Developing Layered Occlusion Perception Model: Mapping community open spaces in 31 China cities

Abstract

Community Open Spaces (COS) refer to the fine-grained and micro-open areas within communities that offer residents convenient opportunities for social interaction and health benefits. The mapping of COS using Very High Resolution (VHR) imagery can provide critical community-scale data for monitoring urban sustainable development goals (SDGs). However, the three-dimensional structure of COS often results in layered occlusion in two-dimensional satellite imagery, leading to the invisibility and fragmentation of ground COS features in VHR images. This study presents a novel Layered Occlusion Perception Model (LOPM) to address these challenges by accurately modeling and reconstructing the intricate layered structure of COS. Our approach involves the automatic generation of a comprehensive COS database and the joint training of a deep learning network to decompose occlusion relationships. The developed dual-layer map product, COS-1m, includes various elements and their coupled spaces, with a resolution of 1 m, covering 31 major cities in China. The results demonstrate that the proposed method achieved an overall accuracy of 86.39% and an average F1-score of 77.47% across these cities. COS-1m reveals that, on average, 60.51 km² of COS area per city is occluded, constituting 10.18% of the total COS area. This research advances the technology for layered monitoring of COS, fills a critical gap in community-scale SDG assessments by providing fine-grained COS data products, and offers valuable insights for urban planners and policymakers to promote more effective and sustainable urban development.