Resumen
Here we present a machine-learning-based method for utilizing traditional ocean-viewing satellites to perform automated atmospheric correction of nanosatellite data. These sensor convolution techniques are required because nanosatellites do not usually possess the wavelength combinations required to atmospherically correct upwelling radiance data for oceanographic applications; however, nanosatellites do provide superior ground-viewing spatial resolution (~3 m). Coincident multispectral data from the Suomi National Polar-Orbiting Partnership Visible Infrared Imaging Radiometer Suite (Suomi NPP VIIRS; referred to herein as ?VIIRS?) were used to remove atmospheric contamination at each of the nanosatellite?s visible wavelengths to yield an estimate of spectral water-leaving radiance [Lw(l)], which is the basis for surface ocean optical products. Machine learning (ML) algorithms (KNN, decision tree regressors) were applied to determine relationships between Lw and top-of-atmosphere (Lt)/Rayleigh (Lr) radiances within VIIRS training data, and then applied to test cases for (1) the Marine Optical Buoy (MOBY) in Hawaii and (2) the AErosol RObotic Network Ocean Color (AERONET-OC), Venice, Italy. For the test cases examined, ML-based methods appeared to improve statistical results when compared to alternative dark spectrum fitting (DSF) methods. The results suggest that ML-based sensor convolution techniques offer a viable path forward for the oceanographic application of nanosatellite data streams.