Optimizing roofing strategies to mitigate urban climate risks using a deep learning-based surrogate method

The anticipated increases in urban climate vulnerabilities resulting from climate shifts have led to various mitigation efforts, such as adopting green or cool roofs in urban areas. Therefore, determining the most effective roofing strategies, including type and distribution, is crucial given associated costs. This preliminary assessment typically involves employing statistical or numerical models with diverse scenario simulations, which are computationally intensive and time-consuming. In this study, we introduce a deep learning-based surrogate model designed to optimize roofing strategies for urban climate risk mitigation in the Greater Seoul area, South Korea. First, we implemented the Weather Research and Forecasting model coupled with Urban Canopy Modeling (WRF-UCM) while assigning one type of roof (e.g., 100% green roof scheme or cool roof schemes ranging from 25% to 100%) to urban grids within the study region under the business-as-usual climate scenario (RCP8.5) during the period 2090-2099. Using the outputs from the WRF-UCM model, we calculated three objective indices (heat stress index, flash flood index, and wind speed index), and trained a deep learning algorithm, Multi-residual networks (Multi-ResNet), to construct a surrogate model of the WRF-UCM. To reduce the total number of scenarios, we applied the Mini Batch K-mean method to cluster 379 urban grids into nine. Afterwards, we generated multi-type roof scenarios by assigning each roof scheme to every urban cluster (four roof schemes across nine clusters, totaling 0.3M scenarios). We then employed the surrogate model to compute the three objectives (i.e., heat, flood, and wind) for each multi-type roof scenario. Finally, we present the optimal roof configurations lying on the Pareto front, reflecting the trade-offs among objectives including cost reduction, heat mitigation, flash flood prevention, and wind speed enhancement. Our results demonstrate the potential of deep learning-based surrogate models as an effective framework for urban planning to mitigate climate risks.

This study is supported by the National Research Foundation of Korea (NRF) grants funded by the Korean government (MSIT) (2020R1A2C2007670, 2020R1C1C1014886, and 2022R1C1C2009543) and Korea Environment Industry & Technology Institute (KEITI) through the R&D Program for Innovative Flood Protection Technologies against Climate Crisis funded by the Korean Ministry of Environment (MOE) (No. RS-2023-00218873).