1. ESA UNCLASSIFIED – For Official Use Intercomparison of Proba-V observations to simulations over the Libya-4 site M. Bouvet – ESA/ESTEC Proba-V QWG#2

2. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 2 Outline Proba-V observations vs. ProbaV simulated observations (on the MERIS radiometric scale) over Libya-4 Proba-V observations vs. simulated VGT-2 observations over Libya-4

3. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 3 Objectives Two questions: Can we directly compare the Proba-V and MERIS observations (supposedly a well calibrated instrument)? Can we directly compare the Proba-V and VGT-2 observations (a sensor interesting for Proba-V users)? Issues: Different in RSRs Fly/flew at different times Observe the Libya-4 site under different viewing/illumination geometries Third question: Does Proba-V radiometry evolve in time?

4. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 4 Approach Q: Proba-V vs. MERIS, what’s more interesting to compare? 1.Compare the radiometric scale of Proba-V to the radiometric scale of MERIS (i.e., differences in RSR accounted for and only differences in absolute calibration are left)? 2.Compare the Proba-V observations to the MERIS observations (i.e., differences in RSR and differences in calibration add up)? A: Point 1 => We address the first point using a hyperspectral (1 nm) TOA reflectance model ‘calibrated’ on MERIS (3 rd reprocessing) observations of Libya in the 2006-2009 period that can simulate the TOA reflectance in the Proba-V spectral bands, for any illumination/viewing geometry.

5. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 5 Approach Q: Proba-V vs. VGT-2: what’s more interesting? 1.Compare the radiometric scale of Proba-V to the radiometric scale of VGT-2? (i.e., decouple the differences in RSR from the differences in absolute calibration) 2.Compare the Proba-V observations to the VGT-2 observations? (i.e., have both differences in RSR and differences in calibration added up) A: Point 2 (assuming users are mostly interested in time series continuity) => We address the first point using a TOA reflectance model ‘calibrated’ on VGT-2 observations of Libya in the 2006-2009 period that can simulate the TOA reflectance in the VGT-2 spectral bands … but we will still address Point 1

6. ESA UNCLASSIFIED – For Official Use Proba-V observations vs. Proba-V simulations on the MERIS radiometric scale

7. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 7 Approach (from QWG#1) We start from: a hyperspectral TOA reflectance model based on the Monte Carlo MYSTIC radiative transfer code A pre-defined atmosphere that requires O3 and WV as input A surface BRDF model with 4 free parameters. ‘Calibration’ of TOA model on MERIS = inversion of the 4 BRDF free parameters that best explain the MERIS observations at each spectral band. The inverted BRDF models obtained in each MERIS spectral band are spectrally interpolated using a sand BRF measurement database => hyperspectral surface BRDF model TOA observations from Proba-V can be simulated using the hyperspectral BRDF model and the same pre-defined atmosphere + the Proba-V spectral responses See Bouvet M., Radiometric comparison of multispectral imagers over a pseudo-invariant calibration site using a reference radiometric model, Remote Sensing of Environment 140 (2014) 141–154

8. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 8 The radiative transfer model MYSTIC (http://www.libradtran.org)http://www.libradtran.org Monte Carlo Full treatment of polarisation Full treatment of absorption and scattering processes Coupling between atmosphere and surface Spectral parameterisation LOWTRAN 7 (Pierluissi et al. (1986)) adopted from the SBDART code (Ricchiazzi et al., 1998) - resolution of 0.32 nm at 400 nm and 2 nm at 1 micrometer Spectral sampling chosen: 1 nm Atmosphere Summer mid-latitude profile scaled with ERA-Interim O3 and WV corresponding to each acquisition to be simulated So-called ‘continental average’ aerosol model described in Hess (1998) with AOT=0.2

9. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 9 The surface BRDF parameter optimisation BRDF model: a 4 parameter model RPV model proposed by Rahman et al. (1993) Modified to allow flexible representivity of the hot spot =f(ρ 0, k, θ, ρ c ) The difference between modelled and measured MERIS TOA observation is optimized using the cost function: The optimization is driven by a downhill simplex (Nelder et al. 1965)

10. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 10 MERIS times series

11. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 11 MERIS observations vs. simulations

12. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 12 Inversion results: model capability at describing the TOA signal variability The spectral variations of the RMSE between the MERIS TOA reflectance simulations and the observations is shown in red. In black, the MERIS TOA reflectance standard deviation for the 4 year period. In blue, the mean photon noise associated to a single simulation of a MERIS TOA reflectance observation.

13. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 13 BRDF inversion result: surface BRDF model

14. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 14 BRDF inversion result: BRDF model 1.Visible hotspot Inverted BRDF model at MERIS 620 nm band. Plotted with Anisview: software developed by P. Vogt and M. Verstraete Bowl shape Backscatter stronger Hot spot

15. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 15 Spectral validity of the model: a sand spectral database 83 BRF sand spectra measured on samples collected at the Simpson desert (Australia), Muleshoe (Texas, USA) and in the Namibian desert. Illumination is at 45 degrees and measurement at nadir (courtesy J. E. Bullard and K. H. White )

16. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 16 Spectral validity of the model: spectral interpolation based on the sand database The BRF interpolation between the MERIS spectral bands is achieved by: Principal Component Analysis of the 83 hyperspectral BRF measurements Identification of the first 8 hyperspectral eigen vectors Least square minimisation of difference between a linear combination of the 8 hyperspectral eigen vectors and 12 MERIS spectral BRFs (where gaseous absorption is minimal)

17. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 17 MERIS, Proba-V (Center/Left/Right) and VGT-2 RSR

18. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 18 Applying the hyperspectral TOA model to MERIS full archive

19. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 19 Application of the hyperspectral TOA model to simulate various sensor observations Bias between model and 10 years of MERIS data for all spectral bands < 1% RMSE between model and MERIS < 1.5 % outside absorption bands (not shown here) VGT-2 radiometric scale about 4 % lower than MERIS radiometric scale

20. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 20 Current methodology vs. the IVOS/WG4 (PICS) results Comparison involving CNES, VITO, ESA, RAL using MERIS as reference sensor over 3 sites (of which Libya- 4) and 4 sensors.

21. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 21 Is Libya-4 a radiometrically stable site? MERIS trends over 10 years < 0.1% / year Similar temporal trend estimates found in the Red/NIR for other sensors Largest trends for the blue where largest instrument degradation are expected This might point towards Libya-4 being stable to at better than 1%/decade (at least in the red/NIR and likely in the BLUE too)

22. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 22 Proba-V observations vs. ProbaV simulated observations on the MERIS radiometric scale PROBA-V ICP modifications => XCEL spreadsheet ftp://ftp.estec.esa.int/pub/xe/anonymous/ProbaV_vs_MERIS_and_VGT2_results.html

23. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 23 Proba-V SRD requirements

24. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 24 Proba-V observations vs. Proba-V simulations on MERIS radiometric scale Band name TMA name Nb acquisitions Nb acquisitions discarded by 3-sigma filtering Mean of (rho_meas- rho_sim)/r ho_sim in % 95% CI assuming normal distribution in % Standard deviation associated to (rho_meas- rho_sim)/r ho_sim in % 1-sigma CI associated to each simulation of MERIS in % Temporal linear trend in %/year 95% CI assuming normal distribution in %/year BLUELEFT1723-1.520.251.631.50.410.43 BLUECENTER966-0.220.331.621.5-1.550.48 BLUERIGHT15013-2.180.130.771.50.180.22 REDLEFT1697-1.820.1511.5-0.580.26 REDCENTER984-0.370.271.341.5-1.690.32 REDRIGHT14914-0.980.130.821.5-0.420.23 NIRLEFT1697-2.810.322.061.5-1.170.53 NIRCENTER957-1.550.331.61.5-1.20.53 NIRRIGHT1549-2.180.281.741.5-1.30.45

25. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 25 Proba-V observations vs. Proba-V simulations on MERIS radiometric scale The comparison of PV observations and PV simulations indicates that: Assuming MERIS 2% absolute radiometry uncertainty is true (so far not demonstrated otherwise), the absolute accuracy requirement < 5 % (VI-PER- 220) is met for all VIS-NIR bands. On average for all bands and TMA in the VNIR, Proba-V radiometry is about 1.5 % below MERIS radiometry Stability < 3 % (VI-PER-240) over 6 months is met although statistically trends up to about 1.7 % / year are found but these are explained partially by the operational changes to the calibration Interband < 3 % (VI-PER-250) is also met if we restrict the interband to the VNIR bands

26. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 26 Proba-V observations vs. Proba-V simulations on MERIS radiometric scale Remaining issues (of the model?): NIR oscillation due to residual error in water vapour absorption modelling? What about the SWIR?

27. ESA UNCLASSIFIED – For Official Use Proba-V observations vs. VGT-2 simulations

28. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 28 Proba-V observations vs. VGT-2 simulations: the approach The reference TOA model outputs VGT-2 like observations over the Libya-4 site (in 4 spectral bands only) The approach is similar to the one used previously: a TOA model with 4 free parameters (surface BRDF) is fitted to the VGT-2 observations in the 2006-2009 period. BUT There is no attempt to further account for the differences in RSR between VGT-2 and Proba-V (i.e., no spectral interpolation of the obtained surface BRDF models in the VGT-2 bands)

29. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 29 Performance of the VGT-2 TOA model

30. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 30 Proba-V observations vs. VGT-2 simulations ftp://ftp.estec.esa.int/pub/xe/anonymous/ProbaV_vs_MERIS_and_VGT2_results.html

31. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 31 Proba-V observations vs. VGT-2 simulations: synthesis of results Band name TMA name Nb acquisitions Nb acquisitions discarded by 3- sigma filtering Mean of (rho_meas- rho_sim)/rho_ sim in % 95% CI assuming normal distribution in % Standard deviation associated to (rho_meas- rho_sim)/rho_ sim in % 1-sigma CI associated to each simulation of VGT-2 in % Temporal linear trend in %/year 95% CI assuming normal distribution in %/year BLUELEFT17231.990.271.7920.710.47 BLUECENTER9664.860.331.642-1.210.53 BLUERIGHT148152.140.171.0520.390.3 REDLEFT17061.40.181.152-0.770.28 REDCENTER9752.330.281.392-1.650.35 REDRIGHT148151.790.160.982-0.550.27 NIRLEFT16790.420.3122-1.250.51 NIRCENTER9482.410.331.612-1.370.52 NIRRIGHT153103.010.311.932-1.60.48 SWIR1LEFT964-1.290.321.562-1.70.44 SWIR1CENTER4330.640.441.432-1.070.68 SWIR1RIGHT4120.020.541.742-1.150.93 SWIR2LEFT6630.20.311.272-1.780.38 SWIR2CENTER3620.610.551.652-0.740.96 SWIR2RIGHT682-0.970.331.362-1.260.47 SWIR3LEFT5030.370.411.442-1.160.6 SWIR3CENTER4710.620.642.192-0.531.16 SWIR3RIGHT867-3.860.170.7920.380.32

32. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 32 Proba-V observations vs. VGT-2 simulations: synthesis of results The comparison of PV observations and VGT-2 simulations indicates that: Mean relative difference between PV and the simulated VGT-2 in VNIR spectral bands is +2.3 %. It was -1.5% between PV observations and PV simulations based on the MERIS radiometric scale. Mean difference PV and the simulated VGT-2 is on average over the SWIR detectors ~0% Interband < 3 % (VI-PER-250) is not met if we now take VGT-2 as reference for interband. BLUE/CENTER and SWIR3/RIGHT appear to be outliers. The stability requirement < 3 % (VI-PER-240) over 6 months is met Remaining issues (of the model?) => Larger than expected seasonal oscillations in the SWIR possibly due again to a modeling of the water vapour absorption?

33. ESA Presentation | Marc Bouvet | ESRIN| 28/10/2015 | Slide 33 Next steps Continue the monitoring of Proba-V over Libya-4 Use Proba-V calibration data over ocean and apply the Rayleigh, Glint and Desert methodologies implemented in ESA/DIMITRI radiometric tool. Estimated start time: end 2015