Effects of spatial scale on water surface delineation with satellite images

Abstract

Lake area is an important indicator for lake-related research. Accurate extraction of water surface from remote sensing images with various spatial resolutions still remains unsettled. To investigate the effects of spatial resolution on lake surface monitoring,two Landsat ETM +( Enhanced Thematic Mapper Plus) images were acquired with one at high water level and the other at low water level. These images were resampled to lower resolutions using Nearest Neighboring( NN) and Pixel Aggregation( PA) methods,and extracted for water surface area using NDWI( Normalized Difference Water Index) thresholding method. Variation trend and error distribution of water surface area with spatial resolution were thoroughly investigated. In addition,the influence of different factors on water surface extraction was discussed. The main conclusions are as follows:( 1) the accuracy decreases gradually at lower spatial resolution relative to 30 m resolution. However,the overall accuracy is higher with a minimum of 67. 64%;( 2) NN has minor impact on DN values,whereas PA reduces mean digital number values and standard values. The reduction is gradual and smoother than NN; and( 3) the threshold varies for PA and remains stable for NN. Thus,use of threshold determined on 30 m image should introduce larger errors for PA-resampled images than NN-resampled images. This study provides invaluable guidance for water surface mapping using remote sensing methods under global changes.