Research on the construction and distribution model of spatio-temporal database of medicinal plant resources based on GIS technology

Abstract

Multi-source remote sensing data fusion is a core pathway to overcome the bottleneck of single-sensor bathymetric capability. This review summarizes representative fusion methods published between 2024 and 2026, covering four technical routes: active-passive optical fusion, wave-based indirect bathymetry, gravity-sonar multi-source fusion, and deep learning-driven fusion. Results indicate that using ICESat-2 laser altimetry as an accuracy anchor fused with Sentinel-2 multispectral imagery achieves RMSE better than 1.3 m in shallow water (0-30 m), while spherical radial basis function (SRBF)-based gravity-sonar fusion improves accuracy by 18-68% in deep water. However, a critical, cross-cutting deficiency is identified: most methods report only point-wise accuracy metrics (e.g., RMSE)and lack predictive uncertainty characterization. Furthermore, physical constraints are weakly integrated, typically as additional input features rather than as regularization terms within the loss function. Consequently, this review proposes a unified Bayesian inference framework that explicitly models observation noise and prior uncertainty while embedding the water optical transfer equation as a physical regularization term into the loss function.