Sensor-generic adjacency-effect correction for remote sensing of coastal and inland waters

The adjacency effect distorts the top-of-atmosphere (TOA) spectral signals of coastal and inland waters and is a major challenge for optical remote sensing of nearshore aquatic environments. We introduce a closed-form expression that corrects for the adjacency effect prior to atmospheric correction. The method is included in an open-source Python tool, which ingests level-1 imagery and calculates the point-spread function of the atmosphere to convolve the input imagery. For each band, the difference between the observed and convolved reflectances is used to quantify and correct for the adjacency effect, i.e., pixels are corrected to the TOA reflectance they would have if surrounded by pixels of identical reflectance. Validation was conducted for Sentinel-2 MSI and Landsat 8 OLI imagery against a global dataset of coincident in situ radiometric measurements. Results showed improved accuracy of water-leaving reflectance derived by atmospheric correction processors, including ACOLITE, POLYMER, and l2gen, when these were applied following adjacency-effect correction. For matchups within 200 m of shorelines (n = 212), adjacency-effect correction resulted in an average 16.7 % reduction in root mean square error, a 32.4 % reduction in symmetric signed percentage bias, and a 36.8 % reduction in median symmetric accuracy for the three processors. The improvements were more significant in the near-infrared (NIR) range for ACOLITE, visible wavelengths for l2gen, and evenly distributed across the visible-NIR spectrum for POLYMER. We anticipate that this physics-based approach to adjacency-effect correction will lead to improved satellite-derived aquatic products for coastal and inland waters under diverse atmospheric and aquatic conditions.