Open AccessArticle

A Novel Urban Heat Vulnerability Analysis: Integrating Machine Learning and Remote Sensing for Enhanced Insights

Fei Li

¹,

Tan Yigitcanlar

^1,*

Madhav Nepal

¹,

Kien Nguyen Thanh

and

Fatih Dur

City 4.0 Lab, School of Architecture and Built Environment, Faculty of Engineering, Queensland University of Technology, 2 George Street, Brisbane, QLD 4000, Australia

School of Electrical Engineering and Robotics, Faculty of Engineering, Queensland University of Technology, 2 George Street, Brisbane, QLD 4000, Australia

Author to whom correspondence should be addressed.

Remote Sens. 2024, 16(16), 3032; https://doi.org/10.3390/rs16163032

Submission received: 20 June 2024 / Revised: 12 August 2024 / Accepted: 17 August 2024 / Published: 18 August 2024

(This article belongs to the Special Issue Estimating the Ecological Services of Urban Green Infrastructures Using Remote Sensing)

Download

Browse Figures

Versions Notes

Abstract

Rapid urbanization and climate change exacerbate the urban heat island effect, increasing the vulnerability of urban residents to extreme heat. Although many studies have assessed urban heat vulnerability, there is a significant lack of standardized criteria and references for selecting indicators, building models, and validating those models. Many existing approaches do not adequately meet urban planning needs due to insufficient spatial resolution, temporal coverage, and accuracy. To address this gap, this paper introduces the U-HEAT framework, a conceptual model for analyzing urban heat vulnerability. The primary objective is to outline the theoretical foundations and potential applications of U-HEAT, emphasizing its conceptual nature. This framework integrates machine learning (ML) with remote sensing (RS) to identify urban heat vulnerability at both long-term and detailed levels. It combines retrospective and forward-looking mapping for continuous monitoring and assessment, providing essential data for developing comprehensive strategies. With its active learning capacity, U-HEAT enables model refinement and the evaluation of policy impacts. The framework presented in this paper offers a standardized and sustainable approach, aiming to enhance practical analysis tools. It highlights the importance of interdisciplinary research in bolstering urban resilience and stresses the need for sustainable urban ecosystems capable of addressing the complex challenges posed by climate change and increased urban heat. This study provides valuable insights for researchers, urban administrators, and planners to effectively combat urban heat challenges.

Keywords:

urban heat island; urban heat vulnerability; remote sensing; machine learning; artificial intelligence; urban sustainability; sustainable urban development

1. Introduction and Background

1.1. Introduction

The profound impacts of climate change are increasingly apparent. In the face of rapid urbanization, cities are grappling with an intensified urban heat island effect, a climate change phenomenon that elevates temperatures in urban areas relative to their rural counterparts [1,2]. Such temperature spikes exacerbate the risks of extreme heat, contributing to a surge in heat-related illnesses and fatalities. In Australia, for instance, extreme heat has emerged as the leading cause of weather-related injuries and deaths, with a startling 7104 hospitalizations and 293 fatalities over the span of a decade, as reported by the Australian Institute of Health and Welfare [3].

Historically, the nuanced aspects of urban heat vulnerability received limited research attention [4,5]. The introduction of the Heat Vulnerability Index, by Reid et al. [6], marked a turning point, catalyzing a wave of studies that utilize demographics, socio-economic status, health conditions, and environmental data to assess heat vulnerability [7,8,9]. While index models incorporating these indicators have gained popularity, they often suffer from subjective judgments and potential inaccuracies. This is largely due to the absence of universal criteria and references for selecting indicators and developing methodologies, especially in the diverse contexts of various urban environments [10,11,12,13]. Moreover, the reliance on static data in these models poses a limitation, as they may not fully capture the dynamic and evolving nature of urban vulnerabilities to heat [14].

1.2. The Current State of Urban Heat Vulnerability Assessment

The increasing urbanization under changing global climate conditions has made the study of urban heat vulnerability critically important [15,16]. As cities expand and weather patterns change, the well-being of urban residents is increasingly at risk. It is essential to understand the complexities of vulnerability to mitigate the issue [7,8,17]. Despite concerted efforts, achieving methodological standardization in this area remains a challenge [14]. Researchers are grappling with multiple factors contributing to urban heat vulnerability in their pursuit of a comprehensive understanding [18,19].

In the realm of urban heat vulnerability indicator selection, the absence of universal standards poses significant challenges. The indicators are often influenced by local contexts and researcher preferences [11,12,20]. While this diversity reflects the complexity of the issue, it complicates the comparison of studies and the establishment of best practices [10,13]. Furthermore, finding fitting data proxies for these indicators is challenging. Many studies build their own indicators and data systems, but the reliability varies, leading to inconsistencies in study outcomes [21,22,23,24].

Spatial and temporal considerations add further complexity to these assessments. Many studies use datasets from census tracts and administrative areas [9,25,26], but this approach has been criticized for potential biases. The Modifiable Areal Unit Problem (MAUP) is a significant issue, leading some researchers to adopt grid-scale spatial analyses for improved accuracy [27,28]. On the temporal front, there is a predominance of short-term, post-1990s assessments, highlighting a gap in longitudinal research that could provide insights into evolving vulnerabilities [14].

The development of heat vulnerability indexing models has been a popular approach among researchers. These models often focus on specific aspects like biophysical or social vulnerabilities. However, they overlook the complex interactions between the natural/built environment, individual biophysical and psychological conditions, and socio-economic status, underscoring the need for further exploration. Integrative methods, including composite heat vulnerability indices that combine biophysical and socio-demographic indicators, have been proposed. However, the integration of these indicators brings up issues related to potential cumulative effects [15,29,30]. In these models, the method of assigning weights is vital, with popular approaches including Equal Weighting (EW) and Principal Component Analysis (PCA). Using PCA, however, presents unique difficulties, particularly in terms of naming components and maintaining stringent data accuracy [31,32].

Result validation in urban heat vulnerability research is paramount. Unfortunately, only a few studies incorporate validation procedures, raising questions about their authenticity and credibility. Studies that perform validations often use heat-related health outcomes for validation data, including emergency hospital admissions, as well as morbidity and mortality rates. However, the use of these outcomes as validation variables presents its own challenges, for example, the unavailability of direct data and reliance on proxies, which may introduce inaccuracies [33,34]. Additionally, there is an emerging call to consider the mental health impacts of extreme heat, an aspect often neglected in current studies [35,36].

In summary, while the research on urban heat vulnerability has considerably advanced, numerous challenges remain. Firstly, a lack of a rigorous approach to specifying relevant indicators for heat vulnerability assessment leads to inconsistencies in the study outcomes and their comparisons, as well as the reliability of the results. Secondly, an adaptive, dynamic modelling approach designed to sufficiently capture both rich historical data and real-time urban planning/environment-related data for predictive urban heat vulnerability analysis to address the rapidly changing urban environment for urban resilience and sustainability is needed. Lastly, multifaceted validation methods should be the essential element of urban heat vulnerability modelling to ensure its reliability and accuracy. The integrative framework proposed in this study is designed to address these challenges.

Machine learning (ML) and remote sensing (RS) have shown great promise in urban studies for applications such as land cover/use classification and disaster risk management [37,38]. Their potential in urban heat vulnerability assessment, however, remains underutilized. RS, with its unparalleled spatial and spectral data from platforms like Landsat and MODIS, is invaluable for land use classification and heat mapping [39,40]. Complementing this, ML’s prowess in data analytics can provide deeper insights into urban environments [41,42]. Although there has been some incorporation of ML in urban heat vulnerability research [21,43], the focus has predominantly been on data processing and preparation, with less emphasis on urban heat vulnerability identification and its in-depth analysis.

1.3. The Emerging Trend of Machine Learning and Remote Sensing Integration

Urban studies are witnessing a remarkable transformation, marked by the fusion of ML and RS. This integration is largely powered by ML algorithms like Support Vector Machines (SVMs), Random Forests (RFs), and Convolutional Neural Networks (CNNs) [44]. Each of these methods has its unique strengths: SVMs are adept at classification tasks; RFs excel in decision-making processes; and CNNs are unparalleled in processing visual data. These algorithms are particularly effective in urban applications. CNNs, for example, have shown outstanding performance in analyzing satellite imagery for land use classification [37,45]. The versatility of RFs and SVMs extends across diverse ML scenarios [46,47,48,49], with their adoption driven by reliability and tool accessibility [50].

In the realm of urban studies, supervised learning techniques are prominent, primarily due to their effectiveness with labelled datasets. These supervised learning approaches iteratively train models on input–expected output (also known as labels) pairs to learn the function that maps the input to the output, and adapt to new, unseen data—a crucial advantage in urban contexts with abundant labelled data [51]. Conversely, the laborious nature of data labelling has instead steered researchers towards unsupervised learning. These unsupervised learning techniques, which discern the hidden patterns and insights from the given data without pre-set labels, offer alternative methodologies for urban data analysis [25,52]. RS has revolutionized urban studies by offering dynamic, high-resolution images that reveal a range of spatial characteristics. When coupled with ML, these tools surpass traditional/static models. Such integration is essential for precise urban land use classification, monitoring urban expansion, and evaluating environmental influences on urban well-being [53,54,55].

Most recently, Li et al. introduced a pivotal framework to harmonize ML and RS in urban studies [44]. This framework (Figure 1) promotes a more structured approach for comprehensive urban analyses. It emphasizes increased automation in image processing and data analytics, a critical move away from conventional manual methods [25,56], while acknowledging some inherent challenges related to algorithmic biases and the need for precise modelling of parameters [57]. Despite its transformative potential, the integration of ML and RS in urban studies faces some challenges, with the major ones being the biases inherent in algorithms and an ongoing need for clear, well-defined parameters to guide the use of these advanced tools.

In line with the United Nations Sustainable Development Goals (SDGs) established by the United Nations [58], the integration of ML and RS is poised to significantly enhance urban sustainability due to its potential to produce increased accuracy [59,60], real-time monitoring capability [50,61], and adaptability to dynamic urban environments [62,63]. As urban settings rapidly evolve and regenerate, the adaptive and precise capabilities of these integrated technologies will be invaluable for sustainable and resilient urban development, including urban heat vulnerability studies.

Urban heat vulnerability assessment is critical due to rising urbanization and climate change impacts, yet current approaches face significant challenges and limitations. These include the lack of standardized criteria for selecting indicators, leading to inconsistencies and inaccuracies; insufficient spatial resolution and temporal coverage that fail to capture the dynamic nature of urban heat vulnerability; and subjective judgments in indicator selection, which introduce biases. Additionally, while ML and RS have potential, they are underutilized, focusing more on data processing than comprehensive analysis. The validation of results also remains a challenge, with few studies incorporating robust procedures, often relying on proxies. Furthermore, many models do not fully integrate the complex spatial and temporal dynamics of urban environments, including interactions between natural and built environments and socio-economic factors. This study aims to bridge the gaps by integrating ML techniques with RS data.

Building upon our research findings, this paper proposes a novel framework called the ‘Urban Heat vulnErability Analysis Tool (U-HEAT)’. U-HEAT is designed to assess urban heat vulnerability through an innovative conceptual framework that leverages the strengths of both ML and RS. It develops universal criteria for indicator selection, reducing subjectivity and enhancing reliability. By leveraging high-resolution RS data and advanced ML techniques, U-HEAT provides detailed, grid-level evaluations, capturing the spatial and temporal dynamics of urban heat vulnerability. This approach allows for the precise identification of vulnerable areas and improves temporal coverage through historical and predictive mapping. U-HEAT’s active learning capabilities ensure continuous improvement in predictive accuracy, and its robust validation framework using heat-related health outcomes enhances result reliability. This innovative framework equips urban planners and policymakers with actionable insights for identifying vulnerable areas, developing mitigation strategies, and enhancing urban resilience and sustainability.

The primary objective of this study is to outline the theoretical underpinnings and potential applications of U-HEAT, emphasizing its conceptual nature. The authors intend for future research to empirically validate and analyze the framework’s performance. This framework aims to provide actionable tools for policymakers and urban planners to identify vulnerable localities and communities. By focusing on the creation of well-informed, precise, and effective mitigation strategies, U-HEAT aspires to play a pivotal role in enhancing the well-being and future of urban populations. The integration of ML’s advanced data analytic capabilities with RS’s high-resolution spatial data promises to improve urban heat vulnerability assessments, setting a solid foundation for future empirical studies and practical implementations [64,65,66].

1.4. The Focus of This Study

This study presents significant contributions to addressing urban heat vulnerability through the establishment of the U-HEAT framework, characterized by four major advancements. First, it contributes by developing a comprehensive set of universal criteria, creating a benchmark for the selection and categorization of indicators in this field. Second, this study demonstrates the effective application of ML and RS in conducting comprehensive and precise long-term urban heat assessments. This integration allows for detailed evaluations at a granular, grid-level scale, ensuring that the spatial and temporal dynamics of urban heat vulnerability are accurately captured and analyzed. Third, it introduces an innovative predictive approach, filling a crucial gap in existing methodologies, and enabling the forecasting of future trends in urban heat vulnerability. Finally, this study offers a robust, enduring, and sustainable framework, specifically designed for the continuous, precise, and focused monitoring and management of the challenges associated with urban heat vulnerability.

Following the introduction in Section 1, Section 2 outlines the U-HEAT methodology, covering its components, features, and contributions. Section 3 details the approach for mapping and managing urban heat vulnerability. Section 4 discusses challenges, future applications, limitations, and broader implications. Finally, Section 5 concludes by emphasizing the necessity and innovation of U-HEAT in addressing climate issues and advancing urban sustainability and resilience.

2. Research Design

An overarching methodology proposed in this study consists of a structured, three-step process. First and foremost, this study synthesizes results from two systematic reviews conducted in accordance with PRISMA guidelines. The first review critically evaluates methodologies used in prior research on urban heat vulnerability, identifying the key areas needing refinement [14]. The second review extends this analysis to the combined application of ML and RS in urban studies [44]. These reviews have laid the foundation for a comprehensive understanding of the current practices and emerging trends in the field.

Building upon these reviews, the second part of this study introduces U-HEAT for assessing urban heat vulnerability. U-HEAT integrates ML with RS techniques to achieve the precise, grid-scale mapping of urban heat impacts. The framework is unique due to its loop structure, guiding users from the selection of indicators to the continuous monitoring of urban heat vulnerability evolutions, a critical effort towards standardizing urban heat vulnerability assessments. Detailed descriptions of the U-HEAT components, their interactions, and the specific ML techniques employed are described below, including U-HEAT’s advancements over the existing methodologies.

Lastly, a structured workshop was conducted with domain experts to validate the coherence and alignment of the proposed U-HEAT with established theories and best practices. This process was essential in guiding and designing the presentations and discussions for the subsequent workshops. Expert reviews involve professionals in urban planning, environmental sustainability, climate change, and ML offering practical insights and diverse perspectives on the relevance and applicability of U-HEAT in real-world scenarios and its refinement.

The Urban Heat Vulnerability Analysis Framework

The primary goal of U-HEAT is to mitigate urban heat vulnerability by providing a structured approach and guidance for future research in the area, as well as a tool for heat vulnerability prediction, intervention, and policy making. The proposed framework is presented in Figure 2.

A key aspect of U-HEAT is its structured, component-based approach, which is crucial for mapping urban heat vulnerability effectively. This approach can be broken down into four main components:

Component 1—Indicator and Data Selection: The first component of U-HEAT involves a meticulous process of selecting relevant indicators and gathering proxy data. This phase is critical as it establishes the foundation for accurate vulnerability assessment. By analyzing various socio-economic, environmental, and health-related indicators, U-HEAT ensures a comprehensive understanding of the factors contributing to urban heat vulnerability. The integration of RS data enhances this process by providing high-resolution spatial and temporal data that captures the dynamic nature of urban environments. This combination of indicators and RS data allows for a nuanced analysis, identifying vulnerable populations and areas with precision. The rigorous selection and collection of data in this phase are essential for creating a robust and reliable vulnerability assessment framework, which is crucial for informing urban planning and policy decisions.

Component 2—Urban Heat Vulnerability Mapping: Mapping is a core component of U-HEAT, divided into two crucial phases:

Historical Mapping: Historical mapping in U-HEAT utilizes advanced ML and RS techniques to reconstruct detailed heat maps from diverse data sources. This phase is pivotal as it provides a comprehensive understanding of past and present heat vulnerabilities, allowing for the identification of long-term trends and spatial patterns in urban heat exposure. By fusing various socio-economic, environmental, and health-related data with high-resolution RS imagery, U-HEAT generates a nuanced portrayal of how urban heat vulnerability has evolved over time. This retrospective analysis is essential for establishing a baseline, understanding the historical context of current vulnerabilities, and identifying persistent hotspots that require targeted interventions. The detailed historical maps produced in this phase serve as a foundational reference for subsequent predictive mapping efforts, ensuring that future projections are grounded in a robust empirical understanding of past conditions.
Predictive Mapping: Building on the insights gained from historical mapping, predictive mapping in U-HEAT integrates urban planning data to forecast future trends and distributions of urban heat vulnerability. This phase leverages the predictive power of ML models to simulate how urban heat patterns might evolve under various scenarios, such as climate change, population growth, and urban development. By incorporating forward-looking data, such as planned infrastructure projects and anticipated demographic shifts, U-HEAT can generate projections that inform proactive urban planning and policymaking. The predictive mapping capability is crucial for identifying emerging areas of concern and guiding the implementation of preventative measures. This forward-thinking approach ensures that cities can anticipate and mitigate future heat risks, enhancing their resilience and adaptability to climate change. Predictive mapping transforms U-HEAT from a reactive tool into a proactive planning resource, enabling urban planners to design cities that are better equipped to handle the challenges of rising temperatures.

Component 3—Strategy Recommendation: The strategy recommendation component leverages U-HEAT’s ability to identify vulnerable areas and populations across different time frames. By simulating various scenarios, such as climate change impacts and potential interventions, U-HEAT can predict how urban heat vulnerability will evolve. This feature allows for a comparative analysis of different strategies, helping urban planners and policymakers to identify the most effective heat mitigation and adaptation measures. The ability to adjust data inputs and simulate hypothetical scenarios provides a dynamic tool for exploring various intervention strategies, making U-HEAT an invaluable resource for informed decision-making. This component ensures that strategies are evidence-based and tailored to address the specific needs of vulnerable populations, promoting effective urban heat management.

Component 4—Continuous Monitoring and Updating: To maintain its relevance and accuracy, U-HEAT incorporates a dynamic model that continuously adapts to new data. This ongoing monitoring and updating process is critical for ensuring that the framework remains responsive to evolving urban heat challenges. By integrating fresh data, U-HEAT can refine its models and improve its predictive accuracy over time. This adaptability makes U-HEAT a sustainable tool for long-term use, capable of providing up-to-date insights into urban heat vulnerability. Continuous monitoring ensures that urban planners and policymakers have access to the latest information, enabling them to respond proactively to emerging heat risks. This component underscores the importance of a flexible and iterative approach to urban heat management, ensuring that strategies remain effective in the face of changing environmental conditions.

The U-HEAT framework’s active learning capabilities are integral to its effectiveness and adaptability in managing urban heat challenges. These capabilities are designed to operate within an iterative loop, where continuous monitoring and feedback play crucial roles. As new data becomes available, U-HEAT actively learns from it, refining its indicator selection, enhancing mapping accuracy, and updating strategies accordingly. This means that U-HEAT does not require complete system re-training each time new data are introduced. Instead, the model incrementally improves its predictive accuracy, becoming more precise and reliable over time.

Active learning within U-HEAT ensures that the framework remains dynamic and responsive to changing conditions. This approach allows the model to adapt to new patterns and emerging trends in urban heat vulnerability, making it a continually evolving tool. By incorporating fresh data, U-HEAT can adjust its assessments and recommendations in real-time, providing urban planners and policymakers with the most up-to-date insights. This capability is vital for addressing the unpredictable and often rapid changes associated with urban heat variations, driven by factors like climate change and urban development.

The cyclical process facilitated by U-HEAT’s active learning involves constant refinement and improvement. As the model processes new data, it fine-tunes its parameters and recalibrates its predictions, ensuring that the framework remains accurate and relevant. This ongoing adaptation enhances U-HEAT’s ability to identify and mitigate urban heat risks effectively. By maintaining an up-to-date and accurate understanding of urban heat vulnerabilities, U-HEAT supports the development of proactive and informed strategies, helping cities to better prepare for and respond to the challenges posed by rising temperatures. In summary, the active learning capabilities of U-HEAT provide a robust mechanism for continuous improvement and adaptability. This ensures that U-HEAT is not just a static tool for current analysis, but a forward-thinking solution that evolves to meet future urban heat challenges, making it an essential resource for sustainable urban planning and resilience-building.

The U-HEAT framework is innovative in its granular, grid-scale approach to evaluating long-term urban heat vulnerability. The U-HEAT framework achieves granular, grid-scale urban heat vulnerability mapping by meticulously aggregating and analyzing data using advanced ML techniques in conjunction with high-resolution RS data. This integration allows U-HEAT to segment urban areas into fine, grid-scale segments, incorporating RS data such as Land Surface Temperature (LST), land use patterns, population density, the Normalized Difference Vegetation Index (NDVI), and the Normalized Difference Built-up Index (NDBI). Each grid cell is analyzed individually for a detailed assessment of urban heat vulnerability, helping solve the MAUP problem (Figure 3). ML enhances this process by intelligently processing and interpreting the RS data, identifying complex patterns such as the correlation between built-up areas and increased heat vulnerability. This combination of ML and RS not only facilitates the processing of vast datasets, but also enables nuanced analysis at a granular level, crucial for pinpointing specific hotspots of heat vulnerability and facilitating targeted intervention strategies. It harnesses the power of ML and RS to offer the following advantages:

Relevance: Rooted in established frameworks and empirical research, U-HEAT’s approach to selecting indicators and gathering data is both relevant and representative.
Precision: By transitioning from broad statistical areas to a more detailed grid-scale, U-HEAT provides a finer-grained and accurate depiction of urban heat vulnerability, benefitting from the integration of ML and RS.
Comprehensiveness: U-HEAT not only maps historical data, but also predicts future urban heat trends, resulting in spatially detailed and temporally extensive outcomes.
Sustainability: The U-HEAT framework’s ability to recommend mitigation strategies, adapt to new data, and provide ongoing monitoring highlights its sustainability.

In sum, this study aims to establish U-HEAT as a benchmark framework for addressing urban heat vulnerability, with the following four distinct objectives and potential contributions:

Criteria Development: To formulate a universal set of criteria for the selection and categorization of indicators, establishing a reference framework.
Feasibility Demonstration: To showcase the practicality of conducting long-term, grid-scale, and precise assessments by integrating ML and RS technologies.
Predictive Methodology: To bridge the existing gap in predictive methods by introducing an innovative approach for forecasting urban heat vulnerability trends in future decades.
Framework Proposal: To offer a robust, enduring, and sustainable framework for the continuous, accurate, and focused monitoring and management of urban heat vulnerability challenges.

Figure 4 provides a detailed workflow of U-HEAT’s implementation, with various steps involved within each component. Section 4 delves into each of these components.

3. Integrated Urban Heat Vulnerability Analysis with Machine Learning and Remote Sensing

As mentioned before, this study has developed a systematic framework for urban heat analysis by leveraging ML and RS. This framework has five key components, each of which is discussed below.

3.1. Indicators and Data Selection

3.1.1. Popular Reference Frameworks

Based on the findings from our recent study [14] on current urban heat vulnerability assessment research, the selection of heat-related indicators and the determination of data proxies are crucial for quantitative assessment of urban heat vulnerability [13,67]. Typically, indicators are chosen based on two popular disaster evaluation frameworks, the Population Vulnerability Framework and the Risk Tringle Framework.

The Population Vulnerability Framework was proposed by The Intergovernmental Panel on Climate Change (IPCC) [68]. The IPCC emphasizes three key elements when addressing vulnerability: exposure, sensitivity, and adaptive capacity. Exposure to extreme heat, whether in daily life or at work, is a primary cause of heat vulnerability [42,69]. Differences in individual sensitivities to extreme heat—often determined by demographics and socio-economic status—mean that people experience varying levels of heat vulnerability [48,70]. As a result, individuals pursue mitigation solutions to reduce the risk of exposure to extreme heat, such as seeking green spaces, using air conditioning, and accessing medical relief. The availability of these heat-relief services and facilities constitutes adaptive capacity [5,71].

The other widely used theoretical framework is the Risk Triangle Framework, proposed by Crichton [72]. This framework also encompasses three elements: hazard, exposure, and vulnerability, with vulnerability viewed as an integral component [73,74]. While ‘exposure’ bears some similarities in definition across both frameworks, in the Risk Triangle Framework it is often represented by population density. Conversely, ‘hazard’ in this framework closely aligns with the concept of ‘exposure’ in the Population Vulnerability Framework, representing the spatio-temporal distribution of harmful heat [26,75].

As depicted in Figure 5, despite the different descriptions of their underlying elements, both indicator selection frameworks exhibit some overlap. This similarity often results in confusion during the indicator selection process. Cheng et al. [9] and Li et al. (2022) [14] noted that the same indicators are frequently interpreted as different elements due to the lack of generic criteria. Additionally, even within studies adopting the same theoretical framework, interpretations of identical indicators can differ. To eliminate this confusion, this paper has organized the indicators used in current studies and recategorized them more distinctly. These categories include demographic and socio-economic characteristics, health conditions, and environmental factors (see Table 1). Table 1 also shows the descriptions and data sources of the indicators.

3.1.2. Indicator Collection and Categorization

This study gathered all heat-related indicators from the findings from our research [14]. To ensure a broad acceptance and clear understanding of the chosen indicators, this study classified them into three primary categories, focusing on those that are employed most in the field. These categories were formulated with a specific focus on human health and well-being in response to harmful thermal conditions. This approach to categorization marks a significant improvement over previous methods, which were often less explicit. The framework for selecting indicators enables us to pinpoint specific areas of daily life that are more likely to be affected by heat-related issues, thereby guiding the implementation of preventive strategies against hazardous heat conditions. Each category of indicators reflects, either directly or indirectly, the vulnerability of urban populations to heat exposure. This vulnerability is assessed through various lenses, including physical characteristics, physiological responses, psychological factors, and environmental conditions. For instance, while individuals over 65 years old are generally at a higher risk in extreme heat situations [76,77], access to public health services and cooling facilities can significantly reduce this risk [38,78]. Such external support systems can effectively offset physiological constraints.

Validation indicators, which are as crucial as assessment indicators, ensure the accuracy and reliability of methodologies in research. The systematic review found that less than 30% of studies included a validation step, often hindered by data unavailability. This lack of validation can significantly undermine the credibility of the results [47,71]. To mitigate this issue, this paper compiled a list of validation indicators frequently used in previous studies, specifically focusing on urban heat vulnerability assessments. The analysis revealed that data on heat-related morbidity and mortality are regularly employed in these validation processes [79,80]. Importantly, Reid et al. established a clear positive correlation between chronic diseases, such as cardiovascular, respiratory conditions, and diabetes, and heat-related mortality rates [81]. This insight led the following research to focus on using heat-related mortality data for validation purposes. This paper further categorized diseases contributing to mortality as either directly or indirectly related to heat exposure (see Table 2). This categorization allows for a more nuanced understanding of the impacts of heat on health. The morbidity and mortality data from these diseases will be integral in the ML model evaluation, providing robust assessment results of urban heat vulnerability.

The indicator selection process in the U-HEAT framework is carefully detailed to ensure a thorough understanding of urban heat vulnerability. Indicators are prioritized based on their relevance, representativeness, data quality, and alignment with established theoretical frameworks. Demographic factors like age and economic status are critical, as they affect individuals’ sensitivity and adaptive capacity to heat stress. Health conditions and access to medical resources are also key, influencing an individual’s ability to cope with extreme heat. Indicators are chosen based on the availability of reliable data, with socio-demographic data sourced from census bureaus and environmental data from high-resolution satellite imagery.

The selected indicators align with frameworks like the Population Vulnerability and Risk Triangle Frameworks, ensuring consistency with other studies. This approach allows for a nuanced analysis of urban heat vulnerability, incorporating socio-demographic, health, and environmental factors. High-resolution remote sensing data provide precise spatial information, while longitudinal data enable the analysis of temporal trends. The rigorous selection and data collection process is crucial for creating a reliable vulnerability assessment framework, essential for informing urban planning and policy decisions. This careful process ensures the U-HEAT framework offers a robust assessment of urban heat vulnerability, supporting informed decision-making and effective policy development.

3.1.3. Data Collection and Pre-Processing

Determining the right indicators is the first step of U-HEAT, followed by collecting corresponding data proxies from reliable sources. Data quality, encompassing integrity, accuracy, consistency, and more, is critical for the framework’s operation. Table 1 shows the authentic data sources utilized in the reviewed articles. All socio-demographic data are freely available from census datasets provided by national/local census bureaus. Additionally, for population density data, it is not only available from statistical datasets, but also from some satellite imagery products, such as WorldPop, which offers detailed data at a 100 m resolution, invaluable for scaling urban heat vulnerability from census tracts to grids. Health-related data, crucial for assessment as well as validation, including individual health conditions, public medical infrastructures, and morbidity and mortality statistics, come from public datasets provided by health departments or public well-being organizations and programs. Census datasets also include some information on pre-existing illnesses and disabilities distributions in terms of ages, genders, and regions. To assess accessibility to medical infrastructure, Google Maps offers a promising way to measure the distance to the nearest available medical infrastructure through Point of Interest data. It is also a surprisingly valuable resource in the context of measuring accessibility to cooling spaces.

For environmental factors, temperature (LST), vegetation cover (NDVI), and building density (NDBI) are considered. These metrics are often derived from satellite images, such as Landsat, MODIS (Moderate Resolution Imaging Spectroradiometer), and Sentinel, which offer resolutions up to 30 m, 250 m, and 10 m, respectively. Landsat images provide high-resolution thermal data crucial for studying surface temperature dynamics and identifying urban heat islands. The long historical record of Landsat enhances its capability to analyze trends over extended periods. MODIS offers frequent, high-resolution thermal images essential for capturing diurnal variations in urban heat patterns, helping to understand how heat accumulates and dissipates in urban areas. Sentinel-2 provides data on the NDVI and the NDBI, which are crucial for assessing vegetation cover and built-up areas. These indicators help understand the factors contributing to urban heat vulnerability and validate the model’s predictions. Additionally, Geoscience Australia provides a public 25 m land cover/use dataset for free download, saving time on classification. Digital Earth Australia offers detailed land cover/use information, vital for analyzing how different land uses contribute to urban heat vulnerability. OpenStreetMap provides data on the availability of cooling spaces and medical resources, which are integrated into the analysis to evaluate adaptive capacities. These diverse RS data sources collectively enhance the understanding of urban heat vulnerability and support the comprehensive assessment of environmental factors.

One key environmental data in text format is air temperature, observed by various sites and released by meteorological bureaus and related institutions and organizations. The collected data come in diverse formats, including text, shape, and raster. Standardizing this into a raster grid format is crucial for grid-scale assessment. One potential approach involves inputting statistical and record data into corresponding administrative unit shapes as properties, and then interpolating these into raster grids. Zhang et al. utilized raster land cover data to evenly interpolate census data into urban area grids [8]. This is a good reference based on the assumption of the same distribution of demographic features, but it ignores the spatial heterogeneity of population distribution. Therefore, this paper proposes to involve population density data as another reference for interpolation, which enables the generation of more reasonable raster socio-demographic data.

While the U-HEAT framework comprehensively lists its data sources, it is important to acknowledge potential limitations such as availability, quality, and representativeness, which can impact the accuracy of urban heat vulnerability assessments. A significant issue is the variability in data quality and availability across different regions and time periods. Socio-demographic data, though generally reliable, may suffer from being outdated or inconsistently collected, as census data are typically gathered every five to ten years, potentially missing recent changes. To address this, the framework incorporates supplementary data from more frequently updated sources like WorldPop, which provides higher resolution and current estimates of population density.

Environmental data from satellite imagery, while precise, may face challenges like atmospheric conditions, sensor malfunctions, and temporal resolution inconsistencies. For instance, Landsat offers high spatial resolution, but is limited by a 16-day temporal resolution, which might miss short-term temperature variations. To mitigate this, the framework uses multiple sources, including MODIS for frequent thermal imaging and Sentinel-2 for detailed vegetation and built-up area assessments, ensuring comprehensive environmental data representation.

Health-related data, crucial for validating vulnerability assessments, may encounter issues such as data privacy, limited access, and inconsistencies in records, with reliance on proxies like hospital admissions potentially introducing inaccuracies. The framework addresses this by utilizing well-established correlations, such as those between chronic diseases and heat-related mortality, and integrating health infrastructure data from Google Maps to assess medical service accessibility.

The framework also recognizes the challenge of integrating diverse data formats and employs interpolation methods and machine learning algorithms to standardize and accurately transform data for grid-scale assessment. By using multiple data sources, advanced processing techniques, and continuous updates, the U-HEAT framework effectively mitigates these limitations, ensuring robust and reliable urban heat vulnerability assessments.

3.2. Historical Mapping of Urban Heat Vulnerability

3.2.1. Two Scenarios of Historical Mapping

In the U-HEAT framework, historical mapping involves using past data to analyze and map urban heat vulnerability patterns.

Scenario 1—Spatial and Temporal Extrapolation: The first scenario is training a ML model using data from a selected area spanning over time to identify correlations between urban heat vulnerability measures and inputs from collected data areas. These inputs include multiple factors such as socio-demographic characteristics, health conditions, and environmental factors. Then, this trained model can be applied: (a) spatially to new similar geographical areas that lack direct urban heat vulnerability measures; and (b) temporally to historical time periods when such measures are absent. This approach allows the model to extrapolate and estimate urban heat vulnerability for different areas and times, providing insights into periods and locations not directly observed or measured. This offers a comprehensive historical view of urban heat vulnerability.

Scenario 2—Fine-tuning for a Tailored Area with Limited Data: Training a ML model usually requires a large amount of data, which may not be always available for a specific area. In this scenario of limited data availability for a specific area, a ML model can be first trained using a general dataset (e.g., from other areas). Once trained, the ML model can be further fine-tuned or adapted with the small amount of localized data available in the specific area of interest. This fine-tuning process enables the model to adapt its predictions to the unique conditions of each new area and time frame. For instance, if detailed urban heat vulnerability measures from specific urban sectors for recent years are available, these are integrated into the model, refining its predictive capabilities. This scenario highlights the dynamic and adaptive nature of the U-HEAT framework, enabling it to continually learn, adapt, and improve its accuracy with each new dataset. This provides a more precise and context-specific mapping of urban heat vulnerability across different times and locations.

3.2.2. Challenges and Algorithm Selection for Historical Mapping

Mapping urban heat vulnerability accurately across large areas and extensive historical periods is a multifaceted challenge. This task requires the navigation of the complexities of data granularity, the completeness of historical records, and the dynamic nature of urban environments [21,48]. For fine-scale mapping, high-quality, detailed data is essential to pinpoint minor vulnerability variations within cities. However, longitudinal datasets often suffer from temporal inconsistencies and spatial inhomogeneities due to evolving data collection methodologies, leading to potential uncertainties in historical trend analysis [82]. Additionally, the dynamic nature of cities, shaped by ongoing developments, infrastructural changes, and socio-economic shifts, complicates historical comparisons. As urban landscapes and populations evolve, vulnerability assessments must account for historical shifts in population density, land use, and socio-economic conditions, which influence how people experience and adapt to heat [64,74].

To ensure data accuracy and reliability, several mitigation strategies are employed. First, data preprocessing techniques like normalization and interpolation address temporal inconsistencies and spatial inhomogeneities. Normalization standardizes data from different sources, making them comparable, while interpolation estimates missing data points, enhancing dataset completeness. Additionally, integrating supplementary data sources, such as high-resolution RS imagery and detailed socio-economic data, fills gaps and provides a comprehensive view of urban heat vulnerability. RS data offers insights into land surface temperature and vegetation cover, while socio-economic data highlights population vulnerability factors like income, age, and access to medical facilities. Advanced ML algorithms, such as CNNs and LSTMs, further enhance analysis granularity and accuracy. CNNs effectively extract spatial features from high-resolution RS images, enabling precise urban heat island mapping, while LSTMs excel at capturing temporal patterns, making them ideal for analyzing changes in urban heat vulnerability over time. These algorithms help identify complex patterns and interactions within the data, leading to more accurate predictions and assessments.

In mapping historical urban heat vulnerability, selecting appropriate RS images and ML algorithms is vital. High-resolution satellite imagery, such as from Landsat’s Thermal Infrared Sensor (TIRS) and MODIS, is critical for providing historical thermal data to study surface temperature dynamics and urban heat [83,84]. Using a series of Landsat images can illuminate changes in land use and urban expansion, offering insights into shifting heat vulnerability. Meanwhile, CNNs stand out for their capabilities in managing spatial complexities in RS data and maintaining the hierarchical structure of images [37,48]. For studying temporal patterns, LSTM networks, a type of recurrent neural network (RNN), are effective, capturing the progression of changes over time with precision [85,86,87,88,89].

In the U-HEAT framework, spatial mapping is innovatively approached as a semantic segmentation problem in machine learning. Unlike traditional methods like linear regression, which estimate urban heat vulnerability based on individual locations in satellite images, semantic segmentation considers both individual locations and their neighboring regions. This method enhances the accuracy and consistency of mapping by accounting for the interconnected nature of urban environments and the influence of adjacent areas on heat vulnerability. U-HEAT’s flexible design allows for the integration of various ML algorithms suited to semantic segmentation in historical urban heat vulnerability mapping.

Modern semantic segmentation approaches fall into two main categories: CNN-based and Transformer-based. CNNs, like U-Net, excel at processing spatial hierarchies in complex urban landscapes, making them ideal for high-precision segmentation in dense urban environments with fine-grained details. U-Net’s encoder-decoder architecture captures both context and spatial information, making it particularly effective for urban heat vulnerability mapping where precise boundaries are essential. On the other hand, Transformers, such as UNetFormer, are superior in modeling global information and context, making segmentation both accurate and efficient. Their self-attention mechanism captures long-range dependencies within data, enhancing the understanding of complex urban landscapes and their thermal characteristics.

The U-HEAT framework leverages these algorithms to ensure robustness and versatility. U-Net can be used for detailed segmentation, providing high-resolution mapping of urban heat islands, while Transformers can model broader contextual information, identifying patterns and trends over larger spatial extents. By incorporating both historical and real-time RS data, U-HEAT allows for the continuous monitoring and dynamic updating of urban heat vulnerability assessments. For instance, historical Landsat data can be used to train U-Net to recognize long-term trends in land surface temperature changes, while current Sentinel-2 data can be processed using UNetFormer for the up-to-date segmentation of urban green spaces and built-up areas. This adaptability makes U-HEAT a comprehensive tool for urban heat vulnerability assessment, enabling urban planners to select the most suitable methods based on the specific characteristics of the urban area and the nature of the historical data, ensuring precise and actionable insights for mitigating urban heat risks.

3.2.3. Model Development, Validation and Effective Communication of Results

Developing an effective training and validation process for ML models like U-Net, used in mapping urban heat vulnerability, is crucial for their reliability and practicality in sustainable urban planning and policymaking [90]. A comprehensive training phase requires a meticulously organized dataset encompassing a variety of data types such as socio-demographic information, health statistics, and environmental factors [9,14]. The objective is to capture the complex and detailed aspects of urban heat vulnerability, which vary across different urban layouts and environmental conditions. A stratified sampling method is key to ensuring the dataset accurately represents diverse urban environments [91], from densely populated residential areas to industrial zones, green spaces, and water bodies, reflecting their unique thermal profiles and seasonal variations.

Additionally, it is essential to combine the strengths of RS and ground-based data. Integrating high-resolution satellite imagery with detailed on-the-ground measurements creates a robust training set, offering both extensive coverage and precision [92]. This approach lays a solid foundation for the algorithms to learn and interpret the complex patterns of urban heat disparities. A crucial component of the training set is labelled data, particularly from the validation dataset, which includes statistics on heat-related health issues. This strategic inclusion of real-world health outcomes enables the model to associate thermal anomalies with potential health risks, enhancing its predictive accuracy and making its insights highly relevant to public health and urban policymaking [93].

The validation phase in urban heat vulnerability mapping is essential for ensuring the model’s rigor and accuracy. We use k-fold cross-validation as a key technique to enhance the model’s robustness [94]. In k-fold cross-validation, the dataset is split into k equally sized subsets (folds). The model is trained on k-1 of these subsets and tested on the remaining one. This process is repeated k times, with each subset serving as the test data once. The results from each fold are averaged to produce a single performance metric. This technique maximizes data use and helps detect and reduce overfitting and underfitting by exposing the model to different data subsets. The evaluation metrics include accuracy, precision, recall, and the F1 score, offering a balanced measure of performance. Accuracy measures the proportion of correctly predicted instances, providing an overall performance overview. Precision indicates the proportion of true positive predictions among all positive predictions, reflecting the model’s exactness. Recall measures the proportion of true positive predictions among all actual positives, highlighting the model’s ability to identify all relevant cases. The F1 score, the harmonic mean of precision and recall, balances these two metrics, providing a single measure that considers both false positives and false negatives.

The primary validation indicators are heat-related morbidity and mortality rates, chosen for their relevance. These rates are direct measures of the health impacts of extreme heat, making them highly suitable for validating urban heat vulnerability models. They reflect the real-world consequences of heat exposure on human health, with a well-established positive correlation between chronic diseases (like cardiovascular, respiratory conditions, and diabetes) and heat-related mortality rates [79,80]. This correlation makes these indicators especially useful for assessing the model’s accuracy and reliability. Using heat-related morbidity and mortality rates allows for a direct assessment of how well the model predicts real-world health outcomes, linking model predictions to actual impacts.

These indicators provide a solid benchmark for evaluating the model’s performance, enabling comparisons between predictions and observed health outcomes to assess accuracy and generalizability. Validating the model against independent datasets with real-world health outcomes avoids bias and overfitting, ensuring the model’s results are generalizable to new, unseen data [93]. Advanced validation techniques like confusion matrices, receiver operating characteristic (ROC) curves, and area under the curve (AUC) metrics also provide detailed evaluations of the model’s predictive capabilities, measuring the trade-offs between true and false positive rates. This comprehensive approach ensures that the U-HEAT framework offers reliable and actionable insights for urban heat vulnerability assessment, equipping planners and policymakers with accurate, up-to-date information for mitigating heat-related risks [93].

The expected grid-scale map of urban heat vulnerability is presented in Figure 6. This map categorizes vulnerability into five levels, as established by prior studies: Very Low, Low, Medium, High, and Very High [95,96]. Unlike earlier mappings, this grid-scale approach offers a more detailed view of heat vulnerabilities within administrative boundaries. This approach aids urban planners in pinpointing vulnerabilities more precisely. The grid-based maps enable us not only to identify areas of high vulnerability through anomalies, but also to gain a deeper understanding of the spatial coherence in the distribution of these vulnerabilities.

For visualizing long-term and grid-scale historical urban heat trends, geographic information system (GIS) platforms are indispensable. They enable the integration and analysis of multi-temporal, multi-source data, applying spatial statistics to identify trends, patterns, and hotspots. Enhanced with visualization techniques like heat maps, change detection overlays, and temporal sliders, these maps vividly illustrate the dynamics of urban heat vulnerability over time. The insights gained are actionable for urban planners and policymakers, aiding in identifying areas vulnerable to heat that need mitigation efforts, such as increased greenery or reflective surfaces. Temporal analysis helps evaluate the effectiveness of past policies and guides future initiatives [97]. By analyzing the progression of heat vulnerability, policies can be tailored to specific phases of urban development, promoting dynamic and sustainable urban planning.

3.3. Future Prediction of Urban Heat Vulnerability

3.3.1. Lack of Future Prediction

A research gap in current studies is the lack of effective methods for predicting future urban heat vulnerability [14]. This shortfall in predictive methods limits the ability to anticipate and respond to future urban challenges, affecting preparedness and urban resilience. The difficulty in creating forward-looking assessment methods for urban heat vulnerability stems from several factors. Firstly, increasing urbanization and climate change bring more uncertainty to prediction research compared to historical identification. Urban heat vulnerability is influenced by a complex mix of biophysical, socio-demographic, and environmental factors, each with its own set of complexities and interconnections. Capturing these in future predictions may be more challenging because these phenomena have not fully occurred yet. In this context, the ever-changing nature of urban ecosystems, coupled with the multifaceted nature of vulnerability, makes forecasting particularly challenging [75].

Secondly, the lack of methodologies leads to a noticeable absence of methodological reference and uniformity in forecasting research. This might be due to path dependence [98], which describes how researchers or policymakers are inclined to make decisions based on previous studies and policies. As a result, forecasting urban heat vulnerability receives little attention. However, the importance of forecasting in managing urban heat vulnerability issues cannot be denied. Thirdly, the lack of a consistent historical understanding hinders the forecasting process. Many studies focus on data post-1990, offering a limited view and reducing the potential for a long-term understanding and prediction of evolving vulnerabilities [14]. A comprehensive understanding of historical patterns and changes is missing, which is essential for predicting future evolution.

The combination of ML and RS presents a promising solution to these challenges in forecasting urban heat vulnerabilities. RS provides detailed insights into urban landscape dynamics, crucial for identifying changes in urban microclimates [99,100]. ML, with its ability to handle large datasets, helps identify patterns in these insights [101,102]. Together, they might effectively identify evolving urban heat trends, pinpoint vulnerable areas, and predict emerging concerns. The importance of this approach is further heightened when combined with historical urban heat vulnerability maps, which act as empirical references, depicting past trends in urban heat vulnerability. Using these maps to train ML models improves their predictive accuracy, ensuring predictions are grounded in historical data. Moreover, ML’s flexibility in integrating varied data from multiple sources in a more comprehensive understanding of urban heat vulnerability enhances the accuracy of predictions [103]. In summary, the integration of ML and RS may aptly fill the gaps in predicting urban heat vulnerability. This approach shifts the focus from just understanding current vulnerabilities to actively preparing for future challenges, equipping urban areas with the tools to handle upcoming complexities.

3.3.2. Challenges and Algorithm Selection for Future Prediction

Predicting future urban heat vulnerability at a grid scale, even with historical maps and urban planning data, faces significant challenges. The main issue is the ever-changing nature of urban environments [75]. Factors like urban expansion, land use changes, and different building materials can significantly alter heat absorption and release, making it difficult to rely on past patterns to forecast future conditions. Moreover, urban planning efforts like increasing green spaces and upgrading building standards can greatly reduce heat effects, reducing the reliability of historical trends for future predictions.

Climate change exacerbates these challenges by unpredictably altering baseline temperatures and the frequency of extreme heat events [29]. Socio-economic changes, such as shifts in demographics and economic growth, add further complexity [78]. These changes can impact population density and resource distribution, and intensify the urban heat island effect. Technological advancements in construction and environmental policies can also change established patterns of heat vulnerability [2]. Therefore, while historical and planning data are valuable, their application to future scenarios must be approached cautiously, considering the various evolving factors that impact urban heat dynamics.

Selecting the right ML algorithms for long-term urban heat vulnerability prediction requires a strategic approach. Techniques like RF and Gradient Boosting Machines effectively handle nonlinear patterns and interactions among socio-environmental factors [104,105], accommodating diverse urban data and climate change impacts. Conversely, deep learning algorithms like Convolutional Neural Networks, which excel in analyzing complex spatial data, are better suited for examining RS imagery and detecting urban landscape changes [106]. We propose framing future urban heat vulnerability prediction as a spatio-temporal ML challenge, integrating geographical and time-based dimensions. This approach is particularly suited for dynamic urban environments, where spatial layout and temporal changes are crucial. Spatio-temporal models synthesize spatial dimensions (like city layouts) and temporal dimensions (such as seasonal variations) to predict patterns shaped by urban expansion, land use changes, and socio-economic shifts. ConvLSTM, which combines the spatial processing of CNNs with the temporal sequencing of LSTM networks, is ideal for scenarios demanding high precision in handling temporal dynamics [107,108].

ConvLSTM captures spatial features from RS data, like satellite images, and processes them over time to predict changes in urban heat patterns. For instance, it can analyze historical land surface temperature data to forecast future urban heat vulnerability, considering seasonal and long-term trends. Another example is STCNN (Spatio-Temporal Convolutional Neural Network), which excels in analyzing complex urban landscapes and processing large-scale geospatial data [109,110]. STCNNs leverage convolutional layers to extract spatial features from high-resolution RS imagery, such as vegetation cover and urban density, while temporal layers track changes over time. This makes STCNNs effective in diverse cities, allowing detailed analysis of how various urban features contribute to heat vulnerability over time.

Additionally, STGCNs (Spatio-Temporal Graph Convolutional Networks) are particularly useful for structured spatial data analysis, such as urban networks, assessing the impact of urban development and infrastructural changes on heat vulnerability [111,112]. STGCNs extend CNNs and LSTMs by incorporating graph structures to model complex relationships between urban entities. For instance, STGCNs can evaluate how changes in transportation networks, building layouts, and green spaces influence heat distribution and vulnerability, enabling a more comprehensive analysis of urban environments by considering both spatial connections and temporal evolution.

Integrating these advanced ML algorithms within the U-HEAT framework ensures robust and versatile heat vulnerability assessments. ConvLSTM, with its spatio-temporal data handling, is particularly valuable for predicting future urban heat trends based on historical RS data. STCNNs offer detailed spatial analysis of current urban conditions, while STGCNs provide insights into the broader impact of urban planning decisions. By leveraging these algorithms, U-HEAT can dynamically adapt to new data inputs, continuously refining its predictions and offering actionable insights for urban planners and policymakers. This comprehensive approach enhances the framework’s ability to address the complexity of urban heat vulnerability, making it a powerful tool for sustainable urban development and climate resilience planning.

3.3.3. Model Development, Validation and Presentation of Results

To effectively train models for predicting future urban heat vulnerability, it is crucial to consider several key factors for accuracy and effectiveness. Initially, a deep understanding of historical urban heat vulnerability patterns is necessary [83]. This involves examining past maps to identify trends and establish a baseline for future comparisons. Additionally, incorporating urban planning or projected data is critical [113]. This includes projected changes in land use, infrastructure, and vegetation cover, which heavily influence urban heat vulnerability dynamics. Combining these data sources is challenging but essential to accurately represent the complex interactions affecting urban heat vulnerability.

For instance, in training a ConvLSTM network, these elements are pivotal. ConvLSTM, adept in handling both spatial and temporal data due to its fusion of convolutional neural networks and Long Short-Term Memory networks [114,115], is ideal for this task. Training involves using historical heat maps for spatial understanding and urban planning data to add a temporal perspective, anticipating how heat distribution might evolve. It is vital to expose the model to a diverse array of scenarios, including varying climate conditions, urban layouts, and time scales. This diversity is key to enabling the model to generalize and make reliable predictions in changing urban environments. As urban areas grow and evolve, the adaptability and predictive capabilities of models like ConvLSTM become increasingly important for effective urban planning and heat management.

Validating predictions of future urban heat vulnerability is crucial for ensuring accuracy and reliability [116]. Without future real-world data, the process involves examining the model’s ability to replicate historical patterns and apply them to future scenarios. This includes using recent data (e.g., the 2020 map) for validation, ensuring the model captures both short-term fluctuations and long-term trends. Validation assesses the model’s skill in capturing the temporal and spatial dynamics of urban heat vulnerability. Consistency over time and spatial precision are key—the model must accurately identify areas in urban landscapes prone to heat vulnerability. Sensitivity analyses under different urban planning scenarios also provide insights into the model’s robustness and flexibility.

When applying these principles to a ConvLSTM model for predicting urban heat vulnerability, the process involves complex steps. The LSTM components are tested against temporal sequences of real-world historical data to ensure accurate time-related pattern capture. The convolutional parts, responsible for spatial feature detection, are validated by comparing predictions with historical maps, confirming the model’s ability to interpret spatial distribution variances. The model is also tested under hypothetical urban development scenarios, evaluating its adaptability to changes in urban planning. Its effectiveness is measured with statistical metrics, offering a thorough assessment and identifying areas for improvement. This ongoing validation and refinement cycle is crucial for enhancing performance, ensuring it serves as a reliable resource for urban planners and policymakers.

Predictive urban heat vulnerability maps are essential for advancing urban planning and policymaking. These maps offer more than historical data, providing a forward-thinking view crucial for adapting to rapid urban changes [83]. They project future heat vulnerabilities, enabling planners and policymakers to prepare for upcoming challenges. This proactive approach is vital for identifying future high-risk areas, guiding interventions, and shaping resilient infrastructure. Predictive maps offer insights for developing urban policies, from heat-resilient building regulations to public health plans for heatwaves [75]. Adopting predictive heat maps marks a significant shift in urban planning, emphasizing long-term sustainability and resilience against increasing heat-related challenges.

3.4. Strategy Recommendation

The impact of certain research results on the strategy design is often not fully acknowledged. Many studies highlight their potential to influence policy decisions, yet their influence is sometimes limited by the constraints of specific space and time frames [14]. A considerable number of these studies focus on evaluations within a single year, constrained by administrative boundaries. While these studies provide foundational insights, they may not offer the comprehensive analysis needed for definitive policymaking decisions [18,65]. It is important to recognize that assessments based on administrative scales might suit broad management strategies, but the diverse needs of individuals within these areas call for more personalized policy approaches. Therefore, a one-size-fits-all policy might not be the most effective approach.

To develop truly thorough and influential policies, ongoing monitoring and detailed data analysis are essential [117,118], features that U-HEAT embodies. Combining ML’s analytical power with RS’s ability to capture LST in high resolutions, U-HEAT is an innovative and promising approaches. RS allows for long-term and accurate temperature data collection through thermal bands, while ML excels in fusing various data types, offering deep insights into urban heat vulnerability with sophisticated computational analysis. U-HEAT sets a high bar for urban policy development and is also a ground-breaking framework for assessing strategy effectiveness. Its framework supports ongoing self-validation and refinement of parameters as data evolve. Strategies are reassessed by comparing outcomes before and after implementation, ensuring continuous improvement and relevance.

3.5. Continuous Monitoring and Updating

Remarkably, many existing studies have not highlighted the critical importance of continuous monitoring for effective disaster prevention and management in urban planning [66]. This gap is partly due to the lack of long-term focus in these studies, as well as their methodological constraints, which are often tied to specific datasets, time frames, and contexts. Current methods typically lack the flexibility needed to remain relevant across different scenarios. The notable absence of robust predictive techniques hinders the establishment of continuous monitoring, a key element for informed policy development [75]. Both retrospective analysis and forward-looking insights are crucial for crafting wise policies. Achieving ongoing monitoring requires a combination of sustained research, methodological diversity, and iterative processes.

U-HEAT stands out as a sophisticated and efficient tool for the continuous assessment of urban heat vulnerability and its changes over time. Constructed as an iterative system, U-HEAT is designed to be adaptable. As new data become available, U-HEAT actively incorporates this information, allowing for model refinements based on updated datasets and facilitating comparisons between earlier predictions and recent results. This adaptability ensures that U-HEAT remains a dynamic tool, capable of responding to evolving urban heat challenges. By integrating fresh data, U-HEAT continuously improves its predictive accuracy, thereby providing urban planners and policymakers with the most up-to-date insights.

The continuous monitoring and updating process in U-HEAT involves several key elements. Firstly, it emphasizes the importance of an iterative approach to data integration. By regularly updating its models with new socio-economic, environmental, and health-related data, U-HEAT ensures that its assessments reflect the latest conditions and trends. This real-time data integration is critical for maintaining the relevance and accuracy of the framework, enabling it to detect and adapt to new patterns of urban heat vulnerability.

Secondly, U-HEAT’s ability to refine its models based on new data inputs allows for the ongoing enhancement of mapping accuracy. This means that as new information is fed into the system, U-HEAT recalibrates its predictions, improving the precision of its vulnerability assessments. This continuous refinement process ensures that urban heat maps remain current and reliable, providing a solid foundation for decision-making.

Moreover, U-HEAT’s active learning capabilities enable the model to learn from new data without the need for complete system re-training. This incremental learning approach allows U-HEAT to adjust its parameters and improve its performance efficiently, making it a sustainable tool for long-term use. By incorporating the latest data and refining its models accordingly, U-HEAT can offer proactive insights that help urban planners anticipate and mitigate future heat risks.

The continuous monitoring and updating feature of U-HEAT also extends to strategy evaluation and policy adaptation. By comparing expected trends with actual developments, U-HEAT provides a robust mechanism for evaluating the effectiveness of implemented policies. This comparative analysis ensures that policies are not only thoroughly evaluated, but also appropriately modified when necessary, enhancing their impact and efficacy. This feedback loop is essential for creating resilient urban environments that can adapt to changing heat conditions over time.

In summary, U-HEAT’s continuous monitoring and updating process is fundamental to its ability to provide ongoing, accurate assessments of urban heat vulnerability. By incorporating new data, refining its models, and enabling proactive policy adjustments, U-HEAT ensures that urban planners and policymakers have access to the most current information. This dynamic and iterative approach makes U-HEAT a vital tool for long-term urban heat management, helping cities to effectively address and adapt to the challenges posed by rising temperatures.

4. Findings and Discussion

4.1. Key Challenges and Limitations in Existing Approaches

This study highlighted the key challenges and limitations of existing urban heat vulnerability assessment approaches as follows:

Lack of Standardized Criteria and References: Many existing studies on urban heat vulnerability assessment suffer from the absence of standardized criteria and references for selecting indicators and developing methodologies. This lack of standardization leads to inconsistencies and potential inaccuracies in the assessment results, making it challenging to compare findings across different studies and contexts.

Insufficient Spatial Resolution and Temporal Coverage: Traditional approaches often do not meet the needs of urban planning due to their insufficient spatial resolution and temporal coverage. Many studies rely on static data from census tracts or administrative areas, which can introduce biases and fail to capture the dynamic and evolving nature of urban heat vulnerability. Additionally, short-term assessments predominate, leaving a gap in longitudinal research that could provide insights into evolving vulnerabilities over time.

Subjective Judgments in Indicator Selection: The selection of indicators in many urban heat vulnerability studies is often based on subjective judgments. The absence of universal criteria for selecting and categorizing indicators complicates the comparison of studies and the establishment of best practices. This subjectivity can lead to potential biases and variations in the reliability of the assessment outcomes.

Limited Integration of Machine Learning and Remote Sensing: While ML and RS have shown great promise in urban studies, their potential in urban heat vulnerability assessment remains underutilized. Existing approaches predominantly focus on data processing and preparation, with less emphasis on in-depth analysis and identification of urban heat vulnerability. This limitation hinders the development of comprehensive and precise long-term urban heat assessments.

Challenges in Validating Results: Result validation is paramount in urban heat vulnerability research, yet only a few studies incorporate robust validation procedures. The use of heat-related health outcomes, such as emergency hospital admissions and mortality rates, as validation variables presents challenges due to the unavailability of direct data and reliance on proxies, which may introduce inaccuracies. Additionally, the mental health impacts of extreme heat are often neglected in current studies.

Complexity of Spatial and Temporal Dynamics: Urban heat vulnerability assessments must account for the complex interactions between natural and built environments, individual biophysical and psychological conditions, and socio-economic status. Many existing models focus on specific aspects, such as biophysical or social vulnerabilities, but fail to integrate these dimensions comprehensively. The MAUP and the need for grid-scale spatial analyses for improved accuracy further complicate these assessments.

U-HEAT aims to overcome these challenges by integrating ML with RS to establish a standardized and sustainable approach for urban heat vulnerability assessment. U-HEAT develops universal criteria for selecting and categorizing indicators, reducing subjectivity, and enhancing the comparability and reliability of assessments. By leveraging high-resolution RS data and advanced ML techniques, U-HEAT provides detailed evaluations at a granular grid-level scale, accurately capturing the spatial and temporal dynamics of urban heat vulnerability. The integration of ML and RS significantly enhances urban heat vulnerability analysis by improving spatial resolution, temporal coverage, and accuracy. ML algorithms such as CNNs and LSTM networks process high-resolution RS data, including LST and NDVI, to create detailed grid-scale maps.

This approach allows for the precise identification of vulnerable areas by considering the influence of neighboring regions. Temporal coverage is improved through historical and predictive mapping, enabling the assessment of long-term trends and future scenarios. The framework’s active learning capabilities ensure continuous improvement in predictive accuracy by integrating new data. Additionally, U-HEAT incorporates a robust validation framework using heat-related health outcomes and other relevant data to ensure the accuracy and reliability of results. Its innovative predictive approach enables forecasting future trends in urban heat vulnerability, providing a structured component-based framework that integrates multiple dimensions of vulnerability for detailed mapping and management. This comprehensive tool aims to equip urban planners and policymakers with actionable insights for identifying vulnerable localities, developing mitigation strategies, and enhancing urban resilience and sustainability.

4.2. Prospective Applications

The framework proposed is at the conceptual stage and will need to be applied in pilot testing for recalibration once all datasets are ready. This step will provide invaluable insights and possible improvements, guiding further actions. Then, further testing and extensive applications on relevant case studies are needed to understand U-HEAT’s effectiveness and adaptability in different local contextual settings. The results at this initial case testbed stage also require feedback from local authorities and expert professionals. This feedback is critical for validating the results and making them accurate, locally relevant, and widely applicable. Data consistency and integration are also significant challenges, given the variety of data sources. This needs to be tackled by using data from reliable sources, standardizing it through proven processing techniques, and applying ML for multi-modal data fusion. Additionally, there are questions about U-HEAT’s scalability and its effectiveness in larger, more complex urban areas. The likely approach is to start with smaller, representative regions in the pilot study and, upon its success, expand it to a broader case study. The step-by-step approach aims to validate and refine U-HEAT and explore its impact on urban research. By explaining the methods, addressing concerns, and proposing solutions, it is important that we work towards creating a robust, scalable tool for analyzing urban heat vulnerability in various contexts. Ultimately, collaboration with the academic and urban planning communities to keep improving and broadening U-HEAT’s utility is a necessity.

4.3. Contributions to Sustainable Development

The development of U-HEAT, aimed at assessing and mitigating urban heat vulnerability, is closely tied to several SDGs. U-HEAT supports these goals by providing critical insights and tools for urban planning and policymaking. It primarily aligns with SDG 11, which focuses on creating sustainable cities and communities. This goal emphasizes the need for urban areas to be inclusive, safe, resilient, and sustainable. Urban heat vulnerability affects these aspects, as increased temperatures can compromise urban livability and safety. U-HEAT aids in designing resilient and sustainable urban spaces that better cope with rising temperatures, integrating green spaces, sustainable materials, and heat-mitigating urban designs. Additionally, U-HEAT is closely related to SDG 13, centered on climate action. The increasing instances of urban heat islands, driven by climate change, make their mitigation vital to climate action strategies [119,120,121]. U-HEAT helps cities adapt to and mitigate a significant impact of climate change. It also ties into SDG 3, which focuses on good health and well-being, as heatwaves pose major health risks, particularly to vulnerable populations [122,123,124]. By reducing heat-related health risks, U-HEAT promotes public health and well-being, ensuring safer urban environments for all, especially the vulnerable. Moreover, U-HEAT addresses SDG 10, which seeks to reduce inequalities. Urban heat often disproportionately impacts lower-income communities, making its assessment and mitigation crucial for reducing social disparities. U-HEAT helps identify and protect those most at risk from urban heat, contributing to reducing inequalities. By offering a comprehensive approach to urban heat vulnerability, U-HEAT not only improves immediate urban living conditions but also plays a crucial role in sustainable development, ensuring a more equitable and resilient future for urban populations.

4.4. Implications in Policy and Public Engagement

The U-HEAT framework has significant implications for sustainable urban development and climate change adaptation, as well as for planning, policymaking, and public engagement.

Enhanced Urban Planning: U-HEAT enables urban planners to design cities that are more resilient to heatwaves by providing high-resolution, grid-scale data on urban heat vulnerability. This detailed information facilitates the precise identification of heat-vulnerable areas, allowing for targeted interventions such as increasing green spaces and optimizing building materials to reduce heat absorption. The framework supports comprehensive methodological guidance throughout the disaster management process, aiding in the preparation, mapping, strategic recommendations, and ongoing monitoring of urban heat issues.

Informed Policymaking: The predictive capabilities of U-HEAT are crucial for proactive policymaking. By forecasting future urban heat vulnerability trends, policymakers can implement strategies to mitigate these risks before they become critical. This includes zoning regulations, urban greening initiatives, and infrastructure improvements to enhance urban resilience against rising temperatures. The high-resolution, long-term datasets provided by U-HEAT support evidence-based governance and planning, pinpointing highly vulnerable areas that require preventive interventions or targeted resource allocations.

Public Health and Well-being: U-HEAT focuses on socio-economic and health-related indicators to prioritize vulnerable populations, such as the elderly and those with pre-existing conditions, in heat mitigation strategies. This approach supports sustainable development by promoting equitable urban environments where all residents have access to necessary resources and protections against extreme heat. By integrating datasets like green infrastructure, housing conditions, and cultural backgrounds, U-HEAT can analyze how these indicators affect inequity, leading to more targeted strategy development and resource allocation [99]. For example, planting vegetation in areas with green inequity and distributing heat-relief resources to disadvantaged groups can help create a livable, health-friendly environment for all communities.

Long-term Sustainability: U-HEAT’s continuous monitoring and updating process ensures that urban heat vulnerability assessments remain current and relevant. This dynamic approach allows cities to adapt their strategies over time, maintaining resilience in the face of evolving climate conditions. The framework’s ability to refine its models based on new data inputs allows for the ongoing enhancement of mapping accuracy, ensuring that strategies remain effective and responsive to changing urban heat challenges.

Public Engagement: U-HEAT promotes public engagement by offering online map resources that help residents understand the heat vulnerability of their areas. These resources, implemented by various national and state governments, raise awareness, and provide tools to mitigate heat exposure risks in daily life. The framework also advocates for more public engagement opportunities, such as training modules and discussion workshops, to effectively integrate policy interventions and public involvement [93]. This collective effort aids in managing urban heat vulnerability more efficiently and sustainably.

The U-HEAT framework is adaptable to various global urban contexts, but it requires further testing in diverse local settings to gauge its effectiveness. Implementing U-HEAT in cities with different climates, socio-economic conditions, and urban layouts is essential to recalibrate the model, ensuring it accurately reflects the unique heat vulnerability factors of each region. Collaboration with local planners, scientists, and health officials is crucial for refining the framework with region-specific data, making U-HEAT’s recommendations more relevant and effective. Comparative studies across regions can also identify global patterns and regional nuances, enhancing our understanding of urban heat risks and informing tailored intervention strategies. By addressing these broader implications, U-HEAT aims to be a versatile tool for global urban resilience and sustainable development.

Implementing U-HEAT involves addressing several critical considerations. First, data accuracy and consistency are essential; high-quality datasets are necessary for reliable assessments, but integrating them from various sources can be challenging. Continuous updates and robust algorithms are required to maintain predictive accuracy, demanding long-term commitment. Second, technical expertise and training for urban planners are crucial to effectively use U-HEAT outputs; without proper training, its potential may not be fully realized. Third, ethical data handling and equitable resource access are vital; protecting socio-economic and health data privacy while prioritizing vulnerable populations aligns with sustainable development goals. Finally, effective public engagement is necessary for U-HEAT’s success; raising awareness and ensuring inclusive access to digital tools fosters a collective effort to manage urban heat vulnerability. Addressing these key considerations will enable U-HEAT to significantly enhance urban resilience to heatwaves and support sustainable urban development.

4.5. Assumptions and Limitations

The development of U-HEAT is based on several assumptions and has potential limitations. The selection of indicators and data assumes that they accurately reflect urban heat factors and that the data sources are reliable and representative [9,14]. This paper compiles frequently used indicators and data sources as a standard reference, but supplementing suitable indicators for different urban environments may encounter biases and subjective preferences. The process is also limited by data availability and quality. Urban environments are complex, and chosen indicators may not capture all nuances, especially in areas with limited data collection infrastructure. Moreover, data quality can vary significantly due to socio-economic and technological factors, potentially leading to skewed assessments.

In integrating ML and RS, RS provides high-resolution images of urban environments, essential for data on LST and infrastructure. The assumption is that these images are consistent and representative. However, data quality can be affected by atmospheric conditions and technological constraints. For example, Landsat provides 30 m spatial resolution images but only captures data every 16 days at 10 AM, possibly missing peak temperatures and heat variations during heatwaves. MODIS offers daily LST data but with a 1 km spatial resolution [71]. Interpreting RS data within complex urban landscapes can introduce errors or misinterpretations.

ML algorithms are expected to uncover patterns and correlations within large datasets, providing accurate predictions based on success in other urban domains. However, their effectiveness depends on the quality of input data. Issues like overfitting or underfitting can limit the models’ ability to generalize, leading to inaccuracies in new or evolving scenarios. The ‘black box’ nature of complex ML models poses challenges in transparency and interpretability [125], crucial for justifying assessments. Additionally, the computational demands of processing large datasets require significant resources, which may not always be available.

Continuous monitoring is assumed to be feasible and integral for U-HEAT’s adaptability. This involves regularly collecting and integrating new data to refine models and adjust strategies. However, it requires consistent funding and manpower, which may not always be available. Challenges in timeliness and accuracy of data updates could impact the framework’s responsiveness to evolving urban heat challenges [75]. Despite these potential limitations, U-HEAT remains a promising tool for assessing and mitigating urban heat vulnerability. By limiting bias and enhancing the feasibility of each component—selecting suitable indicators, RS images, and ML algorithms—we can improve the framework’s integrity and build a more consolidated tool.

5. Conclusions

The growing challenge of urbanization and climate change has intensified the focus on urban heat, particularly its disproportionate impact on vulnerable groups like the elderly and those with pre-existing health conditions. While previous studies have examined the spatial distribution and key drivers of urban heat vulnerability, they often lacked rigor in indicator selection, spatio-temporal coverage, and validation techniques.

The primary objective of this study is to outline the theoretical underpinnings and potential applications of U-HEAT, emphasizing its conceptual nature. U-HEAT introduces a novel approach by integrating machine learning with remote sensing, enabling the precise mapping and forecasting of urban heat vulnerability at a detailed resolution. This approach is crucial for informed urban planning and disaster management. U-HEAT also combines historical data with predictive mapping, offering valuable insights for ongoing monitoring and mitigation, thereby aiding urban planners and policymakers in addressing urban heat island effects and protecting communities.

U-HEAT marks a significant step forward in understanding urban heat vulnerabilities, with important implications for urban sustainability and resilience. By melding advanced technologies and data, U-HEAT enriches academic research and serves as a crucial resource for cities looking to enhance their ability to adapt to climate-related challenges. This framework underscores the importance of interdisciplinary research in strengthening urban resilience and highlights the need for sustainable urban ecosystems capable of facing the complex challenges posed by climate change and increased urban heat.

Lastly, we conclude by highlighting the following key contributions of this study: (a) the introduction of an innovative predictive method for forecasting urban heat vulnerability; (b) the development of universal criteria and a reference framework for selecting urban heat vulnerability indicators; (c) the demonstration of the feasibility of long-term, grid-scale, precise urban heat vulnerability assessments; (d) the proposal of a robust, enduring, and sustainable framework for monitoring and managing urban heat vulnerability; and (e) support for urban sustainability through the integration of ML and RS in the urban heat vulnerability framework.

Future research should explore additional indicators, such as psychological factors and detailed socio-economic variables, to enhance vulnerability assessments. Refining the ML algorithms in the U-HEAT framework, by incorporating advanced deep learning techniques and improving data handling, could further boost predictive accuracy and model robustness. Empirical validation and broad application across various urban contexts are also necessary to ensure the framework’s adaptability and effectiveness. By refining and expanding U-HEAT, future studies can contribute to more resilient and sustainable urban environments, better equipped to protect vulnerable populations and ensure urban well-being amid rising temperatures and climate change.

Author Contributions

F.L.: data collection, processing, investigation, analysis, and writing—original draft; T.Y., M.N., K.N.T. and F.D.: supervision, conceptualization, and writing—review and editing. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

No new data were created or analyzed in this study. Data sharing is not applicable to this article.

Acknowledgments

We appreciate the editor and anonymous referees for their invaluable comments on an earlier version of this manuscript. Additionally, the authors acknowledge the financial support from the Chinese Scholarship Council (No. 202106420006) and QUT towards the postgraduate research scholarship of the first named author.

Conflicts of Interest

The authors declare no competing financial interests or personal relationships that could have influenced the work reported in this paper.

References

Kamruzzaman, M.; Deilami, K.; Yigitcanlar, T. Investigating the Urban Heat Island Effect of Transit Oriented Development in Brisbane. J. Transp. Geogr. 2018, 66, 116–124. [Google Scholar] [CrossRef]
Degirmenci, K.; Desouza, K.; Fieuw, W.; Watson, R.; Yigitcanlar, T. Understanding Policy and Technology Responses in Mitigating Urban Heat Islands: A Literature Review and Directions for Future Research. Sustain. Cities Soc. 2021, 70, 102873. [Google Scholar] [CrossRef]
Australian Institute of Health and Welfare. Let’s Talk about the Weather: Injuries Related to Extreme Weather. 2023. Available online: https://www.aihw.gov.au/reports/injury/extreme-weather-injuries/contents/about (accessed on 19 June 2024).
Vescovi, L.; Rebetez, M.; Rong, F. Assessing Public Health Risk Due to Extremely High Temperature Events: Climate and Social Parameters. Clim. Res. 2005, 30, 71–78. [Google Scholar] [CrossRef]
Harlan, S.; Brazel, A.J.; Prashad, L.; Stefanov, W.; Larsen, L. Neighborhood Microclimates and Vulnerability to Heat Stress. Soc. Sci. Med. 2006, 63, 2847–2863. [Google Scholar] [CrossRef]
Reid, C.E.; O’Neill, M.S.; Gronlund, C.J.; Brines, S.J.; Brown, D.G.; Diez-Roux, A.V.; Schwartz, J. Mapping Community Determinants of Heat Vulnerability. Environ. Health Perspect. 2009, 117, 1730–1736. [Google Scholar] [CrossRef]
Boumans, R.; Phillips, D.; Victery, W.; Fontaine, T. Developing a Model for Effects of Climate Change on Human Health and Health–Environment Interactions: Heat Stress in Austin, Texas. Urban Clim. 2014, 8, 78–99. [Google Scholar] [CrossRef]
Zhang, W.; McManus, P.; Duncan, E. A Raster-Based Subdividing Indicator to Map Urban Heat Vulnerability: A Case Study in Sydney, Australia. Int. J. Environ. Res. Public Health 2018, 15, 2516. [Google Scholar] [CrossRef]
Cheng, W.; Li, D.; Liu, Z.; Brown, R. Approaches for Identifying Heat-Vulnerable Populations and Locations: A Systematic Review. Sci. Total Environ. 2021, 799, 149417. [Google Scholar] [CrossRef] [PubMed]
Adger, W.N. Vulnerability. Glob. Environ. Change 2006, 16, 268–281. [Google Scholar] [CrossRef]
Tate, E. Social Vulnerability Indices: A Comparative Assessment Using Uncertainty and Sensitivity Analysis. Nat. Hazards 2012, 63, 325–347. [Google Scholar] [CrossRef]
Chen, Q.; Ding, M.; Yang, X.; Hu, K.; Qi, J. Spatially Explicit Assessment of Heat Health Risk by Using Multi-Sensor Remote Sensing Images and Socioeconomic Data in Yangtze River Delta, China. Int. J. Health Geogr. 2018, 17, 13. [Google Scholar] [CrossRef]
Niu, Y.; Li, Z.; Gao, Y.; Liu, X.; Xu, L.; Vardoulakis, S.; Liu, Q. A Systematic Review of the Development and Validation of the Heat Vulnerability Index: Major Factors, Methods, and Spatial Units. Curr. Clim. Chang. Rep. 2021, 7, 87–97. [Google Scholar] [CrossRef]
Li, F.; Yigitcanlar, T.; Nepal, M.; Nguyen, K.; Dur, F. Understanding Urban Heat Vulnerability Assessment Methods: A PRISMA Review. Energies 2022, 15, 6998. [Google Scholar] [CrossRef]
Kazak, J. The Use of a Decision Support System for Sustainable Urbanization and Thermal Comfort in Adaptation to Climate Change Actions: The Case of the Wrocław Larger Urban Zone (Poland). Sustainability 2018, 10, 1083. [Google Scholar] [CrossRef]
He, C.; Ma, L.; Zhou, L.; Kan, H.; Zhang, Y.; Ma, W.; Chen, B. Exploring the Mechanisms of Heat Wave Vulnerability at the Urban Scale Based on the Application of Big Data and Artificial Societies. Environ. Int. 2019, 127, 573–583. [Google Scholar] [CrossRef] [PubMed]
Holec, J.; Šveda, M.; Szatmári, D.; Feranec, J.; Bobáľová, H.; Kopecká, M.; Šťastný, P. Heat Risk Assessment Based on Mobile Phone Data: Case Study of Bratislava, Slovakia. Nat. Hazards 2021, 108, 3099–3120. [Google Scholar] [CrossRef]
Weber, S.; Sadoff, N.; Zell, E.; de Sherbinin, A. Policy-Relevant Indicators for Mapping the Vulnerability of Urban Populations to Extreme Heat Events: A Case Study of Philadelphia. Appl. Geogr. 2015, 63, 231–243. [Google Scholar] [CrossRef]
Wilson, B.; Chakraborty, A. Mapping Vulnerability to Extreme Heat Events: Lessons from Metropolitan Chicago. J. Environ. Plan. Manag. 2019, 62, 1065–1088. [Google Scholar] [CrossRef]
Heaton, M.; Sain, S.; Greasby, T.; Uejio, C.; Hayden, M.; Monaghan, A.; Boehnert, J.; Sampson, K.; Banerjee, D.; Nepal, V.; et al. Characterizing Urban Vulnerability to Heat Stress Using a Spatially Varying Coefficient Model. Spat. Spatio-Temporal Epidemiol. 2014, 8, 23–33. [Google Scholar] [CrossRef]
Yin, Y.; Grundstein, A.; Mishra, D.R.; Ramaswamy, L.; Tonekaboni, N.; Dowd, J. DTEx: A Dynamic Urban Thermal Exposure Index Based on Human Mobility Patterns. Environ. Int. 2021, 155, 106573. [Google Scholar] [CrossRef]
Bu, S.; Smith, K.L.; Masoud, F.; Sheinbaum, A. Spatial distribution of heat vulnerability in Toronto, Canada. Urban Clim. 2024, 54, 101838. [Google Scholar] [CrossRef]
Barron, L.; Ruggieri, D.; Branas, C. Assessing Vulnerability to Heat: A Geospatial Analysis for the City of Philadelphia. Urban Sci. 2018, 2, 38. [Google Scholar] [CrossRef]
Prosdocimi, D.; Klima, K. Health Effects of Heat Vulnerability in Rio de Janeiro: A Validation Model for Policy Applications. SN Appl. Sci. 2020, 2, 1948. [Google Scholar] [CrossRef]
Oh, K.; Lee, M.; Jeon, S. Development of the Korean Climate Change Vulnerability Assessment Tool (VESTAP): Centered on Health Vulnerability to Heat Waves. Sustainability 2017, 9, 1103. [Google Scholar] [CrossRef]
Dong, J.; Peng, J.; He, X.; Corcoran, J.; Qiu, S.; Wang, X. Heatwave-Induced Human Health Risk Assessment in Megacities Based on Heat Stress-Social Vulnerability-Human Exposure Framework. Landsc. Urban Plan. 2020, 203, 103907. [Google Scholar] [CrossRef]
Ho, H.; Knudby, A.; Huang, W. A Spatial Framework to Map Heat Health Risks at Multiple Scales. Int. J. Environ. Res. Public Health 2015, 12, 16110–16123. [Google Scholar] [CrossRef] [PubMed]
Krstic, N.; Yuchi, W.; Ho, H.; Walker, B.; Knudby, A.; Henderson, S. The Heat Exposure Integrated Deprivation Index (HEIDI): A Data-Driven Approach to Quantifying Neighborhood Risk During Extreme Hot Weather. Environ. Int. 2017, 109, 42–52. [Google Scholar] [CrossRef] [PubMed]
Tomlinson, C.; Chapman, L.; Thornes, J.; Baker, C. Including the Urban Heat Island in Spatial Heat Health Risk Assessment Strategies: A Case Study for Birmingham, UK. Int. J. Health Geogr. 2011, 10, 42. [Google Scholar] [CrossRef] [PubMed]
Macintyre, H.; Heaviside, C.; Taylor, J.; Picetti, R.; Symonds, P.; Cai, X.; Vardoulakis, S. Assessing Urban Population Vulnerability and Environmental Risks Across an Urban Area During Heatwaves: Implications for Health Protection. Sci. Total Environ. 2017, 610–611, 678–690. [Google Scholar] [CrossRef] [PubMed]
Johnson, D.; Stanforth, A.; Lulla, V.; Luber, G. Developing an Applied Extreme Heat Vulnerability Index Utilizing Socioeconomic and Environmental Data. Appl. Geogr. 2012, 35, 23–31. [Google Scholar] [CrossRef]
Kamal, N.; Ash’aari, Z.; Abdullah, A.; Kusin, F.; Mohamat Yusuff, F.; Sharaai, A.; Mohd Ariffin, N. Extreme Heat Vulnerability Assessment in Tropical Region: A Case Study in Malaysia. Clim. Dev. 2021, 4, 472–486. [Google Scholar] [CrossRef]
Conlon, K.; Mallen, E.; Gronlund, C.; Berrocal, V.; Larsen, L.; O’Neill, M. Mapping Human Vulnerability to Extreme Heat: A Critical Assessment of Heat Vulnerability Indices Created Using Principal Components Analysis. Environ. Health Perspect. 2020, 128, 97001. [Google Scholar] [CrossRef] [PubMed]
Estoque, R.; Ooba, M.; Seposo, X.; Togawa, T.; Hijioka, Y.; Takahashi, K.; Nakamura, S. Heat Health Risk Assessment in Philippine Cities Using Remotely Sensed Data and Social-Ecological Indicators. Nat. Commun. 2020, 11, 1581. [Google Scholar] [CrossRef]
Chen, N.; Lin, P.; Guo, Y. Long-Term Exposure to High Temperature Associated with the Incidence of Major Depressive Disorder. Sci. Total Environ. 2019, 659, 1016–1020. [Google Scholar] [CrossRef] [PubMed]
Yoo, E.; Eum, Y.; Roberts, J.; Gao, Q.; Chen, K. Association Between Extreme Temperatures and Emergency Room Visits Related to Mental Disorders: A Multi-Region Time-Series Study in New York, USA. Sci. Total Environ. 2021, 792, 148246. [Google Scholar] [CrossRef]
Huang, F.; Yu, Y.; Feng, T. Automatic Extraction of Impervious Surfaces from High Resolution Remote Sensing Images Based on Deep Learning. J. Vis. Commun. Image Represent. 2019, 58, 453–461. [Google Scholar] [CrossRef]
Wang, P.; Bayram, B.; Sertel, E. A Comprehensive Review on Deep Learning Based Remote Sensing Image Super-Resolution Methods. Earth-Sci. Rev. 2022, 234, 104110. [Google Scholar] [CrossRef]
Osaku, D.; Nakamura, R.Y.M.; Pereira, L.A.M.; Pisani, R.J.; Levada, A.L.M.; Cappabianco, F.A.M.; Falcão, A.X.; Papa, J.P. Improving Land Cover Classification through Contextual-Based Optimum-Path Forest. Inf. Sci. 2015, 324, 60–87. [Google Scholar] [CrossRef]
Kafy, A.A.; Saha, M.; Faisal, A.-A.; Rahaman, Z.A.; Rahman, M.T.; Liu, D.; Fattah, M.A.; Al Rakib, A.; AlDousari, A.E.; Rahaman, S.N.; et al. Predicting the Impacts of Land Use/Land Cover Changes on Seasonal Urban Thermal Characteristics Using Machine Learning Algorithms. Build. Environ. 2022, 217, 109066. [Google Scholar] [CrossRef]
Uniyal, S.; Purohit, S.; Chaurasia, K.; Rao, S.S.; Amminedu, E. Quantification of Carbon Sequestration by Urban Forest Using Landsat 8 OLI and Machine Learning Algorithms in Jodhpur, India. Urban For. Urban Green. 2022, 67, 127445. [Google Scholar] [CrossRef]
Wu, X.; Ou, J.; Wen, Y.; Liu, X.; He, J.; Zhang, J. Developing a Data-Fusing Method for Mapping Fine-Scale Urban Three-Dimensional Building Structure. Sustain. Cities Soc. 2022, 80, 103716. [Google Scholar] [CrossRef]
Kim, Y.; Kim, Y. Deep Learning-Based Analytic Framework Using Comprehensive Urbanization Index for Heat Vulnerability Assessment in Urban Areas. Expert Syst. Appl. 2024, 235, 121140. [Google Scholar] [CrossRef]
Li, F.; Yigitcanlar, T.; Nepal, M.; Nguyen, K.; Dur, F. Machine Learning and Remote Sensing Integration for Leveraging Urban Sustainability: A Review and Framework. Sustain. Cities Soc. 2023, 96, 104653. [Google Scholar] [CrossRef]
Zhu, Q.; Sun, Y.; Guan, Q.; Wang, L.; Lin, W. A Weakly Pseudo-Supervised Decorrelated Subdomain Adaptation Framework for Cross-Domain Land-Use Classification. IEEE Trans. Geosci. Remote Sens. 2022, 60, 1–13. [Google Scholar] [CrossRef]
Jamali, A. Evaluation and Comparison of Eight Machine Learning Models in Land Use/Land Cover Mapping Using Landsat 8 OLI: A Case Study of the Northern Region of Iran. SN Appl. Sci. 2019, 1, 1527. [Google Scholar] [CrossRef]
Frantz, D.; Schug, F.; Okujeni, A.; Navacchi, C.; Wagner, W.; van der Linden, S.; Hostert, P. National-Scale Mapping of Building Height Using Sentinel-1 and Sentinel-2 Time Series. Remote Sens. Environ. 2021, 252, 112128. [Google Scholar] [CrossRef]
Guo, H.; Huang, J.J.; Chen, B.; Guo, X.; Singh, V.P. A Machine Learning-Based Strategy for Estimating Non-Optically Active Water Quality Parameters Using Sentinel-2 Imagery. Int. J. Remote Sens. 2021, 42, 1841–1866. [Google Scholar] [CrossRef]
Jato-Espino, D.; Manchado, C.; Roldán-Valcarce, A.; Moscardó, V. ArcUHI: A GIS Add-In for Automated Modelling of the Urban Heat Island Effect through Machine Learning. Urban Clim. 2022, 44, 101203. [Google Scholar] [CrossRef]
Abunnasr, Y.; Mhawej, M. Towards a Combined Landsat-8 and Sentinel-2 for 10-m Land Surface Temperature Products: The Google Earth Engine Monthly Ten-ST-GEE System. Environ. Model. Softw. 2022, 155, 105456. [Google Scholar] [CrossRef]
Casali, Y.; Aydin, N.Y.; Comes, T. Machine Learning for Spatial Analyses in Urban Areas: A Scoping Review. Sustain. Cities Soc. 2022, 85, 104050. [Google Scholar] [CrossRef]
Demarchi, L.; Canters, F.; Cariou, C.; Licciardi, G.; Chan, J.C.-W. Assessing the Performance of Two Unsupervised Dimensionality Reduction Techniques on Hyperspectral APEX Data for High Resolution Urban Land-Cover Mapping. ISPRS J. Photogramm. Remote Sens. 2014, 87, 166–179. [Google Scholar] [CrossRef]
Deng, C.; Wu, C. The Use of Single-Date MODIS Imagery for Estimating Large-Scale Urban Impervious Surface Fraction with Spectral Mixture Analysis and Machine Learning Techniques. ISPRS J. Photogramm. Remote Sens. 2013, 86, 100–110. [Google Scholar] [CrossRef]
Priem, F.; Okujeni, A.; van der Linden, S.; Canters, F. Comparing Map-Based and Library-Based Training Approaches for Urban Land-Cover Fraction Mapping from Sentinel-2 Imagery. Int. J. Appl. Earth Obs. Geoinf. 2019, 78, 295–305. [Google Scholar] [CrossRef]
Yigitcanlar, T. Australian Local Governments’ Practice and Prospects with Online Planning. URISA J. 2006, 18, 7–17. [Google Scholar]
Patino, J.E.; Duque, J.C. A Review of Regional Science Applications of Satellite Remote Sensing in Urban Settings. Comput. Environ. Urban Syst. 2013, 37, 1–17. [Google Scholar] [CrossRef]
Galaz, V.; Centeno, M.A.; Callahan, P.W.; Causevic, A.; Patterson, T.; Brass, I.; Baum, S.; Farber, D.; Fischer, J.; Garcia, D.; et al. Artificial Intelligence, Systemic Risks, and Sustainability. Technol. Soc. 2021, 67, 101741. [Google Scholar] [CrossRef]
United Nations. Global Indicator Framework for the Sustainable Development Goals and Targets of the 2030 Agenda for Sustainable Development; United Nations: San Francisco, CA, USA, 2023. [Google Scholar]
Touzani, S.; Granderson, J. Open Data and Deep Semantic Segmentation for Automated Extraction of Building Footprints. Remote Sens. 2021, 13, 2578. [Google Scholar] [CrossRef]
Chi, Y.; Fan, M.; Zhao, C.; Yang, Y.; Fan, H.; Yang, X.; Yang, J.; Tao, J. Machine Learning-Based Estimation of Ground-Level NO₂ Concentrations over China. Sci. Total Environ. 2022, 807, 150721. [Google Scholar] [CrossRef]
Peng, B.; Meng, Z.; Huang, Q.; Wang, C. Patch Similarity Convolutional Neural Network for Urban Flood Extent Mapping Using Bi-Temporal Satellite Multispectral Imagery. Remote Sens. 2019, 11, 2492. [Google Scholar] [CrossRef]
Ghaffarian, S.; Roy, D.; Filatova, T.; Kerle, N. Agent-Based Modelling of Post-Disaster Recovery with Remote Sensing Data. Int. J. Disaster Risk Reduct. 2021, 60, 102285. [Google Scholar] [CrossRef]
Zhang, P.; Liu, L.; Yang, L.; Zhao, J.; Li, Y.; Qi, Y.; Ma, X.; Cao, L. Exploring the Response of Ecosystem Service Value to Land Use Changes under Multiple Scenarios Coupling a Mixed-Cell Cellular Automata Model and System Dynamics Model in Xi’an, China. Ecol. Indic. 2023, 147, 110009. [Google Scholar] [CrossRef]
Li, F.; Yigitcanlar, T.; Li, W.; Nepal, M.; Nguyen, K.; Dur, F. Understanding urban heat vulnerability: Scientometric analysis of five decades of research. Urban Clim. 2024, 56, 102035. [Google Scholar] [CrossRef]
Yigitcanlar, T. Sustainable Urban and Regional Infrastructure Development: Technologies, Applications and Management; IGI Global: Hersey, PA, USA, 2010. [Google Scholar]
Koumetio Tekouabou, S.C.; Diop, E.B.; Azmi, R.; Jaligot, R.; Chenal, J. Reviewing the Application of Machine Learning Methods to Model Urban Form Indicators in Planning Decision Support Systems: Potential, Issues and Challenges. J. King Saud Univ.-Comput. Inf. Sci. 2022, 34, 5943–5967. [Google Scholar] [CrossRef]
Karanja, J.; Kiage, L. Perspectives on Spatial Representation of Urban Heat Vulnerability. Sci. Total Environ. 2021, 774, 145634. [Google Scholar] [CrossRef]
Intergovernmental Panel On Climate Change. Climate Change 2007: Impacts, Adaptation and Vulnerability; Intergovernmental Panel On Climate Change: Genebra, Switzerland, 2001. [Google Scholar]
Rinner, C.; Patychuk, D.; Bassil, K.; Nasr, S.; Gower, S.; Campbell, M. The Role of Maps in Neighborhood-Level Heat Vulnerability Assessment for the City of Toronto. Cartogr. Geogr. Inf. Sci. 2010, 37, 31–44. [Google Scholar] [CrossRef]
Depietri, Y.; Welle, T.; Renaud, F. Social Vulnerability Assessment of the Cologne Urban Area (Germany) to Heat Waves: Links to Ecosystem Services. Int. J. Disaster Risk Reduct. 2013, 6, 98–117. [Google Scholar] [CrossRef]
Hulley, G.; Shivers, S.; Wetherley, E.; Cudd, R. New ECOSTRESS and MODIS Land Surface Temperature Data Reveal Fine-Scale Heat Vulnerability in Cities: A Case Study for Los Angeles County, California. Remote Sens. 2019, 11, 2136. [Google Scholar] [CrossRef]
Crichton, D. The Risk Triangle. In Natural Disaster Management; Ingleton, J., Ed.; Tudor Rose: London, UK, 1999; pp. 102–103. ISBN 0-9536140-1-8. [Google Scholar]
Buscail, C.; Upegui, E.; Viel, J. Mapping Heatwave Health Risk at the Community Level for Public Health Action. Int. J. Health Geogr. 2012, 11, 38. [Google Scholar] [CrossRef]
Dugord, P.; Lauf, S.; Schuster, C.; Kleinschmit, B. Land Use Patterns, Temperature Distribution, and Potential Heat Stress Risk: The Case Study Berlin, Germany. Comput. Environ. Urban Syst. 2014, 48, 86–98. [Google Scholar] [CrossRef]
Paranunzio, R.; Dwyer, E.; Fitton, J.; Alexander, P.; O’Dwyer, B. Assessing Current and Future Heat Risk in Dublin City, Ireland. Urban Clim. 2021, 40, 100983. [Google Scholar] [CrossRef]
Luber, G.; McGeehin, M. Climate Change and Extreme Heat Events. Am. J. Prev. Med. 2008, 35, 429–435. [Google Scholar] [CrossRef]
Cheng, W.; Brown, R. An Energy Budget Model for Estimating the Thermal Comfort of Children. Int. J. Biometeorol. 2020, 64, 1355–1366. [Google Scholar] [CrossRef]
Alonso, L.; Renard, F. A Comparative Study of the Physiological and Socio-Economic Vulnerabilities to Heat Waves of the Population of the Metropolis of Lyon (France) in a Climate Change Context. Int. J. Environ. Res. Public Health 2020, 17, 1004. [Google Scholar] [CrossRef]
Hu, K.; Yang, X.; Zhong, J.; Fei, F.; Qi, J. Spatially Explicit Mapping of Heat Health Risk Utilizing Environmental and Socioeconomic Data. Environ. Sci. Technol. 2017, 51, 1498–1507. [Google Scholar] [CrossRef] [PubMed]
Kwon, Y.; Lee, D.; Kwon, Y. Is Sensible Heat Flux Useful for the Assessment of Thermal Vulnerability in Seoul (Korea)? Int. J. Environ. Res. Public Health 2020, 17, 963. [Google Scholar] [CrossRef]
Reid, C.; Mann, J.; Alfasso, R.; English, P.; King, G.; Lincoln, R.; Balmes, J. Evaluation of a Heat Vulnerability Index on Abnormally Hot Days: An Environmental Public Health Tracking Study. Environ. Health Perspect. 2012, 120, 715–720. [Google Scholar] [CrossRef]
Xiao, Y.; Wang, Y.; Yuan, Q.; He, J.; Zhang, L. Generating a Long-Term (2003−2020) Hourly 0.25° Global PM2.5 Dataset via Spatiotemporal Downscaling of CAMS with Deep Learning (DeepCAMS). Sci. Total Environ. 2022, 848, 157747. [Google Scholar] [CrossRef]
Loughnan, M.; Nicholls, N.; Tapper, N. Mapping Heat Health Risks in Urban Areas. Int. J. Popul. Res. 2012, 2012, 518687. [Google Scholar] [CrossRef]
Hammer, J.; Ruggieri, D.; Thomas, C.; Caum, J. Local Extreme Heat Planning: An Interactive Tool to Examine a Heat Vulnerability Index for Philadelphia, Pennsylvania. J. Urban Health 2020, 97, 519–528. [Google Scholar] [CrossRef] [PubMed]
Lammers, R.; Li, A.; Nag, S.; Ravindra, V. Prediction Models for Urban Flood Evolution for Satellite Remote Sensing. J. Hydrol. 2021, 603, 127175. [Google Scholar] [CrossRef]
Wei, G.; Zhang, Z.; Ouyang, X.; Shen, Y.; Jiang, S.; Liu, B.; He, B.-J. Delineating the Spatial-Temporal Variation of Air Pollution with Urbanization in the Belt and Road Initiative Area. Environ. Impact Assess. Rev. 2021, 91, 106646. [Google Scholar] [CrossRef]
Ronneberger, O.; Fischer, P.; Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. Lect. Notes Comput. Sci. 2015, 9351, 234–241. [Google Scholar] [CrossRef]
Yan, C.; Fan, X.; Fan, J.; Wang, N. Improved U-Net Remote Sensing Classification Algorithm Based on Multi-Feature Fusion Perception. Remote Sens. 2022, 14, 1118. [Google Scholar] [CrossRef]
Li, Z.; Demir, I. U-Net-Based Semantic Classification for Flood Extent Extraction Using SAR Imagery and GEE Platform: A Case Study for 2019 Central US Flooding. Sci. Total Environ. 2023, 869, 161757. [Google Scholar] [CrossRef]
Yigitcanlar, T. Rethinking Sustainable Development: Urban Management, Engineering, and Design; IGI Global: Hersey, PA, USA, 2010. [Google Scholar]
Stehman, S.V.; Hansen, M.C.; Broich, M.; Potapov, P.V. Adapting a Global Stratified Random Sample for Regional Estimation of Forest Cover Change Derived from Satellite Imagery. Remote Sens. Environ. 2011, 115, 650–658. [Google Scholar] [CrossRef]
Amani, S.; Shafizadeh-Moghadam, H. A Review of Machine Learning Models and Influential Factors for Estimating Evapotranspiration Using Remote Sensing and Ground-Based Data. Agric. Water Manag. 2023, 284, 108324. [Google Scholar] [CrossRef]
Wang, S.; Sun, Q.C.; Huang, X.; Tao, Y.; Dong, C.; Das, S.; Liu, Y. Health-Integrated Heat Risk Assessment in Australian Cities. Environ. Impact Assess. Rev. 2023, 102, 107176. [Google Scholar] [CrossRef]
Vu, H.L.; Ng, K.T.W.; Richter, A.; An, C. Analysis of Input Set Characteristics and Variances on k-Fold Cross Validation for a Recurrent Neural Network Model on Waste Disposal Rate Estimation. J. Environ. Manag. 2022, 311, 114869. [Google Scholar] [CrossRef]
Azhar, G.; Saha, S.; Ganguly, P.; Mavalankar, D.; Madrigano, J. Heat Wave Vulnerability Mapping for India. Int. J. Environ. Res. Public Health 2017, 14, 357. [Google Scholar] [CrossRef]
Guo, X.; Huang, G.; Jia, P.; Wu, J. Estimating Fine-Scale Heat Vulnerability in Beijing through Two Approaches: Spatial Patterns, Similarities, and Divergence. Remote Sens. 2019, 11, 2358. [Google Scholar] [CrossRef]
Shang, W.L.; Chen, Y.; Yu, Q.; Song, X.; Chen, Y.; Ma, X.; Chen, X.; Tan, Z.; Huang, J.; Ochieng, W. Spatio-Temporal Analysis of Carbon Footprints for Urban Public Transport Systems Based on Smart Card Data. Appl. Energy 2023, 352, 121859. [Google Scholar] [CrossRef]
Matthews, T.; Lo, A.Y.; Byrne, J.A. Reconceptualizing Green Infrastructure for Climate Change Adaptation: Barriers to Adoption and Drivers for Uptake by Spatial Planners. Landsc. Urban Plan. 2015, 138, 155–163. [Google Scholar] [CrossRef]
Mitchell, B.; Chakraborty, J. Landscapes of Thermal Inequity: Disproportionate Exposure to Urban Heat in the Three Largest US Cities. Environ. Res. Lett. 2015, 10, 115005. [Google Scholar] [CrossRef]
Ferreira, F.; Pereira, E.; Gonçalves, A.; Costa, R.; Bezerra, F. An Explicitly Spatial Approach to Identify Heat Vulnerable Urban Areas and Landscape Patterns. Urban Clim. 2021, 40, 101021. [Google Scholar] [CrossRef]
Wieland, M.; Pittore, M. Large-Area Settlement Pattern Recognition from Landsat-8 Data. ISPRS J. Photogramm. Remote Sens. 2016, 119, 294–308. [Google Scholar] [CrossRef]
Das, T.; Jana, A.; Mandal, B.; Sutradhar, A. Spatio-Temporal Pattern of Land Use and Land Cover and Its Effects on Land Surface Temperature Using Remote Sensing and GIS Techniques: A Case Study of Bhubaneswar City, Eastern India (1991–2021). GeoJournal 2021, 87, 765–795. [Google Scholar] [CrossRef]
Tao, C.; Jia, M.; Wang, G.; Zhang, Y.; Zhang, Q.; Wang, X.; Wang, Q.; Wang, W. Time-Sensitive Prediction of NO₂ Concentration in China Using an Ensemble Machine Learning Model from Multi-Source Data. J. Environ. Sci. 2023, 137, 30–40. [Google Scholar] [CrossRef]
Tella, A.; Balogun, A.-L.; Adebisi, N.; Abdullah, S. Spatial Assessment of PM10 Hotspots Using Random Forest, K-Nearest Neighbour and Naïve Bayes. Atmos. Pollut. Res. 2021, 12, 101202. [Google Scholar] [CrossRef]
Al-Aizari, A.R.; Al-Masnay, Y.A.; Aydda, A.; Zhang, J.; Ullah, K.; Islam, A.R.M.T.; Habib, T.; Kaku, D.U.; Nizeyimana, J.C.; Al-Shaibah, B.; et al. Assessment Analysis of Flood Susceptibility in Tropical Desert Area: A Case Study of Yemen. Remote Sens. 2022, 14, 4050. [Google Scholar] [CrossRef]
Singleton, A.; Arribas-Bel, D.; Murray, J.; Fleischmann, M. Estimating Generalized Measures of Local Neighbourhood Context from Multispectral Satellite Images Using a Convolutional Neural Network. Comput. Environ. Urban Syst. 2022, 95, 101802. [Google Scholar] [CrossRef]
Papadomanolaki, M.; Vakalopoulou, M.; Karantzalos, K. A Deep Multitask Learning Framework Coupling Semantic Segmentation and Fully Convolutional LSTM Networks for Urban Change Detection. IEEE Trans. Geosci. Remote Sens. 2021, 59, 7651–7668. [Google Scholar] [CrossRef]
Zhu, Y.; Geiß, C.; So, E.; Jin, Y. Multitemporal Relearning with Convolutional LSTM Models for Land Use Classification. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 3251–3265. [Google Scholar] [CrossRef]
He, Z.; Chow, C.Y.; Zhang, J.D. STCNN: A Spatio-Temporal Convolutional Neural Network for Long-Term Traffic Prediction. In Proceedings of the 2019 20th IEEE International Conference on Mobile Data Management (MDM), Hong Kong, China, 10–13 June 2019; pp. 226–233. [Google Scholar] [CrossRef]
Xiao, Z.; Li, H.; Jiang, H.; Li, Y.; Alazab, M.; Zhu, Y.; Dustdar, S. Predicting Urban Region Heat via Learning Arrive-Stay-Leave Behaviors of Private Cars. IEEE Trans. Intell. Transp. Syst. 2023, 24, 10843–10856. [Google Scholar] [CrossRef]
Yu, B.; Yin, H.; Zhu, Z. Spatio-temporal Graph Convolutional Networks: A Deep Learning Framework for Traffic Forecasting. arXiv 2017, arXiv:1709.04875. [Google Scholar] [CrossRef]
Xu, Z.; Li, J.; Lv, Z.; Wang, Y.; Li, H. A Prediction Method for Population Density in Key Areas. In Proceedings of the International Symposium on Artificial Intelligence and Robotics 2021, Fukuoka, Japan, 28 October 2021; SPIE: Bellingham, WA, USA; Volume 11884, pp. 13–28. [Google Scholar] [CrossRef]
Kim, Y.; Newman, G. Advancing Scenario Planning through Integrating Urban Growth Prediction with Future Flood Risk Models. Comput. Environ. Urban Syst. 2020, 82, 101498. [Google Scholar] [CrossRef] [PubMed]
Ma, Y.; Hu, Y.; Moncrieff, G.R.; Slingsby, J.A.; Wilson, A.M.; Maitner, B.; Zhou, R.Z. Forecasting Vegetation Dynamics in an Open Ecosystem by Integrating Deep Learning and Environmental Variables. Int. J. Appl. Earth Obs. Geoinf. 2022, 114, 103060. [Google Scholar] [CrossRef]
Dehghani, A.; Moazam, H.M.Z.H.; Mortazavizadeh, F.; Ranjbar, V.; Mirzaei, M.; Mortezavi, S.; Ng, J.L.; Dehghani, A. Comparative Evaluation of LSTM, CNN, and ConvLSTM for Hourly Short-Term Streamflow Forecasting Using Deep Learning Approaches. Ecol. Inform. 2023, 75, 102119. [Google Scholar] [CrossRef]
Maier, G.; Grundstein, A.; Jang, W.; Li, C.; Naeher, L.; Shepherd, M. Assessing the Performance of a Vulnerability Index during Oppressive Heat across Georgia, United States. Weather. Clim. Soc. 2014, 6, 253–263. [Google Scholar] [CrossRef]
Mushore, T.; Mutanga, O.; Odindi, J.; Dube, T. Determining Extreme Heat Vulnerability of Harare Metropolitan City Using Multispectral Remote Sensing and Socio-Economic Data. J. Spat. Sci. 2017, 63, 173–191. [Google Scholar] [CrossRef]
Khan, A.; Sudheer, M. Machine Learning-Based Monitoring and Modeling for Spatio-Temporal Urban Growth of Islamabad. Egypt. J. Remote Sens. Space Sci. 2022, 25, 541–550. [Google Scholar] [CrossRef]
Tuczek, M.; Degirmenci, K.; Desouza, K.; Watson, R.; Yigitcanlar, T.; Breitner, M. Mitigating urban heat with optimal distribution of vegetation and buildings. Urban Clim. 2022, 44, 101208. [Google Scholar] [CrossRef]
Liu, Y.; Yigitcanlar, T.; Guaralda, M.; Degirmenci, K.; Liu, A. Spatial Modelling of Urban Wind Characteristics: Review of Contributions to Sustainable Urban Development. Buildings 2024, 14, 737. [Google Scholar] [CrossRef]
Liu, Y.; Guaralda, M.; Yigitcanlar, T.; Limb, M.; Garcia-Hansen, V. Navigating urban climate design implementation challenges: Insights from Brisbane’s built environment experts. J. Urban Des. 2024, 1–20. [Google Scholar] [CrossRef]
Guo, F.; Zheng, R.; Zhao, J.; Zhang, H.; Dong, J. Framework of street grid-based urban heat vulnerability assessment: Integrating entropy weight method and BPNN model. Urban Clim. 2024, 56, 102067. [Google Scholar] [CrossRef]
Kitchley, J.L. A framework to assess the contextual composite heat vulnerability index for a heritage city in India-A case study of Madurai. Sustain. Cities Soc. 2024, 101, 105119. [Google Scholar] [CrossRef]
Wu, H.; Xu, Y.; Zhang, M.; Su, L.; Wang, Y.; Zhu, S. Spatially explicit assessment of the heat-related health risk in the Yangtze River Delta, China, using multisource remote sensing and socioeconomic data. Sustain. Cities Soc. 2024, 104, 105300. [Google Scholar] [CrossRef]
Wang, H.; Yang, J.; Chen, G.; Ren, C.; Zhang, J. Machine Learning Applications on Air Temperature Prediction in the Urban Canopy Layer: A Critical Review of 2011–2022. Urban Clim. 2023, 49, 101499. [Google Scholar] [CrossRef]

Figure 1. ML and RS integration framework for urban sustainability [47].

Figure 2. Framework of urban heat vulnerability analysis with ML and RS.

Figure 3. Likely differences of mapping urban heat vulnerability in census tracts and grid scale.

Figure 4. Flowchart of U-HEAT implementation.

Figure 5. Integrated vulnerability and risk indicator framework.

Figure 6. Visualization of a hypothetical urban heat vulnerability mapping.

Table 1. Frequently used indicators and categories in urban heat vulnerability [14].

Categories	Indicators	Descriptions	Data Sources
Socio-demographic characteristics	Age	% of population over 65, below 5 or in a specific range	Census and demographic data
	Economic status	% of population with high/low incomes; local financial status	Census and demographic data
	Social isolation	% of elderly population living alone or living in a group	Census and demographic data
	Education	% of population with a low education level	Census and demographic data
	Population density	Number of population/households per study unit	Census and demographic data, satellite imagery data
Health conditions	Personal illness status	% population with pre-existing physical/mental illness	Census and demographic data, health and medical data
	Medical infrastructure	Number of medical workers/facilities/institutions; or distance to medical institutions	Health and medical data, Google Maps
	Disability	% population with a disability	Census and demographic data, health and medical data
Environmental factors (natural)	Land surface temperature	Daytime/night-time land surface temperature	Satellite imagery data
	Vegetation cover	%/area of vegetation	Satellite imagery data
	Air temperature	Daytime/night-time mean/maximum/minimum air temperature	Meteorological data
Environmental factors (built)	Accessibility to cooling space	Area of or distance to green space/open space/water body/cooling facilities	Satellite imagery data and Google Maps
	Land cover/use	Area of developed urban land cover	Satellite imagery data
	Building information	Building density/height/type	Satellite imagery data

Table 2. Heat-related health conditions for validation indicators.

Type of Condition	Diseases	ICD-10 Codes
Direct Heat-Related Conditions	Heat Stroke	X30
	Dehydration	E86
	Hyperpyrexia	R50.9
Indirect Heat-Related Conditions (the impact of heat on pre-existing conditions)	Cardiovascular Diseases	I00-I99
	Respiratory Diseases	J00-J99
	Diabetes	E10-E14
	Renal Disease	N00-N29
	Nervous Disorders	G00-G99
	Cerebrovascular Disease	I60-I69
	Mental Health Conditions	F00-F99

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

Share and Cite

MDPI and ACS Style

Li, F.; Yigitcanlar, T.; Nepal, M.; Thanh, K.N.; Dur, F. A Novel Urban Heat Vulnerability Analysis: Integrating Machine Learning and Remote Sensing for Enhanced Insights. Remote Sens. 2024, 16, 3032. https://doi.org/10.3390/rs16163032

AMA Style

Li F, Yigitcanlar T, Nepal M, Thanh KN, Dur F. A Novel Urban Heat Vulnerability Analysis: Integrating Machine Learning and Remote Sensing for Enhanced Insights. Remote Sensing. 2024; 16(16):3032. https://doi.org/10.3390/rs16163032

Chicago/Turabian Style

Li, Fei, Tan Yigitcanlar, Madhav Nepal, Kien Nguyen Thanh, and Fatih Dur. 2024. "A Novel Urban Heat Vulnerability Analysis: Integrating Machine Learning and Remote Sensing for Enhanced Insights" Remote Sensing 16, no. 16: 3032. https://doi.org/10.3390/rs16163032

APA Style

Li, F., Yigitcanlar, T., Nepal, M., Thanh, K. N., & Dur, F. (2024). A Novel Urban Heat Vulnerability Analysis: Integrating Machine Learning and Remote Sensing for Enhanced Insights. Remote Sensing, 16(16), 3032. https://doi.org/10.3390/rs16163032

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Menu