Impact of optical heterogeneity on satellite-derived bathymetry in reservoirs: Case of Lake Uiam; [위성영상을 이용한 호소 수심 추정에 광학적 이질성이 미치는 영향: 의암호 사례]

초록

This study estimates the bathymetry of Uiam Lake using Sentinel-2 satellite imagery and investigates the influence of optical heterogeneity on depth-prediction accuracy. The spectral signal of natural inland waters exhibits substantial spatial heterogeneity due to variations in suspended sediment concentration, algal biomass, and bottom reflectance, which increases the nonlinearity and local variability in the depth-reflectance relationship. The global model, Random Forest (RF), effectively reproduced the overall depth pattern; however, prediction errors increased considerably in areas where optical properties changed rapidly. In contrast, Geographically Weighted Regression (GWR) locally estimated the spatially varying depth-reflectance relationship and successfully captured this heterogeneity. As a result, the GWR model achieved a 22.3% reduction in MAPE and a 9.3% improvement in R2compared with the RF model. The substantial spatial variation in GWR coefficients further indicates strong optical heterogeneity within Lake Uiam. This study highlights the structural limitations of the global-model approaches in optically complex inland waters and demonstrates that the spatially adaptive methods such as GWR provide an effective alternative for improving satellite-derived bathymetry. © 2026 Korea Water Resources Association.

키워드