ABSTRACTSatellite imagery constitutes an affordable solution to map coastal bathymetry at very high resolution (VHR). Bathymetry retrieval is commonly based on regression analysis linking depth response with remotely-sensed spectral predictors. Most studies have focused on a single regressor using conventional three visible bands over clear waters. However the interdependence of added visible bands in respect to linear and non-linear regressors is poorly studied and strongly lacks for turbid waters. Here we investigate the single and joint contribution of both spectral bands (visible 1.24 m WorldView-3, WV-3) and regressor types on the performance of retrieval. A case study over Saint-Malo (Brittany, France) megatidal turbid waters enables to quantify the accuracy gain related to Coastal and yellow bands as well as ordinary least squares (OLS), generalized linear model (GLM) and artificial neural network (ANN). Correlation analysis reveals that both Coastal and yellow bands do not significantly increase conventional performance for the three regressors. ANN surpassed GLM and OLS for both conventional and boosted WV-3 visible spectral datasets, reaching 0.94 correlation coefficient and 0.52 m accuracy. Comparisons’ significance indicates that selecting a robust regression method (including parameterization) is more efficient than adding spectral bands for mapping VHR bathymetry of coastal turbid waters.