Session: 02-01: AI for Energy Sustainability I

Paper Number: 156691

156691 - Quantifying and Zoning Urban Heat Island Effects Using Unsupervised Machine Learning 

Abstract: 

The Urban Heat Island (UHI) effect, characterized by elevated urban temperatures compared to rural areas, poses critical challenges for urban sustainability, public health, and energy management. This study investigates the UHI effects in Maricopa County, Arizona, leveraging a simulation-based approach that integrates large-scale building energy modeling with advanced spatial and machine learning analyses. By utilizing the Automatic Building Energy Modeling (AutoBEM) software suite and the Model America database, we conducted energy simulations for approximately 1.35 million buildings, providing detailed insights into their energy consumption and anthropogenic heat emissions.

Our methodology incorporates multi-scalar spatial analysis, examining emissions at individual building levels, cluster zones derived from K-means clustering, and broader zip code-based aggregations. Anthropogenic Emission Density, a newly developed metric that normalizes emissions by building footprint area, offers a refined lens for understanding emission intensity. Using unsupervised machine learning, we identified high-emission zones, highlighting spatial heterogeneity in UHI contributors. Results reveal that certain zones and building types, such as hospitals and large hotels, disproportionately contribute to the UHI effect due to their higher average emissions.

The spatial analysis was enhanced through clustering and zip code-based heat mapping. K-means clustering facilitated the identification of urban emission zones based on longitude, latitude, and emission density, unveiling emission hotspots and enabling data-driven rezoning. Complementary zip code-level analyses provided a macroscopic view, validating clustering results and aligning them with urban planning contexts. These methods revealed significant variation in emission patterns, with residential buildings dominating total emissions due to their volume, while large hotels and hospitals exhibited higher per-building emissions.

Key findings emphasize the impact of building attributes, such as type, vintage, and energy systems, on emission levels. Older buildings with outdated standards showed higher emissions, underscoring the need for targeted retrofitting strategies, particularly in high-emission zones. The study also highlights the positive impact of recent energy standards in reducing emissions from newer buildings, demonstrating the value of stringent building codes and energy-efficient technologies.

A key outcome of this research is the development of an interactive web-based platform that visualizes building-level emissions and facilitates stakeholder engagement. The platform enables users to explore the spatial distribution of anthropogenic emissions, evaluate retrofitting options, and assess cost-effectiveness in mitigating UHI effects. Such tools are vital for urban planners and policymakers aiming to create equitable, data-driven energy policies and sustainable urban designs.

This research establishes a robust framework for UHI analysis, integrating energy modeling, spatial analytics, and machine learning to address the socio-spatial complexities of urban heat emissions. It provides actionable insights for reducing UHI impacts through strategic interventions, offering a replicable approach for cities worldwide to enhance urban resilience and sustainability.

Presenting Author: Shovan Chowdhury Oak Ridge National Laboratory

Presenting Author Biography: Shovan Chowdhury is a data scientist with expertise in mechanical and energy engineering. He holds a position as a Research Fellow in the Grid-Interactive Controls group at the Building Technologies Research and Integration Center (BTRIC) of Oak Ridge National Laboratory. Prior to this role, he earned a master’s degree in Mechanical Engineering from Idaho State University. At ORNL, Shovan focuses on large-scale building energy modeling and simulation using the AutoBEM software suite. His current projects involve the development of the AutoBEM software suite and the Model America dataset. He specializes in handling big data and implementing machine learning techniques for building energy modeling. Additionally, he has developed future typical meteorological year (fTMY) files for every Shared Socioeconomic Pathway (SSP) and Representative Concentration Pathway (RCP) scenario for each county in the United States. Shovan contributed to key data releases including data and energy models for 141.5 million U.S. buildings.