ABSTRACT Spatial distribution map of water depth inversion Accurate estimation of pond water storage (PWS) is essential for water resource management, ecological conservation, and agricultural planning. However, there is no systematic method for estimating PWS in large regions. In this study, an innovative fusion method (SDB-SM) combining satellite-derived bathymetry (SDB) and a statistical model (SM) was proposed to improve the accuracy of PWS estimation at the regional scale. First, the ponds were categorized into ponds suitable for inversion (PSI) and ponds unsuitable for inversion (PUI) on the basis of shape parameters and spectral features. For PSI, the SDB model was constructed using optical images and field depth data, comparing four machine learning algorithms: Random Forest (RF), Gradient Boosting Decision Tree (GBDT), Extreme Gradient Boosting (XGBoost), and Backpropagation Neural Network (BP). The water storage was then calculated via the volume equation. For PUI, a linear area-storage statistical model (SM) was applied. By integrating water storage estimates derived from two complementary methods, the total water storage volume of all ponds across the region was estimated with high precision. Compared to the use of a single statistical model (SSM), which means the regression model, this integrated approach reduced estimation errors by 37 to 88%. The fusion method not only preserves the high inversion accuracy for large ponds but also enhances the estimation reliability for small ponds. Meanwhile, the application of the SDB method enables a more detailed characterization of the water depth of ponds. This research represents a methodological advancement in the estimation of PWS in complex environments. The proposed dynamic grid sampling strategy offers a practical framework for field surveys of decentralized water bodies. The findings have practical applications in agricultural water accounting and ecological water demand assessments, thereby supporting the sustainable development of water resources.
Wang et al. (Wed,) studied this question.