Remote Sensing of Environment · 2020 · 36 citations · 32 references
Natural SciencesEarth SciencesMeteorologyAtmospheric ScienceCloud MonitoringRemote SensingCloud PropertiesCloud Remote SensingInfrared Remote SensingRemote Sensing Image AnalysisRemote Sensing ImagerySatellite SensorsCloud Cover ObservationCloud Fraction EstimationCloud Mask ValidationData Access PortalsMisleading ImagesRemote Optical SensingSatellite Cloud ObservationsSpectral Signatures
Cloud detection in optical remote sensing imagery is crucial because undetected clouds can produce misleading results in analyses. Almost all optical remote sensing data access portals rely to some degree on a cloud cover filter. Here we show that cirrus as well as opaque cloud cover in Sentinel-2 Level-1C (L1C) and Level-2A (L2A) imagery is systematically and significantly overestimated in very high altitude areas (e.g. Himalayas, Andes). We argue that this systematic bias is created by applying simple thresholds to single bands instead of using a multi-band spectral signature in the cloud detection process. This results in a lot of “hidden” data for very high altitude areas when each image's estimated cloud cover is used as an automated selection criterion for analysis (e.g. global analyses, cloud-free mosaic production). We show geographic locations exemplifying this overestimation, and compare the L1C and L2A cloud masks produced by ESA to cloud masks generated by an expert system that uses comprehensive spectral signatures, showing that reliable cloud estimations are possible in very high altitudes. Based on this comparison, we argue for changes to L1C and L2A cloud detection algorithms in order to improve initial querying and selection of big EO data, where reliable yet automated quality indicators are necessary to handle an overwhelming data volume and velocity. Our contribution raises awareness of potential bias when pre-selecting images based on reported cloud cover in very high altitude areas for researchers and users of Sentinel-2 imagery in the environmental domain.
32
Google Earth Engine: Planetary-scale geospatial analysis for everyone
Noel Gorelick, M. Hancher, Mike Dixon et al. · Remote Sensing of Environment · 2017 · 14K citations · Full text
The Harmonized Landsat and Sentinel-2 surface reflectance data set
Martin Claverie, Junchang Ju, Jeffrey G. Masek et al. · Remote Sensing of Environment · 2018 · 1.4K citations · Full text