High-Temperature Risk Assessment and Adaptive Strategy in Dalian Based on Refined Population Prediction Method

Abstract

Extremely high temperatures can severely impact urban livability and public health safety. However, risk assessments for high temperatures in cold-region cities remain inadequate. This study focuses on Dalian, a coastal city in northeastern China. Utilizing multi-source data, we established a population density prediction model based on the random forest algorithm and a heat vulnerability index (HVI) framework following the “Exposure-Sensitivity-Adaptability” paradigm constructed using an indicator system method, thereby building a high-temperature risk assessment system suited for more refined research. The results indicate the following: (1) Strong positive correlations exist between nighttime light brightness (NL), Road Density (RD), the proportion of flat area (SLP), the land surface temperature (LST), and the population distribution density, with correlation coefficients reaching 0.963, 0.963, 0.956, and 0.954, respectively. (2) Significant disparities exist in the spatial distribution of different criterion layers within the study area. Areas characterized by high exposure, high sensitivity, and low adaptability account for 13.04%, 8.05%, and 21.44% of the total area, respectively, with exposure being the primary contributing factor to high-temperature risk. (3) Areas classified as high-risk or extremely high-risk for high temperatures constitute 31.57% of the study area. The spatial distribution exhibits a distinct pattern, decreasing gradually from east to west and from the coast inland. This study provides a valuable tool for decision-makers to propose targeted adaptation strategies and measures based on the assessment results, thereby better addressing the challenges posed by climate change-induced high-temperature risks and promoting sustainable urban development.

1. Introduction

In the context of global climate change, extreme heat events are becoming more frequent worldwide, especially in urban areas [1]. According to the Sixth Assessment Report (AR6) of the United Nations IPCC (IPCC) [2], compared with the period from 1850 to 1900, the global average temperature has risen by 1.45 °C, and extreme heat events have become more frequent, seriously affecting the economy, society, and people’s lives. Since the Industrial Revolution in the 19th century, human activities have become the main driver of global warming [3], and high temperatures also seriously threaten human health [4], with the most direct harm being heat-related diseases such as heat cramps, heat exhaustion, and heatstroke [5,6]. Koken P J M, Lin S, et al. found that a high average temperature significantly increases the hospitalization rate of acute myocardial infarction and congestive heart failure [7,8]. Wahid S et al. found that a 1 °C increase in temperature increased the risk of anxiety by 21% and depression by 24% [9].

According to previous studies, the frequency of high-temperature heat waves in cold cities is relatively low [10], past studies have mostly focused on traditional high-temperature cities, and historical meteorological data show that the number of days of high-temperature heat waves in cold cities is small, but as long as there is a high-temperature heat wave, it can easily be extreme. For example, in the summer of 2016, severe high-temperature and drought disasters occurred in Northeast China, causing economic losses of up to 15.6 billion yuan [11]. From late July to early August 2018, Dalian experienced continuous extreme high temperatures, with multi-day maximum temperatures exceeding 40 °C [12], posing a serious threat to residents’ lives and health [13]. At present, the shortcomings of cold cities in dealing with high-temperature risks are reflected in many aspects, such as infrastructure and residents’ adaptability. First, infrastructure is mainly designed to cope with cold climates, lacking adaptability to high temperatures, such as Harbin and other cold cities, with thick exterior building walls and small windows, insufficient ventilation design, and indoor temperatures significantly higher than outdoor temperatures [14]. For example, in Quebec, Canada, during the summer heat of 2018, hundreds of people sought medical attention for heat stroke due to a lack of experience in coping with heat [15]. Therefore, how to improve urban resilience in cold regions to cope with the multiple hazards caused by urban high-temperature disasters and ensure the safety of residents’ lives and sustainable urban development has become the focus of current research.

In the field of research on high-temperature risk, the selection and application of assessment methods are crucial for accurately judging the degree of risk and formulating effective strategies to deal with high-temperature risks. The index system method is the most widely used method in the field of high-temperature risk assessment; it selects various indicators related to high-temperature risk, constructs an evaluation index system, assigns corresponding weights to each index, and finally judges the degree of risk based on the comprehensive calculated assessment results [16]. Among them, principal component analysis (PCA) [17], the entropy weight method (EWM) [18,19], and the CRITIC method [20] can be used to determine weights.

The development of the high-temperature risk assessment system has undergone an evolution process from simple to complex, from a single disciplinary perspective to multidisciplinary integration, and from basic data statistics to an in-depth analysis with the help of advanced technology. Early studies on high-temperature risk mainly focused on the analysis of meteorological elements [21,22], and the assessment method only started from a single dimension of meteorology, ignoring many other aspects involved in high-temperature risk. In 2014, the IPCC proposed a “Hazard-Exposure-Vulnerability” (HEV) risk assessment system in its fifth assessment report [23]. In 2019, Wu Xilin et al. [24] proposed a (HEVA) risk assessment framework based on “Hazard-Exposure-Sensibility-Adaptability”. In addition, the Thermal Vulnerability Index Framework (HVI), which is based on the three dimensions of “Exposure, Sensibility, and Adaptability”, has been widely used in high-temperature risk assessment studies [25,26] because the HVI framework explains the structure of the physical environment and socio-economy and its feedback in the face of risks, and it can more accurately assess the thermal vulnerability of the region compared with the HEV and HEVA assessment frameworks.

In the study of high-temperature risk assessment systems, the accuracy and timeliness of population data directly affect the scientific nature of the assessment results and the effectiveness of response strategies. The most common population distribution data are census data for administrative units [27], as well as global and national-scale population spatial databases, such as the Gridded Population of the World (GPW) [28], the Global Rural-Urban Mapping Project (GRUMP) [29], WorldPop [30], and the China 1 km Grid Population Database [31]. However, most of them face problems such as insufficient timeliness, large accuracy errors, or a lag in update frequency. Some scholars have also tried to construct population prediction models, such as multivariate fusion models [32], multiple regression models [33,34], and zoning density models [35], but there are still problems such as subjective determination of fusion weights and insufficient selection of indicators. The method of population prediction through the random forest model offers the advantage of faster data update, meets the needs of dynamic detection, can obtain more accurate population distribution, and greatly compensates for the shortcomings of traditional data.

In summary, the use of random forest models can predict the distribution of urban population more objectively and accurately, while the heat vulnerability index framework (HVI) of the three dimensions of “exposure, sensitivity, and adaptability” (ESA) can more accurately assess the thermal vulnerability of the study area. Therefore, this study takes the four districts of Dalian City center (Ganjingzi District, Shahekou District, Zhongshan District, and Xigang District) as the research area, breaks through the limitations of relatively static and extensive population data in traditional high-temperature risk assessment, and constructs a refined population prediction method based on random forests. At the same time, the dynamic change characteristics of the population are integrated into the high-temperature risk assessment system, changing the previous evaluation mode of single elements, and constructing a multi-dimensional assessment framework covering high-temperature exposure, sensitivity, and adaptability, so as to provide a new perspective for a comprehensive and in-depth understanding of the formation mechanism of high-temperature risk in Dalian.

2. Data and Methods

The random forest regression model used in this study for population data acquisition was proposed by Breiman [36]; it introduces a feature randomization mechanism to randomly select features from the current candidate feature set as a local subset during the node-splitting process of each decision tree, and each decision tree not only uses a different training set but also uses a different feature subset, which effectively avoids the risk of the overfitting of decision tree integration and ensures the effectiveness and efficiency of the ensemble algorithm [37]. The HVI framework explains the structure of the physical environment and socio-economic conditions and their feedback in the face of risks, which can more accurately assess the thermal vulnerability of the region. Compared with subjective weighting methods such as the analytic hierarchy process (AHP), the entropy weight method (EWM) calculates the weight entirely based on the degree of variation of the data itself, avoids the bias caused by subjective judgment, and is suitable for objective, quantitative, and multi-source data, which is more in line with the data-driven and scientific requirements of natural disaster risk assessment.

The framework of this study is shown in Figure 1, and the main steps are as follows: (1) Select evaluation indicators, including physical environmental and social factors, and construct a population prediction model based on the random forest algorithm to realize the dynamic monitoring of population data. (2) Using comprehensively simulated population data and other multi-source data, a high-temperature risk assessment system was constructed based on HVI in the three dimensions of “exposure, sensitivity, and adaptability”. (3) The high-temperature risk map of the study area was obtained after EWM was used to weight various indicators. (4) By analyzing the spatial distribution characteristics and dominant risk types of high-temperature risks in the study area, the urban adaptability strategy based on risk orientation was explored.

2.1. Study Area

Dalian (38°43′−40°10′ N, 120°58′−123°31′ E) is located at the southernmost tip of China’s Liaodong Peninsula, with an average annual temperature of about 10.5 °C. In recent years, the number of high-temperature days above 35 °C in Dalian has increased significantly, and there was a week of extreme high temperatures (up to 37 °C) in the summer of 2022 from 29 July to 4 August, which is rare in the past 40 years. In this study, the four districts of the Dalian City center were selected as the study areas, including Zhongshan District, Xigang District, Shahekou District, and Ganjingzi District (Figure 2), with an area of about 624.33 km². The terrain of the study area is generally higher in the south than in the north, the built area is mainly concentrated in the plain area and extends in the north–south direction, and a small number of areas are distributed near the estuary. The total permanent population reached 2,908,900, accounting for about 38.58% of the permanent population of Dalian, of which the child population (0–14 years old) accounted for about 12.04%, the elderly population (over 65 years old) accounted for about 466,900, and the female population accounted for about 44.45%, and the overall population density was at a high level in the city. The research results of Zhao Jun on the intensity of urban heat islands in Dalian show that the spatial distribution of heat fields in Dalian is significantly different [38], the regional thermal environment problem is severe, and systematic research and targeted governance are urgently needed.

2.2. Preliminary Assessment Indicator Library

Based on the analysis of existing studies, the evaluation indicators of this study are preliminarily clarified, and on this basis, a pre-selected index database for high-temperature disaster risk assessment composed of the “exposure index”, “sensitivity index”, and “adaptability index” is constructed (

Table 1. A database of pre-selected indicators for high-temperature disaster risk assessment.

Indicator Type	Metric Content	Function
Exposure Metrics	Daytime Surface Temperature	It directly reflects the degree to which the ground receives and stores solar radiant heat.
	Population Density	Relevant studies have shown that higher population density is associated with higher thermal health risks [40].
	Road Network Density	Visually reflects the density of roads in the study area.
	Elevation Data	Affect population distribution and infrastructure construction.
	Plant Coverage	A large number of studies have shown that green spaces and other areas with abundant vegetation have natural cooling functions, which can effectively alleviate the impact of the thermal environment on the human body and provide free summer space for the public [41,42].
Sensitivity Indicators	Distribution of Geriatric Population	The body functions deteriorate, and the ability to adjust and adapt to high temperatures is weakened.
	Female Demographics	The physiological structure and hormone level characteristics also make it less tolerant to high temperatures.
	Demographic Distribution of Children	The body is not yet fully developed, and the body temperature regulation mechanism is relatively fragile.
	GDP	Cities with higher GDP have more sound facilities and are less affected by high temperatures. Regions with low GDP are more at risk of extreme heat due to less economic resilience [43].
Adaptability Indicators	Health Care Facilities	The accessibility of medical and health facilities significantly affects the thermal adaptability of individuals [44], which can intuitively reflect the ability of different regions to cope with the health risks of high temperatures.
	A Place to Avoid High Temperatures	Reasonable layout and configuration of high-temperature shelter sites can significantly reduce the risk of high temperature and improve the adaptability of residents during high-temperature heat waves, while vice versa may exacerbate the risk [45].The geometry of the space also has an impact on outdoor thermal comfort [46].
	Housing Values	Areas with higher housing values tend to have well-planned and maintained green landscapes, high-quality community property services, and well-developed infrastructure [47], and residents are more resilient.
	Electricity Consumption	To a certain extent, it reflects the cooling capacity of the region in the face of high-temperature risk, and the higher the power consumption within the load range of the grid, the stronger the cooling capacity.

). Among them, “exposure” refers to the comprehensive measurement of the exposure intensity, duration, frequency, and spatial distribution of people, buildings, facilities, and other factors exposed to high temperatures. “Sensitivity” is a quantitative index used to measure the sensitivity of a system or object to high-temperature events, which reflects the degree to which the system or object may be affected under high-temperature conditions. “Adaptability” refers to the ability of human society to adapt to high-temperature disasters [39], which usually plays a negative role in the risk assessment of high-temperature disasters and can effectively reduce the impact of high-temperature disasters.

2.3. Data Acquisition and Processing

(1) Landsat-8 OLI/TIRS data offer the advantages of a high spatial resolution, multispectral characteristics, and long-term series [48]. In this study, the original image data was downloaded from the USGS official website (https://www.usgs.gov/) and cropped according to the scope of the study area, and the cloudless image of the study area was obtained using radiation calibration and atmospheric correction (Figure 3a).

(2) Nighttime light images can reflect the brightness of the nighttime light on the earth’s surface, thereby characterizing the intensity of human activities to a certain extent, and the National Earth System Science Data Center has recently released the global 500-m resolution “NPP-VIIRS” nighttime light dataset [49]. In this study, the annual average of NPP/VIIRS night light data in 2022 was selected, the data was first converted into a coordinate system so that it was under the projected coordinate system, and the spatial resolution was adjusted to 100 m via resampling. At the same time, in order to improve the data quality, the data needs to be denoised to remove the noise and outliers (Figure 3b).

(3) POI (point of interest) data covers the core information, such as the name, coordinates, and categories of various places; it can accurately locate specific locations, and its data sources are diversified. The distribution of children’s population and the distribution of high-temperature shelter places and medical and health places in this study are based on POI data acquisition, with the help of Python (Python 3.10.0) crawler technology to obtain POI data from AutoNavi map, which covers 9 categories, each POI belongs to at least one category, and the collected data content includes name, classification, type, longitude, latitude, and administrative region, and finally, a total of 73,703 data points were collected within the study scope (Figure 3c).

(4) The road network data is processed by the road data obtained from the website of the Resource and Environmental Science and Data Platform (https://www.resdc.cn/) of the Chinese Academy of Sciences. The GDP distribution data was obtained from the resource and environment data cloud platform of the Chinese Academy of Sciences, and a spatial grid data with an accuracy of 1 km × 1 km was generated through the spatial interpolation technology. The population data was spatialized with a resolution of 100 m using a random forest model.

2.4. Evaluation System Construction

2.4.1. Calculate the Metrics

(1)

Exposure indicators.

In this paper, the exposure criterion layer in the high-temperature risk assessment framework includes five indicators: land surface temperature, population distribution, road network density, plant coverage, and elevation during high temperatures and heat waves.

① Surface temperature during high temperature and heat waves.

The band-10 data of Landsat 8 OLI/TIRS remote sensing images were selected to carry out land surface temperature inversion based on the ENVI 5.6 (The Environment for Visualizing Images) remote sensing data processing platform and the single-window algorithm proposed by Qin Zhihao et al. [50].

Brightness temperature calculation:

$$T_{10}={\displaystyle \frac{K_2}{ln\frac{K_1}{R}+1}}$$

(1)

$$\mathrm{R}=M_{10}{Q}_{10}+A_{10}-O_{10}$$

(2)

Atmospheric mean acting temperature:

$$T_a=16.0110+0.9262T_0$$

(3)

Estimation of atmospheric transmittance:

$$\mathrm{\omega }=0.0981\times 6.1078\times {10}^{{\displaystyle \frac{7.5T_0}{T_0+273.15}}}RH+0.1697$$

(4)

② Spatialization of population distribution based on data such as night lights.

The formation of the population distribution pattern of Dalian has significant natural geographical base characteristics, and the constraint coefficient of its topography and landform on population distribution is 0.68 (2024 data), in which 73% of the permanent population is concentrated in the coastal plain area below 50 m above sea level, while the population density in the mountainous and hilly area is less than 1/5 of the plain area. In this paper, NPP/VIIRS luminous remote sensing data, DEM digital elevation data, slope, land surface temperature, land cover type, and road network density were selected as the modeling characteristics (

Table 2. Variable modeling.

Data Type	Calculation Formula	Remark
NPP/VIIRS-like nighttime light data	$D_{socioeconomic}={\left\{{light}_i\right\}}_{i=1}^{34}$	$\mathrm{By} \mathrm{calculating} \mathrm{the} \mathrm{average} \mathrm{brightness} \mathrm{of} \mathrm{nighttime} \mathrm{lights} \left({light}_{\mathrm{i}}\right) \mathrm{for} \mathrm{each} \mathrm{street}, \mathrm{a} \mathrm{dataset} D_{socioeconomic}$ containing multi-dimensional socio-economic characteristics was created.
Digital elevation	$D_{dem}={\left\{{dem}_i\right\}}_{i=1}^{34}$	$\mathrm{By} \mathrm{calculating} \mathrm{the} \mathrm{average} \mathrm{elevation} \mathrm{of} \mathrm{each} \mathrm{township}, \mathrm{the} \mathrm{elevation} \mathrm{dataset} D_{dem}$ reflecting the topographic and geomorphological characteristics was constructed.
slope	$D_{slope}={\left\{{slope}_i/{area}_i\right\}}_{i=1}^{34}$	$\mathrm{Based} \mathrm{on} \mathrm{the} \mathrm{DEM} \mathrm{data}, \mathrm{the} \mathrm{ArcGIS}10.8 \mathrm{spatial} \mathrm{analysis} \mathrm{tool} \mathrm{was} \mathrm{used} \mathrm{to} \mathrm{calculate} \mathrm{the} \mathrm{proportion} \mathrm{of} \mathrm{the} \mathrm{area} \mathrm{of} \mathrm{the} \mathrm{gentle} \mathrm{area} \mathrm{with} \mathrm{a} \mathrm{slope} \mathrm{of} \mathrm{less} \mathrm{than} 5° \mathrm{within} \mathrm{the} \mathrm{four} \mathrm{streets} \mathrm{of} \mathrm{the} \mathrm{four} \mathrm{districts} \mathrm{of} \mathrm{Dalian} \mathrm{city} \mathrm{center}, \mathrm{and} \mathrm{the} \mathrm{slope} \mathrm{dataset} D_{slope}$ reflecting the terrain suitability was constructed.
Type of land cover	$D_{landcover}=\left\{P_1,P_2,...,P_J\right\}$	$\mathrm{Based} \mathrm{on} \mathrm{the} \mathrm{land} \mathrm{cover} \mathrm{type} \mathrm{data} \mathrm{of} \mathrm{Dalian} \mathrm{in} 2022, \mathrm{the} \mathrm{land} \mathrm{use} \mathrm{structure} \mathrm{of} \mathrm{each} \mathrm{street} \mathrm{was} \mathrm{systematically} \mathrm{sorted} \mathrm{out}, \mathrm{and} \mathrm{the} \mathrm{land} \mathrm{composition} \mathrm{proportion} \mathrm{dataset} D_{landcover}$ was constructed to quantify the regional land use characteristics.
Road network density	$\mathrm{R}\mathrm{o}\mathrm{a}\mathrm{d} \mathrm{D}\mathrm{e}\mathrm{n}\mathrm{s}\mathrm{i}\mathrm{t}\mathrm{y}={\mathrm{L}}_{\mathrm{r}\mathrm{o}\mathrm{a}\mathrm{d}}/{\mathrm{S}}_{\mathrm{r}\mathrm{o}\mathrm{a}\mathrm{d}}$	${\mathrm{L}}_{\mathrm{r}\mathrm{o}\mathrm{a}\mathrm{d}}$$\mathrm{represents} \mathrm{the} \mathrm{sum} \mathrm{of} \mathrm{road} \mathrm{lengths} \mathrm{within} \mathrm{each} \mathrm{street} \mathrm{area}, \mathrm{and} {\mathrm{S}}_{\mathrm{r}\mathrm{o}\mathrm{a}\mathrm{d}}$ represents the area of each street.

), 34 streets in the four districts of Dalian City center were taken as units, and the random forest regression model proposed by Breiman [36] was used to establish the population density prediction model of the four districts of Dalian City center, in which 100 decision trees were selected as the basic classifier of the random forest. Finally, the spatial distribution data of population density with a spatial resolution of 100 m in the four districts of Dalian City center were obtained.

Variable selection:

In this study, the feature combination optimization strategy was used to screen the variables: Firstly, the top three core variables with the strongest linear correlation with population density were screened out by the Pearson correlation coefficient, the urban land area ratio (LC_urban), the road network density (RD), and the mean light brightness (NL) of each street, and then the remaining 8 independent variables were fully combined to generate 255 combination results. The results of each combination were combined with three core variables to form a candidate feature set, and a validation matrix containing 255 feature combinations was constructed. Ten-fold cross-validation was used to perform multiple experiments on the feature set, and the minimum MAE of the results of multiple experiments was calculated according to the number of increasing variables. It is found that, when the number of variables is 7, the minimum MAE has a minimum value. The minimum value of MAE appeared in the model with the mean brightness of streetlights (NL), road network density (RD), flat area ratio (SLP), land surface temperature (LST), urban land (LC_urban), forest (LC_forest), and grassland (LC_grassland) as variables, and the minimum MAE value was 3317.52 people (Figure 4).

$$MAE={\displaystyle \frac{1}{n}}{\sum }_{i=1}^n\left|x_i-\overline{x}\right|$$

(5)

where n is the number of independent variables, x is the independent variable, and the mean of the independent variables is i = 1, 2, 3, …

$$r\left(X,Y\right)={\displaystyle \frac{Cov\left(X,Y\right)}{\sqrt{Var\left[X\right]Var\left[Y\right]}}}$$

(6)

showing the covariance and variance.

In this study, the feature combination optimization strategy was used to screen variables: Firstly, the top three core variables with the strongest linear correlation with population density, namely the urban land area ratio (LC_urban), road network density (RD), and light brightness average (NL) of each street, were screened out via the Pearson correlation coefficient, and then the remaining 8 independent variables were fully combined to generate 255 combined results. Each combination result was combined with three core variables to form a candidate feature set, and a validation matrix containing 255 feature combinations was constructed. After random shuffling, the feature set was tested multiple times (number of repeats K ≥ 30) through ten-fold cross-validation, and the minimum MAE of the multiple test results was counted according to the number of increased variables. It was found that, when the number of variables was 7, the minimum MAE had a minimum value. The minimum MAE value appeared in the model with the average light brightness (NL), road network density (RD), flat area ratio (SLP), surface temperature (LST), urban land (LC_urban), forest (LC_forest), and grassland (LC_grassland) as variables, and the minimum MAE value was 3317.52 people (

Table 3. Correlation between variables and population data.

Variable Type	Variable	R
Core Variables	NL	0.963
	RD	0.963
	SLP	0.956
	LST	0.954
Other Variables	ALT	0.945
	$\mathrm{LC}\_\mathrm{urban}$	$0.945$
	$\mathrm{LC}\_\mathrm{water}$	$0.634$
	$\mathrm{LC}\_\mathrm{forest}$	$0.797$
	$\mathrm{LC}\_\mathrm{barren}$	$0.214$
	$\mathrm{LC}\_\mathrm{grassland}$	$0.797$
	$\mathrm{LC}\_\mathrm{cropland}$	$0.818$

and Figure 4).

Population modeling and result correction (

Table 4. Population data of the four central districts of Dalian.

Administrative Divisions	Permanent Resident Population of Dalian Seventh National Population Census (10,000 People)	Area (Km2)	Population Density (10,000 People/Km2)
Shahekou District	67.031	42.56	1.574977
Zhongshan District	38.8564	48.45	0.80199
Ganjingzi District	153.4722	529.66	0.289756
Xigang District	30.5317	27.99	1.090807

and Figure 5):

Based on ArcGIS, a rule grid was constructed, and the vector boundary of administrative divisions in Dalian City, Liaoning Province, was used to retain 63,814 polygon cells in the complete coverage area. The 100 m × 100 m grid was used to cut the road network density data, surface temperature data, night light data, etc., and calculate the average value in each grid. The projection coordinate system of all spatial data was redefined, the UTM projection (51 divisions) based on the WGS84 datum was used to unify the raster data with spatial resolution of 100 m × 100 m, the application dataset was constructed and substituted with the random forest population prediction model established above, and the population distribution results of the four districts of Dalian City center were obtained (Table 4). The relationship between the statistics of each street (${\mathrm{D}}_{\mathrm{t}\mathrm{j}}$) and the model prediction data (${\mathrm{D}}_{\mathrm{m}\mathrm{n}}$) is as follows:

$${\mathrm{D}}_{\mathrm{t}\mathrm{j}}=\mathrm{m}\times {\mathrm{D}}_{\mathrm{m}\mathrm{n}}$$

(7)

where m is the correction coefficient, which is the ratio of the number of people in each street to the number of people in the simulation.

③ Plant coverage.

In this study, the vegetation coverage was obtained by using the remote sensing image estimation method [51], and the Normalized Difference Vegetation Index (NDVI) image of Dalian was calculated through the inversion of satellite remote sensing images. Secondly, with the help of the band math tool in the ENVI software, the calculation was completed according to the calculation formula of FVC. Finally, the vegetation coverage image was created in ArcGIS software. The calculation formula is as follows:

$$NDVI={\displaystyle \frac{NIR-Red}{NIR+Red}}$$

(8)

$$FVC=\left[\left(NDVI-{NDVI}_{Soil}\right)/\left({NDVI}_{Veg}{NDVI}_{Soil}\right)\right]$$

(9)

④ Elevation.

The Copernicus DEM launched by the European Space Agency (ESA) in 2016 was selected, the basic COP-DEM data was cut using the ArcGIS software to obtain the elevation data of the four districts in the center of Dalian, and the image was drawn.

⑤ Road network density.

The road data obtained from the website of the Resource and Environmental Science and Data Platform of the Chinese Academy of Sciences (https://www.resdc.cn/) were selected for processing, imported into ArcGIS, and cropped according to the boundaries of the study area, and the road network density distribution map of the four districts in Dalian City center was obtained by analyzing the road data in the study area through a kernel density analysis (Figure 6).

(2)

Sensitivity indicators.

The population distribution raster dataset in WorldPop was selected and combined with the data of the seventh population census of China for spatial calibration. Through ArcGIS, the data of each age group in the dataset were fused with the basic geographic information layer, and finally, the spatial distribution data for the female population of each age group and the elderly population aged 65 and above were extracted and generated. The correction factor was calculated by combining the Qipu data and the WorldPop data, and the formula is as follows:

$${\omega }_a={\displaystyle \frac{Q_{ag}}{W_{ag}}}$$

(10)

In this study, the spatial distribution model of the child population was constructed, and the educational institution points in the POI data were innovatively used as the core data source. These point data were imported into the ArcGIS geographic information system platform, and the kernel density analysis algorithm was used to quantify the spatial distribution characteristics of the children’s population through spatial interpolation and density estimation.

(3)

Adaptability indicators.

In this study, 6820 data points were selected as the data of high-temperature shelter in the POI data, 6820 data points were selected in ArcGIS, and a kernel density analysis was used to obtain the distribution of high-temperature shelter in the four districts of the Dalian City center.

In terms of the distribution of medical and health places, the POI data in the study area were classified and screened, the information of sub-categories of medical and health places was extracted, and 6044 valid data points were obtained after standardized processing. ArcGIS was used to analyze the distribution of medical and health facilities in the four districts of the Dalian City center.

In this study, the housing price information of Python crawling shells was used for analysis, the preprocessed housing price data was introduced into ArcGIS, and the ordinary kriging interpolation method was used for analysis to obtain the distribution of housing value in the study area.

The power consumption data used in this paper was calculated by Chen Jiandong et al. [52], using the National Defense Meteorological Satellite Program Line of Service Scan System (DMSP/OLS) and NPP/VIIRS nighttime light data as key data sources. The power consumption data were imported into ArcGIS, cropped, and resampled into a 100 m resolution, and the power consumption distribution in the study area was obtained (

Table 5. Processing methods for index factors at each criterion layer.

Criterion Layer	Evaluation Factors	Influence Relationships	Source Data	Data Processing
Exposure Factor	Surface temperature during a heat wave	+	Landsat 8 OLI/TIRS Satellite digital products	ENVI surface temperature inversion + ArcGIS drawings
	Vegetation coverage	−	Landsat 8 OLI/TIRS Satellite digital products	ENVI calculates vegetation cover + ArcGIS drawings
	population density	+	Nighttime light brightness, elevation, area proportion of flat area (slope less than 5°), urban function type, road network density, and land surface temperature, population density of Qipu	Population estimation models were constructed and modified using random forests
	Road network density	+	BBBike website	ArcGIS Kernel Density tool drawing
	Elevation	−	COP-DEM 30 m resolution elevation data	ArcGIS drawings
Sensitivity Factors	Distribution of the population aged 65 years and over	+	WorldPop website	Obtain the overlay of population data for the age group of 65 and above
	Demographic distribution of children	+	AutoNavi map Dalian children’s agency POI data	ArcGIS Kernel Density tool drawing
	Female demographics	+	WorldPop website	Obtain the overlay of female population data in the 20–65 age group
	GDP	−	Dalian 2023 Statistical Yearbook	Excel Statistics + ArcGIS Plotting
Adaptation Factors	Housing values	−	Housing data of Dalian City	ArcGIS ordinary kriging interpolation + drawing
	Distribution of medical and health facilities	−	AutoNavi map Dalian health institution POI data	ArcGIS kernel density analysis + drawing
	Electricity consumption data	−	DMSP/OLS and NPP/VIIRS nighttime light data	ArcGIS cropping
	Distribution of high-temperature shelter places	−	AutoNavi map Dalian City park green space POI data	ArcGIS kernel density analysis + drawing

Note: “+” represents a positive influencing relationship.“−” represents a reverse influence relationship.

).

2.4.2. Indicator Weights Are Determined

In this study, the entropy weight method was selected as the weight determination method for the assessment of high temperature and heat wave in Dalian, and the weight calculation of the index system was realized through three key steps: the range standardization index, the definition of information entropy, and the calculation of entropy weight (

Table 6. Entropy weight method confirmation process.

Steps	Calculation Formula	Remark
Standardized Metrics	(Normalization of positive indicators) ${\mathrm{X}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{{\mathrm{x}}_{\mathrm{i}\mathrm{j}}-\mathrm{m}\mathrm{i}\mathrm{n}\left({\mathrm{x}}_{\mathrm{j}}\right)}{\mathrm{m}\mathrm{a}\mathrm{x}\left({\mathrm{x}}_{\mathrm{j}}\right)-\mathrm{m}\mathrm{i}\mathrm{n}\left({\mathrm{x}}_{\mathrm{j}}\right)}}$ (Normalization of negative indicators) ${\mathrm{X}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{\mathrm{m}\mathrm{a}\mathrm{x}\left({\mathrm{x}}_{\mathrm{j}}\right)-{\mathrm{x}}_{\mathrm{i}\mathrm{j}}}{\mathrm{m}\mathrm{a}\mathrm{x}\left({\mathrm{x}}_{\mathrm{j}}\right)-\mathrm{m}\mathrm{i}\mathrm{n}\left({\mathrm{x}}_{\mathrm{j}}\right)}}$	The original values of each indicator are normalized to dimensionless values in the interval [0, 1] by a standardized method.
Define Information Entropy	${\mathrm{Z}}_{\mathrm{j}}=-\mathrm{k}{\sum }_{\mathrm{i}=1}^{\mathrm{m}}{\mathrm{P}}_{\mathrm{i}\mathrm{j}}\mathrm{l}\mathrm{n}{\mathrm{P}}_{\mathrm{i}\mathrm{j}}\left(\mathrm{j}=1,2,...,\mathrm{n}\right)$ ${\mathrm{P}}_{\mathrm{i}\mathrm{j}}={\displaystyle \frac{{\mathrm{X}}_{\mathrm{i}\mathrm{j}}}{\sum _{\mathrm{i}=1}^{\mathrm{m}}{\mathrm{X}}_{\mathrm{i}\mathrm{j}}}}\left(\mathrm{i}=1,2,...,\mathrm{m}\right)$ $\mathrm{k}=1/\mathrm{l}\mathrm{n}\mathrm{m}$	A system with an ordered state has a low information entropy; In disordered or chaotic systems, the information entropy value is significantly increased.
Calculate Entropy Weights	${\mathrm{W}}_{\mathrm{j}}={\displaystyle \frac{1-{\mathrm{z}}_{\mathrm{j}}}{\mathrm{n}-{\sum }_{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{z}}_{\mathrm{j}}}}$	Calculations are carried out separately at the criterion level.

).

2.4.3. Build a Model

In this study, a high-temperature risk assessment system was constructed in the study area using the comprehensive index calculation method [53], in which RI was the risk index of high temperature and heat wave, EI was the exposure index, SI was the sensitivity index, and AI was the adaptability index. The natural breakpoint method was used to divide the risk index of high temperature and heat wave into five grades: extremely low, low, medium, high, and extremely high, and the risk of high-temperature disaster in the study area was quantitatively and qualitatively assessed. The specific calculation formula is as follows:

$$RI=EI\times SI÷AI$$

(11)

3. Results

3.1. Population Simulation and Result Verification

In the population simulation data, the population of the study area is mainly concentrated in the eastern part of Ganjingzi District, the northeastern part of Shahekou District, the northern part of Xigang District, and the northwestern part of Zhongshan District, and the population density of these areas is generally more than 209 people/100 square meters. The minimum population density is mainly located in forested areas. From the perspective of specific street-level units, the degree of the spatial heterogeneity presented by the simulation data is much higher than that of statistical data. The accuracy of the population simulation results was verified by taking Xingong Street in Shahekou District and Xinzhaizi Street in Ganjingzi District as examples (Figure 7). The results show that the traditional population data with streets as the statistical unit encounters limitations, and it is difficult to accurately present the detailed population distribution characteristics within urban blocks. The introduction of remote sensing data into the research field of spatial distribution of population density can effectively fill this gap and greatly make up for the shortcomings of traditional data (Figure 7).

3.2. Distribution of High-Temperature Heat Wave Risks in Dalian City

The weights were calculated according to each index (

Table 7. Calculation of indicator weights.

Criterion Level	Evaluation Indicators	Information Entropy (Zj)	Entropy Weight (Wj)
Exposure	Surface temperature	0.967	0.208
	Population density	0.967	0.207
	Plant coverage	0.969	0.198
	Elevation data	0.970	0.190
	Road network density	0.969	0.197
Sensitivity	Elderly population over 65 years old	0.961	0.309
	Female population	0.969	0.246
	Child population	0.958	0.334
	GDP	0.986	0.111
Adaptability	Electricity	0.977	0.165
	Housing value	0.969	0.224
	High-temperature shelter place	0.958	0.302
	Medical Places	0.957	0.309

), the “exposure, sensitivity, and adaptability” in the study area were divided into five levels (very low/low/moderate/high/very high), and finally, the distribution of the high-temperature heat wave risk in Dalian was obtained.

3.2.1. Exposure

The exposure levels of the research area were categorized into five levels: extremely low exposure (0.198–0.364), low exposure (0.365–0.444), moderate exposure (0.445–0.534), high exposure (0.535–0.629), and extremely high exposure (0.630–0.837). The analysis reveals that the spatial distribution of exposure levels in the entire study area shows a trend of decreasing from east to west. The proportion of exposed areas by level is 13.04%, 13.95%, 23.44%, 29.68%, and 19.89%, respectively. Notably, the combined area of extremely high and high exposure constitutes 26.99%, indicating that more than one-quarter of the research area is at risk of high exposure (Figure 8a).

3.2.2. Sensitivity

The sensitivity of the research area is classified into five levels: extremely low sensitivity (0.045–0.165), low sensitivity (0.166–0.281), medium sensitivity (0.282–0.435), high sensitivity (0.436–0.606), and extremely high sensitivity (0.607–0.917). The spatial distribution of sensitivity across the entire research area shows a trend of higher sensitivity in the east and lower sensitivity in the west. A statistical analysis indicates that the total area of extremely high and high sensitivity accounts for 17.73%, with high-sensitivity risk areas constituting less than one-fifth of the research zone. The core urban area, due to its high degree of urbanization and intensive human activities, exhibits higher sensitivity, while the periphery, which is relatively natural, demonstrates lower sensitivity (Figure 8b).

3.2.3. Adaptability

The adaptability of the research area is categorized into five levels: extremely high adaptability (0.242–0.546), high adaptability (0.547–0.681), medium adaptability (0.682–0.865), low adaptability (0.866–0.988), and extremely low adaptability (0.607–0.917). The analysis indicates that, in the high-temperature risk assessment system of this study, all adaptability indicators are negative indicators. After normalization, a smaller value indicates higher adaptability. The spatial distribution of adaptability across the entire research area demonstrates a trend of lower adaptability in the east and higher adaptability in the west (Figure 8c).

3.2.4. High-Temperature Risk

The natural breakpoint method was used to divide the high-temperature risk in the study area into five levels: a very-low-risk area, a low-risk area, a medium-risk area, a high-risk area, and an extremely high-risk area, with the maximum value being 2.215, the minimum value being 1.044, and the distribution of high-temperature risk in the overall city showing a trend of gradually decreasing from east to west and gradually decreasing from coastal to inland regions (Figure 8d). According to the research results, there are obvious differences in the distribution between different criterion layers in the main urban area of Dalian, with the high-exposure area accounting for 13.04%, the high-sensitivity area accounting for 8.05%, and the low-adaptability area accounting for 21.44%, and exposure is the main cause of high-temperature risk.

4. Discussion

4.1. A New Method for Population Prediction Based on Random Forests

There are obvious spatial distribution differences in the population densities of the four districts of Dalian City center, which is obviously lower in the northwest of the study area, while the density in the southeast is relatively high, showing a trend of gradually decreasing from east to west, and gradually decreasing from coastal to inland, which is generally consistent with the results of high-temperature risk assessment, and there are local differences. Population data can intuitively reflect the size of the population exposed to high temperatures and, by analyzing the population density and age structure, accurately locate high-risk areas and vulnerable groups, which is of great help in improving the accuracy of urban high-temperature risk assessment.

In the current population data acquisition methods, census data [27] takes national, provincial, municipal, and county administrative divisions as spatial units, and it adopts household inquiry, on-the-spot reporting, and the independent reporting of census subjects to form data, with a time accuracy of 10 years. The population density prediction model constructed by using the random forest algorithm offers the advantage of faster data updates, meets the needs of dynamic detection, and can effectively supplement the traditional data. The population data of the geospatial data platform (taking WorldPop as an example [30]) uses a semi-automated spatial distribution model method based on a random forest regression tree to construct data with a spatial accuracy of 1 km or 100 m, which offers the advantages of timeliness and resolution compared to census data. In this study, the random forest algorithm model can calculate the average of the multi-source data grid and construct a dataset based on the fixed pixel size of ArcGIS, and it can bring the model to obtain the spatial distribution results of population with a spatial resolution of 100 m, providing more accurate population distribution data for high-temperature risk assessment. Moreover, there are inherent deficiencies in the compatibility of current population data, and there is a spatial dislocation between the statistical unit and the physical geography unit of the administrative region, which restricts the integration and analysis of population data with other disciplines. The method of population prediction through the random forest model and the use of dynamic remote sensing data can obtain a more accurate population distribution and greatly make up for the shortcomings of traditional data. Some scholars have built population prediction models to achieve more accurate and time-sensitive population data acquisition. Taking the habitat index model as an example [54], the model is constructed by synthesizing NPP/VIIRS night light data and NDVI data as modeling parameters to achieve the spatialization of population data with a spatial resolution of 100 m.

Based on the above analysis, the population data generated via the random forest population prediction model in this study achieves a higher spatial resolution, which can integrate multiple types of data, such as NL, RD, SLP, etc., capture the complex nonlinear relationship between multiple data, and achieve a stronger dynamic update ability and better timeliness.

4.2. High-Temperature Risk Assessment System Under the HVI Framework

In the context of climate change and urbanization, high temperature is a common problem faced by every city, the urban population is deeply threatened by high-temperature disasters, and people have begun to realize that high-temperature risk is not a simple meteorological problem but a complex systemic problem involving meteorological, geographical, socio-economic, ecological environment, and other fields [55]. Risk assessment systems such as HEV [23] and HEVA [24] have appeared in the field of high-temperature risk assessment. In addition, the Thermal Vulnerability Index (HVI) framework based on the three dimensions of “exposure-sensitivity-adaptability” has been widely used in high-temperature risk assessment studies [26,56,57]. At present, many scholars have used high-temperature risk assessment methods to establish evaluation systems (

Table 8. Selection of risk research indicators and weighting methods.

Article Title	Author	Evaluation Framework	Select Metrics	Empowerment Method	Reference
Heat Vulnerability and Street-Level Outdoor Thermal Comfort in the City of Houston: Application of Google Street View Image Derived SVFs	Kim Y J, Li D, Xu Y, et al.	HVI Risk Assessment Framework	Exposure: thermal comfort during the day, thermal comfort at night, SVF, normalized vegetation index Sensitivity: hypertension, asthma, diabetes, obese population, population density, child population, elderly population, non-white population, single population, disabled population Adaptability: unemployed people, poor people, people not in high school, buildings built before 1970, buildings built before 1980	Principal component analysis	[25]
Heat vulnerability caused by physical and social conditions in a mountainous megacity of Chongqing, China	Xiang Z, Qin H, He B J, et al.	HVI Risk Assessment Framework	Exposure: Surface temperature, population density Sensitivity: plot ratio, urban surface roughness, skyscape factor, sensitive population, year of construction Adaptability: normalized vegetation index, housing value, medical and health services, living places	Equal-weight method	[58]
Beijing based on multi-source remote sensing data Comprehensive assessment of the risk of high temperature and heat wave in the city	He Miao, Xu Yongming, Mo Yaping, et al.	HVI Risk Assessment Framework	Exposure: population density, normalized vegetation index, normalized water body index, normalized building index Danger: average daily maximum temperature, number of high-temperature days Vulnerability: proportion of elderly population, proportion of people living alone, per capita income, proportion of construction personnel, average ownership rate of air conditioners	Subjective and objective weighting method	[59]
Identification and assessment of heat disaster risk: a comprehensive framework based on hazard, exposure, adaptation and vulnerability	Luo Y, Cheng X, He JB, et al.	HEVA Risk Assessment Framework	Hazard: rainfall, geography, surface temperature, air quality, air temperature, humidity, etc. Exposure: census data, outdoor activities, population density, working hours, living environment, landscape coverage, etc. Adaptability: population density, preparedness and prevention, cooling facilities (density and scale), medical facilities (density, size, personnel, beds), etc. Vulnerability: economy, population, health education facilities, high-density conditions of residential buildings, etc.	Analytic hierarchy process + entropy weight method	[60]

), such as Kim Y J et al. using principal component analysis based on the HVI risk assessment framework [25], Xiang Z et al. using the equal-weight method based on the HVI risk assessment framework [58], and He Miao et al. using the subjective and objective weighting method based on the HVI framework [59]. Luo Y et al. integrated hazards, exposures, adaptations, and vulnerabilities to systematically assess thermal risk in summer cities [60], using a weighting method that combines analytic hierarchy process (AHP) and entropy method (EM).

This study constructed a high-temperature risk assessment system in Dalian based on the HVI framework of the index system method, constructed evaluation criteria from three levels, exposure, sensitivity, and adaptability, by collecting meteorology, urban morphology, physical geography, and socio-economic and demographic characteristics, used a machine learning correlation analysis to screen indicators, used an entropy weight method to determine weights, used the comprehensive index method to construct a thermal risk assessment model, drew a high-temperature risk map of the study area, formed a complete set of high-temperature risk assessment methods and workflows, obtained objective and accurate high-temperature risk distribution data in Dalian, improved the research in the field of high-temperature risk assessment in Dalian, and provided ideas for references for cold cities.

4.3. Adaptive Strategies

The risk of high temperature and heat wave is closely related to urban planning, which greatly affects the impact of high temperature and heat wave on the city, and the risk of high-temperature heat wave also provides new ideas and directions for urban planning. This study formulates multi-level planning strategies and specific measures from three dimensions: urban master planning, urban detailed planning, and special planning (

Table 9. Multi-level planning of response measures.

Planning Level	Coping Aspect	Relevant Factors	Specific Measures	Time Frame	Responsible Party
Master Plan	Climate pattern	Building/population/structure/land	(1) Build a spatial climate model to accurately understand climate characteristics and change laws (2) Combine the natural pattern of Dalian’s mountains and seas to avoid the risk of high-temperature disasters	Long term	Government agencies
	Greenfield system	Urban scale/functional structure/land use/green square	(1) Build a network ecological green space system, improve the ecological network, and improve the urban heat regulation function (2) Build diversified ecological corridors to form ecological cooling channels (3) Build a park and green space system with regional characteristics (such as coastal city characteristics) to enrich urban green space	Medium term	Government agencies
	Space layout	Construction/roads/infrastructure/industry/layout/land	(1) Build a mitigation zone with specific roads to optimize the urban spatial structure (2) Take important nodes as the main improvement core to improve the comfort of the regional thermal environment (3) Expand the east–west development space and optimize the horizontal layout of the city (4) Set up an open space in the north–south direction to guide the airflow	Medium term	Government agencies
	Industrial function	Industry/layout/land	(1) Build a cascade industrial layout, optimize industrial spatial distribution, and promote industrial collaborative cooling (2) Upgrade and transform traditional industries, cultivate and develop emerging industries, and reduce the heat production of industrial operations	Long term	Government agencies
Detailed planning	Road traffic	Construction/population/roads/land/industry	(1) Road system planning: optimize the road system; design orientation, density, and traffic organization in combination with ventilation needs (2) Road section planning: improve road permeability; Work together to build a protective green space	Short term	Government agencies and Communities
	Building control	Building/land	(1) Building height: evaluate high-rise buildings and super-high-rise buildings; control the building height on both sides of the ventilation duct (2) Building density: high-rise and low-density distribution are adopted	Medium term	Government agencies
	Public spaces	Building/land/population	(1) Arrange urban public open space nodes according to high-temperature risk levels (2) Form a highly accessible open space network	Short term	Government agencies and Communities
Special planning	High-temperature disaster shelter	Greenspace/infrastructure/industry	(1) Build a four-level disaster shelter system of “community-street-district-municipal level” (2) Rational use of public open spaces such as green spaces and squares (3) Establish a thermal environment monitoring system to monitor the temperature, humidity, and other environmental parameters in the site in real time	Short term	Government agencies and Communities
	Ventilation corridors	Green spaces/roads	(1) Use ecological cold sources such as large parks and green spaces as important nodes of ventilation corridors (2) Make full use of roads and rivers to guide the airflow	Medium term	Government agencies and Communities
	Urban cold island	Green space/land use	(1) Reasonably control the intensity of development near the cold island (2) Delineate prohibited construction areas	Medium term	Government agencies

).

Currently, there are a limited number of planning strategies specifically for high-temperature risks, and most of the relevant adaptation strategies, measures, and technical methods are scattered in the study of climate change and urban heat island effects [61]. From an urban perspective, cities in the United States and Europe are the most active in exploring the risk of high temperatures, mainly by enacting laws, regulations, and policy documents to limit greenhouse gas emissions and thereby mitigating urban heat disasters. Although the United States has not issued a national unified action plan, local governments are actively formulating and implementing relevant programs, such as New York City’s well-served community planning for parks, cultural resources, and shared spaces mentioned in OneNYC 2050 [62]; the City of Chicago proposed solar energy projects and green roofs in its climate action plan [63]. The City of San Francisco proposed transportation and land use in its climate action plan [64]. European countries have achieved fruitful results in the formulation of action plans, response policies, and development strategies, such as Denmark’s CPH2025 climate plan [65]. Sweden has proposed the Swedish Climate Act and Climate Policy Framework [66], emphasizing the advancement of renewable products in the energy sector to reduce CO₂ emissions. These planning measures have commonalities with the strategies proposed in this study, but this study is mainly focused on Dalian to conduct high-temperature risk assessment and adaptation strategy research, and the research results can be extended to other cold cities in the future, further verifying and improving the research methods and strategies, and providing a broader reference for global cold cities to cope with high-temperature disasters.

4.4. Study Limitations

This study involved several limitations that are worth further exploration: (1) In terms of data, there are still certain limitations in the accuracy or acquisition time of some data, and more scientific data acquisition methods should be explored in the future, in order to provide better data guarantee and theoretical support for high-temperature risk assessment. A time series model can be built around part of the data to achieve an accurate prediction of data and improve data timeliness. (2) In terms of the evaluation model, although the current high-temperature risk assessment model takes into account multiple factors, there is still room for improvement. Some complex factors, such as the dynamic changes of urban microclimate and the uncertainty of extreme heat events, have not been fully considered. Follow-up research can try to introduce more advanced models and technologies, such as combining artificial intelligence algorithms, to more accurately simulate the formation and development process of high-temperature risks and improve the scientific approach and reliability of the evaluation model. (3) In terms of strategy implementation, the adaptive strategy proposed in this study may face many challenges in practical application, such as insufficient policy support, lack of capital investment, and difficulties in departmental coordination. In the future, it is necessary to strengthen cooperation with government departments, enterprises, and social organizations, promote the implementation of strategies, and continuously optimize and improve strategies through actual monitoring and feedback.

5. Conclusions

This study focused on the assessment of high-temperature risks and adaptive strategies in Dalian City. It innovatively proposes a refined population density prediction method based on the random forest algorithm in terms of research methods and system construction. It constructs evaluation criteria from three levels: exposure, sensitivity, and adaptability, and utilizes the comprehensive index method to create a high-temperature risk assessment model. The main conclusions are as follows:

• In population density prediction, there is a strong positive correlation between night light brightness (NL), road network density (RD), the flat area ratio (SLP), surface temperature (LST), and demographic data. The correlation coefficients for NL and RD both reach 0.963, while the correlation coefficient for SLP is 0.956, and for LST, it is 0.954.
• In the high-resolution population distribution map of the study area generated by integrating multi-source data, the population in the four districts of the Dalian City center is mainly concentrated in the eastern part of Ganjingzi District, the northeast of Shahekou District, the northern part of Xigang District, and the northwest of Zhongshan District, with population densities generally exceeding 209 people per 100 square meters.
• In the high-temperature risk map, the spatial distribution of high-temperature heatwave risks within the study area shows a significant pattern of decreasing from east to west and from the coast to the inland region. There are notable differences in the distribution of different criteria levels within the main urban area of Dalian, with areas of high exposure accounting for 13.04%, high sensitivity areas for 8.05%, and low adaptability areas for 21.44%. Exposure is the primary cause of high-temperature risks.

Based on population forecasts and assessments of high-temperature risks, this study proposes globally relevant and targeted optimization strategies to enhance the city’s capacity to cope with extreme high-temperature climates, ensuring the quality of life and the safety of residents’ lives and property. This not only contributes to the sustainable development of Dalian but also provides reference experiences and methods for other cold-region cities. Future research outcomes could be expanded to other cold-region cities to further verify and refine research methods and strategies, providing broader references for global cold-region cities in addressing high-temperature disasters.