Résumé
Accurate global bathymetry mapping underpins natural hazard prediction or marine habitat management. However approximately 80% of the seafloor remains unmapped. Traditional waterborne (multi-/single-beam sonar) and airborne (lidar, spectral inversion) methods provide high precision and complementary coverage but are constrained in cost, accessibility, and temporal resolution. Spaceborne approaches, leveraging high- to very high-resolution multispectral and lidar sensors, increasingly enable shallow-water bathymetry retrieval at regional to global scales. This study investigates the potential of the VENμS mission for deriving shallow bathymetry over Glorioso Archipelago (Indian Ocean). VENμS superspectral imagery (12 bands, 4 m resolution, daily revisit) was collected in 2022 and attempted to predict ~83,000 lidar illuminations acquired in 2009. Three predictor series were investigated: surface reflectance, ln-transformed surface reflectance, and band ratios of ln-transformed surface reflectance. Nine depth ranges, from 0 to -45 m, were modeled using stepwise three-factor linear regressions, with performance assessed across calibration, validation, and test subsets. Results indicate excellent performance with ln-transformed surface reflectance achieving the highest predictive skill. The [0; -10 m] interval was optimal, with R2test reaching 0.93 using blue, green, and yellow-2 bands, even if the [0; -30 m] range was satisfactorily modelled (R2test = 0.78). VENμS-derived bathymetry maps show strong concordance with lidar to ~5 m depth, though increasing divergence suggests potential sediment redistribution over the 13-year period. These findings demonstrate that simple, transferable linear models applied to VENμS imagery can yield accurate, scalable shallow-water bathymetry, highlighting the mission’s value for cost-effective coastal mapping and supporting global seabed initiatives such as Seabed 2030.