Pléiades Neo-derived bathymetry in coastal temperate waters: the case study of Saint-Malo

Despite the growing interest in seabed mapping in the context of sea level rise and storm intensification, only 10% of the global bathymetry has been sampled using reliable technologies, such as sonar or lidar, primarily due to the high cost of associated campaigns (waterborne and airborne). Consequently, satellite-derived bathymetry has thrived rapidly in recent decades, given its affordability for the user, and its ongoing gain in radiometric, spatial, spectral and temporal resolutions.

As a flagship of the very high spatial resolution sensor, the Pléiades Neo (PNEO) multispectral imagery, acquired by PNEO 3 and 4 sensors, leverages 6 bands : 1 purple (so-called deep blue), 3 visible, 1 red edge, and 1 infrared, provided with a spatial resolution of 1.2 m. This new sensor thus outperforms the Pléiades-1 multispectral imagery endowed with 4 bands (3 visible and 1 infrared) at 2 m pixel size.

The contribution of the novel bands of the PNEO to the bathymetry retrieval was innovatively quantified over an optically-challenging body of coastal seawater (0.2 m^-1 of vertical light attenuation). The importance of the level of the radiometric correction was tested based on the bathymetric lidar bathymetry predicted by a neural network (1 hidden layer and three neurons).

A PNEO 4 imagery was collected over the megatidal Bay of Saint-Malo (Brittany, France) on December 7, 2022. Following the orthorectification, the multispectral imagery was processed for the radiometric correction using the PNEO 4-specific spectral sensitivity, yielding five outputs: digital numbers (DN), top-of-atmosphere (TOA) radiance, TOA reflectance, bottom-of-atmosphere (BOA) maritime-modelled reflectance, and BOA tropospheric-modelled reflectance. The lidar response dataset, ranging from 0 to 20 m depth, was statistically stratified at the rate of 90 random samples per bathymetric slice of 1 m, every one divided into calibration, validation and test sub-samples.

The best predictions, reaching R²test of 0.81, were obtained for the full PNEO 4 dataset when uncorrected for the radiometry (namely, DN), corrected at both the TOA radiance and reflectance. For both BOA full-dataset products, the results were slightly less satisfactory: R²test of 0.75 (maritime) and 0.76 (tropospheric).

Taking the reference of the blue-green-red-infrared (simulating Pléiades-1 imagery), gains in R²test attained 0.05 for DN and TOA radiance datasets when the deep blue band replaced the blue one; 0.07 for maritime BOA reflectance when both deep blue and blue bands were integrated; and even 0.11 for that BOA reflectance when all PNEO 4 bands were used as predictors.