Mapping potential wetlands by a new framework method using random forest algorithm and big Earth data: A case study in China's Yangtze River Basin

Abstract

Mapping potential wetlands provides a promising approach to get such information rapidly, and thus is of great significance to understanding ecosystem sustainability and support wetland conservation and restoration. This study proposed a new processing pipeline to map potential wetlands in the Yangtze River Basin, the largest basin in China, by combining a random forest (RF) algorithm and an indicator system constituted by several indicators, including vegetation, soil, terrain, and climatic features. Results reveal that slope, annual precipitation (APRE), digital elevation model (DEM), normalized difference vegetation index (NDVI), and annual mean temperature (AMT) are the most important variables affecting the distribution of potential wetlands, with a relative importance value of 7.5 %, 5.9 %, 5.5 %, 5.2 %, and 5.2 %, respectively. Mapping potential wetlands in the Yangtze River Basin was achieved using the RF model with overall accuracy of 79.31 % and Kappa coefficient of 0.58. The estimated total area of potential wetlands in this basin is approximately 39,677 km2, mainly distributed in the Yalong River watershed, the Dongting Lake watershed, and the regions bordering main streams of the Yangtze River. The proposed approach in this study evidenced its generalizability in terms of the good accuracy and distribution consistency with the natural wetlands observed from satellites and field investigation. We expect that this approach can be further used to generate potential wetland datasets at a broader scale in a long time series and benefit the evaluation of Sustainable Development Goals (SDGs).

Keywords

• Potential wetland
• Random forest
• Sustainable Development Goals (SDGs)
• The Yangtze River Basin
• China