Data-Driven Adjacency Correction from High-Contrast Satellite Imagery Using DeepWater as a Natural Reference Target

The adjacency effect introduces spatial biases in high-resolution satellite imagery through atmospheric scattering of radiance from adjacent surfaces, particularly near sharp albedo contrasts such as ice-ocean boundaries. This study presents a data-driven approach for estimating the atmospheric point spread function directly from spaceborne observations over marine environments in both polar regions. Using optically deep ocean water as a natural reference target with known spatial homogeneity, an iterative inversion procedure minimizes reflectance variance to retrieve double Gaussian kernel parameters characterizing multi-scale scattering processes. The correction is sensor-agnostic and requires no external ground measurements or climatological aerosol profiles. The methodology is validated across Sentinel-2 MSI and EnMAP imagery capturing ice-ocean interfaces in diverse conditions over multiple acquisition dates. Results demonstrate that the adjacency effect extends over characteristic length scales of 2–12 km, with a short-range aerosol component (2 km) and long-range molecular scattering contribution (12 km). Corrected deep water reflectance exhibits improved spatial homogeneity consistent with expected values for open ocean waters, with RMSE reductions of 28–82 % across both sensors. Oceanographic features affected by the adjacency effect become apparent after correction, while snow surfaces exhibit improved uniformity. These findings have implications for polar region monitoring and processing pipelines, as current atmospheric correction schemes may underestimate adjacency effects at moderate to large distances from high-contrast boundaries.