Tom Narock , Anna Kulikova, Sakib Hussen, Marcellus Mwangi, Abdul Siam

Urban Heat Islands (UHIs) are areas in cities that experience higher temperatures than surrounding areas due to construction features such as buildings, roads, and a general lack of vegetation. UHIs, which pose a threat to public health while also increasing energy usage, are often defined using land surface temperatures. Our study demonstrates how a community sensor network from the Baltimore Social-Environmental Collaborative (BSEC) can be paired with advances in machine learning to provide near real-time predictions of UHIs with uncertainty estimates. While prior research has investigated the use of neural networks in UHI assessment, those efforts did not include uncertainty quantification. We demonstrate how uncertainty quantification can be incorporated into UHI prediction in Baltimore, Maryland. We show how a machine learning approach using community weather stations compares to ground-truth values from Landsat 9 spacecraft measurements in summer 2023. Moreover, we show how the capturing of different types of uncertainty are informative for future smart city planning. This information can empower city planners and residents to proactively address the risks associated with UHIs, potentially reducing heat-related illnesses.