A higher accuracy forest cover product of China by fusing heterogeneous forest-related datasets using Dempster-Shafer theory

Abstract

Forests are of vital importance in maintaining the balance of ecosystems, and accurate forest cover maps provide basic data for research. Existing forest cover products suffer from substantial discrepancies in forest definition, area estimation, statistical accuracy, and spatial consistency, complicating their use. This paper uses the Dempster-Shafer theory to create a precise forest cover map of China using forest datasets from 2020. The basic probability assignment (BPA) function calculation adopts a combination of global constraints and local adaptive approach. The experimental results reveal that the accuracy of the fused forest cover map (FFCM) is significantly higher than that of the reference maps, with 91.62% user accuracy, 91.61% producer accuracy, 91.61% F1 score, and 94.65% overall accuracy. The area of China’s forests in 2020 is estimated at 229.43 million ha. The higher-precision forest cover map provides reliable data for research and a reference for the use of existing products.

Keywords

• Forest cover
• Data fusion
• Dempster-Shafer theory
• LULC products