Fabrizio Niro - ESA/ESRIN, Italy
Erminia Degrandis - SERCO, Italy
Validation is the process of assessing by independent means the quality (uncertainty) of satellite derived products. The validation process is key to ensure that the mission is actually meeting its goals. Proba-V higher-level products consist of atmospherically-corrected Top-Of-Canopy (TOC) reflectances, provided as multi-days composites at various spatial resolution (1 km, 333 m and 100 m). The validation of such products is a rather complex topic, since no real “ground truth” data is available at the medium spatial resolution scale. One of the approaches for validating surface reflectance at the hectometric scale is to use AERONET ground-based network. The process consists in using a Radiative Transfer Model (RTM) and the input atmospheric data retrieved from the AERONET measurements, i.e., the aerosol optical thickness (AOT) and its spectral dependency (Angstrom exponent) and the column water vapour content (WV), to compute “ground-truth” surface reflectances from TOA satellite measurements. These “synthetic” surface reflectances can then be compared against the collocated operational TOC products in order to estimate their precision and accuracy. This process should be systematically available within the Proba-V MEP as a predefined workflow (e.g., through the Jupyter Notebook), so that a Cal/Val or expert user can estimate the accuracy of the current products version, as well as test new approaches (e.g., for AOT retrieval) to explore potential data quality improvements.
The project is performed in the frame of the ESA-ESRIN Sensor Performance Products and Algorithm (SPPA) section and it is part of the activities supporting Proba-V Data Quality and Cal/Val. The project is run by Serco with the support of the VITO MEP team.
Prepare and publish a notebook for scientific users
The main Proba-V MEP feature used within this project is the Virtual Machine, where a development environment was set-up consisting on a series of Python programs, the associated mathemetical libraries and the required input data, i.e., the AERONET retrieved atmospheric values (AOT, Angstrom exponent and WV total column) over a predefined set of ground-based stations, globally spread and covering a wide range of land cover and aerosol conditions. Additionally, the same environment is being replicated in a Jupyter Notebook, allowing to fully share the validation approaches and the associated results. This Notebook could be provided to the scientific users with the intention of automating the process of algorithm validation and evolution.
EO-Data archive combined with ground-based data
Validation of satellite dataset requires synoptic reference “ground truth” in order to derive a representative estimate of the accuracy at global scale. Furthermore, evolution of the accuracy with time is crucial to detect trend due to drifting calibration or to sensor ageing. The Proba-V MEP eliminates the barreer of having a global long-term archive of collocated satellite and ground-based dataset, allowing therefore to perform an extensive and global validation of the relevant satellite products.
The same validation exercise without the MEP would have required a enormous investment on the user side, e.g. to download and store the ground-based dataset, harvest to full Proba-V archive to look for space and time collocation, retrieve the match-up dataset and finally perform the validation analysis.
There are several ideas on how to continue the study, by extending the investigation to all spatial resolution and all AERONET stations, by developing a Jupyter notebook to be shared among Cal/Val experts, by consolidating a database of “ground-truth” surface reflectances to be used as reference for subsequent validation activities, by including inter-comparison with similar satellite products, such as the S3 Synergy surface directional reflectances. Besides, the full processing capabilities of the MEP will be explored, for instance, accessing the cluster computing facility, through Spark, in particular for the analysis of high spatial resolution (100 m) dataset.
References - results
The project effectively started during August 2017, though, initial results are presented herebelow. The following plots show the location of the AERONET stations currently used and the scatter plot of the Proba-V TOC operational (OP) reflectances in the four bands (Blue, Red, NIR, SWIR) with respect to the reference AERONET “synthetic” reflectances for various AOT regimes.