Identifying local climate zones in two U.S. Midwestern cities: a comparison of methods

Accurately identifying heat exposures within cities is essential for addressing the Urban Heat Island (UHI) effect. Local Climate Zones (LCZs) provide a framework for classifying urban landscapes based on physical and morphological characteristics that influence heat exposure. Despite recent advances, including near-global LCZ classifications and cloud-based tools, methodologies vary widely and are often limited to large cities, with few applications in Midwestern U.S. cities. Moreover, the resolution and accuracy of these maps may be insufficient for local-scale decision-making, particularly in small and mid-sized cities. In this study, we focus on Bloomington and Indianapolis, Indiana, where we evaluate the utility of different datasets, including satellite imagery and urban form variables, and test whether incorporating neighborhood context using a kernel-based moving window improves LCZ classification accuracy. We then compare these results to two widely used large-scale LCZ models: (1) the Global Map of Local Climate Zones and (2) the cloud-based LCZ Generator with locally trained samples. We assess overall accuracy scores, analyze feature importance, and validate classifications against satellite-derived land surface temperature (LST) during an extreme heat event. Our findings highlight challenges unique to small and mid-sized cities and underscore trade-offs between accessibility, data requirements, and classification accuracy. By clarifying the strengths and limitations of these approaches, this work advances computational urban climate methods.