Effects of Sampling and Interpolation Methods on Accuracy of Extracted Watershed Features

Abstract

High-accuracy watershed features can be used as hydrological parameters of a distributed hydrological model and a nonpoint source pollution model. A total of 3,863 elevation points in Haizi Watershed, China, are collected to produce a basic digital elevation model (DEM). Spline, kriging, and inverse distance weighting methods are selected to interpolate different sample sizes obtained by different sampling methods. A total of 105 DEMs are constructed to extract river networks, outfalls, and watersheds. The accuracies of watershed features are evaluated by using root-mean-square error, and the errors of outfall position, river network closure (crossings of the actual and the extracted river networks), river network density, and watershed area. The results show that the accuracies of DEMs and watershed features increase with an increase in sample size. Sample size, sampling method, and interpolation method have significant impacts on the accuracies of DEMs, outfall position, river network closure, and watershed area. Sample size is most important for deriving watershed features. The optimal combinations of sample size, sampling method, and interpolation method can improve the accuracies of watershed features.