Wetland Mapping Using Subpixel Analysis and Decision Tree Classification in the Yellow River Delta Area

A lot of wetlands exist in the Yellow River Delta area, Shandong China and exhibit significance for the delta ecological environmental evolution and animal habitat protection. It is important for wetland protection and management to accurately map the wetland distribution by remote sensing due to the fact that the estuary area changes very fast. This study tries to combine subpixel analysis of remotely sensed data with decision tree classification to improve the wetland mapping in the delta area. A two-step classification technique was proposed trying to make use of a priori knowledge of wetland compositions. First, a subpixel analysis technique was applied to remotely-derived reflectance data using in-site spectra to extract abundance images for each component material of the wetlands. A suitable set of endmembers were selected for the unmixing analysis. A classification scheme of wetlands was then created and used to classify the abundance images into various wetlands and non-wetlands. The approach combines quantitative analysis (subpixel unmixing) with qualitative decision (classification), and incorporates with human knowledge during the whole process of analysis. An ASTER image of surface reflectance was acquired for the study area and used to evaluate the proposed approach. The approach was implemented in RSI ENVI/IDL environment. Twelve wetland and non-wetland covers were identified using the technique and the result was then assessed with field verification and the land use data that was acquired by other method. The overall accuracy of the wetland mapping is about 88.7% and the Kappa coefficient is 0.86. ∗ Corresponding author. Email: [email protected]; Address: Rm 427, Remote Sensing Building, Peking University, 100871