Deep learning-based spatial optimization of green and cool roof implementation for urban heat mitigation

Abstract

Intensifying urban heat extremes require efficient mitigation strategies; therefore, we propose a methodological framework for optimizing the implementation of urban green and cool roofs to reduce heat stress while maximizing their cost-effectiveness. In particular, we develop a surrogate model based on the deep learning algorithm Multi-ResNet, which is trained on data generated by the physically-based Weather Research and Forecasting model coupled with an urban canopy model (WRF-UCM). We applied this framework to the Greater Seoul region under the SSP585 climate scenario for 2090–2099 with projected 2100 land cover and evaluated 262,144 scenarios for cool and green roof allocation across 379 urban grids. Our results showed that, at the current cost of green roofs, the Pareto optimal scenario involves implementing cool roofs over 89.2 % of urban areas. This scenario would reduce the total effective heat stress index by 8.8 % compared to the business-as-usual scenario while decreasing costs by 19.6 %. We identified an optimal cost range of 117.4–146.1 $/m over 40 years for green roofs to become cost-effective and more widely adopted. Our approach demonstrates the potential of deep learning techniques to provide efficient quantitative assessments with lower computational demands (from 3561 h with the WRF-UCM to 72 h), potentially supporting climate-resilient urban building planning.