Is Sensible Heat Flux Useful for the Assessment of Thermal Vulnerability in Seoul (Korea)?

Climate change has led to increases in global temperatures, raising concerns regarding the threat of lethal heat waves and deterioration of the thermal environment. In the present study, we adopted two methods for spatial modelling of the thermal environment based on sensible heat and temperature. A vulnerability map reflecting daytime temperature was derived to plot thermal vulnerability based on sensible heat and climate change exposure factors. The correlation (0.73) between spatial distribution of sensible heat vulnerability and mortality rate was significantly greater than that (0.30) between the spatial distribution of temperature vulnerability and mortality rate. These findings indicate that deriving thermally vulnerable areas based on sensible heat are more objective than thermally vulnerable areas based on existing temperatures. Our findings support the notion that the distribution of sensible heat vulnerability at the community level is useful for evaluating the thermal environment in specific neighbourhoods. Thus, our results may aid in establishing spatial planning standards to improve environmental sustainability in a metropolitan community.


Introduction
Approximately 48% of the global population will face lethal heat waves by 2100 due to rising global temperatures associated with climate change [1]. Moreover, as urbanization accelerates [2], researchers have raised concerns regarding the deterioration of the thermal environment [3,4]. Several studies have investigated the urban heat island (UHI) effect [5][6][7] in efforts to address global warming due to urbanisation [8]. Additional studies have examined the influence of heat waves on heat-related mortality [9,10]. Ebi et al. argued for the need to improve analytical techniques for exploring and identifying vulnerabilities due to urban warming [11]. The UHI effect represents a serious threat to citizens of population-concentrated metropolitan areas [12,13] where urban canyons exacerbate thermal conditions and increase heat concentration [14,15]. Moreover, recent studies have assessed heat vulnerability mainly in Europe and the United States [16] in regional scale and shown the trend of using various methods; a principal component analysis, a regression, and multi-criteria outranking approach [17][18][19]. Therefore, it is necessary to review thermal vulnerability among residents in urban and metropolitan areas at the community level [20,21]. outskirts, whereas hazardous thermal areas may be scattered throughout the city [53]. The degree of urban sprawl also influences the distribution of hazardous thermal areas [54]. However, an integrated socioeconomic response strategy for climate change is critical [55]. To cope with the progress in urbanisation, it is becoming increasingly necessary to elucidate environmental characteristics that influence thermal vulnerability to reduce the impact of the UHI effect and heat waves.
Currently, studies regarding thermal vulnerability in a given area are derived from: (1) a sensitivity based on population, socioeconomic indices, (2) an ability to adapt to high temperatures according to the level of medical infrastructure, and (3) a climate change exposure based on climate-related variables of heat, temperature and spatial attributes. Assessments of thermal vulnerability rely on the three indices; sensitivity, adaptive ability and the exposure, including socioeconomic vulnerability, isolation of older adults, and the number of unrecognised areas, which strongly influence surface temperature [56]. Some research, which includes the three indices, has also suggested that the thermal vulnerability index is related to environmental factors [57], with several researchers attempting to verifying statistical relationships for thermal vulnerability [58][59][60].
One major factor that cannot be overlooked is the response to heat. In other words, heat may differ due to differences in thermal sensitivity based on age and personal characteristics. Such differences complicate the scope and scale of heat waves in urban areas [61], although researchers have developed a web-based tool for combining and mapping the vulnerability index among older adults [62]. Older adults living alone, preschool children, and patients with heat-related illnesses living in a community are considered vulnerable or sensitive to high temperatures or sensible heat flux [63,64]. Sensible heat flux can be mitigated by increases in the number of medical institutions or green spaces in the community. Thus, thermal vulnerability due to external stresses can decrease depending on the level of social efforts. In general, vulnerability can be reduced by mitigating external stresses or by strengthening internal adaptive capacities [65]. The sensitivity of sensible heat flux is important for assessing vulnerability in various populations [66].
In order to identify and mitigate vulnerability, most of all, analyses of extreme urban heat events must take spatial characteristics into account [10]. Spatial mapping of vulnerability can be improved using a meso-scale approach to city-level units [32]. That is, it is necessary to approach vulnerability at the community level [20].
According to previous reviews, we found that the current method for identifying vulnerable areas is based on temperature and therefore does not reflect the heat budget. As urban heat can be influenced by sensible heat flux or heat balance [67], we aimed to compare methods for assessing vulnerability based on existing temperatures after examining whether the current method, which is based on heat budget as determined using spatial characteristics, can be utilised for vulnerability assessments.
For this reason, sensible heat flux is an important variable for estimating thermal vulnerability. Sensible heat flux, especially that related to heat yields, depends of the type of land cover. Therefore, extracting heat-sensitive areas based on the sensible heat flux makes it easy to identify the ratio of land cover. Diverse methods can be used to secure heat comfort [68,69]. Given the relationship between heat balance and thermal comfort [70], heat flux can aid in a spatial action planning to improve the thermal environment at a community scale. However, there are two important questions concerning this approach. First, can sensible heat flux at the community level be used to conduct vulnerability index that accurately reflect the thermal environment? Second, which variable is easier to identify vulnerable areas, sensible heat flux or temperature?
In the present study, we aimed to identify a more reasonable method for evaluating thermal vulnerability by comparing areas with high sensible heat flux and areas with high temperature using two methods. One method relied on heat budget to determine sensible heat, whereas another relied on temperature. If the vulnerable areas identified based on heat budget and temperature are similar, our results should support the adoption of a new vulnerability index that incorporates spatial attributes. Thus, we verified thermal vulnerability based on sensible heat flux by comparing spatial distribution patterns (maps) of mortality rates and vulnerability at a community level.

Case Study
Seoul (37.33 • N 126.58 • E) is a global mega-city with a population density of 27,018 people/km 2 [71]. It is the capital of Korea and home to approximately 19% of the total population (9,780,000/51.47 million) (2017, Statistics Korea). Approximately 65% (366 km 2 ) of the city's surface area is covered by artificial surfaces. Summers in Seoul are characterised by heavy rain (precipitation: 892.1 mm), whereas winters are relatively dry (precipitation: 67.3 mm). Since heatwaves vary in duration, intensity, or temperature depending on where they occur, there are naturally various definitions of heatwaves [72,73]. Therefore, studies that set the threshold through health indicators have recently been performed. Although the average daytime temperature in the summer is 32 • C, temperatures can reach as high as 37 • C. Summer in Seoul begins in mid-June and lasts until early September. In 2015, the number of heat-wave days in Seoul was 5.9 days in August, which was higher than the average number of annual heat-wave days (5.3 days) [74]. In contrast, the numbers of heat-wave days for June and July were 0.6 and 3.2 days, respectively (Korean Meteorological Administration, KMA). The term "heat-wave" does not have a universally consistent definition but we considered heat-waves as temperatures of 33 • C or higher for 2 or more days; in addition, heat-wave days refer to the number of days with the highest daily temperature above 33 • C according to the KMA [75]. Thus, in this study, the number of thermal mortalities was counted from the second day from the start of the heat wave [18]. The number of patients experiencing heat-related illness during the summer in Seoul continues to increase each year. In urban areas, which are vulnerable to the UHI effect, these effects are further amplified. Notably, the UHI effect also continues to increase in Seoul due to the widespread presence of impervious surfaces and decreases in green space [76].

Method for Assessing Thermal Vulnerability
In the present study, we analysed the 2015 Geographic Information System (GIS) spatial data for 438 neighbourhood administrative districts provided by an open Seoul database and the Korean Statistical Information Service (KOSIS); the ratio of five landcovers and administrative boundary ( Figure 1, Appendix B).

Case Study
Seoul (37.33° N 126.58° E) is a global mega-city with a population density of 27,018 people/km 2 [71]. It is the capital of Korea and home to approximately 19% of the total population (9,780,000/51.47 million) (2017, Statistics Korea). Approximately 65% (366 km 2 ) of the city's surface area is covered by artificial surfaces. Summers in Seoul are characterised by heavy rain (precipitation: 892.1 mm), whereas winters are relatively dry (precipitation: 67.3 mm). Since heatwaves vary in duration, intensity, or temperature depending on where they occur, there are naturally various definitions of heatwaves [72,73]. Therefore, studies that set the threshold through health indicators have recently been performed. Although the average daytime temperature in the summer is 32 °C, temperatures can reach as high as 37 °C. Summer in Seoul begins in mid-June and lasts until early September. In 2015, the number of heat-wave days in Seoul was 5.9 days in August, which was higher than the average number of annual heat-wave days (5.3 days) [74]. In contrast, the numbers of heat-wave days for June and July were 0.6 and 3.2 days, respectively (Korean Meteorological Administration, KMA). The term "heat-wave" does not have a universally consistent definition but we considered heatwaves as temperatures of 33 °C or higher for 2 or more days; in addition, heat-wave days refer to the number of days with the highest daily temperature above 33 °C according to the KMA [75]. Thus, in this study, the number of thermal mortalities was counted from the second day from the start of the heat wave [18]. The number of patients experiencing heat-related illness during the summer in Seoul continues to increase each year. In urban areas, which are vulnerable to the UHI effect, these effects are further amplified. Notably, the UHI effect also continues to increase in Seoul due to the widespread presence of impervious surfaces and decreases in green space [76].

Method for Assessing Thermal Vulnerability
In the present study, we analysed the 2015 Geographic Information System (GIS) spatial data for   [77]. The variable "admin_GU" represents 25 administrative districts and "ADMIN_DONG" refers to neighbourhood administrative districts. Source: GIS map (https://sgis.kostat.go.kr).

Figure 1.
Map of Seoul. Note: CBD refers to a traditional business district. YBD indicates Yeoido business district, and GBD stands for Gangnam business district. These three districts are the main urban centres in Seoul [77]. The variable "admin_GU" represents 25 administrative districts and "ADMIN_DONG" refers to neighbourhood administrative districts. Source: GIS map (https: //sgis.kostat.go.kr).
To evaluate thermal vulnerability, we performed a correlation analyses of indices used in previous studies [78]. Our study was focused on the suggestion of other variables for investigating thermal vulnerability for thermal environment improvement. There are three parts of the research; vulnerability index and variables, sensible heat flux (Q h ) estimation and thermal vulnerability index (TVI) variables verification ( Figure 2). Firstly, we tested correlation of variables based on previous researches related to the thermal vulnerability (Appendix B Table A4). Then, we conceived variables of the exposure for the thermal vulnerability from comparing temperature-based vulnerability and sensible heat flux-based vulnerability.
researches related to the thermal vulnerability (Appendix Table A4). Then, we conceived variables of the exposure for the thermal vulnerability from comparing temperature-based vulnerability and sensible heat flux-based vulnerability.
In the second part, we estimated the sensible heat flux in community scale (Equations (3), (A1)-(A4), Tables A1 and A2) for the comparison in the first part. In this step, we rasterised the sensible heat flux based on existing data (Appendices A and B), considering the energy balance, net radiation, latent heat, sensible heat, storage heat, and artificial heat [24]. Then, we used the heat flux as a variable of sensible heat vulnerability.
In the third part, we rasterised three indices maps and found a different pattern of the exposure index between sensible heat flux-based map and temperature-based map ( Figure 3). Then, we derived thermal vulnerability index (TVI) (Equations (1) and (2)), classified in five levels for the rasterization and comparison (Table 1, Appendix Figure A2). To verify the TVI variables, we compared a trend line of sensible heat vulnerability and temperature vulnerability with correlation of two variables' vulnerability ( Figure 4, Equation (4)). Lastly, answering to the research question, we performed validation comparing the sensible heat and temperature vulnerability index maps, and mortality rate map in community scale ( Figure 5).  In the second part, we estimated the sensible heat flux in community scale (Equations (3), (A1)-(A4), Tables A1 and A2) for the comparison in the first part. In this step, we rasterised the sensible heat flux based on existing data (Appendices A and B), considering the energy balance, net radiation, latent heat, sensible heat, storage heat, and artificial heat [24]. Then, we used the heat flux as a variable of sensible heat vulnerability.
In the third part, we rasterised three indices maps and found a different pattern of the exposure index between sensible heat flux-based map and temperature-based map ( Figure 3). Then, we derived thermal vulnerability index (TVI) (Equations (1) and (2)), classified in five levels for the rasterization and comparison (Table 1, Appendix C Figure A2). To verify the TVI variables, we compared a trend line of sensible heat vulnerability and temperature vulnerability with correlation of two variables' vulnerability ( Figure 4, Equation (4)). Lastly, answering to the research question, we performed validation comparing the sensible heat and temperature vulnerability index maps, and mortality rate map in community scale ( Figure 5).

Thermal Vulnerability and Thermal Vulnerability Index (TVI)
IPCC defines the vulnerability as "the degree to which a system is susceptible to, or unable to cope with, adverse effects of climate change, including climate variability and extremes." Thus, heat vulnerability is "a function of the character, magnitude, and rate of heat variation to which a system is exposed, its sensitivity, and its adaptive capacity" [79]. Vulnerability to heat has been estimated based on the combination of climate change exposure (E), sensitivity (S), and adaptive capacity (A) [21,65,[80][81][82]. However, in the present study, we aimed to determine whether indices other than temperature can be used to assess thermal vulnerability. Therefore, in contrast to previous studies, we compared the ability of high temperature and high sensible heat flux and spatial attributes as the exposure variables for determining vulnerability.
In the present study, thermal vulnerability was determined based on external stresses, such as sensible heat flux, temperature [56,83], and the influence of the built environment. External stress is a parameter that indicates the extent to which heat is exposed to space and is among the components used to calculate thermal vulnerability, which reflects human sensitivity and adaptive capacity to heat. Temperature was defined as the average temperature during the day (between 12:00 and 16:00) in a community scale, whereas maximum temperature was defined as the highest temperature during the same period. Sensible heat vulnerability assessments also rely on indices of sensible heat, which refers to the heat experienced by residents. In the present study, we proposed a heat index for evaluating thermal vulnerability according to sensible heat flux, which reflects the degree of the exposure within a given environment.
Assessments of thermal vulnerability involve a series of processes [84] that comprehensively determine how sensitive an area is to the effects of heat change, and how well it is capable of adapting. In the present study, we adopted a formula to evaluate the level of thermal vulnerability. This formula applies the numerical values for the exposure, sensitivity, and adaptive capacity derived from the Analytical Hierarchy Process (AHP) based on questionnaires from disaster-related practicing professionals [81,85]; 35 of managers and employees in various fields of industries such as car, gas, electricity, and researchers in the climate change field. The questionnaire data collected from the professionals were converted using a weighing scale, ranging from 0 to 1, with the AHP [86]. The weights of three indices influencing heat vulnerability were applied to each index and the sum of these were expected to be equal to 1: The exposure, sensitivity, and adaptive capacity are composed of various quantitative variables. Values of each variable are standardised and added to the upper level elements (e.g., sensitivity and adaptive capacity). Each summed value is again normalised and applied to Equation (1) to calculate thermal vulnerability, the standardised value (Equation (2)) which ranges between 0 and 1. The standardisation prevented the generation of spatial deflections induced by very large or small census sites: Finally, the resulting vulnerability values are divided into five classes [82], as follows. Because there are none values on mortality rate, the range of the both sensible heat and temperature vulnerability was classified using hierarchical clustering (Appendix D) to compare with the death rate [87].
All vulnerability maps were generated based on these five classes (Table 1). Various thermal variables we used for the thermal vulnerability are described in detail in the next section. The thermal vulnerability index (TVI) [82] was determined via spatial analysis. Spatial patterns were mapped throughout the city. A set of selected variables related to thermal vulnerability was prepared and correlation analysed. The fundamental goal of utilizing an efficient set of variables was to identify both the obvious, as well as the stable, but important, relationships between the occurrence of heat events and the associated health hazard, in terms of various vulnerability variables. These variables include population size, elderly individuals, elderly individuals living alone, people living below the poverty line, laborers, income, medical insurance budget, spatial attributes, and the number of medical institutes. Spatial attributes include vegetated areas (green), wetland and water surface (water), impervious surface, building, and road (see Appendix B Table A4). Since we researched urbanised areas at the community level, we did not include urban infrastructure, such as access to water or electrical supply and good roads, as an adaptive capacity. Although all the data were available for every year, we use the data for 2015, the latest available meteorological data collected monthly from AWS and a severity of the heat-wave. Table 2 lists the variables of vulnerability assessment from published statistical and census data. The explanations for thermal vulnerability indices are elaborated in each chapter. We analysed data for August 2015 due to the availability of certain data (e.g., health-related diseases and sensible heat) and the severity of the heatwave in 2015.

Sensitivity
Sensitivity is a social factor that increases vulnerability in a heat-intensive space within a city. Sensitivity was quantified based on six variables: (1) population density, (2) population of older adults over age 65 living alone, (3) population of preschool children under age 5, (4) number of heat-related illnesses, (5) population of below poverty line (BPL), and (6) population of heat-related deaths in August(Department of Welfare for Seniors, Seoul Metropolitan Government).
The number of patients with heat-related illnesses (3) corresponds to the sum of: (i) the number of patients with respiratory illnesses, (ii) the number of patients with heart disease, and (iii) the number of patients with cerebrovascular disease experiencing refractory illnesses. Data sources and descriptions are shown in Table 1. However, due to the definition of heat-wave, we also considered the second day of the heat-wave when we counted heat-related disease mortality.
The total number of items per category was weighted by each variable's normalised value of ranking, and the proportion of elements included in the vulnerability analysis varied. Total numbers per category were divided by the total and divided by the population ratio for each administrative area to obtain a relative vulnerability assessment of the area. Demographic data were obtained from a 2015 source (Table 2).

Adaptive Capacity
Adaptive capacity weakens thermal vulnerability in heat-concentrated areas. Thus, there is an inverse relationship between the sum of all adaptive capacity variables, climate change exposure and sensitivity. High adaptive capacity lowers the total vulnerability score and overall spatial thermal vulnerability in areas with high concentration.
Adaptive capacity was determined based on: (1) income, (2) the number of medical institutions, and (3) the annual medical insurance budget by an administrative community (called "dong"). Income was defined as average monthly income in 2015, which was calculated by dividing household income by the population of the neighbourhood unit. We included the number of hospitals (rather than distance from the hospitals) as one factor of adaptive capacity. The reason why the number of beds was considered as an important factor is that there are hospitals at the centre of every "Dong" at the community level, which allows the residents to access the hospitals within a radius of approximately 2 km [88]. In Seoul, the number of hospitals [32] or the number of beds was more likely to affect vulnerability.

Exposure
In this study, the climate change exposure is composed of three variables: air temperature, sensible heat flux and landcover ratio (here, spatial attributes). Air temperature data were obtained from 38 Automatic Weather Stations (AWS) in Seoul and from 249 AWS owned by SKTech X (Seoul, Republic of Korea), a private company. Data were constructed from equipment installed at least 15 m above the ground (i.e., four stories). Sensible heat flux was calculated using the energy budget model adopted from previous studies [24]. The value was calculated using a thermodynamic model that considers the urban canyon structure. It contains landcover ratio of the study site and architectural aspects, such as the height of buildings and altitude. The ratio of five landcovers was extracted from landcover classification shape files. Area statistics for each census area were calculated using ArcGIS Desktop Release 10.2 (ESRI Inc., Redlands, CA, USA, 2014). Exposure values for each census were calculated by averaging the pixel-based estimates for each census; exposure was estimated using the standard deviation of sensible heat flux for the same census area for which atmospheric temperatures were obtained.

Sensible Heat Flux Estimation
Sensible heat flux was derived based on four types of heat flux [24]. We used the following energy budget model to calculate sensible heat flux based on net radiation (Q n ), which was derived by Offerle [89]. All of the heat fluxes have units of W/m 2 : [89][90][91][92]:

of 26
The energy budget is composed of four elements, namely, the anthropogenic heat flux (Q F ), sensible heat flux (Q h ), latent heat flux (Q e ), storage heat flux (Q s ). All the models are explained in Appendix A.

Thermal Disease-Related Mortality
Heat-related diseases were defined based on a previous study [63]. In this study, we included data for patients with respiratory, heat disease (ischaemic) and cerebrovascular disease to calculate thermal vulnerability. Those diseases are mostly higher than 1 (cumulative lags 0-2) of relative risk (RR) under 95% confidence intervals of mortality and extremely hot days by 66 causes of death. The number of deaths in August 2015 due to these three diseases was determined for both sexes, and each distribution was calculated and extracted among various mortality causes by using SPSS (IBM Corp. Released 2013. IBM SPSS Statistics for Windows, Version 22.0.) and R (R Core Team, 2017). In addition, we obtained raw data of heat-related diseases ( Table 3) and mortality rate from Korea Centres for Disease Control and Prevention for estimating the sensitivity and comparison of the vulnerability indices.

Results and Discussion
One aim of the present study was to identify additional design criteria for improving the thermal environment based on vulnerability analysis. In our correlation analysis, we investigated the distribution and numerical value of temperature, which are existing indicators of the sensible heat vulnerability, as well as the distribution and numerical value of sensible heat flux, a novel indicator. We compared these variables in relation to the number of heat-related deaths. We also aimed to determine the spatial distribution of sensible heat vulnerability in relation to mortality rates, and whether these exhibited quantitative correspondence with the spatial distribution of temperature vulnerability in relation to mortality rate.

Maps of Three Indices
We derived three kind of indices maps, sensitivity, adaptive capacity, and exposure ( Figure 2). To find different variables' impact on the indices, we use the exposure map based on sensible heat, we also created sensible vulnerability based on sensible heat flux during the day (map of Figure 3). Because the vulnerability indices are standardised values, a range of class is divided by equal interval in GIS, which is best applied to familiar data ranges, such as temperature.
The vulnerability index, which does not reflect temperature and sensible heat, shows sensitivity and adaptive capacity for both maps a) and b) (Figure 3). However, due to changes in temperature and sensible heat flux, which are included in climate change, maps of temperature or areas vulnerable to heat are different.

Validation Based on the Root Mean Square Error (RMSE)
We compared the distribution of heat-related deaths in August 2015 between areas with high vulnerability based on sensible heat flux and temperature. The two vulnerability distributions are shown in Figure 4. Calculations were performed based on the RMSE for each vulnerable area. We observed a correlation of 0.71 (Table 4) between vulnerable areas based on sensible heat f lux and mortality rate. These findings suggest that deriving vulnerable areas based on sensible heat flux is more objective than deriving them based on temperature. Indeed, the high correlation coefficient indicates the reliability of the sensible heat flux as an index for determining thermal disease-related mortality: and represent death rate and vulnerability, respectively, while n represents the number of administrative regions (Equation (4)). In this study, we investigated the distribution of death rates among patients with febrile illness during vulnerability assessments based on temperature and sensible heat.

Validation Based on the Root Mean Square Error (RMSE)
We compared the distribution of heat-related deaths in August 2015 between areas with high vulnerability based on sensible heat flux and temperature. The two vulnerability distributions are shown in Figure 4. Calculations were performed based on the RMSE for each vulnerable area. We observed a correlation of 0.71 (Table 4) between vulnerable areas based on sensible heat flux and mortality rate. These findings suggest that deriving vulnerable areas based on sensible heat flux is more objective than deriving them based on temperature. Indeed, the high correlation coefficient indicates the reliability of the sensible heat flux as an index for determining thermal disease-related mortality: D i and V i represent death rate and vulnerability, respectively, while n represents the number of administrative regions (Equation (4)). In this study, we investigated the distribution of death rates among patients with febrile illness during vulnerability assessments based on temperature and sensible heat.
The changes in community (Dong)  We adopted this approach to validate the objectivity of these indicators. RMSE analyses provide a numerical value representing the accuracy of the model by comparing the difference between the estimated value and the actual value. For RMSE analyses, we adopted the two vulnerability values as the estimated values, whereas mortality rate was adopted as the actual value. Although RMSE analyses are associated with scale-dependent errors, the influence of such errors depends on the range of the estimated value. However, in this study, changes in sensible heat flux and temperature were standardised for the vulnerability assessment, and the mortality rate was standardised in the same way. We adopted this approach to validate the objectivity of these indicators. RMSE analyses provide a numerical value representing the accuracy of the model by comparing the difference between the estimated value and the actual value. For RMSE analyses, we adopted the two vulnerability values as the estimated values, whereas mortality rate was adopted as the actual value. Although RMSE analyses are associated with scale-dependent errors, the influence of such errors depends on the range of the estimated value. However, in this study, changes in sensible heat flux and temperature were standardised for the vulnerability assessment, and the mortality rate was standardised in the same way.  Our results indicated that the distribution pattern for vulnerability assessments based on a correlation coefficient value of 0.431 (0.734-0.303) for sensible heat flux is higher than that for assessments based on temperature when examining mortality rates (Table 4). In other words, sensible heat flux is more useful for estimating thermal vulnerability considering the mortality from heat-related illness than by considering temperature at the community level.
Sensible heat vulnerability index was associated with mortality rates that were quantitatively similar to those for temperature vulnerability index ( Figure 5). However, the mortality rate was determined based on the average for the administrative unit of Dong, which may not accurately reflect the standard variables for the community. Our results indicated that the distribution pattern for vulnerability assessments based on a correlation coefficient value of 0.431 (0.734-0.303) for sensible heat flux is higher than that for assessments based on temperature when examining mortality rates (Table 4). In other words, sensible heat flux is more useful for estimating thermal vulnerability considering the mortality from heatrelated illness than by considering temperature at the community level.
Sensible heat vulnerability index was associated with mortality rates that were quantitatively similar to those for temperature vulnerability index ( Figure 5). However, the mortality rate was determined based on the average for the administrative unit of Dong, which may not accurately reflect the standard variables for the community.

Findings from Community's Comprehensive Thermal Vulnerability Index
We obtained the indicators to improve the thermal environment based on the sensible heat flux change values of the communities associated with the thermal vulnerability index; and the mean, minimum, and maximum values were calculated, and the following results were noted:

Findings from Community's Comprehensive Thermal Vulnerability Index
We obtained the indicators to improve the thermal environment based on the sensible heat flux change values of the communities associated with the thermal vulnerability index; and the mean, minimum, and maximum values were calculated, and the following results were noted: The mortality rate in August shows that communities with higher sensible heat flux had a higher thermal vulnerability ( Table 5). The mortality rate was calculated considering an annual total number of mortalities in 2015 and number of August mortalities. The Oryu community had the highest number of mortalities in August; however, the annual rate was lower than those in the Wolgye and Noryiangjin communities. The annual number of mortalities in Oryu was the highest at 183.

Spatial Attributes and Patterns Related to Sensible Heat Vulnerability
To examine the spatial causes of sensible heat flux vulnerable communities, we observed the differences in sensible heat vulnerability between older urban areas of the city and urban residential areas ( Figure 6). Areas with higher vulnerability mainly included dense residential areas and compact spatial structures. Areas with lower vulnerability mainly included low-density residential areas. The distribution of sensible heat vulnerability also differed from that for temperature vulnerability. The mortality rate in August shows that communities with higher sensible heat flux had a higher thermal vulnerability ( Table 5). The mortality rate was calculated considering an annual total number of mortalities in 2015 and number of August mortalities. The Oryu community had the highest number of mortalities in August; however, the annual rate was lower than those in the Wolgye and Noryiangjin communities. The annual number of mortalities in Oryu was the highest at 183.

Spatial Attributes and Patterns Related to Sensible Heat Vulnerability
To examine the spatial causes of sensible heat flux vulnerable communities, we observed the differences in sensible heat vulnerability between older urban areas of the city and urban residential areas ( Figure 6). Areas with higher vulnerability mainly included dense residential areas and compact spatial structures. Areas with lower vulnerability mainly included low-density residential areas. The distribution of sensible heat vulnerability also differed from that for temperature vulnerability. Spatial patterns of temperature vulnerability differed significantly among the western, northern, and southern regions of the city. The highest vulnerability values were observed in the southwest and northeast areas of the city. These areas are low-rise residential neighbourhoods with high population densities. The lowest values were observed in the northwest and southeast regions of the city, which include urban areas. Spatial patterns of temperature vulnerability differed significantly among the western, northern, and southern regions of the city. The highest vulnerability values were observed in the southwest and northeast areas of the city. These areas are low-rise residential neighbourhoods with high population densities. The lowest values were observed in the northwest and southeast regions of the city, which include urban areas. Low vulnerability values in the city's core district may have been influenced by high building density or high income, as high-rise buildings exert a cooling effect by providing shade. In some cases, cooling is associated with high plant density and the presence of green infrastructure, parks, or other green areas [24,93]. In other cases, lower levels of impervious surfaces are associated with lower temperatures and reduced thermal vulnerability.
The spatial characteristics of communities associated with thermal vulnerability affecting the health of residents due to changes in living environment in each community are as follows ( Figure 6): (a) Seoul's 438 "dongs" have individual placemarks based on the culture and traditions of each community. Each community expresses the thermal environment in a distinctive space, creating sensible heat mainly in the building and open space among five typical urban land cover factors. These communities ( Figure 6, picture 1 ~6 ) had a land cover attribute that reduced sensible heat flux. According to the previous study, when the area of green surface increased by 1%, the sensible heat flux decreased by 4.9 w/m 2 [24]. However, an increase in impervious surface area contributed to increased sensible heat flux ( Figure 6, picture 7 ~12 ). (b) In this study, we obtained a "street view" that had a symbolic place among the communities with relative uniqueness of thermal vulnerability. By reviewing twelve pictures as shown in Figure 6, the properties of two thermal environmental types, favourable and unfavourable areas, were reflective of the land cover types [24].

Conclusions
In the present study, we compared the distribution of sensible heat vulnerability and temperature vulnerability at the community level in order to identify an appropriate indicator that can be used to improve the thermal environment. Our findings indicated that the distribution of vulnerability as derived based on sensible heat was more similar to the distribution of mortality rate than to the distribution of vulnerability as derived based on temperature. Thus, our results demonstrate that sensible heat flux can be used as an objective indicator in the assessment of thermal vulnerability. Future studies should aim to determine how this indicator can be used to inform spatial planning and design criteria [24].
Adaptation to climate change is a holistic issue affected by social, economic, and environmental factors [94]. There are inherent limitations to using standardised values for correlation coefficients, given the differences in clinical background and personal characteristics among individuals [95]. Therefore, although the correlation between mortality and temperature was under 0.5 in the present study, one cannot conclude that this parameter is meaningless. Additional studies should aim to establish standard indicators for improving the thermal environment at the community level.
Our results suggest that difficulties in reflecting the spatial characteristics that influence thermal vulnerability can be addressed using measures of heat flux. The sensible heat vulnerability index can, therefore, be useful to developing practical methods for improving the thermal environment and estimating another thermal barometer, sensible heat flux in community scale. The proposed variable, sensible heat flux, could be further employed in examining the thermal vulnerability in the community level. Further, a suggested technique on vulnerability to heat at spatial resolutions finer than the regional scale is useful to assist decision makers with mitigation of the vulnerability in the preparation for and response to extreme heat events.
And it helps representing spatial characteristics of a community, where heat vulnerable areas are derived, based on sensible heat flux. This research contributes to considering the way to create an implemental tool, a sensible heat vulnerability index mapping method, and designing standards for thermal environment improvement. Therefore, we encourage the spatial fine resolution and an approach with the new variable to assess the vulnerability. We also suggest that this methodology has an indirect effect on spatial sustainability.  (2) was derived from an equation considering the land cover ratio and empirical coefficients. Here, γ is a psychrometric constant, and s is the slope of the curve of saturation vapour pressure versus temperature: α : an empircal parameter related to the moisture status of the surface. Both sensible heat flux (3) and latent heat flux (4) were estimated using the model developed by Holtsalg and Van Ulden [90]. In these equations, 20 (W/m 2 ) is an empirical constant. Both α and 20 (W/m 2 ) were determined based on the Penman Monteith approach [96]: where T C represents the maximum daily temperature (unit: K) and T d represents the mean daily temperature (unit: K). The anthropogenic heat flux (5) was estimated using a model that considers temperature. Advection (∆Q A ) was negligible at the six investigated sites on the day of measurement because we chose a day with a wind speed of approximately 0 m/s. The land cover ratio ( f i ) was calculated using the relative area occupied by each type of land cover within a 100 m ×100 m grid (Tables A1 and A2.). Empirical land cover coefficients and anthropogenic heat flux at the neighbourhood scale were calculated as previously described [24].

Appendix B. Data
The hourly heat flux was calculated using the data of air temperature, atmospheric pressure, and relative humidity collected from 287 stations in total, including 38 meteorological stations in Seoul and 249 SKTech X meteorological stations. We typified the heat flux and land cover data ( Figure A1, Table A3) on a day with clear weather, low cloud cover, and peak air temperature in August (Table A3) [24]. We also created a 100 m × 100 m grid to map the thermal environment and thermal distribution data.

Appendix B. Data
The hourly heat flux was calculated using the data of air temperature, atmospheric pressure, and relative humidity collected from 287 stations in total, including 38 meteorological stations in Seoul and 249 SKTech X meteorological stations. We typified the heat flux and land cover data ( Figure  A1, Table A3) on a day with clear weather, low cloud cover, and peak air temperature in August (Table A3) [24]. We also created a 100 m × 100 m grid to map the thermal environment and thermal distribution data.     Table 2).