Enhancing Sustainability: Quantifying and Mapping Vulnerability to Extreme Heat Using Socioeconomic Factors at the National, Regional and Local Levels

Abstract

Population-dense urban areas often concentrate high commercial and industrial activity and intricate transportation systems. In crowded cities, extreme events can be even more damaging due to the high population they affect and the social inequalities that are likely to emerge. Extreme heat is a climate hazard that has been linked to high morbidity and mortality, especially in cities with high population densities. The way extreme heat events are felt in the population varies depending on a variety of factors, such as age, employment status, living conditions, air-conditioning, housing conditions, habits, behaviors and other socio-demographic parameters. In this article, we quantify and locate vulnerabilities of populations to extreme heat in order to formulate policy and practice recommendations that will make communities resilient and will shape the transition to a more sustainable future. This work contributes towards the achievement of Sustainable Development Goal 11—Sustainable Cities and Communities—by developing the tools to make cities and settlements resilient and sustainable. To this end, we analyze socioeconomic data at the NUTS3 level for the national case study of Greece and at the census tract level for the local case study of the city of Athens. The target variable for this study is defined as the average daily mortality during heatwaves per 100,000 individuals, and a methodology is developed for constructing this variable based on socioeconomic data available in public databases. The independent variables were selected based on their contribution to socioeconomic vulnerability; they include the percentage of elderly individuals, retirees, unemployed persons, renters, those living alone, those residing in smaller houses, those living in older houses and immigrants from developing countries. An ensemble gradient boosted tree model was employed for this study to obtain feature importance metrics that was used to construct a composite index of socioeconomic heat vulnerability. The socioeconomic heat vulnerability index (SHVI) was calculated for each prefecture in Greece and for each census tract in the city of Athens, Greece. The unique feature of this SHVI is that it can be applied to any geographical resolution using the same methodology and produces a result that is not only quantifiable, but also facilitates a comparison between vulnerability scores across different regions. This application aimed to map the SHVI of both prefecture and city, to examine the significance of scale, to identify vulnerability hotspots, and rank the most vulnerable areas, which are prioritized by authorities for interventions that protect public health.

1. Introduction

In recent years, the phenomenon of extreme heat has emerged as a critical issue, posing significant risks to public health and urban environments worldwide [1]. Heatwaves can be deadly—during the summer of 2003, over 70,000 additional deaths were recorded, while about 20 years later, in the summer of 2022, an estimated number that exceeds 61,500 deaths occurred in Europe [2]. The escalation of such events is closely linked to climate change, leading to increased frequencies and intensities of heatwaves [3,4], affecting resilience, health and wellbeing in cities [5]. In fact, relative to other climate hazards, extreme heat is responsible for the highest morbidity and mortality [6]. Urban areas, due to their dense populations and infrastructure, are especially susceptible to the adverse effects of heatwaves, which can not only lead to heightened mortality rates, but may also strain public health systems [7,8]. Given that over 60% of the world population is expected to live in cities by 2030 [9], it is imperative for urban planners and policymakers to understand, quantify, locate and assess heat-health risks and develop effective adaptation strategies in the cities to protect populations and public health.

The climate change risk assessment to human health presented within the fifth Assessment Report of the IPCC [10] states that risk is considered to result from the interaction of climatic hazard, vulnerability and exposure of humans, ecosystems, economic, social or cultural assets. Climate change vulnerability is a complex issue, as it is not only related to the intensity of extreme events, but also to the socioeconomic status of populations, such as poverty, unemployment, the age of the home, educational level, living conditions, health conditions, etc. The way a heatwave is experienced by society depends on a variety of factors, such as individual attributes, socio-demographics, availability of air-conditioning, individual attributes, health condition, connection with community, having a car, being elderly, ill, being mobility-challenged, being socially isolated, living alone, etc. [11]. There are many factors that define socioeconomic vulnerability to extreme heat, and developing a composite socioeconomic heat vulnerability index (SHVI) that clusters them all is an important tool that allows urban planners and local authorities to locate such vulnerabilities, combine them with other hazard-specific vulnerabilities and develop a plan for timely action.

The 1995 Chicago heatwave serves as a seminal case study for examining the effects of extreme thermal conditions on urban populations, particularly given the projected increase in such events due to climate change. Kaiser et al. [12] employed sophisticated time-series analysis to reassess all-cause and cause-specific mortality during this event, identifying 692 excess deaths over a 50-day period, with 26% attributed to mortality displacement. Their findings emphasize the substantial risk of heat-related mortality, particularly among African-American populations, and indicate the necessity for targeted mitigation strategies. Semenza et al. [13] conducted a case-control study that identified individuals with pre-existing medical conditions, those with limited mobility and socially isolated individuals lacking access to air conditioning as high-risk groups. These studies collectively underscore the importance of comprehending social and health-related vulnerabilities in heatwave preparedness and response strategies. Projections by Hayhoe et al. [14] suggest that by the end of the 21st century, under high-emission scenarios, Chicago could experience heat waves comparable to the 1995 event up to thrice annually, potentially doubling the annual average mortality rates observed during the 1995 heat wave. These forecasts highlight the urgency of implementing both preventive and adaptive measures to reduce population vulnerability to increased thermal extremes. Collectively, these studies provide a comprehensive analysis of historical impacts and future risks associated with heat waves, emphasizing the critical need for enhanced resilience and preparedness strategies in urban environments vulnerable to climate change.

In this study, we delve into the intricate relationship between urban resilience and socioeconomic vulnerability in the face of climate hazards, with a particular focus on extreme heat. A growing body of literature has explored various facets of this issue, emphasizing the importance of statistical and deep learning models in understanding and mitigating these effects. For instance, Founda and Santamouris [15] shed light on the synergies between Urban Heat Island (UHI) effects and heatwaves in Athens. They highlighted the increased thermal risk for urban residents due to a positive feedback loop between UHIs and heatwaves, intensifying the average UHI magnitude by up to 3.5 °C during heatwaves compared to summer background conditions. Heat wave risk has also been linked to age, leading to a variety of adverse health conditions, such as cardiovascular, respiratory and other effects, which means that heatwave mortality will not always be classified as such [16]. Similarly, recent research underscores the role of socioeconomic pathways in shaping future urban heat-related challenges, suggesting that vulnerability is a crucial determinant of heat-related risk and mortality in cities [17]. Their findings indicate that future urban climate risks are significantly driven by population growth and changes in vulnerability, with climate conditions alone being of lesser influence [18,19].

In this article, we develop a methodology for quantifying and assessing socioeconomic vulnerability to extreme heat. It is implemented at the level of prefecture, or regional unit (NUTS3 level) in Greece, with a case study centered in Athens, a densely populated urban environment that exemplifies the challenges and vulnerabilities associated with extreme heat. The primary objective is to quantify the average daily mortality during heatwave periods per 100,000 individuals by employing machine learning, namely the gradient boosting ensemble, for capturing nonlinear relationships within limited data volumes. The methodology encompasses several key steps, including the grouping of NUTS3 regions, identification of heatwave events following established criteria [20] and combining aggregated mortality data with temperature time series to construct an appropriate target variable. The independent variables were selected based on their contribution to socioeconomic vulnerability, emphasizing the negative impact of certain demographic and housing characteristics on heat resilience [11]. The overarching aim of this study is to map vulnerability hotspots, rank the most vulnerable areas and offer a clearer understanding of how socioeconomic factors influence heat-related risks. The construction of a composite index, combined with visualization techniques, provides a novel approach to assessing and addressing the multifaceted issue of extreme heat in urban settings. Contrary to other indices developed on vulnerability to extreme heat, the SHVI can be applied to any geographical resolution, using the same methodology, and produces a result that is not only quantifiable, but also allows for a comparison between vulnerability scores across different regions. The article contributes to the ongoing discourse on urban resilience, offering valuable insights and tools for policymakers, urban planners and public health officials tasked with mitigating the impacts of climate change-induced heatwaves.

2. Materials and Methods

For the analysis of the hazard-dependent vulnerability of the population of Athens to extreme heat, we attempted to quantify and correlate socioeconomic indicators, as shown in

Table 1. Socioeconomic indicators used for the development of the SHVI.

Data Description	Scale and Units	Data Source
Maximum temperature for 10 Greek cities, 2015–2020	Hourly, Celsius	Open Weather Map https://openweathermap.org/ (accessed on 10 December 2022)
Number of people by age group, from the 2011 national census	Per NUTS3 region and per census tract in Athens	Hellenic Statistical Authority
Number of people by country of origin, from the 2011 national census	Per NUTS3 region and per census tract in Athens	Hellenic Statistical Authority
Number of people by different household sizes (e.g., 2-people households, 4-people households, etc.) from the 2011 census	Per NUTS3 region and per census tract in Athens	Hellenic Statistical Authority
Number of people by employment status, (employed, looking for work, first-time looking, student, retired, independent, housework, other) from the 2011 census	Per NUTS3 region and per census tract in Athens	Hellenic Statistical Authority
Number of dwellings by living status (owner, renting, cooperative ownership, communal housing, other) from the 2011 census	Per NUTS3 region and per census tract in Athens	Hellenic Statistical Authority
Number of dwellings by square footage, from the 2011 census	Per NUTS3 region and per census tract in Athens	Hellenic Statistical Authority
Number of dwellings by age of building, from the 2011 census	Per NUTS3 region and per census tract in Athens	Hellenic Statistical Authority
Number of deaths by age group, 2015–2020	Per NUTS3 region, weekly	Eurostat

, with a composite-constructed target predictor for extreme heat impact. The predictor is the average number of daily deaths per 100,000 people during heatwaves. To develop it, we used weekly mortality rate timeseries from Eurostat for all 52 prefectures (NUTS3 regions) in Greece from the 2011 national census and the corresponding temperature time series for the 2015 to 2020 time period. Since this study evaluates mortality, all data during and after the COVID-19 pandemic were removed. Assuming that some of the temperature time series of the 52 prefectures are quite similar, as temperatures will not vary so dramatically within the NUTS3 regions, we grouped the country in 10 groups, following adjacency rules and the zonal demarcations established by the Greek Ministry of Environment, to estimate the needs of households for heating or cooling. For each of the 10 groups, we chose one main urban center and used the same temperature time series for all prefectures that are included in the group. The delineation of the 10 groups is shown in Figure 1. The analysis was conducted at two levels: (i) the prefecture (NUTS3) level, on which the model was trained with the mortality and temperature timeseries, and (ii) the census tract level for the city of Athens. The city is divided in 494 census tracts, providing a granularity that is finer than that of the zip code level, and socioeconomic data are available for each tract, thus enabling us to quantify, differentiate and map the SHVI at the city level. The results of this study are presented both at the national level (NUTS3 granularity) and at the city level (census tract granularity).

Next, we identified the heatwave events in the given time series. According to Russo et al. (2015), heatwave events are defined as a period of at least three consecutive days where the maximum temperature (Tmax) exceeds the threshold set at the 90th percentile of daily maximum temperatures, calculated over a 31-day centered window [20]. The heatwave events were defined for each of the ten NUTS3 grouping timeseries and were then used to aggregate the mortality rate of each NUTS3 region (see Figure 1 for the number of heatwave events identified per group). We proceed to identify which socioeconomic variables are the most important ones for predicting the number of deaths during heatwaves. These dependent variables of the analysis were chosen based on their contribution to socioeconomic vulnerability, as per the literature [11], as well as based on data availability. Several such variables were considered, and a final group of eight was selected, based on their contribution in predicting the number of deaths. These eight variables are the following: (i) % elderly persons (people older than 65 years old); (ii) % retirees; (iii) % people living in houses built before 1980; (iv) % of people living alone; (v) % those renting; (vi) % of houses smaller than 60 m²; (vii) % of those unemployed; and (viii) % immigrants from developing countries. Values of these 8 variables, along with their spatial distribution both for the national case study of Greece and the city case study of Athens are shown in Figure 2 and Figure 3, respectively.

When it comes to model selection, tree-based machine learning methods, including ensemble approaches, have shown superior solutions compared to traditional analysis in health research, being potentially applicable to environmental health studies [21]. Specifically for urban resilience, a holistic approach that considers both aspects of environmental and socioeconomic variables aligns with the flexible modeling potential of ensemble decision trees [22]. Many different ensemble decision tree methods have been used in studies in the fields of health, environment and urban resilience, including Random Forest [23], Extreme Gradient Boosting (XGB) [24] and Categorical Boosting (CatBoost) [25]. These tree-based machine learning methods have been used in the field of health for statistical inference, variable selection and the estimation of causal effects. Hu and Li [21] highlight the use of ensemble trees to select important predictors, demonstrating the application of feature importance in statistical analysis.

For the socioeconomic heat vulnerability model, three methods, namely Random Forest [26], XGB [24] and CatBoost [27], were examined and compared to identify the strongest one in predicting the average heatwave mortality rate from the socioeconomic indicators. Due to the relatively small amount of data (coming from only 52 NUTS3 regions), each model was evaluated, with cross validation for a range of hyperparameters (such as the model’s learning rate, max depth and regularization parameters), using the Python package hyperopt [28]. With hyperopt, after defining the hyperparameter space and its distribution, hyperparameters are selected using Bayesian Optimization, by modelling their response surface using probabilistic methods (such as Gaussian Processes and Tree Parzen Estimator). For each hyperparameter selection, a 5-fold cross validation is performed, and the average mean squared error of all 5 folds is used as a comparison metric for each hyperparameter grouping. Hyperopt is then used for 1000 runs for each of the three methods, resulting in XGB being the most accurate method overall, ranking first on all coefficients of determination, namely mean squared error (MSE), mean absolute error (MAE) and R². All regression results are depicted in

Table 2. Regression results on the prediction of average number of daily deaths per 100,000 people during heatwaves for three methods.

Method	MSE	MAE	R2
XGB	2.1437	0.9	0.737
CatBoost	3.724	1.688	0.726
Random Forest	4.882	1.9	0.642

. The hyperparameter values chosen for the XGB model are shown in

Table 3. Hyperparameter values for the selected XGB method.

Hyperparameter Name	Hyperparameter Value
Column sample by tree	0.862
Gamma	1.309
Learning rate	952 × 10−5
Max depth	17
Minimum child weight	2
Number of estimators	1372
Alpha regularization	0.2086
Lambda regularization	1.007
Subsample	0.96

.

XGB’s capability to provide feature importance metrics was utilized to construct the composite index of socioeconomic heat vulnerability, as shown in Equation (1). The index was formulated as a weighted average, where each variable’s contribution was multiplied by its respective feature importance value. This method offers a percentage-based index, aligning with the structure of the input variables.

$$SHV{I}_j={\sum }_{i=0}^n{x}_i\ast v_i, \forall j in C,$$

(1)

where x is the feature importance value of feature i, and v is the value of the feature i, for each j region in the C case study area. The SHVI ranges from 0 to 1, with 0 being the least possible vulnerable area, with no vulnerable population, and 1 the most possible vulnerable area, with the entire population having all possible vulnerabilities. In this specific case, a value of 1 is unattainable, as some variables are mutually exclusive (unemployed and retired), but depending on the feature importance values of each variable, the maximum possible SHVI value will be 1 − min(x_unemployed,x_retired) or 0.986.

Further quantifying the link between the SHVI, temperature and the mortality risk is difficult due to the lack of data for mortality by each socioeconomic variable. However, the Eurostat mortality dataset is separated by age group, allowing us to further analyze the connection between age, temperature and the mortality risk. In order to identify the breakpoints between temperature increases and the increase in the mortality rate, we use a piecewise regression. It is a statistical technique used to model different phases within data by fitting multiple lines, each representing a distinct segment of the data. It is particularly useful when the relationship between variables changes at certain points, known as “breakpoints”. Toms and Lesperance [29] demonstrated the use of piecewise regression to model ecological thresholds. They used breakpoints in their analysis to estimate the width of edge effects in understory plant communities, comparing various models to determine the most accurate representation of ecological thresholds [29]. Xiang et al. [30] applied multiple linear regression models, including piecewise regression, to study the impact of environmental visual exposure on pedestrian emotion in the high-density urban areas of Hong Kong [30]. Using piecewise regression on the data from central Athens, we identify the breakpoints where temperature increases correspond to increases in mortality rates for the different age groupings: elderly people (age ≧ 65) and young people (age < 65). Even though this analysis and the breakpoint identification only includes age and excludes all other socioeconomic variables that compose the SHVI, it is a good quantifiable indicator between temperature increases and the mortality risk, especially since age plays the most significant role out of all socioeconomic variables in the SHVI calculation. A step-by-step flowchart of the overall methodology is shown in Figure 4. All data relevant for this analysis can be found in the ARSINOE project, Zenodo community [31].

3. Results

The results from the piecewise regression are shown in

Table 4. Piecewise regression data—young opulation.

No. Observations			2983
No. Model Parameters			4
Degrees of Freedom			2979
Res. Sum of Squares			1.29253 × 105
Total Sum of Squares			1.55617 × 105
R Squared			0.169419
Adjusted R Squared			0.168303
Converged			True
	Estimate	Std Err	t-value	p > |t|	95% CI (LL)	95% CI (UL)
Constant	31.2981	0.395	79.296	0	30.524	32.072
Alpha1	0.23707	0.0159	14.957	7.94 × 10−49	0.206	0.26815
Beta	3.99292	0.387	10.326	-	3.2347	4.7511
Breakpoint	37.3017	0.253	-	-	36.806	37.797
These alphas (gradients of segments) are estimated from betas (change in gradient)
Alpha2	4.23	0.386	10.948	2.23 × 10−27	3.4724	4.9875

CI: Confidence interval; LL: Lower level; UL: Upper level.

and

Table 5. Piecewise regression data–Elderly Population.

No. Observations			2983
No. Model Parameters			6
Degrees of Freedom			2977
Res. Sum of Squares			1.85371 × 106
Total Sum of Squares			2.65208 × 106
R Squared			0.301034
Adjusted R Squared			0.299624
Converged			True
	Estimate	Std Err	t-value	p > |t|	95% CI (LL)	95% CI (UL)
Constant	274.319	2.33	117.8	0	269.75	278.89
Alpha1	−3.34262	0.121	−27.629	8.9 × 10−150	−3.5798	−3.1054
Beta1	5.68021	0.573	9.9095	-	4.5563	6.8041
Beta2	3.92391	0.759	5.1673	-	2.4349	5.4129
Breakpoint1	26.8632	0.418	-	-	26.043	27.683
Breakpoint2	32.9795	0.649	-	-	31.707	34.252
These alphas (gradients of segments) are estimated from betas (change in gradient)
Alpha2	2.33759	0.56	4.1721	3.1 × 10−5	1.239	3.4362
Alpha3	6.2615	0.513	12.216	1.6 × 10−33	5.2565	7.2665

CI: Confidence interval; LL: Lower level; UL: Upper level.

and in Figure 5. In each table (young and elderly population, respectively), we see the breakpoints, measured in °C—there are two breakpoints for the elderly and one for the young population. The first elderly breakpoint corresponds to cold temperatures and is not included in our analysis. The second breakpoint is located at 32.9795 °C, ranging from 31.707 to 34.252 for the 95% confidence interval (CI). The single breakpoint for the young is located at 37.3017 °C (36.806 to 37.797, 95% CI). We can see that the mortality rate breakpoints greatly differ between the elderly and the young population of Athens, with a mean difference 4.32 °C (2.55 to 6.09, 95% CI). From this, we can conclude that being elderly corresponds to an increase in the mortality rate at temperatures 4.32 degrees lower than if those experience by younger individuals, showing a strong correlation between age, temperature increases and the mortality risk. While the R² scores of the two regressions are low (0.169 and 0.30), meaning that temperature and age do not adequately capture all the variance in mortality rates, the breakpoints identified are statistically significant (p scores of ~0). While age and the mortality risk are always highly correlated, temperature extremes demonstrably exacerbate the mortality rate. However, due to lack of data on mortality rates by socioeconomic category, we utilize the SHVI to demonstrate the extreme heat socioeconomic vulnerability risk in Greece and Athens. It is important to note that the R² scores between the XGB model (0.737) and the piecewise regression models (0.169 and 0.30 for young and elderly populations, respectively) differ substantially due to the models used and the ability of ensemble boosted tree-based models to better capture the nonlinearity and collinearity in the dataset. While the piecewise regression models here are also useful for visually displaying the results, the XGB model is overall more robust and, as such, is the model used to produce the socioeconomic heat vulnerability index (SHVI).

For the calculation of the SHVI, as per Equation (1), we need the feature importance values. According to the XGB analysis, the feature importance results from XGB are shown in

Table 6. Feature importance values from the resulting XGB model.

Feature Name	Feature Importance Value
% Elderly	0.455
% Retired	0.245
% Living in houses built before 1980	0.130
% Living alone	0.063
% Renting	0.058
% Living in houses smaller than 60 m2	0.023
% Unemployed	0.014
% Immigrants from developing countries	0.010

. Overall, the most important variables for the model were the percentage of the population that is elderly and the percentage of the population that is retired (45.5% and 24.4%, respectively). This is expected for a dependent variable such as mortality. However, age was not the only significant variable used in the model’s predictions. The third most important variable, houses built before 1980, had a notable feature importance of 13%, followed by the percentage of people living alone and the percentage of people renting their homes (6.3% and 5.8%, respectively). The percentage of the population that is unemployed, immigrants from developing countries and people living in small houses (<60 m²) were almost insignificant in the overall model’s predictive ability (1.4%, 1% and 2.3%, respectively). Based on the feature importance values, the SHVI is then calculated as shown in Equation (1) for each NUTS3 region in Greece and each census tract area in Athens, as shown in Figure 6 and Figure 7, respectively. In Figure 6, we also include in part (b) a bivariant graph indicating the overall exposure of the population to extreme heat, by plotting both vulnerability and the population of each prefecture. In the latter plot, dark-blue areas signify both high population and high socioeconomic vulnerability scores. Areas that appear with the highest vulnerability are not necessarily colored blue if they are not highly populated as well. On the other hand, areas with medium high vulnerability scores appear blue due to the high exposure that results from the combination of vulnerability and population. In Figure 7, part (b) is a 3-D bivariant plot, in which the height of each census tract column corresponds to the population, and the color of the column corresponds to the SHVI score. In Athens, the highest columns appear in the center, with a medium high SHVI (orange).

In addition to the NUTS3 level analysis for the national case study of Greece, we group prefectures by administrative regions (NUTS2) to see the overall socioeconomic vulnerability to heat by region. This is shown in Figure 8, where the SHVI by prefecture is grouped at the NUTS2 level. The highest SHVI value at the NUTS3 level is found in the NUTS2 region EL52 (Kentriki Makedonia). The second largest urban center in Greece, the city of Thessaloniki, is found in this region (EL523), with a large percentage of mainly old houses and a large number of immigrants found in EL526.

When plotting all SHVIs for the city of Athens (one value per census tract) as a function of distance from the center of the city, we see that there is a strong correlation between vulnerability and distance, which is the highest at the center of the city, and falls to lower values as we move away from the center. This is shown in Figure 9.

4. Discussion

Following up on the presentation of our results, the importance of scale should be emphasized, i.e., the difference between viewing a region overall and examining it in depth. When examining the vulnerability of Greece as a whole, we see that the SHVI ranges between [16.4–29.13%], with a mean of 22.48% (a population-weighted mean of 21.63%) and a standard deviation of 2.94. For an indicator that ranges from 0 to 100%, this provides an overall picture of the socioeconomic vulnerability of Greece that might be considered medium to good. However, when examining the vulnerability of the city of Athens and each of its census tracts, as shown in Figure 7, we see a different picture, with an SHVI ranging from [14.3–42.32%], with a mean of 32.64% (a population-weighted mean of 32.98%) and a standard deviation of 5.79. Overall, the city of Athens accounts for around 657,000 people, out of the 1,029,000 people in the central Athens’ sector NUTS3 region (NUTS3 code EL303, shown in Figure 8), with a mean SHVI 8.97% higher than that of the NUTS3 region’s SHVI. Additionally, the maximum SHVI in the city center reaches 42.32%. For reference, with the current SHVI calculation, a region consisting entirely of a population over 65 years old would achieve an SHVI score of 45.5% with regard to heatwave mortality. This disparity shows the concentration of vulnerable people in the city of Athens and the importance of examining an urban environment in depth, instead of the aggregated socioeconomic indices of the entire region or country.

It is important to note that the socioeconomic features used in the analysis (as shown in Table 6) are correlated with each other, which is normally avoided, especially when using linear models. However, due to the nature of data availability in socioeconomic variables, especially high-resolution data such as the data used in this article, and the highly correlated nature of socioeconomic variables overall, it is impossible to avoid some collinearity. This is why we derived the socioeconomic heat vulnerability index (SHVI) from the XGB model, which, as a boosted ensemble tree model, can handle nonlinear and collinear relationships between the variables better than other linear models.

Regarding the results of Figure 9, this is mainly due to the fact that in the city center buildings are older, which increases the socioeconomic vulnerability of the residents, as shown in Table 6. Of course, while this may be the case in older historical cities such as Athens [32], it might not be the case for every urban center, since other cities might have different concentrations of vulnerable populations (such as elderly people), less urban fabric density and/or a smaller building age differential.

5. Conclusions

The heat vulnerability index developed in this article uses solely socioeconomic data to record, quantify and locate the geographical variability of vulnerability at both the national and the city levels for Greece and Athens, respectively. This variability is observed due to differences in the socioeconomic characteristics at the regional and local levels, as well as differences in demographics, living standards and other cultural characteristics. The highest vulnerabilities were spotted in urban areas with a high density of older houses and high proportions of elderly residents, immigrants from developing countries and lower income households. This analysis can help the nation and the city prioritize where the citizen needs are more urgent in order to plan interventions, or adaptation measures, such as cooling stations, green shaded areas, neighborhood watches for the elderly and the people living alone, etc. Identifying vulnerabilities can better prepare authorities and can help streamline efforts towards adaptation and mitigation measures to address the effect of heat on public health. In conclusion, the following policy and practice recommendations are natural outcomes of this work: (i) socioeconomic factors are critical to defining population vulnerabilities to heatwaves; (ii) being elderly, unemployed, an immigrant and living in an old house increases vulnerability; and (iii) prevention measures should include house retrofitting and the training/education of the vulnerable populations.