Accuracy comparison and analysis of methods for water area extraction of discrete lakes

Abstract

Remote sensing is a widely used technology for water surface monitoring. Currently,there exists no general method for water surface extraction at a dispersed lake,which affects the accuracy of lake area monitoring. For investigation of different methods,this paper uses ALOS multi-spectral data to extract lake surface in Lake Poyang. The reference lake area is obtained by unsupervised classification( ISODATA,Iterative Self-Organizing Data Analysis Technique Algorithm) based on fused 2. 5 m ALOS multi-spectral data. Subsequently,three methods are investigated based on 10 m ALOS data,including Normalized Difference Water Index( NDWI) thresholding,NDWI-based ISODATA method and Near Infrared( NIR)-based ISODATA method. Differences between these extractions are analyzed for intrinsic reasons. The main findings are:( 1) all three methods can be used for water surface identification,ISODATA method provides more detailed information and larger water surface area;( 2) NDWI-based method performs better than NIR-based method and the extraction accuracy increases;( 3) NDWI-based ISODATA gets the most reliable result with R2= 0. 993 and RMSE = 0. 115 km2,followed by NIR-based ISODATA with R2= 0. 992 and RMSE = 0. 144 km2,and NDWI thresholding with R2= 0. 938 and RMSE = 0. 443 km2. The results provide important references for selection of method and data source for water extraction of lake in dry season.