Land cover mapping of China based on an improved bi-temporal comprehensive change detection method from Landsat MSS observations

Abstract

The limited availability of satellite data during the early period of satellite-based land cover mapping, has led to land cover datasets from before 1985 being scarce. In this study, we developed a novel 60 m land cover dataset with a fine classification system for China in 1980, termed CLC_FCS60_1980, based on the improved bi‐temporal Comprehensive Change Detection Method (CCDM) and the GLC_FCS30D product. First, an improved CCDM was proposed to identify temporally stable and changed pixels in the early 1980s. Subsequently, high-confidence samples were obtained from temporally stable pixels and used to train a local adaptive random forest classifier to classify changed pixels by tile, with post-classification processing rules applied to mitigate spurious changes between 1980 and 1985. Finally, the CLC_FCS60_1980 dataset was validated using 6,550 ground truth points, achieving an overall accuracy (OA) of 85.56% (±0.87%). Statistical analysis identified minimal land cover changes in China between 1980 and 1985, accounting for only 0.82%. Specifically, forest loss and cropland gain dominated, with a net decrease of 32,100 km2 (−1.32%) in forest and an increase of 30,500 km2 (+1.64%) in cropland. Hence, the proposed method provides an efficient and practical solution for national-scale land cover mapping during the early satellite era.

Affiliated Institutions

• Xi'an University of Science and Technology CN
• Beijing Institute of Big Data Research CN
• University of Chinese Academy of Sciences CN
• Chinese Academy of Sciences CN
• Nanjing University CN
• Aerospace Information Research Institute CN

Related Publications