A Portable Algorithm to Retrieve Bottom Depth of Optically Shallow Waters from Top-Of-Atmosphere Measurements

Bottom depth ( H ) of optically shallow waters can be retrieved from multiband imagery, where remote sensing reflectance ( R rs ) are commonly used as the input. Because of the difficulties of removing the atmospheric effects in coastal areas, quite often, there are no valid R rs from satellites for the retrieval of H . More importantly, the empirical algorithms for H are hardly portable to new measurements. In this study, using data from Landsat-8 and ICESat-2 as examples, we present an approach to retrieve H directly from the top-of-atmosphere (TOA) data. It not only bypasses the requirement to correct the effects of aerosols but also shows promising portability to areas not included in algorithm development. Specifically, we use Rayleigh-corrected TOA reflectance ( ρ rc ) in the 443–2300 nm range as input, along with a multilayer perceptron ( MLP H ρ rc ), for the retrieval of H . More than 78,000 matchup points of ρ rc and H (0–25 m) were used to train MLP H ρ rc , which resulted in a Mean Absolute Percentage Difference (MARD) of 8.8% and a coefficient of determination ( R 2 ) of 0.96. This MLP H ρ rc was further applied to Landsat-8 data of six regions not included in the training phase, generating MARD and R 2 values of 8.3% and 0.98, respectively. In contrast, a conventional two-band ratio algorithm with R rs as the input generated MARD and R 2 values of 31.6% and 0.68 and significantly fewer H retrievals due to failures in atmospheric correction. These results indicate a breakthrough of algorithm portability of MLP H ρ rc in sensing H of optically shallow waters.