Assessing Influence of Urban Scene Characteristics on Urban Heat Island: Interpretable ML in NYC

Abstract

Examines how street-level urban scene characteristics relate to urban heat island intensity across New York City, using interpretable machine learning so that the contribution of individual scene attributes can be read from the fitted model. Reported findings include a canopy threshold — land surface temperature falls consistently once tree canopy coverage exceeds roughly 28% — and about 0.40 °C of cooling per additional building floor, strongest where built form and greenery occur together. Heat exposure is distributed unevenly, with high-poverty neighbourhoods facing elevated exposure through low vegetation cover.