large, resulting in higher high-temperature danger. Shanghai is affected by the ocean monsoon in summer and has few high-temperature days, strong high-temperature preventability, and low vulnerability indicators in populated areas, which will lower the overall high-temperature risk index in Shanghai. Comparing Nanjing with Hefei, the population scale of the two cities is similar, while the built-up area of Nanjing is larger, the number of high-temperature days and intensity in Hefei are relatively larger, and the high-temperature preventive ability of Nanjing is stronger than that of Hefei, so the high-temperature risk of Hefei is generally higher than that of Nanjing.

According to the calculation model of the high-temperature risk index and the influence of each factor and through comprehensive comparison and analysis of Fig. 3 to Fig. 8, high-temperature danger factors are the most important when it comes to the influence of multiple factors on the high-temperature risk index. The high-temperature danger in different return periods in regions including southern Chuzhou, southwest Jinhua, southwest Xuancheng, and northern Anqing is high, so the calculated high-temperature risk is at a high level. Among them, the high-temperature risk of southern Chuzhou is the highest in the study area, and the maximum and average values of the high-temperature risk index in different return periods are maintained at approximately 0.32 and 0.30, respectively, while the high-temperature risk index changes little with increasing return period. Although the high-temperature vulnerability of the central urban areas in each city is relatively higher, the high-temperature risk level of these areas has been reduced to a certain extent due to strong preventability.

This paper uses multisource data, including daily maximum temperature data observed from 1951 to 2018 by meteorological stations, socioeconomic statistical data of the Yangtze River Delta urban agglomeration in 2018, remote sensing image data and basic geographic information data, and uses the generalized extreme value method to calculate the average annual high-temperature intensity and days with return periods of 5, 10, 20 and 30 yr, and then spatializes 8 high-temperature vulnerability factors and 11 prevention capacity factors. On this basis, the high-temperature risk assessment model is constructed by considering the influence of each factor determined from the analytic hierarchy process and expert scoring method, and the spatial distribution characteristics of high-temperature risk in different return periods of the study area are analyzed.

Analyzing the spatial distribution of high-temperature risk in different return periods and comparing the spatial distribution of high-temperature danger, vulnerability and preventability, the high-temperature risk index and its spatial distribution difference decrease with the increase in return period. The high-temperature risk in the central urban area is relatively higher than that in the surrounding suburban areas, and it gradually decreases as latitude and distance from the ocean decrease. On the pixel scale, the high-temperature risk in some areas of southern Chuzhou is the highest. On the urban scale, Jinhua, Hangzhou and Xuancheng in the southwestern part of the study area occupy the top three positions in terms of average high-temperature risk. Among the provincial capitals and municipalities, Hangzhou has the largest high-temperature risk, Shanghai has the smallest risk, and Nanjing and Hefei are at the intermediate risk level.

By comprehensively considering the influencing factors of high-temperature risk, this paper presents the spatial distribution of high-temperature danger, vulnerability and preventability in different return periods and then analyzes the spatio-temporal distribution characteristics of high-temperature risk in the study area. Due to there no reliable data to evaluate the high-temperature risk in different return periods, the research tried to estimate spatial distribution of each intermediate influencing factor reasonably to ensure its higher precisions. The calculation of high-temperature days and intensity in different return periods requires a time series of observed air temperature data, while the number of stations with long-term observations is limited, which may introduce some errors to the spatial distribution of high-temperature danger with 1 km spatial resolution, especially the estimation of daily maximum temperature in areas with large topographic fluctuations and complex underlying land surface types. In addition, some socioeconomic factors used in assessing high-temperature vulnerability and preventability are difficult to quantified and accurately spatialized. At the same time, the determination of the influence of different factors has a certain human subjectivity, which affects the accuracy of the calculation results of the high-temperature risk index to a certain extent. These deficiencies need to be further studied and analyzed.