Mapping heat vulnerability in Australian capital cities: A machine learning and multi-source data analysis

Heat vulnerability has emerged as a global concern amidst ongoing urbanisation and climate change. While numerous studies have examined heat vulnerability, a gap remains in the application of machine learning to this field. This study aims to address this gap by evaluating the effectiveness of various machine learning algorithms in assessing heat vulnerability across Australian capital cities using heat-related indicators. The findings reveal that: (a) The Random Forest algorithm outperforms the others, achieving a training R2 of 0.9179 and a testing R2 of 0.9089, indicating its superior performance in assessing heat vulnerability in Australian capital cities; (b) The spatial analysis reveals significant regional disparities, with higher vulnerability in densely populated urban areas and lower vulnerability in green, less developed suburban and rural areas, necessitating tailored heat mitigation strategies; (c) Heat vulnerability analysis reveals that Australian Capital Territory (ACT) and Greater Darwin (GDRW) have the lowest proportions of highly vulnerable Statistical Area Level 1 (SA1) units, whereas Greater Hobart (GHBA) and Greater Adelaide (GADL) have the highest ones, a clear indication of significant regional disparities, again pointing to tailored mitigation and adaptation strategies; and (d) The sensitivity analysis reveals that personal health conditions and socio-demographic characteristics, such as personal illness status, age, and education level, play dominant roles in determining heat vulnerability, overshadowing the impact of environmental and infrastructural factors. This research provides a comprehensive understanding of heat vulnerability in Australian capital cities and offers valuable insights for urban planners and policymakers to develop data-driven mitigation and adaptation strategies for enhanced urban sustainability and climate resilience.