A comprehensive generalizability assessment of data-driven Urban Heat Island (UHI) models

• DOI: 10.1016/j.scs.2023.104701
• Publication Date: 2023-06-07T19:53:59Z
• Authors: Monica Pena Acosta; Marco Dikkers; Faridaddin Vahdat...; João Santos; Andries G. Dorée
• Keywords: Urban Heat Island; Similarity (geometry)

Data-driven models serve as valuable tools for understanding and tackling the UHI phenomenon that can provide user-friendly platforms urban planners incorporating considerations in their decisions. This study aims to assess generalizability of data-driven at street-level resolution, particularly considering various similarity degrees contexts between training testing cities. Five cities from three countries were selected encompass a diverse range similarities this comparative study. Random Forest developed. The lowest-performing model has an R2 value 0.56 MAE 0.07, highest-performing 0.71 0.05. While these proved be accurate they trained for, cross-validation different revealed low capabilities, irrespective degree datasets. Small changes feature importance resulted significant variation derivation mechanisms behavior, which contributes models’ generalizability. findings research indicate universal mitigation strategies may not yield consistent outcomes worldwide, one-size-fits-all approach inefficient addressing UHI. Hence, it's vital tackle locally.