Assessment of atmospheric correction methods for Sentinel-2 images in Mediterranean landscapes

• Published in: International journal of applied earth observation and geoinformation Vol. 73; pp. 63 - 76
• Main Authors: Sola, Ion, García-Martín, Alberto, Sandonís-Pozo, Leire, Álvarez-Mozos, Jesús, Pérez-Cabello, Fernando, González-Audícana, María, Montorio Llovería, Raquel
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.12.2018
• Subjects: Atmospheric correction; Field spectrometry; iCOR; MAJA; Mediterranean shrub and grasslands; SEN2COR; Sentinel 2A-MSI; Atmospheric correction; SEN2COR; Field spectrometry; MAJA; Mediterranean shrub and grasslands; Sentinel 2A-MSI; 6S; iCOR
• ISSN: 1569-8432; 1872-826X
• DOI: 10.1016/j.jag.2018.05.020

•Performance of 4 atmospheric correction methods evaluated on Sentinel-2 images.•4 dates, 6 plots and 9 spectral bands assessed.•Results suggest the suitability of the applied corrections.•Minor differences observed between the four methods (MAJA, SEN2COR, 6S, iCOR). Atmospheric correction of optical satellite imagery is an essential pre-processing for modelling biophysical variables, multi-temporal analysis, and digital classification processes. Sentinel-2 products available for users are distributed by the European Space Agency (ESA) as Top Of Atmosphere reflectance values in cartographic geometry (Level-1C product). In order to obtain Bottom Of Atmosphere reflectance images (Level-2A product) derived from this Level-1C products, ESA provides the SEN2COR module, which is implemented in the Sentinel Application Platform. Alternatively, ESA recently distributes Level-2A products processed by SEN2COR with a default configuration. On the other hand, the conversion from Level-1C to Level-2A product can be generated using alternative atmospheric correction methods, such as MAJA, 6S, or iCOR. In this context, this paper aims to evaluate the quality of Level-2A products obtained through different methods in Mediterranean shrub and grasslands by comparing data obtained from Sentinel-2 imagery with field spectrometry data. For that purpose, six plots with different land covers (asphalt, grass, shrub, pasture, and bare soil) were analyzed, by using synchronous imagery to fieldwork (from July to September 2016). The results suggest the suitability of the applied atmospheric corrections, with coefficients of determination higher than 0.90 and root mean square error lower than 0.04 achieving a relative error in bottom of atmosphere reflectance of only 2–3%. Nevertheless, minor differences were observed between the four tested methods, with slightly varying results depending on the spectral band and land cover.