Optimized LCZ mapping with automated machine learning reveals thermal disparities during a heatwave event

Source abstract

Cities are increasingly challenged by intensifying extreme heat, underscoring the need for accurate characterization of urban thermal environments. Urban climate modeling is central to this, yet its fidelity is often limited by the complexity and heterogeneity of urban landscapes. The Local Climate Zone (LCZ) scheme provides a standardized framework for describing urban morphology and has enhanced the representation of urban areas in numerical models. However, the widely used LCZ mapping framework developed by the World Urban Database and Access Portal Tools (WUDAPT) uses an algorithm with fixed hyperparameters across all study areas. This one-size-fits-all strategy promotes general usability but often leads to only moderate accuracy. Such limitations can propagate into urban climate simulations and ultimately hinder the representation of the urban thermal environment, yet this impact remains insufficiently explored. To address these gaps, we present an automated machine learning-driven LCZ classification approach that enhances the WUDAPT framework by generating locally optimized LCZ maps. Using Greater Manchester (GM) as a testbed, we investigate how locally optimized LCZ maps influence urban climate simulations under extreme heat conditions. Results show that the proposed method consistently outperforms the WUDAPT approach in classification accuracy. Discrepancies in LCZ maps affect urban climate simulations in a scale-dependent manner, with the strongest impacts at the sub-seasonal level. The locally optimized LCZ maps enhance simulation accuracy and reveal critical spatial thermal disparities. This approach exposes overlooked differences in population heat exposure during a heatwave event in GM, offering valuable insights for climate-responsive urban planning and targeted heat mitigation strategies.