Assessment of a generalized linear model for satellite-derived bathymetry in turbid waters using Sentinel-2 multi-temporal images

Climate change is reshaping the world’s coastlines through coupled dynamic processes. The increased importance of monitoring coastal changes over time can be partially addressed using satellite derived bathymetry (SDB), which is more cost effective than traditional methods and allows for monitoring capabilities. In this study we developed a two-step methodology aiming to improve shallow water depth estimates from multi-temporal Sentinel-2 satellite images. The pilot area lies in north-east Ireland in optically complex waters. A threshold criterion was applied to identify 10 suitable Sentinel-2 images over one year time (2021). Lyzenga and Stumpf empirical models were evaluated followed by the application of an empirical generalized linear model (GLM). The performance of atmospherically corrected composite images, created using a reducer function (mean and median), was also evaluated, and compared with the performance of single images. Validation results confirmed the outperformance of the GLM model compared to Lyzenga and Stumpf empirical models. The optimum combination of multi temporal images outperformed the single images regression scores, with a reduction of 45 % in RMSE and a MAE as low as 31 cm in the 0 to 10 m depth. The application of empirical models on the multi-temporal image analysis results in a reduction of error outliers. These results enhance the potential of SDB and Sentinel-2 data in a range of potential coastal monitoring applications, such as repetitive bathymetric changes, ecosystem mapping and environmental management.