LANDSCAPE CHANGES OF THE YANCHENG WETLAND IN THE PAST 20 YEARS: EVIDENCE FROM RANDOM FOREST

To study wetland landscape and its dynamic changes can effectively monitor the changes in wetland ecosystem. In this paper, we selected the core wetland of the Yancheng National Rare Birds Nature Reserve in Jiangsu Province as a research focus, and used the methods of maximum likelihood, CART decision tree and random forest to extract information from the core area of the wetland. Our results suggest that the method of random forest has the highest accuracy. The overall classification accuracy is 91.02% with a Kappa coefficient up to 0.89. The method of random forest is adopted as the mean to extract landscape information from the entire coastal wetland of Yancheng and obtained are the six landscape categories of wetland in 1995, 2006, and 2016 respectively. The annual changes of the six types of landscapes are analyzed by transfer matrix. The results show that the area of Yancheng wetland is increasing in an overall trend and the aquaculture areas is continuously expanding to shallow waters. Although the vegetation, construction areas and crop areas were all increased during the period of 1995—2006, they had basically remained stable during the latter period of 2006—2016, especially the areas of nature reserves were kept unchanged.