Evaluating hyperspectral Secchi depth retrieval through hybrid modeling and regression

Abstract

This work compares different regression models combined with hybrid modeling to estimate water clarity using hyper-spectral remote sensing data. The Secchi depth, a proxy of water clarity, can be modeled using first principles bio-optical modeling and other static pre-processing steps are used to generate four different feature sets. The different feature sets and regression models are evaluated using cross-validation on the recently published GLORIA dataset, representing a vast set of Secchi depth measurements from various aquatic environments (N = 3914). The best-performing feature generation and regression model combination can provide promising Secchi depth inference from hyperspectral data (RMSE = 1.543, AP D = 39.419, R 2 = 0.636). The study demonstrates the potential of hyperspectral remote sensing data for monitoring and managing aquatic ecosystems.