On Sentinel-2 Scene Classification Layer for Global Surface Water Occurrence Mapping

In this presentation, we provide a detailed analysis of the Scene Classification Layer (SCL) at 20 m distributed with the Sentinel-2 Level 2A products, focusing on the water class omissions and commission errors. The motivation for this analysis originates in the use of the SCL for the efficient mapping of global water occurrence, a fundamental variable for environment and climate change research. By providing a specific label for water pixels, the SCL is directly suitable for the computation of water occurrence maps. The generated Sentinel-2 SCL-based water occurrence maps applied to the 10-year Sentinel-2 time series are promising, offering cost-effective alternatives to algorithms based on the processing of the reflectance bands. However, the results also reveal a series of quality issues that require attention. The primary finding is that the most prominent source of SCL water class error stems from the use of the Climate Change Initiative global map of open Water Bodies (CCI WB v4.0) as supplementary land cover data in the SCL algorithm. We demonstrate that these errors can be alleviated by utilising a more reliable persistent water body mask. Illustrative examples highlighting the strengths and weaknesses of the initial results are presented while discussing possible enhancements including the spatial resolution refinement from 20 m to 10 m. We also show that the proposed developments can be directly transposed for the generation of vegetation occurrence maps, demonstrating the potential for a more comprehensive and efficient environmental monitoring system.