E. Tilmans ESA C. Baijot - ESA F. Niro -ESA J. Vandenabeele BELSPO. Maisongrande CNES D. Clarijs -VITO W. Dierckx -VITO E.

Transcription

1 Proba V 6 th Quality Working Group (QWG): Summary Report The 6 th Proba-V QWG took place in Redu on 09 th - 10 th Nov, 2017 Partecipants: R. Biasutti - ESA S. Dransfeld - ESA E. Tilmans ESA C. Baijot - ESA F. Niro -ESA J. Vandenabeele BELSPO Maisongrande CNES D. Clarijs -VITO W. Dierckx -VITO E. Swinnen - VITO I. Benhadj-VITO S. Sterckx -VITO S. Adriaensen - VITO J. Leon Tavarez -VITO C. Tote VITO I. Moreau UCL J. Redoux -UCL C. Henocq ACRI Y. Govaerts Rayference E. De Grandis SERCO R. Lacaze HYGEOS C. Brockmann Brockmann Consult Objectives of the meeting The mission status and the two main points of discussion for the current QWG meeting are recalled: 1. How to correctly handle mission s drifting phase, in particular, how to mitigate the effect of changing SZA, which will introduce spurious long-term trend in TOC reflectances and NDVI time series, as already demonstrated for AVHRR-NOAA series. 2. How to justify a further extension of the mission beyond end of 2019, considering that the mission gap-filling requirement to S3 will be met by that time and that the LTDN will continue drifting at increasing pace, reaching the 9:00 AM limit toward the end of Concerning the first question, the identified mitigation actions include the improvement of the current Atmospheric Correction (AC) approach and the provision of Normalized Bottom of Atmosphere Reflectances (correction of BRDF effects). About the second question the rationale is to identify key user requirements and existing or innovative scientific applications, which will justify the extension of the mission beyond Flight and Ground Segment Status On the Flight Segment side, the platform and the payloads are performing extremely well with no sign of degradation. The satellite is most of the time in the Nominal Observation Mode, with all sub-systems nominally working, and with the Attitude and Orbit Control System (AOCS) in the operational Nominal Pointing Mode. The AOCS performances requirements are met by far, platform-pointing accuracy very good. The thermal performances of the instrument are excellent and very stable: small increase of temperature of the optical bench due to the Antarctica acquisition, which imply skipping the sun bathing mode; impact on the data quality (dark current) was checked by VITO with no sign of change or degradation. The number of decompression errors suddenly decrease after the first yaw maneuver in July 2017 allowing to continue operate in the primary lane, the reason for such decrease is not understood yet. Two anomalies occurred since May 2017: Start Tracker 1 no valid attitude received for long time, the event was due to the blinding induced by the angular drift caused by 2 very close consecutive on-board reconfigurations (no action for this anomaly); the GPS sent erroneous velocities by leading to satellite safe mode for 6 hours, it has been concluded that a simple on-board monitoring to power cycling of the GPS could resolve the problem with limited impact on the operations The yaw maneuvers for the left and right camera was planned on July 2017 and a new instance for the left camera, not executed due to an error in the uploaded attitude values, was successfully acquired in October. The Antarctica acquisition is currently performed using the calibration planning and processing chain, the extra data downlink is accommodated within the margin of the X-band downlink. Since beginning of November 2017 the data are

2 acquired and ingested within the calibration data processing chain but the data handling and processing in this scenario is considered however cumbersome, since a significant amount of manual work is required. The update of the integration time matrix would allow automatizing the process allowing to redirect the data to the nominal chain. The investigation on how to implement such change without impacting the nominal acquisition over continental landmasses is on-going and final feedback from QinetiQ is expected. A special request was received from UCL to repeat the measurement campaign carried out during 2013 for the EPT (Energetic Particle Telescope) sensor. There are no special constraints from VITO to proceed with the special off-pointing maneuver during December 2017, although it was requested UCL to provide a detailed planning of the required maneuver 2 weeks in advance in order to check for potential conflict with calibration activities performed at night (dark current). On the Ground Segment side, all segment operations are running nominally and the performances are well within requirements in terms of data availability and timeliness, both for nominal processing and for the MEP. The number of users which was steadily increasing in the past years now reached a plateau, while an increasing number of users is now moving to the MEP infrastructure. The UCL study on 100m time series exploitation for Locust Habitat Monitoring on the MEP is presented. VITO is investigating on the interest in providing Proba-V data in UTM projection, in addition to the currently used Plate Carrée global projection. The idea is to use the same tiling and projection scheme, which is currently used for S2, this will allow to ease the usage of Proba-V data in combination and in synergy with S2. This option will be investigated as potential extension of the Proba-V Toolbox. Radiometric and Geometric calibration The results of radiometric calibration are presented: no significant trend is observed for the VNIR bands, confirmed by the long term stability monitoring using monthly lunar observation for the VNIR central camera and by the DCC intercalibration results for the Blue and NIR bands. However, since May 2017 a slight degradation trend of about 0.5% is observed in the Blue Center and Left camera, confirmed by Desert, DCC and Moon calibration, starting from that date a degradation model is applied for these cameras. The SWIR band shows a slight degradation trend in the order of 1%- 1.5%/year relatively uniform for all camera strips. The degradation in the SWIR is taken into account within the degradation model, which is applied for the Collection 1 dataset in order to avoid step-wise update of the calibration coefficients. The latest results from the analysis of the special Yaw maneuvers calibration performed in order to improve inter-pixel calibration are presented: the coefficients derived for all cameras yield to improved pixel uniformity for the SWIR strips, allowing to reduce striping effects. The coefficients derived from the yaw maneuver will be used in the operational calibration and applied to the full archive for next reprocessing. It is agreed to repeat a similar campaign during mid A very good behavior for the geometric error is observed, the Absolute Accuracy is excellent (<100m for all cameras) and well within mission mandatory and goal requirements. No sign of degradation is observed; geometric performances remain stable in time. Inter-band geometric accuracy is also extremely good and well within the specifications with values ranging around 50m and no sign of degradation with time. Multi-temporal accuracy is very good with specification requirements met for about 90% of the considered GCPs. A geometric shift anomaly, a large shift of multiple pixels, was detected in February The shift was too big (>1km) to be detected automatically, therefore the data were not appropriately flagged in the processing. The shift occurred for about 24h between two orbits during 18/02/2016. After a deep analysis, the issue has been forwarded to QinetiQ and they found that the anomaly was due to erroneous coordinates sent by the on-board GPS receiver to the AOCS. The on-board AOCS SW rejected these erroneous GPS values falling back into the usage of the TLE for propagating the orbit; depending on the age of the used TLE, an error of few km can be expected, this explain the anomaly observed in the data. It has been discussed on the fact that a similar anomaly could have been solved implementing an automatic power cycling of the GPS receiver, to be triggered in case of erroneous GPS values. It is suggested that a close monitoring of the performances of the GPS should be implemented at REDU. This is in principle possible using the house-keeping telemetry, although the required effort to operationally implement such monitoring system should be assessed.

3 Algorithm Baseline Definition for C2 Despite the improvements obtained with C1, issues related to cloud screening are still present, in particular a latitude dependent anomaly (high latitude) appearing every year during Dec/Jan. The issue is caused on the one side by the presence of several pixels classified as undefined in the input land cover map used to drive the cloud algorithm, on the other side, by the use of the GlobAlbedo map as gap-filling in case of missing pixels in the background surface reflectance climatology, which was built from MERIS data. The effect is both over-detection due to the pixels classified as undefined and under-detection in case of the use of the GlobAlbedo brighter background reflectance. Looking at a definitive and accurate solution, the most consistent approach would be to build climatology of surface reflectances based on the full archive of Proba-V data and use this climatology as input to the cloud-screening algorithm. The adoption to the latest ESA CCI Land Cover map would additionally improve cloud classification algorithm, owing to the more accurate pixel labelling. For the validation purpose, the Pixbox database generated by Brockmann Consult will be reused. In addition an underdetection of thin clouds in particular for the 100m dataset, and an over-detection of clouds for bright land pixels are pointed out; the omission of cloud shadows was also reported to impact some downstream applications. Investigation paths in order to address all these issues were identified and improvements will need to be incorporated within C2 baseline. Concerning the BRDF correction, four different semi empirical models, Roujean, MODIS, MODIS+Hot Spot and RPV, are compared and discussed. The assessment is performed with respect to the consistency between SPOT-VGT and Proba-V dataset during the overlapping period. The results show that the MODIS with Hot Spot model performs better than the others, although differences are limited. The RPV model seems the most sensitive to residual undetected clouds in the TOC reflectances. The results are presented and the proposed choice of a MODIS-like approach showed some clear advantage, since it is widely used within the EO community. It is agreed to proceed with the recommendations provided in the relative TN, i.e. to consider the MODIS + Hot Spot algorithm as baseline for C2. It is stressed that the implementation of such algorithm in the frame of C2 baseline means the provision of new products to the users, i.e., multi-days Angular Normalized TOC reflectances with associated BRDF kernel parameters. The need for Angular Normalized TOC reflectances was recognized as crucial in order to provide the users with a long-term dataset free of any spurious trend induced by the orbital drift and to solve the remaining inconsistencies between SPOT-VGT and Proba-V archive. Additional effort will be required to VITO to optimize computation time for the BRDF correction algorithm and the compositing strategy should be verified with the expectations of the user community. The Atmospheric Correction will become progressively more challenging in the orbit-drifting scenario moving toward higher SZA. A comparative analysis of different atmospheric correction approaches is carried out providing an assessment of AOT and SR estimation. The following datasets are considered: the AOT estimated within the operational processing chain, the CAMS NRT, the CAMS climatology, the AOT estimated with icor processor, and the AOT retrieved by the CISAR algorithm. The estimated AOT is compared against AERONET over 7 stations, representative of different land cover types and aerosol loading. The SR estimated using as input the different AOT datasets are compared to reference SR built using as input the AOT provided by AERONET. The main conclusion of the investigation is that the CAMS NRT provides the best estimate of AOT and of SR for the considered validation dataset. The required effort in order to implement any of these changes in the operational chain is also presented. CAMS NRT will require effort mainly for the implementation of the data flow. A more general question was addressed: whether the AOT should be retrieved from the images or taken from an external dataset, the preferred approach within the Proba-V QWG is to retrieve AOT from the image itself and use CAMS as fallback. The approach for the SR validation was discussed, a more suitable approach would be to use for the reference SR a different and more accurate RTM respect to SMAC which is known to be an approximation of the radiative transfer with some known limitations, namely for high SZA condition. It is agreed that the preferred option is to implement both BRDF and AC at the same time and be consistent at all resolution, namely to have a single AC scheme for all products: 1km, 300m and 100m. The possibility to provide within the output TOC products the estimated AOT, and the used WV, as it is currently done for S2 and S3, is recalled. This has obviously a strong impact on the PDGS, since the size of the products will increase significantly. It is agreed that the possibility to provide AOT, and WV, within the TOC products should be investigated for the definition of the C2 baseline.

4 In conclusion it is agreed that the results of SR validation, and partly also the AOT validation, are not mature enough in order to draw final conclusion on which AC processor will be used for C2 baseline. It is therefore recommended to extend the considered reference AERONET dataset, mostly in time, and to use a more accurate RTM for the computation of the reference SR, the preferred approach is to use CISAR for this purpose, 6SV could be an alternative. The assessment should mainly focus on SMAC, CAMS and icor performances. Copernicus Global Land Service (GCLS) feedback on C1 data A review of the current available CGLS downstream products is presented, the portfolio of products has evolved since last meeting. Proba-V is the key contributor to Vegetation and Energy products (e.g., FAPAR, LAI, NDVI, DMP, BA, TOCr, Albedo), but it is also used for Water Bodies. Proba-V C1 data is now the baseline for the CGLS operational production. With regard to C1 data quality, additional feedback are reported in particular on the quality of the cloud screening at 100m, which seems to be less accurate than for 300m products especially for thin cloud. Concerning the mission extension the availability of Proba-V data is key for CGLS to ensure a smooth transition to Sentinel- 3. To this end, Proba-V mission should be operated at least to end of 2019, when the constellation of S-3 satellite (A+B) is expected to be fully operational, including 1 year of overlap between the two missions. Any unforeseen delay in the S-3 B satellite s launch or on the availability of the relevant S3-SYN data, will provide justification to further extend Proba-V mission beyond end of 2019 and at least until full confidence is obtained on the consistency of S-3 data. Final Validation report on C1 dataset The final validation results for C1 archive are presented. The quality assessment is based mainly on the consistency between Proba-V and SPOT-VGT archives. Overall, the validation exercise demonstrated that the recent SPOT-VGT and Proba-V reprocessing activities yield to improved consistency between the two datasets and with respect to the external MODIS and AVHRR datasets. Concerning Proba-V C1 the improvements are mostly related to a more conservative cloud screening as well to improved inter-band calibration. Some remaining issues were identified with the cloud screening, such as commission error for bright land pixels, omission error of the first pixel around the cloud border and the latitude dependent artefacts due to the used background reflectance climatology and Land Cover Map. All these issues were already communicated to the VITO quality team and were addressed and discussed during this meeting with associated AI. PV-LAC project highlights The Proba-V Advanced Land Atmosphere and Coastal project (PV-LAC) was funded by ESA/BELSPO to prototype new algorithms for improving atmospheric correction and for demonstrating the interest of 100m products for coastal water applications. The CISAR algorithm, developed by Rayference, allows to jointly retrieving AOT and Surface BRF from the accumulation of Proba-V 1km observation over 16 days period. The resulting AOT is validated against AERONET data, showing overall good agreement, although some over-estimation for low AOT are observed, likely induced by undetected thin clouds. The Surface BRF is validated with respect to similar MODIS products showing a very good consistency. It is recalled the main idea of using this algorithm in the MEP to generate a long term CDR of AOT and BRF from SPOT-VGT and Proba-V dataset. The framework to initiate this project needs still to be defined and a significant effort is associated to this implementation since the algorithm is working on a single pixel and it will need to be extended to support processing of full images. Concerning the use of Proba-V for deriving OC 100m products over coastal areas, a prototype code was developed, allowing to retrieve Turbidity Map from Proba-V 100m data and consequently to derive the Total Suspended Matter (TSM) if regional in-situ data are available for calibrating the algorithm (CEFAS smart buoys measurements). The validation of the AC approach for the Coastal applications relies on AERONET-OC. The icor algorithm was used for AC, adapted over water pixel using the SWIR-based approach (K. Ruddick), the turbidity is retrieved from the Red band using a global algorithm developed by Dogliotti et al. The icor AC and the Turbidity algorithm could be implemented within the MEP and initially included as part of SNAP.

5 S3 SYN-VGT Algorithm status The status of the S3 SYN algorithm was presented, in particular the latest improvements on the cloud masking, aerosol interpolation and gap filling. Recently the SYN cloud-flagging module was updated with the support from Brockmann, it is based on Neural Network and used both OLCI and SLSTR solar channels, SLSTR TIR is not used yet. Example results of the new cloud flagging are shown, demonstrating good cloud detection. Assessment of the SYN Aerosol retrieval shows some AOT over-detection over specific geographical areas with resulting over-correction of the surface directional reflectances (SDR). The new S3 Global Aerosol product is presented; it will include AOD, SSA and SDR over Land and Ocean. The Global Products will be delivered at 4.5x4.5 km 2 resolution. The current planning is to have NRT production based on SLSTR starting during Q and NTC production, based on SYN products, during Q Mission Extension beyond 2019 The extension of the mission beyond the end of 2019 was discussed and a brainstorming session was held. Among the mentioned justifications we report the following: i) the need to consider any potential delay in the availability of S-3 A+B operational data and to have a long (at least 1 year) overlapping period between the two mission for inter-calibration; ii) the need to sustain some of the operational services which could benefit of Proba-V 100m data, such as the Locust monitoring performed at UCL; iii) the interest of Proba-V for preparatory activities of the Surface Water and Ocean Topography (SWAT) mission to be launched during 2020, in particular the complementarity of Proba-V 100m data for water bodies detection. These ideas will be reviewed and users will be consulted to gather any additional feedback, the final goal is to provide a TN with some solid recommendations.