Assessment of atmospheric correction algorithms for the Sentinel-2A MultiSpectral Imager over coastal and inland waters

REMOTE SENSING OF ENVIRONMENT

卷:225

Article

Warren, M. A.¹  Simis, S. G. H.¹  Martinez-Vicente, V.¹  Poser, K.²  Bresciani, M.³  Alikas, K.⁴  Spyrakos, E.⁵  Giardino, C.³  Ansper, A.⁴ 

[1] Plymouth Marine Lab, Plymouth PL1 3DH, Devon, England

[2] WaterInsight, POB 435, NL-6700 AK Wageningen, Netherlands

[3] Natl Res Council Italy, Inst Electromagnet Sensing Environm, CNR IREA, Via Bassini 15, I-20133 Milan, Italy

[4] Univ Tartu, Tartu Observ, Observatooriumi 1, EE-61602 Noo Parish, Tartu County, Estonia

[5] Univ Stirling, Biol & Environm Sci, Stirling FK9 4LA, Scotland

关键词: Atmospheric correction;    Sentinel 2;    Remote sensing reflectance;    Hyperspectral radiometry;    Baltic Sea;    Lakes;    Western Channel Observatory;    Coastal waters;    Inland waters;   

DOI  :  10.1016/j.rse.2019.03.018

来源: Elsevier

【 摘 要 】

The relatively high spatial resolution, short revisit time and red-edge spectral band (705 nm) of the ESA Sentinel-2 Multi Spectral Imager makes this sensor attractive for monitoring water quality of coastal and inland waters. Reliable atmospheric correction is essential to support routine retrieval of optically active substance concentration from water-leaving reflectance. In this study, six publicly available atmospheric correction algorithms (Acolite, C2RCC, ICOR, 12gen, Polymer and Sen2Cor) are evaluated against above-water optical in situ measurements, within a robust methodology, in two optically diverse coastal regions (Baltic Sea, Western Channel) and from 13 inland waterbodies from 5 European countries with a range of optical properties. The total number of match-ups identified for each algorithm ranged from 1059 to 1668 with 521 match-ups common to all algorithms. These in situ and MSI match-ups were used to generate statistics describing the performance of each algorithm for each respective region and a combined dataset. All ACs tested showed high uncertainties, in many cases > 100% in the red and > 1000% in the near-infra red bands. Polymer and C2RCC achieved the lowest root mean square differences (similar to 0.0016 sr(-1)) and mean absolute differences (similar to 40-60% in blue/green bands) across the different datasets. Retrieval of blue-green and NIR-red band ratios indicate that further work on AC algorithms is required to reproduce the spectral shape in the red and NIR bands needed to accurately retrieve the chlorophyll-a concentration in turbid waters.

【 授权许可】

Free