Improved estimation of lake storage variation via DEM reconstruction integrating elevation and morphological constraints

Abstract

High-quality Digital Elevation Models (DEMs) are essential for accurately estimating lake storage variations, but the high cost of in-situ measurements limits their large-scale application. Reconstructing DEMs from remote sensing-derived lake contours is a promising low-cost alternative. However, most existing methods primarily rely on elevation-matched contours and discard those without elevation data. Consequently, the loss of morphological constraints leads to incorrect topographic structures and significantly amplifies reconstruction errors where elevation data are sparse. To overcome this limitation, we proposed a valued and Non-valued contour–based Unified DEM reconstruction model (NUDEM), which integrates contours with elevations and those without elevations to exploit both elevation constraints and morphological constraints. Evaluated on four lakes in the Tibetan Plateau, NUDEM consistently outperformed five existing methods in DEM reconstruction, reducing the RMSE by approximately 60% compared to the second-best method, with an average RMSE of 0.93 m and a minimum RMSE of 0.15 m. For lake storage variation estimation, NUDEM improved R2 by about 30% over other DEM-based methods and nearly doubled the R2 of traditional hypsometric-curve-based methods. This method provides a low-cost, remote sensing-based solution for large-scale lake DEM reconstruction, offering significant potential for improving the accuracy of lake storage variation estimation.