Spatial Optimisation of Local Climate Zones in Kuala Lumpur for Surface Urban Heat Island Mitigation Using ResNet and Evolutionary Algorithms

Abstract Land Use & Land Cover (LULC) optimisation is critical for mitigation of Urban Heat Island (UHI) effects. We use the Local Climate Zones (LCZ) classification system to optimise LULC as it has achieved very strong results in Surface UHI intensity (SUHII) modelling. Two evolutionary algorithms are employed to minimise extreme SUHII in Greater Kuala Lumpur: Genetic Algorithm, and Particle Swarm Optimisation, with total population and GDP set as constraints that should not decrease. Eight local administrative areas LCZ spatial layouts’ were each optimised for 1500 iterations using a trained ResNet model as the objective function to simulate the resulting SUHII patterns, providing an optimal LCZ spatial layout of each area using each algorithm. Genetic Algorithm results showed an average reduction in extreme SUHII of 2.85℃ and a total altered area of 18.76% of the total area, while Particle Swarm showed a slightly better reduction of 3.03℃ and a much smaller altered area of 7.95% indicating better efficiency. Biggest reductions in SUHII happened in Gombak while the smallest reductions happened in Sepang. Three-dimensional urban morphology blocks were generated based on the optimised LCZ layouts using a generative adversarial network illustrating the differences between the original and optimised layouts for two representative areas. These findings provide local governments with surrogate-evaluated candidate configurations that identify potential areas and spatial strategies for adapting land-use planning maps to support UHI mitigation, subject to further evaluation of planning feasibility and implementation constraints.