Explainable AI for effective management of urban heat sources

Abstract

Abstract Many cities around the world face the challenge of polluted air, even when implementing restrictive heat source policies. Analyses of air flows often reveal that particulate matter is primarily carried from neighboring areas. Programs are frequently introduced to encourage the replacement of old, inefficient heat sources with cleaner alternatives, though the use of solid fuels is not always prohibited. In this paper, we analyze the case of Kraków as a testbed due to its unique conditions: strict regulations on heat sources within the city, a lack of such policies in surrounding municipalities, and a dense network of low-cost sensors (LCS) monitoring air quality. Additionally, precise data on the number and type of heating sources in municipalities neighboring Kraków is available. Moreover, tax revenue per capita for these municipalities was incorporated, providing additional socioeconomic context for emissions management. The purpose of the paper is to assess the impact of neighboring municipalities on particulate matter concentrations in Kraków, considering heat source replacement programs and tax revenue data. An analysis of program impacts on air quality is made, and a revision of the project formula addressing sustainability is recommended. To achieve these objectives, a Spatiotemporal Explainable AI (XAI) with Dynamic Time Warping (DTW), Random Forest (RF), and direct interpretability techniques, including H statistics and Accumulated Local Effects (ALE), were applied. This approach identified emission patterns which, combined with emissions data, facilitated developing a model supporting data-based management. The results highlight the necessity of integrating socioeconomic factors to design sustainable air quality policies.