Upscaling of Field Measurements for the Validation of the Sentinel-2 Level-2A Product

Kurzfassung

In times of rapid global environmental change, the free distribution of the Sentinel-2 surface reflectance product has greatly improved the availability of high-resolution satellite imagery capable of capturing such changes. Having previously undergone geometric, radiometric and atmospheric correction, it is essential to determine the data quality before their further use in scientific studies. Preferably, validation of the Sentinel-2 Level-2A surface reflectance (L2A SR) product is performed using higher resolution reference data. In this study, field measurements and HySpex high-resolution airborne imagery for an area near Lake Stechlin, Germany, serve as reference to assess the Sentinel-2 L2A SR data quality. After spectral resampling of both field measurements and airborne imagery to Sentinel-2 spectral resolution, the field measurements are upscaled to match the spatial resolution of the airborne imagery using simple averaging. To account for the geolocation error, this is performed repeatedly with different geographic offsets until correlation with the HySpex pixel values is maximized. The resulting field measurement pixels are used to determine the quality of the airborne imagery, which subsequently undergoes spatial resampling, i.e. upscaling, to Sentinel-2 spatial resolution. Following this, surface reflectance from pixel samples is used to validate the Sentinel-2 L2A SR product with the upscaled HySpex data serving as reference. Accuracy, Precision and Uncertainty (APU) estimates of surface reflectance are reported together with uncertainties arising at different steps in the analysis. Ultimately, this is an assessment of Sen2Cor, the atmospheric correction algorithm used to produce the Sentinel-2 L2A SR product. The upscaling of field measurements using simple averaging produced realistic estimates but could be further improved by applying a more suitable field sampling scheme that allows for sophisticated upscaling techniques. HySpex imagery shows a high correlation with the upscaled field measurements comparable to results in PFLUG (2019). Validation of the Sentinel-2 L2A SR product revealed an overestimation of surface reflectance and APU statistics mostly outside their specifications. In comparison to similar studies (DOXANI et al. 2018; ESA 2019; PFLUG 2019), the Sentinel-2 imagery over Lake Stechlin contains larger uncertainties, which may originate from differences between Sentinel-2 and HySpex imagery in georeferencing, solar and viewing angles, IFOVs and processors as well as the retrieved AOT value at 550 nm as processing input.