Weakly supervised mapping of old and renewed urban areas in China during the recent two decades

Abstract

China has progressively elevated old city transformation and urban renewal to a policy priority, positioning them as new endogenous drivers of urban development. It raises the demand for real-time insight into the spatiotemporal distribution of old and renewed urban areas. We propose a weakly supervised mapping framework with adaptive adjustments city by city without relying on high-precision training samples. It is also convenient for variable spatial range and study period. We combined Landsat imagery during 2000–2021, LandTrendr change detection algorithm and Simple Non-Iterative Clustering image segmentation into a Threshold Voting approach. The overall accuracy and Kappa coefficient of our results are 78.37 % and 0.57, respectively, with interesting global and local patterns. Old urban areas cluster in the early developed city centers, accounting for 22.55 % nationwide, usually interspersed and surrounded with renewed urban areas (77.45 %). Our mapping framework provides an efficient and flexible scheme for ground history detection, and the related results can be applied as helpful references for urban renewal field work in China.