Comparison of Masks of Fmask5, ATCOR and Sen2Cor

Abstract

Thirty-two Sentinel-2 scenes with a worldwide global coverage were processed by three masking algorithms (Fmask5, Sen2Cor version 2.8.0, and ATCOR version 9.3.0) using the following six classes: clear, semi-transparent cloud, cloud, shadow, water, and snow/ice. The evaluation is performed with the default global settings, i.e. without parameter tuning per scene. All Sentinel-2 bands are resampled to a common 20 m resolution. In case of mountainous terrain the 1 arc-sec (30 m) SRTM Digital Elevation Model (DEM) is used and also resampled to 20 m. The comparison was performed qualitatively based on visual interpretation. The results show a similar performance of ATCOR and Sen2Cor, with larger differences to Fmask5. The presentation highlights the advantages and limitations of each processor and compares the performance of some critical scenes by selecting areas that are difficult to classify. Although no statistical, quantitative comparison with reference masks is available, all results still show a consistent trend and allow an assessment of the advantages and drawbacks of each processor.