Simulation and Prediction of Daytime Surface Urban Heat Island Intensity Under Multiple Scenarios Via Deep Neural Network

Abstract

The intensiffcation of the Surface Urban Heat Island (SUHI), driven by urbanization, land use and land cover (LULC) changes, and population growth, poses signiffcant environmental and public health risks in urban areas. Simulating and predicting SUHI, particularly by identifying future high SUHI intensity (SUHII), has been recognized as a critical step in mitigating SUHI. This study employs a Fully Convolutional Neural Network (FCNN) model to simulate current daytime SUHI across three sites in Singapore, utilizing 15 essential independent variables identiffed by previous studies. The model demonstrates high simulation accuracy, achieving 94.4%, 84.3%, and 79.3% for the respective sites. Three projection scenarios, based on projected population and LULC changes, predict a decrease in high SUHII by 2.6%, 2.8%, and 5.4% for the sites, attributed to LULC enhancements proposed in the 2019 Master Plan. Spatial analysis of predicted SUHII maps indicates consistent locations of SUHII classes across scenarios, whereas some high SUHII areas are detected emerging in new locations through quantitative analysis. These ffndings offer valuable insights for urban planners to implement preemptive mitigation measures, addressing potential high SUHII hotspots. Future research should expand input data samples and explore alternative models to address current study limitations.

Keywords: Surface Urban Heat Island, Deep Neural Network, simulation, Multi-scenario prediction, Remote Sensing