A generalized physics-based correction for adjacency effects

Atmospheric scattering occurs over a horizontal scale of several kilometers. This results in influence from neighboring surface features on the signal recorded over a given position, reducing contrast and the accuracy of quantitative retrievals of surface reflectance from satellite imagery. This atmospheric blurring, or adjacency effect, must be accounted for when both contrast in surface reflectance and magnitude of atmospheric scattering are significant. Taking into account the adjacency effect is of particular importance for aquatic remote sensing of inland and coastal waters due to the high contrast between water and different land cover types, as well as the small spatial scale of most inland water bodies. In this paper, we present a physics-based processor to retrieve surface reflectance over all surface types, regardless of the subscene composition and sensor waveband configuration. The processor is implemented in the free and open source ACOLITE software and is composed of two modules: (1) TSDSF for the estimation of aerosol properties and (2) RAdCor for the retrieval of surface reflectance. We demonstrate the performance of the TSDSF + RAdCor processor for the Operational Land Imager (OLI) onboard Landsat 8 and the Multispectral Instrument (MSI) onboard Sentinel-2A and 2B over a set of small (<1km²) inland waters in Belgium, and compare the performance with other common processors for these sensors, including C2RCC, POLYMER, Sen2Cor, iCOR, ACOLITE/DSF, and LaSRC. For clear sky matchups, the relative deviation againstin situ data in the visible wavebands ranged between 6% and 18% for OLI, and between 14% and 31% for MSI, except for the MSI waveband centered at 443 nm where the relative deviation was 70%. In the near-infrared wavebands, the relative deviation varied from 70% to 150%, with the exception of the MSI waveband centered at 704 nm, for which the performance was 17%. Overall, the new processor outperformed the other evaluated processors in the visible range, with the exception of the MSI waveband centered at 443 nm, and was outperformed by C2RCC and POLYMER in the near-infrared wavebands. Recommendations on how to use TSDSF and RAdCor in ACOLITE are provided.