For over 20 years, the processing facility at VITO has been processing, archiving and disseminating satellite data derived from SPOT4-Vegetation1, SPOT5-Vegetation2 and PROBA-V in order to serve a large user community providing vegetation information products with nearly daily global coverage. Since these data are widely used to monitor environmental change and the evolution of vegetation cover at global scale, the continuity of the SPOT-Vegetation and PROBA-V time series is important for a wide range of users.

The Sentinel-3 synergy “VGT-like” (SYN VGT) products were defined to provide continuity to the SPOT-Vegetation and PROBA-V time series. In the frame of the Sentinel-3 Mission Performance Center (S3-MPC), the consistency between SYN VGT products and PROBA-V Level2 products is evaluated.

In subsequent processing baseline updates, important changes were implemented in the Sentinel-3 SYN VGT processing chain, including improvements to the temporal and spatial compositing routines. As the operational phase of PROBA-V has ended in June 2020, direct product comparison is no longer possible. Therefore, for June 2019, in the frame of the S3-MPC SYN VGT data was reprocessed based on the latest processing baseline.

Although of limited size, this SYN VGT reprocessed dataset provides important insights into the current level of consistency between PROBA-V and SYN VGT. The analyses focus on the calculation of measures of similarity (geometric mean regression parameters, R²) and measures of difference (accuracy, precision and uncertainty) between paired observations. As the analyses will be based on the PROBA-V Collection 2 that is currently being processed, the results will be presented at the symposium.

PROBA-V (Project for On-Board Autonomy – Vegetation) was launched on May 7th 2013, with the objective of providing global land-coverage data continuity for the SPOT (Système Pour l’Observation de la Terre)-Vegetation user community. After more than 6 years of operations, the operational phase of PROBA-V has ended on June 30, 2020.

The reprocessing of the entire PROBA-V archive at 1 km, 300 m and 100 m resolutions to Collection 2 is currently ongoing. Compared to Collection 1, the following changes are included: (1) updated radiometric calibration, (2) a new and better cloud detection method and improved cloud shadow detections, (3) an improved atmospheric correction, (4) harmonisation of the compositing among the resolutions, (5) update of the product format and (6) a new catalogue to distribute the data.

The updates in the radiometric Instrument Calibration Parameter (ICP) ) files include the dependence on date since launch, modeled by a 2nd degree polynomial, for trending of the absolute calibration coefficients for the different strips/bands. The model also corrects for the increasing trend observed in some bands. Secondly, improvements are made in the low frequency multi-angular coefficients (i.e. equalization) for the SWIR strips of the LEFT and RIGHT camera form start of mission based on yaw manoeuvre results. In addition, a correction of inter-camera bias in the Blue band is applied.

The newly designed algorithm for cloud detection in PROBA-V data introduces major changes w.r.t. the algorithm used in Collection 1. It uses a Multi-Layer Perceptron (MLP) neural network algorithm. The training and validation data have been gathered on a much larger scale compared to Collection 1, and final performance is greatly improved compared to both Collection 0 and Collection 1. The method also removed the dependency on auxiliary input data, which was a major issue in Collection 1. The cloud shadow detection was improved by removing the 1-pixel border between cloud and cloud shadow.

As in Collection 1, the new atmospheric correction is also based on the Simplified Model for Atmospheric Correction (SMAC). Compared to the previous collections, it also includes an assessment of uncertainties of the atmospheric correction and error propagation. It uses an external dataset, namely the MERRA-2 (Modern-Era Retrospective analysis for Research and Applications, version 2) for the inputs of the atmospheric correction. A validation of the atmospheric correction was done based on the Atmospheric Correction Inter-Comparison Exercise (ACIX) approach. Validation of the atmospheric correction was done based on the ACIX approach. The results show that the TopOf-Canopy (TOC) reflectances are better characterized and that artefacts due to the image based Aerosol Optical Thickness (AOT) retrieval are removed.

The compositing method is harmonized between the different resolutions in Collection 2. Previously, for 100 m and 300 m the radiometric quality of all 4 bands were checked prior to compositing. Since the SWIR band has quite a number of defect detectors, this resulted in composites with a striping effect. For the 1 km, the SWIR radiometric quality was not checked in the compositing. This method is now applied to all resolutions.

For Collection 1, the entire archive is available in two product formats: HDF5 and GeoTIFF. For Collection 2, however, we follow the newly becoming standard which is Cloud Optimized GeoTIFF (https://www.cogeo.org/) instead of the standard GeoTIFF format. It is backwards compatible with current GeoTIFF format. The regular HDF5 format will keep on existing too. This way, the handling of files in cloud environments and visualization of the data gets easier in the future.
The PROBA-V C2 metadata will also receive a update to allow for minimum compliancy with the CEOS Analysis Ready Data for Land (CARD4L). This allows for immediate analysis and interoperability both through time and with other datasets.

The PROBA-V Collection 2 data will be ingested into a new catalogue client instead of the legacy Product Distribution Facility. The user interface will allow several functionalities, such as time series export, viewing capabilities, additional authentication options, etc.

The presentation will focus on the first four changes. Details of the adaptations, together with a summary of verification and validation results, will be presented.

The ESA SPAR@MEP project has been designed to deliver a consistent long-term data record (LTDR) of aerosol optical properties and surface reflectance from SPOT-VGT and PROBA-V observations. The project exploits the Mission Exploitation Platform (MEP) developed by VITO, to access and process the satellite data. The LTDR is obtained by processing satellite images with the CISAR algorithm, developed by Rayference.

The CISAR algorithm has been originally developed for a consistent retrieval of surface reflectance and aerosol single scattering properties. Given the large number of issues associated to clouds in traditional aerosol retrieval algorithms, CISAR has been extended to the retrieval of cloud single scattering properties as well, although the main focus remains the aerosols. Standard aerosol retrieval algorithms make use of a cloud mask to discriminate cloud-free atmosphere and only perform the inversion on these pixels. Within the SPAR@MEP project, the PROBA-V cloud mask is only used by CISAR to build a prior information on the aerosol or cloud optical thickness (AOT/COT), but both cloudy and cloud-free observations are processed.

One interesting advantage of this innovative approach for aerosol retrieval is the possibility of retrieving high aerosol load, that are often discarded by traditional algorithms due to erroneous classification of these pixels as cloudy. This feature will be illustrated during a dust storm moving from North Africa to Greece, flagged as cloud by the PROBA-V Collection 1 cloud mask, but correctly retrieved as high aerosol load by CISAR thanks to its innovative approach.

One year (2019) of PROBA-V data is processed at 1 km resolution over Europe. Such a high-resolution aerosol product provides the opportunity to appreciate in detail aerosol spatial features. This advantage will be illustrated over few case studies, such as during the wildfire season in Turkey and the ship trails in the Biscay region. The CISAR product will be evaluated against AERONET observations in terms of error and correlation over selected stations, and against MODIS MCD19A2 product to observe spatial and temporal cross-sensor consistency. The algorithm capability of discriminating between fine and coarse mode will be evaluated over the selected case studies. Processing both aerosols and clouds, the CISAR algorithm could represent a powerful tool to study aerosol-cloud interactions, such as the Twomey effect, from satellite images at high resolution.

As advances in technology make payloads and instruments for space missions more efficient and accessible, a need for the rapidly developed, low-cost missions on very small satellites emerge. Among that class of spacecraft, the Cubesats have distinguished themselves among several distinct missions. A Cubesat is a class of nanosatellites, built to standard dimensions – or Units, “U” - of 10cm x 10cm x 10cm.
The PROBA-V Companion Cubesat (PV-CC) mission aims at flying the spare Spectral Imager from PROBA-V on a tailored 12U Cubesat. One of the major challenges of this mission is to balance the performances of the payload with the limitations of the cubesat platform. Indeed, the downsizing from a microsatellite towards a 12U Cubesat implies some major impacts in terms of mechanical and thermal path from the environment to the payload is restricted. A lower mass also implies higher mechanical loads and reduced thermal inertia, with less radiative surfaces. Moreover, the altitude limitation for satellites without propulsion prevents from operating the mission at a similar altitude, impacting both GSD and acquisition rates.

This presentation is about the development status of the mission as well as the major design and mission drivers that arise from the mentioned points for the satellite hardware and the ground segment.

The Proba-V mission was designed to cover the gap between the Spot-VGT and ESA Sentinel-3 satellite missions and has exceeded all expectations, providing more than 7 years of data. Now that Proba-V has reached the end of its operations, the new PV-CC mission, to be launched in 2022, aims to extend the Proba-V mission objectives. The number of Earth observation satellites carrying optical sensors with similar characteristics is constantly growing, and these new modest-size space missions may complement institutional missions, such as the Copernicus Sentinels, to develop Earth observation products at unprecedented temporal and spatial resolutions. However, cross-mission applications require a level of data harmonization that is not easy to meet, since differences in the radiance/reflectance products usually hamper data sharing across sensors.

It is worth noting that the second goal of Proba-V was to demonstrate the feasibility to get the same observations as other much bigger satellites with a 1-cubic meter satellite and, on the other hand, that the main goal of PV-CC is to demonstrate that an even smaller 12U CubeSat platform can carry the same Proba-V vegetation instrument. Despite that, the performance of small cost-effective satellites is usually lower in terms of both calibration and processing algorithms. In this context, we expect that the synergistic exploitation of PV-CC in combination with Sentinel-2 might be beneficial for improving and validating the radiometric calibration and atmospheric correction of PV-CC. In particular, we propose a deep learning (DL) method that learns a general transformation model to convert PV-CC top of the atmosphere (TOA) images into S-2 bottom of atmosphere (BOA) images. Basically, it can be considered as a domain adaptation model that learns how to atmospherically correct PV-CC images by learning to reduce the statistical differences between PV-CC TOA images and S-2 TOC images.

Standard atmospheric correction (AC) of multispectral images usually involves the estimation of atmospheric transmittance at the sensor spectral bands using radiative transfer models and taking into account the atmospheric conditions at the time of image acquisition. Therefore, atmospheric correction is a complex, difficult and ill-posed problem that we will approach following a two-stage process. In the first stage, we will develop a data-driven atmospheric correction model using Sentinel-2 MSI data only. This first DL-AC model will be trained using Sentinel-2 Level-1C TOA images as inputs and Sentinel-2 Level-2A BOA images, obtained with Sen2Cor, as reference outputs. It means that we are developing a machine-learning-based emulator of the Sen2Cor processor using the four spectral bands in common with PV-CC, which will be used as a pretrained model for the second stage. In the second stage, we will fine tune the DL-AC model to convert PV-CC TOA images into S-2 BOA images. At this stage, we will use Proba-V data as the best proxy for PV-CC, which presents different but compatible spatio-spectral characteristics. Training these deep learning models requires a comprehensive dataset of TOA/BOA images, so we will create a geographically diverse database for the whole 2019 year covering all seasons and biomes.

Acknowledgements: This work was supported by the Spanish Ministry of Science and Innovation (PID2019-109026RB-I00) and by the European Space Agency (ESA IDEAS grant).

Developed by prime contractor Aerospacelab in Belgium for ESA, PROBA-V Cubesat Companion (PV-CC) In-Orbit Demonstration (IOD) mission is a small low-cost 12-unit CubeSat, satellite built up from standardised 10-cm boxes. It will fly a cut-down version of the vegetation-monitoring instrument aboard the Earth-observing PROBA-V to perform experimental observations. This IOD mission will be used to explore the harmonisation and synergistic exploitation of datasets captured by different platforms, in a different orbit, but with the same sensor. Additionally, it aims at developing and testing innovative Calibration/Validation (Cal/Val) strategies for small satellites. The mission is supported through the Fly in-orbit testing element of ESA’s General Support Technology Programme.

PV-CC will be flying at 550km in a Sun-synchronous orbit providing a similar Earth observation dataset as PROBA-V, but with an improved spatial resolution (65m at nadir). VITO will be in charge of the development and the hosting of the data processing facility and the Cal/Val chain. As PV-CC is carrying the same Vegetation imager, the PV-CC data processing and calibration will largely reuse the existing PROBA-V ground segment infrastructure. Processing adaptation and fine-tuning are however needed to accommodate for the PV-CC specifications and differences.

Considering the smaller size of the PV-CC platform, temperature effects might be more significant for PV-CC causing more scattering in the in-flight radiometric calibration result. Additionally, PV-CC mission will be using Cubesat platform, which is known to be less accurate in terms of geo-pointing performance and thus reducing the on-ground geolocation accuracy. To overcome all these challenges, a Cal/Val plan has been prepared by VITO, where adaptation to the existing PROBA-V Cal/Val methods and tools were detailed.

In this paper, the overall PV-CC Payload Data Ground Segment (PDGS) development status and milestones will be detailed and the proposed Cal/Val strategy presented. The latter will rely on the experience gained with the PROBA-V mission with respect to performance and applicability of the different vicarious Cal/Val methods. It considers furthermore the specific characteristics of this Cubesat Companion mission such as the longer revisit time at nadir (approximately 12 days) and the possible larger sensitivity to thermal environment changes.