Statistically Validated Urban Heat Island Risk Indicators for UHI Susceptibility Assessment

This research proposes a collection of urban heat island (UHI) risk indicators under four UHI risk components: hazard, exposure, sensitivity, and adaptive capacity. There are 46 UHI risk indicators linked to three pillars of sustainability: social equity, economic viability, and environmental protection. In this study, the UHI risk indicators were first validated by experts to determine their relevancy and subsequently applied to randomly sampled dwellers of Thailand’s capital Bangkok. The UHI indicators were further validated with confirmatory factor analysis to determine the factor loadings (0–1) and reliability. Under the hazard component, the percentage of days when the daily minimum temperature is less than the 10th percentile exhibited the highest indicator-level factor loading (0.915). Vehicular traffic was the UHI exposure indicator with the highest factor loading (0.923), and the proportion of green space to build environment was the UHI sensitivity indicator with the highest factor loading (0.910). For the UHI adaptive capacity component, the highest factor loading (0.910) belonged to government policy and action. To effectively mitigate UHI impacts, greater emphasis should be placed on the indicators with highest factor loadings. Essentially, this research is the first to use statistical structural equation modeling to validate UHI indicators.


Introduction
An urban heat island (UHI) is an urban or metropolitan area with considerably warmer temperatures than its surrounding rural areas as a result of human activities [1]. The conversion of natural land cover into pavement, buildings and other impervious surfaces that absorb and retain heat contributes to the UHI phenomenon [2]. The major impacts of UHIs include increased energy consumption, elevated emissions of air pollutants and greenhouse gases, compromised human health and comfort, and impaired water quality [3][4][5].
UHI risk assessment enables the formulation of public policy and action to mitigate UHI-related impacts in an effective manner and also enhances individual-level UHI adaptive capacity [6][7][8]. According to the intergovernmental panel on climate change, IPCC, a higher adaptive capacity to climate change-induced UHIs reduces socioeconomic losses while improving individual resilience to excessive heat events [9].
As shown in Figure 1, risk arises from the interaction of hazard, exposure, and vulnerability (i.e., risk = hazard + exposure + vulnerability). Specifically, risk is conceptualized as an internal property of a system that is a function of its current endogenous lack of adaptive capacity to overcome the adverse impact (its sensitivity) of a stressor, and vulnerability is an endogenous characteristic of a system that is determined by its sensitivity and adaptive capacity [9]. In theory, a low adaptive capacity relative to sensitivity (i.e., adverse impacts) results in higher vulnerability and higher risk. As a result, enhancing adaptive capacity reduces vulnerability and subsequent risk.
UHI risk assessment plays a crucial role in limiting ecosystem degradation, social disruptions and economic losses. Specifically, UHI risk assessment provides information about the existing weaknesses of a natural or a socioeconomic system and the plausible causes of the weaknesses. The assessment outcomes enable the formulation of mitigation strategies to deal with or adapt to UHI risks. According to Change [9], climate risk vulnerability reduction is crucial to limiting losses and building resilience to long-term climate change-induced UHI risks. Figure 1. The IPCC's risk assessment framework [9].
Specifically, this research relied on the IPCC's risk assessment framework to develop a collection of UHI risk indicators linked to three pillars of sustainability: social equity, economic viability, and environmental protection. There are 46 UHI risk indicators under four UHI risk components: hazard (consisting of 13 indicators), exposure (10), sensitivity (12), and adaptive capacity (11).
For the hazard component, the UHI indicator selection is based on 13 future temperature-based extreme climatic indexes, where six indexes are associated with temperature (HT1-HT6) and the remaining seven are associated with duration (HD1-HD7) ( Table 1). Additionally, the UHI indicators under the exposure component are associated with the three pillars of sustainability: social equity (ES1-ES4), economic viability (EE1-EE3), and environmental protection (EV1-EV3) ( Table 2).
These 46 UHI risk indicators were first validated by experts to determine their relevancy using a relevant/irrelevant questionnaire (Supplementary Materials S1), and they were subsequently applied to a sample of Bangkok residents using an agree/disagree questionnaire (Supplementary Materials S2). Based on the agree/disagree survey data, the UHI indicators were further validated with confirmatory factor analysis (CFA) to determine the factor loadings (0-1).
By definition, CFA-based factor loading is the correlation coefficient for the variable (i.e., UHI indicators) and factor (i.e., UHI phenomenon). In structural equation modeling, a factor loading of 0.6 or higher indicates that the factor (i.e., UHI phenomenon) is significantly influenced by the variable (i.e., UHI indicators). In this research, the factor loadings were calculated by using the AMOS (analysis of a moment structures) statistical software. Figure 1. The IPCC's risk assessment framework [9].

Sensitivity
UHI risk assessment plays a crucial role in limiting ecosystem degradation, social disruptions and economic losses. Specifically, UHI risk assessment provides information about the existing weaknesses of a natural or a socioeconomic system and the plausible causes of the weaknesses. The assessment outcomes enable the formulation of mitigation strategies to deal with or adapt to UHI risks. According to Change [9], climate risk vulnerability reduction is crucial to limiting losses and building resilience to long-term climate change-induced UHI risks.
Specifically, this research relied on the IPCC's risk assessment framework to develop a collection of UHI risk indicators linked to three pillars of sustainability: social equity, economic viability, and environmental protection. There are 46 UHI risk indicators under four UHI risk components: hazard (consisting of 13 indicators), exposure (10), sensitivity (12), and adaptive capacity (11).
For the hazard component, the UHI indicator selection is based on 13 future temperaturebased extreme climatic indexes, where six indexes are associated with temperature (HT1-HT6) and the remaining seven are associated with duration (HD1-HD7) (Table 1). Additionally, the UHI indicators under the exposure component are associated with the three pillars of sustainability: social equity (ES1-ES4), economic viability (EE1-EE3), and environmental protection (EV1-EV3) ( Table 2).
These 46 UHI risk indicators were first validated by experts to determine their relevancy using a relevant/irrelevant questionnaire (Supplementary Materials S1), and they were subsequently applied to a sample of Bangkok residents using an agree/disagree questionnaire (Supplementary Materials S2). Based on the agree/disagree survey data, the UHI indicators were further validated with confirmatory factor analysis (CFA) to determine the factor loadings (0-1).
By definition, CFA-based factor loading is the correlation coefficient for the variable (i.e., UHI indicators) and factor (i.e., UHI phenomenon). In structural equation modeling, a factor loading of 0.6 or higher indicates that the factor (i.e., UHI phenomenon) is significantly influenced by the variable (i.e., UHI indicators). In this research, the factor loadings were calculated by using the AMOS (analysis of a moment structures) statistical software. AMOS is an added SPSS module and is used for structural equation modeling, path analysis, and confirmatory factor analysis. In addition, the correlations (R 2 ) between UHI indicators were determined with a structural equation model (SEM). Essentially, the proposed CFA-validated UHI indicators could be deployed to effectively assess the susceptibility of an urban area to UHI risks. Moreover, the proposed UHI indicators are applicable to urbanized areas of varying size (e.g., first-, second-, and third-tier cities) to assess the climate change-induced UHI susceptibility and future impacts. Furthermore, this research is the first to apply the SEM technique to statistically validate UHI indicators.

Research Methodology
In the development of UHI risk indicators, this research first reviewed the literature and publications related to the IPCC's risk assessment framework ( Figure 1). The UHI risk indicators considered in this study are linked to the three pillars of sustainability (i.e., social equity, economic viability, and environmental protection). There are 46 UHI risk indicators associated with four UHI risk components: hazard (consisting of 13 indicators), exposure (10), sensitivity (12), and adaptive capacity (11) [10][11][12][13][14][15].
The hazard indicators were obtained from the open-source RClimDex1.1 program (http://etccdi.pacificclimate.org/software.shtml, accessed on 12 January 2022), which provides a friendly graphical user interface to compute all core climate change indices defined by the joint CCl/WCRP/JCOMM Expert Team on Climate Change Detection and Indices (ETCCDI). For the hazard component, the indicator selection was based on 13 future temperature-based extreme climatic indexes, where six indexes are associated with temperature (HT1-HT6) and the remaining seven are associated with duration (HD1-HD7) ( Table 1).
The 46 UHI risk indicators were converted into a relevant/irrelevant questionnaire. To evaluate the relevancy of the UHI risk indicators, a total of 32 experts in the fields of architecture, climate change, urban environment and urban planning were randomly selected from Google Scholar (https://scholar.google.com/, accessed on 12 January 2022), and copies of the relevant/irrelevant questionnaire were emailed to them. There were seven responses from a total of 32 experts, and a number of expert responses ≥3 is acceptable.
The expert-validated UHI indicators were subsequently converted into an agree/disagree questionnaire based on a 10-point Likert scale, where 1 denotes strongly disagree and 10 denotes strongly agree. According to Bayraktar and Tatoglu [16], Ismail Salaheldin [17], Thanvisitthpon and Shrestha [18], an ordinal scale can be used with agree/disagree questions.
The agree/disagree questionnaire was applied to a random sample of individuals to further validate the 46 UHI risk indicators. The sample size was 400 individuals, calculated with the following equation where n is the sample size, N is the number of people in Bangkok (10.72 million), and e is the sampling error (0.05) (Yamane [19]).
Due to incomplete data, the actual sample size (i.e., questionnaire respondents) was 398 individuals who resided in Thailand's capital Bangkok. In the data collection, the 398 questionnaire respondents were randomly recruited to answer the agree/disagree questionnaire, and the responses were input into the structural equation model (SEM) analysis.
Prior to SEM analysis, the Kolmogorov-Smirnov test was applied to the data from the 398 questionnaire respondents, with the null hypothesis (H 0 ) being that the data are normally distributed [20]. H 0 is accepted if the observed statistic is greater than the critical value (α > 0.05), indicating that the data are normally distributed [21][22][23][24].
In the SEM analysis, the 46 UHI risk indicators and four UHI risk components were validated with confirmatory factor analysis (CFA) to determine the factor loadings and reliability of the UHI risk indicators and dimensions. Indicator-and dimension-level factor loadings (0-1) were used to indicate the degree of relevancy of UHI risk indicators and dimensions. The reliability of the UHI risk indicators and dimensions were measured with indicator-level reliability (R 2 ) and dimension-level composite reliability (CR).
Since the SEM technique requires a large quantitative dataset or sample size (i.e., 398 respondents in this research) (Dawson and Peppe [25], Thanvisitthpon [26]), its main advantage lies in the validity of the association, as indicated by the factor loadings between the UHI risk indicators and the UHI phenomenon. Additionally, SEM-based CFA relies on the means and variance-covariance matrix instead of on the correlation matrix, thereby minimizing both nonuniform and uniform bias [26]. Furthermore, SEM-based CFA is used when there is an array of variables (i.e., UHI indicators) measuring more than one dimension (i.e., four UHI components-hazard, exposure, sensitivity, and adaptive capacity). However, the limitations of SEM-based CFA include the necessity of a large sample size (despite higher reliability) and the required training and skills to use the AMOS statistical software.

UHI Indicator Relevancy Assessment
In the assessment of the UHI risk indicators, the randomly chosen experts (i.e., 32 experts) were asked to evaluate the relevancy of the 46 UHI risk indicators under four UHI risk components (i.e., hazard, exposure, sensitivity and adaptive capacity) on a relevant/irrelevant scale of −1 to 1, where −1, 0, and 1 denote irrelevant, uncertain, and relevant, respectively. Seven experts responded the relevant/irrelevant questionnaire: two architects, one climate change scientist, two urban planners, and two urban environmentalists.
Tables 1-4 tabulate the UHI indicators, definitions, and item-objective congruence (IOC) indexes under the four UHI risk components of hazard, exposure, sensitivity, and adaptive capacity, respectively. The IOC index was used as the basis for screening the item quality, whereby experts were asked to determine the relevancy score of each item (−1, 0, or 1) given that IOC > 0.6 is acceptable [27,28].

Confirmatory Factor Analysis of UHI Hazard Indicators
Two dimensions constitute the UHI hazard component: temperature and duration. The temperature dimension consists of six UHI hazard indicators, and the duration dimension consists of seven indicators. The UHI hazard indicators (13 indicators) were validated with confirmatory factor analysis (CFA) to determine their factor loadings and reliability (first-order) and their dimensions (second-order). Structural equation modeling (SEM) was used to analyze the correlations between the UHI indicators. Figure 2 shows the structural equation model and the factor loadings of the UHI hazard indicators, where chi-squared = 4413.707, degree of freedom (df) = 9, p-value = 0.114, root mean square residue (RMR) = 0.050, root mean square error of approximation (RMSEA) = 0.038, goodness of fit index (GFI) = 0.995, adjusted goodness of fit index (AGFI) = 0.945, normed fit index (NFI) = 0.997, and confirmatory fit index (CFI) = 0.999. According to Baumgartner and Homburg [78], Gatignon [79] and Hooper and Coughlan [80], GFI, AGFI, NFI and CFI should be close to 1, while RMSEA and RMR should not exceed 0.050. to Baumgartner and Homburg [78], Gatignon [79] and Hooper and Coughlan [80], GFI, AGFI, NFI and CFI should be close to 1, while RMSEA and RMR should not exceed 0.050.  Table 5 tabulates the first-and second-order factor loadings of the UHI hazard indicators and dimensions. The factor loadings of the HT1-HT6 and HD1-HD7 indicators were 0.757-0.915 (R 2 = 0.617-0.837) and 0.764-0.888 (R 2 = 0.583-0.788), respectively, while those of the temperature and duration dimensions (b) were 0.937 and 0.991, respectively. In comparison, the factor loading of duration was slightly higher, suggesting that the duration dimension plays a more significant role in the UHI hazard component, consistent with [81,82]. According to Kim and Mueller [83], a factor loading of >0.3 is statistically significant. The composite reliability (CR) of the temperature and duration dimensions was 0.921 and 0.932, respectively, with corresponding average variance extracted (AVE) values of 0.660 and 0.663. According to Fornell and Larcker [84], a CFA construct is valid if CR >0.6 or AVE > 0.5.   Table 5 tabulates the first-and second-order factor loadings of the UHI hazard indicators and dimensions. The factor loadings of the HT1-HT6 and HD1-HD7 indicators were 0.757-0.915 (R 2 = 0.617-0.837) and 0.764-0.888 (R 2 = 0.583-0.788), respectively, while those of the temperature and duration dimensions (b) were 0.937 and 0.991, respectively. In comparison, the factor loading of duration was slightly higher, suggesting that the duration dimension plays a more significant role in the UHI hazard component, consistent with [81,82]. According to Kim and Mueller [83], a factor loading of >0.3 is statistically significant. The composite reliability (CR) of the temperature and duration dimensions was 0.921 and 0.932, respectively, with corresponding average variance extracted (AVE) values of 0.660 and 0.663. According to Fornell and Larcker [84], a CFA construct is valid if CR >0.6 or AVE > 0.5.
On the indicator level, the highest factor loading (0.915) belonged to HT6 (percentage of days when the daily minimum temperature is less than the 10th percentile, TN10). According to Shrestha and Bajracharya [85], Zhang and Aguilar [86], Wong and Mok [87], Manalo and Matsumoto [88], and Dong and Xu [89], TN10 is a key indicator of UHI trends and is commonly used in climate change-induced UHI trend analysis.

Confirmatory Factor Analysis of UHI Exposure Indicators
The UHI exposure component comprises three dimensions: social, economic and environment. The three exposure dimensions are linked to three pillars of sustainability, i.e., social equity, economic viability, and environmental protection. Specifically, the social, economic and environment dimensions consist of four, three and three UHI exposure indicators, respectively. The UHI exposure indicators (10 indicators) were validated with CFA to determine their factor loadings and reliability (first-order) and their dimensions (second-order), and SEM was used to analyze the correlations between the UHI indicators.  b is the factor loading of the second-order CFA. R 2 is the reliability of the indicators.
On the indicator level, the highest factor loading (0.915) belonged to HT6 (percentage of days when the daily minimum temperature is less than the 10th percentile, TN10). According to Shrestha and Bajracharya [85], Zhang and Aguilar [86], Wong and Mok [87], Manalo and Matsumoto [88], and Dong and Xu [89], TN10 is a key indicator of UHI trends and is commonly used in climate change-induced UHI trend analysis.

Confirmatory Factor Analysis of UHI Exposure Indicators
The UHI exposure component comprises three dimensions: social, economic and environment. The three exposure dimensions are linked to three pillars of sustainability, i.e., social equity, economic viability, and environmental protection. Specifically, the social, economic and environment dimensions consist of four, three and three UHI exposure indicators, respectively. The UHI exposure indicators (10 indicators) were validated with CFA to determine their factor loadings and reliability (first-order) and their dimensions (second-order), and SEM was used to analyze the correlations between the UHI indicators.   Table 6 presents the first-and second-order factor loadings of the UHI exposure indicators and dimensions. The factor loadings of ES1-ES4, EE1-EE3 and EV1-EV3 were 0.860-0.923 (R 2 = 0.739-0.853), 0.808-0.908 (R 2 = 0.683-0.825), and 0.879-0.917 (R 2 = 0.772-  Table 6 presents the first-and second-order factor loadings of the UHI exposure indicators and dimensions. The factor loadings of ES1-ES4, EE1-EE3 and EV1-EV3 were 0.860-0.923 (R 2 = 0.739-0.853), 0.808-0.908 (R 2 = 0.683-0.825), and 0.879-0.917 (R 2 = 0.772-0.840), respectively, while those of the social, economic and environment dimensions (b) were 0.997, 0.999, and 0.994, respectively; a factor loading >0.3 is statistically significant (Kim and Mueller [83]. The CR values of the social, economic, and environment dimensions were 0.930, 0.887, and 0.925, respectively, with corresponding AVE values of 0.770, 0.724, and 0.804. According to Fornell and Larcker [84], a CFA construct is valid if CR >0.6 or AVE > 0.5. The highest indicator-level factor loading (0.923) belonged to ES4 (vehicular traffic). Vehicular traffic is the aggregation of vehicles coming and going in a particular locality and it was positively correlated with the UHI intensity and air pollution. According to Hoehne and Chester [90], vehicular traffic and transportation network design play crucial roles in the UHI phenomenon. Hart and Sailor [91] reported that road surface temperatures during weekdays are 2 • C higher than during the weekend. This finding could be attributed to the higher level of vehicular traffic during weekdays. Sailor and Lu [92] reported that heat from vehicles in six U.S. cities accounted for 47%-62% of total heat emissions.
The second highest factor loading (0.917) belonged to EV3 (pervious surface area). By definition, a pervious surface is land not covered by buildings or other man-made infrastructure, thus allowing rainwater to percolate into the soil to filter out pollutants and recharge the groundwater. Pervious surface coverage and the UHI phenomenon were found to be inversely related.
The conversion of pervious to impervious surfaces as a result of economic development and urbanization expansion causes temperatures in urban areas to be significantly warmer than surrounding rural areas [99,100]. Impervious surface expansion raises urban land surface temperatures (Haselbach and Boyer [100], Shuster and Bonta [101], Gorsevski and Taha [102], Taha [103], Sen and Roesler [104], and Cao and Li [105]).
According to Costanzo and Evola [113], Jamei and Chau [114], and Odli and Zakarya [115], the cooling effects provided by rooftop and vertical gardens mitigate UHIinduced dramatic temperature increases. In addition, rooftop and vertical gardens trap atmospheric pollutants, thus improving air quality [108,116].
In Japan's capital Tokyo, several UHI mitigation measures have been implemented, including green premises, green roofs, green building walls, enhanced rooftop reflectivity, water-retentive pavement, and reductions in heat loss from buildings [117]. In Hong Kong, UHI mitigation strategies have primarily focused on preserving vegetation, improving ventilation routes, and reducing heat load in built-up regions by regulating urban morphology factors [118,119].
The second highest factor loading (0.887) belonged to SS5 (traffic congestion). According to [39], increased vehicular traffic and road congestion exacerbate the UHI situation and air pollution. Increased vehicular traffic also results in increased energy consumption, elevated emissions of greenhouse gases, and compromised human health and comfort [120]. Simpson [121] attributed a surge in vehicular traffic to the affordability and ease of ownership of private cars, as well as inefficiency of public transport.

Confirmatory Factor Analysis of UHI Adaptive Capacity Indicators
The UHI adaptive capacity component comprises three dimensions linked to three pillars of sustainability: social, economic and environment. The social, economic, and environment dimensions consist of four, four, and three UHI adaptive capacity indicators, respectively. The 11 UHI adaptive capacity indicators were validated with CFA to determine their factor loadings and reliability (first-order) and their dimensions (second-order), and SEM was used to analyze the relationships between the UHI indicators.

Conclusions
This research proposes a collection of 46 UHI risk indicators under four UHI risk components: hazard, exposure, sensitivity, and adaptive capacity. The UHI risk indicators were first validated by experts to determine their relevancy. The expert-validated UHI indicators were subsequently converted into a Likert-scale questionnaire and applied to randomly sampled residents of Thailand's capital Bangkok. The respondents' answers were input into SEM analysis.
In the SEM analysis, the 46 UHI risk indicators and four UHI risk components were validated with CFA to determine the factor loadings and reliability of the UHI risk indicators and dimensions. The indicator-and dimension-level factor loadings (0-1) were used to indicate the degree of relevancy of the UHI risk indicators and dimensions.
In the UHI hazard component, the CFA factor loadings of the duration and temperature dimensions were 0.991 and 0.937, respectively, indicating that the duration dimension plays a slightly more significant role in the UHI hazard component than the temperature dimension. Furthermore, the highest indicator-level factor loading (0.915) belonged to HT6 (percentage of days when the daily minimum temperature is less than the 10th percentile, TN10).
In the UHI exposure component, the highest indicator-level CFA factor loading (0.923) belonged to ES4 (vehicular traffic). Vehicular traffic is the aggregation of vehicles coming and going in a particular locality and was positively correlated with UHI intensity. The second highest factor loading (0.917) belonged to EV3 (pervious surface area). The conversion of pervious to impervious surfaces as a result of economic development and urbanization expansion causes temperatures in urban areas to be significantly warmer than surrounding rural areas.
Under the UHI sensitivity component, the proportion of green space to build environment indicator (SV1) had the highest factor loading (0.910). Urban green space provides cool and shaded areas while moderating ambient temperatures. Specifically, the UHI impacts were found to be inversely correlated with the proportion of urban green space to build environment. The second highest factor loading (0.887) belonged to SS5 (traffic congestion). Increased vehicular traffic and road congestion exacerbate the UHI situation and air pollution. Increased vehicular traffic also results in increased energy consumption, elevated emissions of greenhouse gases, and compromised human health and comfort.
In the UHI adaptive capacity component, the highest factor loading (0.910) belonged to AS4 (government policy and action), followed by AS3 (multi-agency collaboration) with a factor loading of 0.887. Government policy and action plays a crucial role in the UHI adaptive capacity of urban residents. In addition, multi-agency collaboration on the dissemination of information and UHI impact mitigation enhances adaptive capacity to abrupt and dramatic temperature increases.
In the implementation of the proposed statistically validated risk indicators to assess area-specific UHI susceptibility, the UHI indicators with high or very high factor loadings are converted into a questionnaire (i.e., UHI risk assessment tool) for primary data collection. In the event that the data are publicly available, the information related to the UHI indicators with high or very high factor loadings can be gathered from secondary sources (e.g., textbooks, reports, and news releases). The collected data (both primary and secondary data) are analyzed using normalized scores, percentages, or weighted average indexes to obtain area-specific UHI susceptibility index scores. The UHI susceptibility index scores are subsequently transformed into a UHI risk map that can be used by policymakers to formulate proper UHI mitigation measures.

Research Implications
Although the proposed UHI indicators could be applied to assess UHI susceptibility, UHI risk assessment was outside the scope of this current research given that the aim of this research was to propose the statistically validated UHI risk indicators under four UHI risk components: hazard, exposure, sensitivity, and adaptive capacity. Nevertheless, the UHI risk assessment of 50 districts of metropolitan Bangkok will be carried out in subsequent research.
In addition, since this current research was limited to one metropolitan area (i.e., Thailand's capital Bangkok), future research could expand the study area to encompass diverse geographical locations, i.e., other major capital cities. Furthermore, in UHI risk assessment, the proposed 46 UHI risk indicators could be partially (<46 indicators) or wholly adopted, depending on data availability and the characteristics of the urban area.