Gradient Boosting for the Spectral Super-Resolution of Ocean Color Sensor Data.

We present a gradient boosting framework for reconstructing hyperspectral signatures in the visible spectrum (400-700 nm) of satellite-based ocean scenes from limited multispectral inputs. Hyperspectral data is composed of many, typically greater than 100, narrow wavelength bands across the electromagnetic spectrum. While hyperspectral data can offer reflectance values at every nanometer, multispectral sensors typically provide only 3 to 11 discrete bands, undersampling the visible color space. Our approach is applied to remote sensing reflectance (Rrs) measurements from a set of ocean color sensors, including Suomi-National Polar-orbiting Partnership (SNPP) Visible Infrared Imaging Radiometer Suite (VIIRS), the Ocean and Land Colour Instrument (OLCI), Hyperspectral Imager for the Coastal Ocean (HICO), and NASA's Plankton, Aerosol, Cloud, Ocean Ecosystem Ocean Color Instrument (PACE OCI), as well as in situ Rrs data from National Oceanic and Atmospheric Administration (NOAA) calibration and validation cruises. By leveraging these datasets, we demonstrate the feasibility of transforming low-spectral-resolution imagery into high-fidelity hyperspectral products. This capability is particularly valuable given the increasing availability of low-cost platforms equipped with RGB or multispectral imaging systems. Our results underscore the potential of hyperspectral enhancement for advancing ocean color monitoring and enabling broader access to high-resolution spectral data for scientific and environmental applications.