Mapping the Spatial and Seasonal Details of Heat Health Risks in Different Local Climate Zones: A Case Study of Shanghai, China

Abstract

In the context of global climate change and rapid population growth, more people in cities are facing heat threats. Although health risk assessment is critical for reducing heat-related morbidity and mortality, previous studies have not accurately identified the spatial details of heat risk levels on a fine scale within a complete framework. Therefore, this study developed a systematic method to conduct a spatially explicit assessment of heat-related health risks using local climate zones (LCZs) in Shanghai, China. First, multisource data were used to map LCZs in Shanghai. Second, a modified temperature-humidity index, population density, and ecological parameters were employed to construct a heat hazard–exposure–vulnerability framework for heat risk assessment. Finally, the differences in heat-related health risks among LCZs were compared. The results indicate that in Shanghai (1) the LCZ concept could help estimate the heat health risk (HHR) at the fine block level, and the area proportion of LCZ5 (open mid-rise buildings) accounted for more than 50%; (2) detailed spatial patterns of heat risk levels were similar in spring, summer, and autumn, but different in winter due to seasonal variations in heat hazards; and (3) the built LCZs usually had higher heat risk levels than natural land cover LCZs, with LCZ2 (compact mid-rise), LCZ3 (compact low-rise), and LCZ5 facing the most serious heat risks. The high-rise LCZs might reduce the heat risk level in hot seasons owing to shading effects but add to this risk in winter. These findings contribute to our understanding of HHR assessment.

1. Introduction

Owing to their large number of socioeconomic activities and complete infrastructural facilities, cities have become attractive places for more people to live [1,2]. According to the sixth report released by the Intergovernmental Panel on Climate Change (IPCC), the world surface temperature was approximately 1.1 °C above 1850 to 1900 levels during 2011 to 2020 [3], which had a significantly negative impact on the daily lives of city dwellers [4,5,6,7]. Accordingly, the combination of climate change and rapid urbanization will result in greater heat health risks (HHRs) for more people in the future [8,9,10,11]. Therefore, it is essential to strengthen our understanding of heat-related health risks faced by urban residents and make cities more resilient to disasters and challenges [12,13].

A reasonable and reproducible HHR assessment framework is crucial for obtaining accurate heat risk patterns [14,15,16]. However, the concept of HHR has become complicated, and research methodologies have evolved in recent years [16,17,18,19,20]. To date, most HHR assessment studies have employed Crichton’s risk triangle as a research paradigm, which uses three essential indicators: hazard, exposure, and vulnerability [18]. The IPCC used the same heat risk assessment framework in its fifth and sixth assessment reports [3]. Owing to the profound influence of the IPCC, an increasing number of studies have used a hazard–exposure–vulnerability framework [15,16,19,21,22,23,24,25]. Hazard refers to the potential intensity of the thermal environment that affects human health, where high temperatures are usually selected as the indicator [15]. Exposure refers to the presence of people who might be affected, with population density widely used as a parameter [20,26]. Vulnerability focuses on people who are sensitive to heatwaves, such as the elderly, children, low-income residents, and disadvantaged people [19,27,28]. Some studies have found that heat risk is not equally distributed across urban spaces, resulting in a more serious threat to these people [23,26,29,30].

Although previous heat risk assessment studies employing the hazard–exposure–vulnerability framework have proven reliable, there are still many theoretical and methodological questions that require further exploration. First, the selection of parameters to characterize the hazard, exposure, and vulnerability indicators has not reached consensus because the HHR index is a complex combination of the three. Some studies have employed only a single parameter to represent the indicator or assigned the indicator the same weight [15,29]; however, this approach may have overlooked other important influencing factors. For instance, previous studies have tended to use land surface temperature (LST), air temperature or the wet-bulb globe temperature (WBGT) to characterize heat hazards (exposure), which may overestimate or underestimate their negative effects on human health [31,32,33,34]. As a result, the use of outdoor thermal comfort as a parameter to assess the effect of heat or cold waves on human health may be a better choice because of its close correlation with the perception of humans [35,36,37,38]. Many human comfort indicators have been proposed, such as the discomfort index (DI), universal thermal climate index (UTCI), and modified temperature-humidity index (MTHI) [39,40,41,42]. Although these comfort indices consider meteorological elements that affect human health—such as humidity and wind speed—they have rarely been used in the current literature to assess heat hazards [43,44]. However, a large number of parameters are involved in heat risk assessment, making it difficult for decision makers to identify the dominant driving factor. For example, 28 parameters were selected in a previous study to evaluate heat vulnerability in Chicago [24].

Second, most previous studies were conducted at administrative units where districts were usually selected as the spatial analysis unit, mainly because of the difficulty in accessing population data [8,19,45]. Typically, census data are collected at the administrative level and cannot provide fine-scale information to capture the detailed spatial patterns of heat risks. Fortunately, the local climate zone (LCZ) concept was proposed in 2012 and has been widely adopted as a framework for conducting urban climate studies [46,47,48,49]. The standard LCZ concept has 10 built and seven land-cover LCZ types, obtaining reliable results of relationships between urban morphology and the urban thermal environment [50,51,52]. Therefore, it might be beneficial to use LCZs as a fine spatial analysis unit for heat risk assessment [15,53]. Third, heat risk assessments should not only focus on hot seasons, as cold waves can also pose a severe threat to human health [8,54,55]. Therefore, it is important to explore seasonal variations in HHRs.

Given the current research gaps, Shanghai, the core city in eastern China, was selected as a case study to assess the spatial patterns and seasonal variations in HHRs at a fine spatial unit. The main objectives of this study were to (1) map and analyze the spatial characteristics of the LCZs in Shanghai using multi-source data; (2) construct an innovative heat risk assessment framework to quantify heat-related health risk levels by combining heat hazard, exposure, and vulnerability; and (3) compare the differences in HHR levels among different LCZ types and explore the potential reasons for these. The results will provide a scientific basis for urban management and planning.

2. Materials

2.1. Study Area

Shanghai—one of the largest industrial and commercial cities in China and with an administrative region covering approximately 6340.5 km²—is located south of the Yangtze River Estuary and north of Zhangzhou Bay. By the end of 2021, Shanghai had a population of 24.89 million with an average life expectancy of 84.11 years. Shanghai has a subtropical monsoon climate with an average temperature of 17.9 °C and precipitation of 1474.5 mm. According to the standards of the Shanghai Meteorological Administration, spring (March and April) and autumn (October and November) are usually short in Shanghai, whereas summer (May, June, July, August, September) and winter (December, January, February) are long. The average altitude of Shanghai is 2.19m, and the highest point, on Dajinshan Island, is 103.7 m above sea level. Shanghai is a developed city with a high population and building density, making it an ideal sample to explore the variations in heat-related health risks across different LCZ types. Consequently, an urban core located on the Outer Ring Expressway of Shanghai was selected as the study area (Figure 1).

2.2. Data Sources

Multiple data sources—including satellite imagery, road networks, building data, population density, and other auxiliary data—were integrated to obtain LCZ maps and assess heat-related health risks. Cloud-free seasonal Landsat 8 images from 2020 to 2022 collected from the United States Geological Survey were employed to retrieve seasonal LST, Normalized Difference Vegetation Index (NDVI), and Modified Normalized Difference Water Index (MNDWI), which were the basis for quantifying heat hazards using the MTHI. Additionally, the area proportions of different land cover types—such as water, vegetation, and impervious surface areas (ISAs)—that were required in the process of LCZ mapping were also obtained from Landsat images. All originally downloaded Landsat 8 images were preprocessed using radiation correction. The road network was obtained from OpenStreetMap and modified via visual interpretation to generate an LCZ boundary. Building data, including information on building outline and height, were gathered from the Resource and Environment Science and Data Center, China. The WorldPop dataset (2020) was used to characterize the population distribution and density of each LCZ. Detailed information on the data used in this study is presented in

Table 1. Data sources in this study.

Type	Data	Time	Resolution	Cloud Cover	ID	Data Source	Description
Remote sensing data (tif)	Landsat 8/9 images	21 January 2020	30 m	5.01%	LC81180382020021LGN00	https://www.gscloud.cn/, accessed on 1 September 2023.	For the calculation of NDVI, LST, NDMI, ISA, and MNDWI. (1) NDVI, ISA, and MNDWI are used for LCZ mapping. (2) LST and NDMI are used for assessing hazard. (3) ISA is used for assessing exposure. (4) NDVI and MNDWI are used for assessing vulnerability.
		16 August 2020	30 m	2.14%	LC81180382020229LGN00
		23 March 2022	30 m	0.27%	LC09_L2SP_118038_20220323_20220424_02_T1	https://earthexplorer.usgs.gov/, accessed on 1 September 2023.
		2 November 2022	30 m	12.73%	LC08_L2SP_118038_20191102_20200825_02_T1
	Population data	2020	100 m	-	-	https://www.worldpop.org/, accessed on 1 September 2023.	For the calculation of population density and the proportion of sensitive groups in assessing exposure and vulnerability.
Vector data (shp)	Road network	2020	-	-	-	https://export.hotosm.org/en/v3/, accessed on 1 September 2023.	For the partition of block boundaries.
	Building data	2020	-	-	-	https://www.resdc.cn/, accessed on 1 September 2023.	For the calculation of building density and height in LCZ mapping.

.

3. Methods

The framework and working steps used in this study are illustrated in Figure 2. First, the heat-related health risk assessment framework was used as a guideline. The workflow was then divided into four steps: (1) the input of multi-source data; (2) the generation of LCZ maps using building height, area proportion of land cover, and decision tree based on the road network; (3) the application of the HEV model to conduct the HHR assessment; and (4) the spatial and temporal analysis for the LCZs and health risk using ArcGIS 10.8.

3.1. LCZ Mapping

3.1.1. LCZ Classification Adjustment

Based on a survey of Shanghai’s current situation, the industrial zone is typically located outside the study area, and the proportion of lightweight low-rise areas is quite low [56]. Therefore, we reclassified the original standard-built LCZ 7–10 types to LCZ 7–9 according to the values of the geometric and surface cover properties for standard LCZ 1–6. Consequently, in our study, LCZ 7, 8, and 9 refer to sparse high-rise, sparse mid-rise, and sparse low-rise areas, respectively [37]. However, no changes were observed in the original land-cover LCZ types in this study. Detailed information regarding the classification criteria for the LCZ types can be found in a previous study [46,57].

3.1.2. LCZ Parameter Calculations

Building height and land cover area fraction were the two basic elements in the LCZ mapping process. The mean building height is the average height of all buildings located inside a block boundary and is calculated based on the building data, which can be expressed as follows:

$$BH={\displaystyle \frac{{{\displaystyle \sum }}_{i=1}^{n}B{H}_i}{n}}$$

(1)

where n is the number of buildings in the block, $B{H}_i$ is the height of building $i$. The BH thresholds to distinguish high-rise, mid-rise, and low-rise were >25 m, 10–25 m, and <10 m, respectively.

The linear spectral mixture analysis (LSMA) model was applied to Landsat 8 images to estimate the area fractions of ISA, vegetation, and soil in each block [58]. LSMA can be described as follows:

$$R_b={{\displaystyle \sum }}_{i=1}^n{f}_i{R}_{\left(i,b\right)}+e_b$$

(2)

where $R_b$ is the reflectance of band b; n is the number of endmembers; $f_i$ is the area proportion of endmember $i$; $R_{\left(i,b\right)}$ is the reflectance of endmember $i$ in band b. In addition, NDVI was employed to identify LCZ A-D, and MNDWI was used to obtain the areas of water bodies [59]. The formulae for NDVI and MNDWI are as follows:

$$\mathrm{NDVI}={\displaystyle \frac{NIR-RED}{NIR+RED}}$$

(3)

$$\mathrm{MNDWI}={\displaystyle \frac{GREEN-SWIR}{GREEN+SWIR}}$$

(4)

where NIR, R, GREEN, and SWIR in Landsat 8 refer to band5, band4, band6, and band3, respectively. The most widely used method in GIS-based LCZ is decision-making algorithms, which is to match the calculated LCZ parameters with their reference ranges [60]. Therefore, the decision tree method was used to classify the LCZ in this paper.

3.2. Hazard–Exposure–Vulnerability Assessment

3.2.1. Heat Hazard Assessment

Urban outdoor thermal comfort was used to describe the degree of human perception of the current thermal environment. Compared with traditional air temperature or LST, outdoor thermal comfort is more closely related to human health [25,61]. In addition, MTHI has higher spatial resolution and the required data are more easily accessible compared to other thermal comfort indicators such as PET, UTCI, and H. Therefore, the MTHI was used to assess thermal outdoor comfort [39]. Typically, outdoor thermal discomfort is considered a high heat hazard. The MTHI is estimated as follows:

$$\mathrm{MTHI}=1.8\times \mathrm{LST}+32-0.55\times \left(1-\mathrm{NDMI}\right)\times \left(1.8\times \mathrm{LST}-26\right)$$

(5)

$$\mathrm{NDMI}={\displaystyle \frac{NIR-SWIR}{NIR+SWIR}}$$

(6)

where NDMI is the Normalized Difference Moisture Index. LST can be retrieved using the Radiative Transfer Equation algorithm [62]. To explore seasonal variations in the MTHI under the same conditions, the mean standard deviation method was applied to represent relative thermal comfort [39,63]. The MTHI was reclassified into five categories, as listed in

Table 2. Classification of outdoor thermal comfort levels.

MTHI Level	Range of MTHI Values
	Spring, Summer, Autumn		Winter
1	MTHI < M − 1.5 a	Cool	M + 1.5 a ≤ MTHI	Warm
2	M − 1.5 a ≤ MTHI < M − 0.5 a	Slightly hot	M + 0.5 a ≤ MTHI < M + 1.5 a	Slightly Warm
3	M − 0.5 a ≤ MTHI < M + 0.5 a	Hot	M − 0.5 a ≤ MTHI < M + 0.5 a	Slightly Cold
4	M + 0.5 a ≤ MTHI < M + 1.5 a	Very hot	M − 1.5 a ≤ MTHI < M − 0.5 a	Cold
5	M + 1.5 a ≤ MTHI	Extremely Hot	MTHI < M − 1.5 a	Very cold

Where M is the mean value of MTHI, and a is the standard deviation of MTHI.

. The higher the MTHI, the greater the heat hazard.

The heat hazard for each LCZ was estimated by the following formula:

$$X_H={{\displaystyle \sum }}_i^n{A}_{mi}\times C_i$$

(7)

where $X_H$ is the heat hazard level for the LCZ, $A_{mi}$ is the area proportion of level i for the MTHI inside the corresponding LCZ, and $C_i$ is the MTHI level. Therefore, the heat hazard is also a relative assessment.

3.2.2. Heat Exposure Assessment

Population density is usually selected as an exposure indicator because if there are no people, there is no exposure [15]. Additionally, there is a strong positive correlation between ISA and LST, indicating that a high ISA fraction affects population exposure [64,65]. Therefore, a modified population ISA-weighted heat exposure assessment framework is proposed here to explore the spatial patterns of human exposure to heat hazards using the following formula:

$$X_E=C_p\times W_p+C_I\times W_I$$

(8)

where $X_E$ is the estimated heat exposure for the LCZ, $C_p$ is the population density level for the LCZ, which is set according to the distribution of population data in Shanghai. $C_I$ is the level for the area proportion of ISA inside the LCZ, which is set according to the setting range of the proportion of building area by Zhou et al. [53]. $W_p$ and $W_I$ are the weights of the population density and area fraction of ISA, respectively. Here, both $W_p$ and $W_I$ were set to 0.5. Similar to the MTHI classification, both the population density and ISA fraction were reclassified into five levels (

Table 3. Classification of population density and ISA fraction.

Level	Range for Population Density (PD) (103/km2)	Range for ISA Fraction (%)
1	PD < 10	0 ≤ ISA < 20
2	10 ≤ PD < 20	20 ≤ ISA < 40
3	20 ≤ PD < 50	40 ≤ ISA < 60
4	50 ≤ PD < 100	60 ≤ ISA < 80
5	100 ≤ PD	80 ≤ ISA < 100

). Finally, the estimated heat exposure levels were classified into five categories: very low, low, medium, high, and very high.

3.2.3. Heat Vulnerability Assessment

Heat vulnerability is defined as an individual’s sensitivity and capacity to adjust, cope with, and recover from outdoor thermal heat risks [29,66]. The elderly and children are more easily affected by an uncomfortable thermal environment [67,68]; therefore, the population density for certain age structures (<14 and >65 years) was selected as a vulnerability indicator. Moreover, vegetation and water can regulate micro-urban climates by cooling high temperatures, which is beneficial for sensitive populations. As a result, the proportions of vegetation and water in the area were chosen as vulnerability indicators as low proportions of vegetation and water can contribute to vulnerability [69]. Heat vulnerability can be calculated as follows:

$$X_V=C_{ps}\times W_{ps}+C_{wv}\times W_{wv}$$

(9)

where $X_V$ is the estimated heat vulnerability for the LCZ, $C_{ps}$ is the sensitive population density level for the LCZ, and $C_{wv}$ is the area proportion level of water and vegetation inside the LCZ. Based on the experience of previous studies [15,25,70], $W_{ps}$ (0.5) and $W_{wv}\left(0.5\right)$ are the weights of population density and area fraction of the ISA, respectively.

3.3. Heat Health Risk Assessment

Based on the hazard–exposure–vulnerability assessment framework, the final HHR was estimated using a combination of normalization indicators with different weights, as follows:

$$HHR=X_H\times W_H+X_E\times W_E+X_V\times W_V$$

(10)

where HHR is the heat health risk level for each LCZ; $X_H,X_E,\mathrm{and}{X}_V$ represent the heat hazard, heat exposure, and heat vulnerability indicators, respectively; $W_H,W_E,\mathrm{and}{W}_V$ refer to the weights of the hazard, exposure, and vulnerability indices, respectively.

To guarantee that the three components were comparable and enabled aggregation, all three HHR indicators were classified into five levels. In addition, all high-level indicators were designed to have a positive relationship with the HHR. Relative weights were identified using the Analytic Hierarchy Process analysis, and the values for $W_H,W_E,\mathrm{and}{W}_V$ were 0.5, 0.2, and 0.3, respectively. Finally, the HHR was classified into five levels: very low (1–1.5), low (1.5–2.5), medium (2.5–3.5), high (3.5–4.5), and very high (4.5–5) in ArcGIS 10.8.

4. Results

4.1. LCZ Spatial Patterns

The spatial distribution of the LCZs in Shanghai is shown in Figure 3. For descriptive statistical features, there were 4501 LCZs in total, and detailed information is presented in

Table 4. Descriptive characteristics of LCZs in Shanghai.

Descriptive Features	Built LCZ Types
	LCZ1	LCZ2	LCZ3	LCZ4	LCZ5	LCZ6	LCZ7	LCZ8	LCZ9	Total
Number	12	203	213	489	2455	306	102	174	40	3994
Area (km2)	0.79	19.24	24.32	52.63	373.75	42.19	12.32	21.21	4.98	551.43
Area Proportion (100%)	0.12	2.90	3.67	7.94	56.37	6.36	1.86	3.20	0.75	83.16
	LCZA	LCZB	LCZC	LCZD	LCZE	LCZF	LCZG	Total
Number	96	111	23	145	94	16	22	507
Area (km2)	19.45	21.21	7.07	26.77	13.33	3.95	19.87	111.65
Area Proportion (100%)	2.93	3.20	1.07	4.04	2.01	0.60	3.00	16.84

. The proportions of built LCZ types and natural land cover LCZ types in the study area accounted for 83.16 and 16.84%, respectively, indicating high artificial building and surface fractions in Shanghai city. Among all LCZ types, LCZ5 (open mid-rise) covered the largest area (52.63 km²), accounting for 56.37% of the study area. LCZ1 (compact high-rise) accounted for only 0.12% of the total area, with the lowest coverage. The vegetation LCZ types (LCZ A–D) accounted for 11.24%, and the proportion of the water body (LCZ G) was 3.00%. For the spatial pattern features, LCZ5 was distributed throughout the study area, whereas the compact LCZ types (LCZ 1–3) were mainly distributed in the city center. Natural LCZs were mainly located in the eastern and southeastern regions.

4.2. Spatial Heat Hazard Patterns

4.2.1. Seasonal Spatial Patterns of LSTs in Shanghai

Figure A1 shows the time series of Landsat LST in the study area from 2020 to 2022, calculated by Climate Engine [71]. Due to cloud cover limitations, the mean LST values for the four seasons were slightly lower than the actual conditions, particularly for the summer image (31 July 2020), which significantly reduced the overall mean (

Table A2. LST of Landsat images available in the study area from 2020 to 2022.

	Data	Mean	Max	Min	Data	Mean	Max	Min	Data	Mean	Max	Min
Winter	13 January 2020	8.69	19.24	2.75	30 December 2020	1.49	18.58	−12.73	2 January 2022	9.07	22.62	4.63
	21 January 2020	13.18	24.91	7.26	15 January 2021	14.17	30.76	−0.07	18 January 2022	9.46	20.71	2.40
	29 January 2020	10.59	26.43	3.58	8 February 2021	−8.77	2.38	−14.26	27 February 2022	18.53	39.13	8.86
	22 February 2020	18.49	36.77	10.12	24 February 2021	18.93	35.57	7.32	11 December 2022	9.81	19.93	5.44
	14 December 2020	8.35	20.94	3.20	1 December 2021	10.64	24.66	−6.38	20 December 2022	12.36	28.56	4.82
	22 December 2020	12.36	27.32	5.38	17 December 2021	10.65	20.11	6.63
Spring	17 March 2020	24.51	49.21	12.55	28 March 2021	25.66	49.90	14.73	7 March 2022	7.03	28.61	−10.01
	25 March 2020	23.26	38.65	0.71	5 April 2021	22.91	45.03	12.60	15 March 2022	24.11	44.27	13.09
	2 April 2020	21.66	41.35	7.51	13 April 2021	26.83	47.28	14.38	23 March 2022	20.50	43.43	4.07
	10 April 2020	15.95	25.20	11.24	21 April 2021	29.23	48.53	16.03	8 April 2022	32.94	55.75	18.24
	26 April 2020	21.58	31.83	13.86	29 April 2021	33.63	58.99	19.90	16 April 2022	24.22	35.28	11.40
Summer	4 May 2020	30.61	49.49	19.67	24 June 2021	32.76	48.44	21.02	27 June 2022	27.02	47.33	10.25
	12 May 2020	37.39	61.83	24.02	10 July 2021	42.08	64.10	29.03	5 July 2022	49.47	68.87	31.77
	20 May 2020	33.36	56.83	18.96	18 July 2021	45.12	66.74	27.53	13 July 2022	46.24	78.14	26.31
	28 May 2020	29.51	35.99	18.64	3 August 2021	30.56	38.02	21.17	28 July 2022	38.42	62.06	30.13
	23 July 2020	42.44	62.68	25.45	19 August 2021	39.56	57.99	28.77	29 July 2022	48.07	64.21	32.83
	31 July 2020	0.90	23.87	−57.64	27 August 2021	32.85	51.21	14.29	6 August 2022	36.52	51.72	2.47
	16 August 2020	45.47	70.29	33.93	20 September 2021	29.14	53.51	11.15	9 August 2022	41.68	62.64	−72.62
	24 August 2020	40.00	67.06	28.47	28 September 2021	36.35	63.98	24.24	14 August 2022	42.06	67.15	30.65
	1 September 2020	40.36	54.84	26.72	2 May 2022	32.80	61.41	18.43	22 August 2022	50.54	77.53	38.29
	9 September 2020	37.56	63.17	23.05	18 May 2022	32.28	50.12	17.01	30 August 2022	42.19	61.75	21.88
	25 September 2020	28.68	40.72	17.54	21 May 2022	29.16	59.77	19.12	7 September 2022	39.41	66.33	29.40
	7 May 2021	33.70	57.76	20.23	7 June 2022	27.03	40.10	16.38	17 September 2022	33.32	54.22	22.48
	15 May 2021	40.10	67.04	20.64	11 June 2022	36.08	54.81	25.83	23 September 2022	32.48	49.67	20.41
Autumn	11 October 2020	29.08	56.79	14.27	4 November 2021	22.46	29.99	16.56	21 October 2022	23.69	39.07	13.22
	27 October 2020	15.85	21.71	11.84	7 November 2021	19.78	26.79	9.62	25 October 2022	25.03	40.36	10.32
	12 November 2020	22.68	41.57	6.26	15 November 2021	18.41	37.84	3.02	2 November 2022	24.78	43.07	17.46
	28 November 2020	14.57	31.33	1.78	23 November 2021	11.24	22.95	4.41	7 November 2022	19.45	33.83	8.79
	6 October 2021	32.52	35.48	26.06	1 October 2022	36.97	52.96	26.47	10 November 2022	28.98	48.31	20.03
	22 October 2021	20.29	35.77	13.11	9 October 2022	26.53	46.75	13.31	26 November 2022	22.54	34.19	16.80
	30 October 2021	23.55	44.15	8.47	17 October 2022	25.03	36.55	18.75

). However, in this study, four clear-sky scenes were selected to represent the corresponding seasons, which better reflect the real conditions. Therefore, the slightly higher LST values are reasonable and meet the needs of thermal risk research.

The seasonal spatial patterns of the LST in Shanghai are shown in Figure 4. For seasonal variations, the distributions of LST in spring, summer, and autumn were similar, with high LSTs located in the northern and northeastern parts, whereas the southeastern region had lower LSTs. In winter, there were no obvious LST distribution features, which differed from those of the other seasons. Overall, the built LCZ types had a higher LST, whereas the vegetation and water LCZ types had a lower LST owing to heat absorption and shading. Descriptive statistics show large LST differences among the LCZ types in spring and summer (

Table 5. LST values of LCZs in Shanghai.

LCZ	Spring			Summer			Autumn			Winter
	Mean	Max	Min	Mean	Max	Min	Mean	Max	Min	Mean	Max	Min
LCZ1	28.93	32.32	25.57	33.82	36.34	31.74	31.51	34.01	29.50	10.37	12.10	8.31
LCZ2	31.06	41.41	20.85	35.13	41.93	30.37	33.10	42.23	28.09	11.35	18.38	7.75
LCZ3	30.14	40.57	17.59	34.59	41.43	30.09	32.43	41.25	25.52	10.95	16.61	3.77
LCZ4	28.17	35.70	22.58	33.08	38.22	30.01	30.86	37.25	27.85	10.06	14.84	6.64
LCZ5	29.45	46.14	19.59	33.90	45.40	29.35	31.95	47.96	27.00	10.67	19.39	2.43
LCZ6	28.69	37.41	20.52	33.15	39.80	29.19	31.10	38.13	27.26	10.54	14.78	2.77
LCZ7	27.15	31.74	23.10	32.08	35.39	29.72	29.96	33.97	27.80	9.86	12.35	8.60
LCZ8	28.14	32.56	23.14	32.52	35.93	29.43	30.71	34.58	27.61	10.41	12.57	8.68
LCZ9	27.58	31.33	23.39	31.85	35.12	29.55	30.02	33.55	27.36	10.47	12.24	8.87
LCZA	26.70	33.46	20.76	31.14	35.72	28.90	29.42	33.73	27.15	10.27	13.04	8.13
LCZB	27.32	34.68	21.35	31.50	36.87	28.80	29.79	35.56	27.03	10.51	13.32	8.22
LCZC	27.77	34.35	24.36	32.20	38.48	28.70	30.52	36.19	27.00	10.82	13.21	8.90
LCZD	27.90	35.00	19.95	32.57	36.89	29.53	30.44	35.08	26.28	10.27	14.15	6.64
LCZE	28.24	35.69	20.68	33.43	38.67	30.00	31.10	38.04	27.34	9.81	14.07	3.89
LCZF	28.97	33.44	22.32	34.21	36.32	31.23	30.80	34.34	28.13	10.44	11.99	8.22
LCZG	20.26	30.30	17.90	29.50	34.42	28.30	27.79	33.94	26.03	8.20	11.47	7.15

). For example, in summer, LCZ2 (compact mid-rise) had the highest LST (35.13 °C), followed by LCZ3, LCZ F, LCZ5, LCZ1, LCZ E, and LCZ6 at 34.59, 34.21, 33.9, 33.82, 33.43, and 33.15 °C, respectively, while the LST of LCZ G (water body) was the lowest (29.5 °C). Typically, the compact LCZ types (LCZ1–3) had higher LSTs than the open and sparse LCZ types (4–9) regarding other spatial features (building height). Another interesting phenomenon was that if the building density was similar, the mid-rise LCZ types had a higher LST than the high- and low-rise LCZ types, possibly due to the shading and cooling effect of the high-LCZ types (LCZ1, 4, and 7) [72,73].

4.2.2. MTHI Spatial Patterns

The characteristics of the seasonal spatial distribution of thermal comfort in Shanghai are shown in Figure 5. As Shanghai has a Cfa (warm temperate, fully humid, hot summer) climate based on the Köppen–Geiger climate classification, the weather was humid and hot except for winter [74]. Therefore, the spatial pattern of the MTHI corresponded to a combination of LST and NDMI. Typically, high and low LST (in winter) generate a high level of heat hazard. The dominant MTHI levels in Shanghai in spring, summer, and autumn were hot and very hot, whereas those in most regions in winter were cold and slightly cold.

The area proportions of different MTHI levels among the LCZ types are shown in Figure 6. It can be seen that the built LCZs had more extremely hot areas than the natural ones. In addition, LCZ 2, 5, and 8 generated a high-level heat hazard more easily, indicating that the mid-rise LCZ types seemed to have the most uncomfortable thermal environments. For seasonal variations, extremely hot levels were found in almost all LCZ types in summer and autumn but were not observed in spring. However, the heat hazard level was mainly estimated using LST and NDWI, and a high hazard did not indicate high heat risk, which required exposure and vulnerability to assess its level.

4.3. Spatial Patterns of Heat Exposure

The spatial patterns of population density, ISA proportion, and heat exposure levels among the LCZs are shown in Figure 7a–c, respectively. Figure 7d shows the area proportions of different exposure levels for the entire study area and among the LCZ types. Figure 7a shows that the population density on the west bank of the Huangpu River was significantly higher than that in the east. The most densely populated area was found in the city center, and the population density of some LCZs was greater than 100,000/km². The proportion of LCZs with high ISA fractions (>60%) was 67.01%, indicating that Shanghai has a dense built environment. For the entire study area, a very high level of heat exposure accounted for only 3.93%, but high-level exposure was 36.01%. For different LCZ types, the proportion of areas with very high heat exposure was higher in LCZ1 5 and LCZ E than in the other areas. Conversely, LCZ9 and LCZ A–C exhibited the lowest heat exposure levels with very low and low heat exposure levels in LCZ A accounting for more than 70% of the total.

4.4. Spatial Patterns of Heat Vulnerability

Figure 8 shows the spatial distributions of sensitive population density, a combination of water and vegetation, heat vulnerability levels, and the area fractions of different vulnerabilities for Shanghai and among different LCZ types. The spatial patterns of the sensitive population were similar to that of the entire population; however, the population density was much lower. Almost 60% of the LCZs have a sensitive population density less than 5000/km². The spatial pattern of water and vegetation was opposite to that of ISA, indicating that the low area of natural land cover for most LCZs in Shanghai might add to the level of heat vulnerability. Figure 8c shows that the very high levels of heat vulnerability were mainly distributed in the city center, where the region has a high sensitive population density and sparse natural water and vegetation. The total area proportions of high and very high levels of heat vulnerability accounted for 27.78%; the value for LCZ1 was 80.52%, followed by 66.15% for LCZ2. Compared to the compact (LCZ1–3) and open (LCZ4–6) LCZ types, the sparse LCZ types (7–9) had lower heat vulnerability owing to their lower population levels.

4.5. Spatial and Seasonal Variations in HHRs

The seasonal spatial patterns of HHRs, derived from the hazard–exposure–vulnerability framework, are shown in Figure 9. The HHR to the west of the Huangpu River was much higher than to the east, and a very high level of risk was concentrated in the northern and western parts, which were very close to the city center. The spatial patterns of the high-risk region (defined as “very high” and “high”) were similar to high-heat hazards—defined as “extremely hot” and “very hot”—characterized by the MTHI in spring, summer, and autumn. In winter, the high-risk region was mainly located in the city center, which was similar to exposure and vulnerability. Notably, the high-risk region in winter had a very cold thermal environment, which was different from that in other seasons.

Figure 10 shows the area proportions of different HHR levels among the seasons. High-risk regions accounted for 26.26, 27.66, and 28.36% of the total area in spring, summer, and autumn, respectively. The high-risk region was 23.07% in winter, indicating that winter had a lower heat-related health risk than other seasons. Overall, based on our results, a level of 26–28% indicated that a substantial number of areas of Shanghai were experiencing high heat risk, even though high-risk areas did not constitute the largest proportion.

5. Discussion

5.1. Selection of Indicators for the HHR Assessment Framework

In the context of global climate change and rapid urbanization, it is estimated that more urban area residents will face severe thermal risks, such as heat or cold waves [6,75]. Therefore, it is essential to construct a scientific and practical framework to assess the heat risk at fine spatial scales, as most current studies focus on administrative or district levels [14,15].

A framework that combines heat hazard, exposure, and vulnerability was proposed in this study to assess the differences in heat risk among the LCZ types. Most previous studies have only used air temperature or LST to characterize heat hazards and ignored thermal reflection [31,76]. Therefore, the thermal comfort characterized by the MTHI was employed in this study to assess the heat hazard, which better reflects human perception in a spatially explicit way owing to the application of remote sensing and enables a more comprehensive heat hazard quantification. Additionally, compared to the limited number of meteorological stations, application of the MTHI made it possible to assess heat hazard seasonal variations for each LCZ at a neighborhood level. For the heat exposure and vulnerability assessment, population density was selected as a key element because we believe that the impacts of heat risk depend not only on hazards but also on the number of residents. Specifically, children and the elderly were both considered in our vulnerability assessment, whereas previous studies selected only the latter. Here, the final HHR was the weighted sum of the three indicators rather than equal weights, because they contributed differently to the heat risk. Overall, the HHR assessment proposed here can help quantify heat risk in a more comprehensive way, especially at a finer scale, owing to the selection of the three indicators and their reasonable multiplication.

5.2. HHR Differences among LCZ Types

The urban thermal environment can change significantly over short distances because of the complicated landscape patterns in cities, which profoundly affects the heat risk of human habitation [37]. Since its emergence in 2012, the new LCZ classification system has become a popular and standard research framework for studying urban heat islands [46]. However, only a few heat risk assessment studies have employed the LCZ concept [15]; therefore, its application here as an analysis unit to quantify heat risk was undoubtedly qualified and innovative.

The area proportions of different HHR levels among the LCZ types across seasons are shown in Figure 11. Overall, the area of high-risk regions—referring to “high” and “very high” levels—in built LCZ types was much larger than that in landcover LCZ types except for LCZ E (bare rock or paved) and LCZ F (bare soil) where there is little ecological land, which was consistent with a previous study [15]. For seasonal variations, spring, summer, and autumn faced similar heat risks owing to their comparable spatial patterns of MTHI, which weighted 50% of the HHR. The very high heat risk levels in winter in LCZ1 and LCZ4 were much higher than those in the other seasons possibly due to the high buildings in these areas that create large areas of shading and generate a low LST, leading to very cold areas in winter. However, the cooling effect of high-rise LCZ types (1, 4, and 7) produced by shading might be helpful for residents in finding cool places to mitigate the negative impacts of heat waves in hot seasons [77]. This result was opposite to that of a previous study, in which compact high-rise buildings increased LST [37]. Our results also provided evidence that LCZ2, 3, and 5 faced the most severe heat risk at the neighborhood level, which was consistent with previous studies where LCZ2 and 5 had negative effects on the inhabitants [15]. The high population density of these LCZ types may be a reason for their high heat risk. However, by comparing Figs. 5 and 9, an interesting phenomenon was noted where a high fraction of high-hazard regions was found in LCZ8 (sparse mid-rise), which had a low proportion of high-risk regions, possibly because LCZ8 has low heat exposure and vulnerability due to its sparse population and more natural land cover [78].

5.3. Implications for Urban Management and Planning

There are contradictory or even opposing results of heat risk assessments in the current literature [15,53]. For instance, Ma suggested that urban planners may select LCZ6 to implement adaptation and mitigation strategies [15], whereas Zhou found that the area of high-risk regions was higher in LCZ6 than in LCZ4 [53]. Therefore, policy and decision makers should have a clear understanding of the specific data, methods, and results of different studies. From the results of this study, the HHR exhibited obvious variations among the LCZs in Shanghai city. Our results indicate that heat health risks increased with an increase in building density. Therefore, in urban renewal demolition or reconstruction—through the intensive use of land—LCZ2 (compact mid-rise) can be changed to LCZ6 (open low-rise) or 7 (sparse high-rise) to reduce heat risks. If the areas cannot be rebuilt, adding more vegetation and water, such as in pocket parks, could also improve living environments [79,80]. Additionally, Shanghai still has some LCZ E (bare paved) and LCZ F (bare soil) areas that can be utilized when planning future developments. Another suggestion is that all three indicators—hazard, exposure, and vulnerability—should be considered during urban planning.

5.4. Study Limitations and Future Work

This study had some limitations wherein, compared to the traditional heat-hazard assessment, the MTHI does not employ meteorological data such as wind speed and humidity, important elements for characterizing human feelings regarding the thermal environment [41]. Furthermore, the results of the heat hazard assessment were limited by the time of the remote sensing data acquisition. To assess heat exposure and vulnerability, more meaningful socio-ecological-economic parameters—unemployed population density, GDP, number of air conditioners, and distance to the closest hospital—should be considered in future studies [81].

6. Conclusions

A heat health risk assessment framework that employed heat hazards, exposure, and vulnerability was proposed in this study to quantify the spatial and seasonal variations in Shanghai city, focusing on the risk differences among different LCZ types. Multisource data, including Landsat 8 images, road networks, building data, and population density data, were employed to generate the LCZ and HHR maps using GIS spatial analysis methods. The main findings are summarized as follows.

(1)

Multi-source data can effectively generate Shanghai LCZ types at the neighborhood level. The results showed that the built LCZ types in Shanghai accounted for 83.16%, whereas natural land cover accounted for only 16.84%. LCZ5 (open mid-rise) and LCZ4 (open high-rise) accounted for the largest area proportions, whereas LCZ1 (compact high-rise) accounted for less than 1%.

(2)

The HHR index used here could not only quantify the spatial details of heat risk levels on a fine scale, but also describe seasonal variations. Spring, summer, and autumn had similar spatial patterns of heat risk, characterized by high temperatures, and they had a higher proportion of high-risk regions (27%) than winter (23%).

(3)

The HHR showed significant differences among different LCZ types. Typically, the built LCZ had a higher heat risk than the natural land cover LCZs, except for LCZ E (bare paved). LCZ2, 3, and 5 posed a more serious heat risk, while LCZ 6–9 were mainly exposed in low-risk regions. Generally, compact LCZs generate severe heat risks; however, in the context of similar building density, high-rise buildings would reduce the heat risk level in hot seasons owing to shading effects but add risks in winter.

Therefore, these findings would aid policymakers in future decisions. Particular attention should be devoted to the heat risks in blocks containing compact mid-rise buildings, while providing guidance for the construction of sparse high-rise blocks. Furthermore, in areas characterized by compact buildings, augmenting blue and green spaces can effectively regulate microclimate temperatures and mitigate heat-related risks. In addition, the proposed HHR assessment framework at the community level could add to our knowledge of the relationship between urban structures and thermal HHRs.