Assessment of the Contribution of Potential Green Roofs to Ecological Network Optimization in High-Density Urban Areas

Urban Farm DB summary

This study, focused on the central urban area of Tianjin, China, presents a method for extracting potential green roofs and assessing ecological network reconstruction. It synthesizes four suitability characteristics — flat roof features, roof color, additional structures, and building height — into a deep learning model with attention mechanisms and multi-scale feature fusion to identify potential green roof areas, then proposes a framework based on a Least-Cost Path model for incorporating potential green roofs into urban ecological network reconstruction. The results identified 21,244 roofs with greening potential in central Tianjin, covering a total of 1,345.22 hectares (average 633 m², largest 24,927 m²), showing a "polycentric and dispersed" spatial distribution. Integrating potential green roofs with existing green space increased the number of source patches from 100 to 131 and total area by 302.35 hectares, with potential green roofs directly contributing 126.28 hectares.