Authors: Qi Dong*, Beijing Normal University, Xuehong Chen, Beijing Normal University, Jin Chen, Beijing Normal University, Dameng Yin, Institute of Crop Sciences, Chinese Academy of Agricultural Sciences/Key Laboratory of Crop Physiology and Ecology, Ministry of Agriculture, Chishan Zhang, Beijing Normal University, Fei Xu, Division Forest, Nature and Landscape, KU Leuven
Topics: Remote Sensing, Land Use, Quantitative Methods
Keywords: land cover/use mapping, subpixel mapping, area estimation correction, abundance-dependent errors, probability distribution of abundances
Session Type: Virtual Paper
Start / End Time: 8:00 AM / 10:20 AM
Room: Virtual 29
Presentation File: No File Uploaded
With the increasingly widespread use of subpixel mapping methods in land cover/use mapping at large spatial scales or in historical periods, more accurate area information is often required for the target land cover types in study regions. However, the extensive existence of area bias, an inevitable but long-ignored issue in subpixel land cover mapping, poses a great challenge to satisfying this requirement. To enhance the understanding of the vital issue, this work proposed a theoretical framework for the generation of the area bias counted from the subpixel mapping results. The framework pointed out that the area bias generally arises from two terms: the abundance-dependent estimation errors and the probability distribution of estimated abundances, which act as a source of error and a weighting function to adjust error source contributions, respectively. Accordingly, we developed a two-term method (TTM) for area bias correction, which estimates the two terms accurately by combining the subpixel map and a stratified random sampling survey. Three validation experiments were conducted to correct the biased areas counted from various subpixel maps by different subpixel mapping methods, with different spatial resolutions, and of regions with different spatial structures, respectively. The results showed that the area biases varied from zero to approximately 20% and TTM effectively corrected the biased areas within the defined uncertainty of 1%, demonstrating the effectiveness and the generality of TTM in various application scenarios.