Time-series urban green space mapping and analysis through automatic sample generation and seasonal consistency modification on Sentinel-2 data: A case study of Shanghai, China

Abstract

Urban green space (UGS) is crucial for the vitality and sustainability of the urban environment. However, the current UGS extraction methods based on satellite images still face the problem of high sample costs and phenological interference, which leads to insufficient efficiency and accuracy in UGS results. In response, this study proposes a robust method for UGS mapping from time-series Sentinel-2 data by automatic sample generation and seasonal consistency modification. Specifically, temporal training samples were selected automatically through anomaly detection and probability filtering. Based on the annual UGS maps obtained by random forest classifier, the seasonal consistency modification approach considering phenological information and category interference is introduced to improve the accuracy of UGS mapping. The UGS maps of Shanghai from 2017 to 2022 extracted by the proposed method demonstrate an overall accuracy of 91.4% and a Kappa coefficient of 81.19%. This indicates that the proposed method can significantly enhance the efficiency and accuracy of extracting time-series UGS maps from Sentinel-2 data. The dynamic results also highlight the spatiotemporal patterns of UGS in Shanghai from 2017 to 2022, offering valuable insights for sustainable urban development.