A Deep Learning-Driven Surrogate Modeling of Optimizing Roofing Strategies for Climate-Resilient Cities

In response to escalating urban climate challenges due to climate change, in this study, we propose an innovative deep learning-based surrogate model to mitigate climate risks by optimizing roofing strategies, including roof type and distribution. We utilized the Weather Research and Forecasting model coupled with Urban Canopy Modeling (WRF-UCM) to generate sample data for various roof schemes (e.g., green roof or cool roofs ranging from 25% to 100% coverage) under the business-as-usual climate scenario (RCP8.5) for the last decade of this century (2090-2099) in the Greater Seoul area, South Korea. After training four deep learning algorithms (UNet, UNet++, UNet3+, and Multi-ResUNet) on WRF-UCM outputs and comparing their performance, we developed a surrogate model based on the best-performing algorithm, which calculates three objective indices: heat stress, flash flood, and wind circulation. The study area was clustered into nine groups using the Mini Batch K-mean method, and a total of 0.3M multi-type roof scenarios were generated by assigning each roof scheme to each cluster. For each scenario, the surrogate model was employed to compute three target objectives, and we also calculated the total cost. The optimal roof configurations were identified among the Pareto front, based on the trade-offs among cost savings, heat mitigation, flash flood reduction, and wind circulation enhancement. This approach using a deep learning-based surrogate model is expected to provide an efficient, agile tool for urban planners and policymakers to address climate risks.

This study is supported by the National Research Foundation (NRF) of Korea grants funded by the Korean government (MSIT) (2022R1C1C2009543, RS-2022-NR072388) and the Basic Science Research Program through the NRF of Korea, which was funded by the Ministry of Science, ICT and Future Planning (RS-2024-00456724).