Exploring air temperature variability and socio-demographic inequalities in heat exposure through machine learning: A case study of Maricopa County, Arizona

Abstract

This study investigates the dynamics of heat exposure in Maricopa County, Arizona, employing a multidimensional approach. Utilizing the Extreme Gradient Boosting machine learning model, we predict census block group (CBG) level air temperatures, revealing the significant influences of land surface temperature (LST) (23 % importance ‘Gain’) and elevation (28 % importance ‘Weight’) on air temperature variability. Even though LST is an important predictor of air temperature variation, for ∼90.0 % of CBGs, LST variations are not associated with air temperature variation in a statistically significant way; rather other relevant factors such as impervious surface, and vegetation played significant roles. Among the minority populations, the Hispanic/Latinx populations are highly exposed to elevated air temperature. There are 181 CBGs with positive association between Hispanic/Latinx and air temperature, based on the local statistical significance test from Geographically Weighted Regression modeling. The study demonstrates the importance of considering local topography, and land use/land cover patterns to characterize UHI and considering socio-demographic characteristics in assessing spatial variation of heat exposure. By addressing socio-demographic disparities in heat exposure, this research contributes valuable insights for urban planning, public health interventions, and climate resilience efforts in Maricopa County, with methods and findings that are widely transferable.