Comprehensive pixel-level Level 1 Uncertainty Characterization for SPOT-VGT1, SPOT-VGT2, and PROBA-V following the FIDUCEO guidelines.

Producing Fundamental Data Records (FDRs) with well-defined uncertainty estimates is crucial for the development of relevant Essential Climate Variables (ECV), as prescribed by the Quality Assurance framework for Earth Observation (QA4EO) and the FIDUCEO guidelines. An FDR is a record, of sufficient duration for its application, of uncertainty-quantified sensor observations calibrated to physical units and located in time and space, together with all ancillary and lower-level instrument data used to calibrate and locate the observations and to estimate uncertainty.

However, satellite operators often do not provide contextual uncertainty products for their missions. This is particularly evident for the PROBA-V, VGT1, and VGT2 sensors, where uncertainty is typically expressed as global upper bounds rather than pixel-specific measures. Furthermore, such uncertainty information is frequently inaccessible to the broader scientific community or inconsistently formatted. This lack of clear uncertainty information constrains researchers’ ability to propagate uncertainty to higher-level ECV models accurately.

The ESA FDR4VGT project led by VITO Remote Sensing addresses this gap by producing pixel-level uncertainty estimates for two decades of harmonized satellite data. The project employs an uncertainty propagation method grounded in metrology principles. Special care has been given to the redaction of FIDUCEO compliant Digital Effect Tables (DETs) for the characterization of digital counts, calibration coefficients and ancillary information. The proposed method exposes the uncertainty estimates as an explicit analytical equation, differentiable and optimized for large scale computing. This comprehensive approach ensures adherence to FIDUCEO guidelines while balancing computational efficiency and accuracy.

Reprocessing 20+ years of Level 1A data to propagate uncertainty estimates to Level 2A projected reflectance images poses several technical challenges. Performance constraints must be considered for the propagation method, and Monte-Carlo uncertainty propagation approaches can only be done for sub-problems limited in terms of time range or scope. An uncertainty quantization using a statistical approach for the less impactful solar geometry uncertainty has been performed on the year 2021 of Proba-V Level 1 data. Level 1A uncertainties have been propagated using an analytical expression of the Law of Propagation of Uncertainty (Guide to the expression of uncertainty in measurement). DETs are prepared to characterise each identified source of uncertainty in the uncertainty diagram.

The uncertainty characterisation at Level-1 is expected to improve the retrieval of ECVs from the VGT and PROBA-V archive, as discussed in previous studies, such as the ESA SPAR@MEP project, underlying the role of improved satellite observation uncertainty characterization in enhancing image inversion performances. This enhancement will be assessed thorough the application of the CISAR algorithm to a diagnostic dataset; a comparison against a reference datases, retrieval uncertainties and ground-based measurements will be performed.

This study demonstrates the feasibility of delivering pixel-level uncertainty maps for Level-1 satellite observations using computationally efficient models. The work highlights the partial or inadequate characterization of several uncertainty contributors, which should be addressed in the preparation of future mission aiming at 1% accuracy. Additionally, this study sets the groundwork for advancing uncertainty analysis in Level 2 and Level 3 data and fulfilling key prerequisites for the delivery of FRM and higher-level CDR.