Gepubliceerd op 11 februari 2021
Missions like Sentinel-2 and Landsat-8 open opportunities to set up operational Earth observation services at local scale. However, due to frequent cloud coverage over Western Europe combined with lower revisit times of higher spatial resolution sensors, operational services such as crop yield forecasts and crop production estimates have to combine data from different missions in order to have sufficiently frequent observations. A seamless combination of EO products coming from different missions is however not straightforward due to differences in the sensor’s spectral characteristics, and in the calibration and processing of the data.
PROBA-V image from the RadCalNet Railroad Valley Playa site (USA)
— Click to view at full resolution —
The STEREO III project BELHARMONY – carried out by researchers from VITO, VUB and the National Research Council of Italy – proposed and evaluated different harmonisation measures in order to improve the consistency of multi-mission high resolution time series.
First the team investigated if a bias exists in the Level 1 radiance data provided by the different sensors. For this, researchers analysed data acquired over bright radiometric calibration sites such as the Libya-4 desert site and the RadCalNet Railroad Valley Playa (USA) and Gobabeb (Namibia) sites. They observed that all sensors are well-calibrated and that over bright targets differences are small between the sensors. The differences turned out to be larger for low radiance targets, as illustrated by the measurements of atmospheric and marine optical parameters provided by AERONET-OC (the Ocean Color component of the Aerosol Robotic Network).
The second step was to derive a correction function to align the spectral responses of different sensors. Here researchers used simulated vegetation spectra derived from a physically-based radiation transfer model that considers the leaf optical properties, the canopy structure and the background reflectance.
Finally the team generated enhanced Level 2 and Level 3 time series by using a common processing chain and by applying intercalibration gains and spectral adjustment functions. An important aspect here is the use of the same atmospheric correction algorithm for all the considered sensors. They used the iCOR tool developed at VITO. For the interested reader, basic versions of iCOR for Landsat-8, Sentinel-2 and OLCI/Sentinel-3 are publicly available as free plugins for SNAP toolbox. An iCOR4S3 processing service is also available at ESA’s Grid Processing on Demand for Earth Observation Application (G-POD).
By applying the relevant harmonisation measures to Sentinel-2A/B, Landsat-8 , PROBA-V and Deimos-1, the researchers were able to generate enhanced level 2 and Level 3 time series over the BELAIR sites. Finally the team used the extensive set of APEX hyperspectral airborne data and in-situ reference data systematically collected over the BELAIR urban, agricultural and coastal waters sites to evaluate the impact of the different harmonisation measures.
In-situ reference measurements at the Belair study site "Sonia"