A framework for fine classification of urban wetlands based on random forest and knowledge rules: taking the wetland cities of Haikou and Yinchuan as examples

Abstract

ABSTRACT Urban wetlands play an important role in sustainable urban development. A wetland city recognized by the Ramsar Convention is a city with a remarkable ecological performance of urban wetlands and can provide an example for urban wetland protection. Therefore identifying the detailed characteristics of wetlands is of great value to the conservation of urban wetlands. However, there is no detailed wetland type extraction for wetland cities using multisource remote sensing data. This paper proposes a framework for fine- urban wetland extraction with multisource and multiperiod remote sensing data, which mainly includes Sentinel 1/2 and Landsat 8 data, and obtains the detailed wetland type data containing 14 wetland categories at a 10-m resolution. The classification methods of random forests were used to extract land cover types, and knowledge rules were used to classify detailed wetland types. The wetland cities, Yinchuan, and Haikou, located in southern and northern China, respectively, were chosen as case studies. The results indicate that the overall accuracy and wetland accuracy of the two cities in 2015 and 2020 were greater than 88% and 87%, respectively. The detailed wetland mapping results were more accurate and had a high resolution, and wetland categories were more diverse compared with other wetland-related products. According to the classification results, the wetland area in Haikou was 326.86 km2 (13.43%) in 2015 and 308.58 km2 (12.66%) in 2020. The wetland area in Yinchuan was 80.39 km2 (4.46%) in 2015 and 74.15 km2 (4.12%) in 2020. The fine urban wetland classification framework we propose and the detailed wetland results could serve for the creation of wetland cities and support the protection and rational use of urban wetlands.