Accurate estimation of surface water volume in tufa lake group using UAV-captured imagery and ANNs

Abstract

Quantitative acquisition of surface water volume of tufa lake group is significant for protecting the tufa landscapes. Limited by low spatial resolution, satellite remote sensing failed to extract the surface water distribution of small and scattered tufa lakes. In recent years, Unmanned Aerial Vehicle (UAV) remote sensing has been increasingly applied to survey aquatic resources and environments, posing great potential for accurately measuring surface water volume of tufa lake group. In this paper, taking the Shuzheng Lakes in Jiuzhai Valley of China as the study site, we used UAV-captured images and Artificial Neural Network (ANN) models to precisely estimate the surface water volume of the tufa lake group. Digital Surface Model (DSM) and orthomosaic were first generated using aerial triangulation. Next, the two classification networks were trained to separate water from vegetation, and then the water extents of individual tufa lakes were extracted. Subsequently, the water depth map of each tufa lake was retrieved using the transferred shallow and deep bathymetry networks. Lastly, the total surface water volume of tufa lake group was achieved by summing the products of pixel unit area and the water depth of each pixel. As results, the surface water volume of tufa lake group estimated by the optimum model combination was approximately 181,360 m3, and the corresponding R-squared was 0.92. These demonstrated the effectiveness of utilizing UAV-based remote sensing and integrated ANN models to accurately estimate surface water volume of tufa lakes.