High‐Resolution Projections of Urban Heat Island Intensity in Greater Kuala Lumpur Using Machine Learning and CMIP6 GCMs

Abstract

Urbanisation has significantly intensified the urban heat island (UHI) effect, particularly in rapidly growing cities like Kuala Lumpur. This study develops 1‐km resolution projections of land surface temperature (LST) using a machine learning‐based downscaling approach that integrates CMIP6 global climate model (GCM) simulations with urbanisation projections under different shared socioeconomic pathways (SSPs). A random forest (RF) model was trained on 1‐km resolution MODIS LST data and urbanisation ratio datasets to estimate GCM‐simulated Earth skin temperature ( T s ) biases. The trained model was then applied to future urbanisation scenarios to project LST anomalies from 2030 to 2090. The analysis revealed a strong positive correlation of 0.785 between the urbanisation ratio and LST anomalies. The RF model accurately predicted LST anomalies with a root mean square error of 1.63°C and a Kling‐Gupta Efficiency of 0.76. The future projections of LST using the multimodel mean of all GCMs revealed an increase in average LST in Kuala Lumpur from 0.7°C (CI: 0.4°C–0.7°C) for SSP1‐2.6 to 1.3°C (CI: 1.2°C–2.9°C) for SSP5‐8.5. The projections of UHI showed an increase ranging from 1°C to 1.9°C for SSP1‐2.6 and from 2.5°C to 3°C for SSP5‐8.5 by 2090 compared to 2030. The results also highlight spatial heterogeneity in UHI expansion, with central urban zones experiencing the most significant warming. This study emphasises the necessity of integrating urban planning strategies, such as increased green spaces, improved urban design, and heat mitigation policies, to address the rising urban temperatures.

Bibliographic Information