Integrating spatial proximity analysis and explainable machine learning to evaluate building energy retrofit potential under urban heat island effects

A comprehensive study of the relationship between climate change, urban morphology, and building performance is essential for advancing urban building energy modeling (UBEM) and achieving the global sustainability goal of carbon neutrality. However, understanding the impact of urban heat island (UHI) effects on building energy retrofit potential during the early stages of urban building design remains a challenge. To predict, assess, and enhance the thermal performance of building envelopes and the retrofit potential of photovoltaic green roofs for nearly a hundred representative buildings in Philadelphia, USA, this study proposes an integrated framework that combines geographic information systems (GIS), physics-based simulation modeling, and machine learning. Urban morphology characteristics are defined using the triangulated irregular network (TIN) method in spatial proximity analysis and implemented in batch UBEM. The Urban Weather Generator (UWG) is employed to adjust typical meteorological year (TMY) data, ensuring that UHI effects are accounted for in energy simulations. The energy retrofit potential is quantified as net energy consumption variations induced by different parameter combinations. A data-driven SHapley Additive exPlanations (SHAP) approach is applied to construct a reliable predictive surrogate model.

The results yield three key insights:
(1) combining spatial proximity analysis with explainable machine learning markedly improves the accuracy of retrofit potential quantification under dynamic UHI scenarios;
(2) among various retrofit parameters, window thermal characteristics exert the most substantial influence on improving energy efficiency; and
(3) neglecting UHI effects can result in significant inaccuracies, notably causing an overestimation of energy-saving potential by approximately 25%.

These findings reveal the intricate interactions between the urban environment and building energy performance, providing practical and actionable insights for sustainable urban planning and building design.