Sentinel-2 and Sentinel-3 absolute calibration monitoring C. Desjardins & V. Lonjou CNES Radiometric S2A Calibration RRQI, 28 January Workshop, 2016, 30-31/08/2017 CNES
SUMMARY CALCON Aug 22-25 2017 VICARIOUS CALIBRATION METHODS SENTINEL-2 RESULTS SENTINEL-3 RESULTS CONCLUSION
Vicarious calibration over natural targets Absolute calibration coefficient estimation Ocean sites (molecular scattering) Desert sites Snow sites Deep Convective Clouds Ocean sites (sunglint) The moon Instrumented sites Absolute Multi-temporal Multi-spectral Cross-calibration MUSCLE (tools) SADE (database)
Aerosols Ocean surface RAYLEIGH Rayleigh calibration on oceanic sites What is Rayleigh calibration? Vicarious absolute radiometric calibration method Statistical approach based on molecular scattering (Rayleigh) Observation of the atmosphere over a dark surface (ocean) Calibration from blue to red spectral bands (440nm to 670nm) Main principles Molecular scattering contributes to 85 to 90% of TOA signal Ocean surface: reflectance estimated with a climatology (filtering of white caps using wind speed) 6 ocean areas used Aerosol filtering using NIR spectral band (865nm) Gaseous absorption: O3 (TOMS), NO2 (climato), H2O (meteo)
Rayleigh calibration on oceanic sites TOA reflectance Total atm. transmission for aerosols and molecules Total gaseous transmittance Molecular and aerosol contribution Marine reflectance Atmospheric albedo Aerosol and molecular scattering contribution Atmospheric functions (ρ A, T, S A ): computed using an accurate radiative transfer model (SOS), including polarization and specular reflection by the wavy surface Surface pressure molecular optical thickness molecular scattering contribution Aerosol optical thickness at 865nm + Maritime 98 aerosol model background aerosol contribution Marine contribution Estimated over the pre-defined oceanic sites through a climatological study, available in SeaWiFS spectral bands (spectral interpolation may be necessary). Bi-directional correction according to Morel et Gentili, 1993 may also be performed in case of geometrical differences between ref. value and viewing and solar observation geometry. Gaseous contribution Main contributors: water vapor, ozone, oxygen, NO2. Correction made with SMAC model, using exponential variation with air mass and gaseous amount. (from Fougnie et al., 2002)
Rayleigh calibration on oceanic sites Error budget for Rayleigh absolute calibration method Errors on parameters directly used as inputs for the computation of Rayleigh scattering contribution: surface pressure, wind speed B3 calibration errors impact the computation of aerosols contribution Gaseous absorption errors Marine reflectance errors Global RMS error inferior to 3.5% (depends on spectral band) Assets Absolute calibration method Validation of the polarization calibration Intra-FOV calibration Limitations and drawbacks Valid only for wavelengths up to 670 nm. Results are less noisy for green and red bands than for the bluest bands. Precise instrument characterization needed (ISRF, FOV variations, NL, SL, noise, )
Cross-calibration over desert sites 19 sites are used Criteria: homogeneity, temporal stability Reference sensors MERIS, MODIS, LANDSAT, SPOT5 Key aspects of the method Geometrical matching Spectral interpolation
Cross-calibration over desert sites For each matchup, [ Sensor_2, Sensor_ref ] measures with close solar and viewing directions Measured TOA reflectance Sensor_ref [ Sensor_ref spectral bands] Simulated TOA reflectance Sensor_ref [ Sensor_2 spectral bands ] Measured BOA reflectance Sensor_2 [ Sensor_2 spectral bands ] Inverse radiative transfer Direct radiative transfer Cross-calibration ΔAk Simulated BOA reflectance Sensor_ref [ Sensor_ref spectral bands ] Simulated BOA reflectance Sensor_ref [ Sensor_2 spectral bands ]
Cross-calibration over desert sites Error budget for Desert cross-calibration method Error on spectral resampling Quality of geometrical matching between measurements to calibrate and measurements from reference sensor Assets Applicable to all spectral bands (better results in the NIR and SWIR, more noise in the shorter wavelengths) Temporal calibration method (precision <1%, <2% in blue bands), independent from the reference sensor Limitations and drawbacks If used as an absolute calibration method: dependent on reference sensor bands+geometry (seasonal effects, spectral interpolation ) System impact : ISRF variations, spectral rejections, The entire FOV can be observed Large number of reference sensors available
Cross-calibration over snow sites of Antarctica 4 sites (120 x 120 km 2 ) DOME 1, DOME 2, DOME 3, DOME C MODIS images (resolution 250m) of the 4 antarctica sites (Nov 17, 2005) DOME 1 DOME 2 Approximative location of the Dome C region (Concordia) DOME 3 DOME C
Cross-calibration over snow sites of Antarctica Assets When measurements are possible, they are extremely frequent, because most sensors have polar orbits The sites are very homogeneous The sites are bright and not very spectrally dependent Sites at high altitude (3300m) not much water vapor and aerosols low impact of the atmosphere (except ozone) Limitations and drawbacks Measurements can only be performed during 3 months of austral summer High solar zenith angles increases the BRDF effects cloud filtering difficult if no dedicated cloud band is available Important ozone variations in this region impact of the choice of exogeneous data Calibration of SWIR bands is not possible due to ice absorption calibration of blue bands is less noisy than for desert sites
Inter-band calibration on Deep Convective Clouds Principle of the method In certain conditions, deep clouds provide a stable radiance Reflectance of the clouds difficult to assess no absolute calibration Reflexion of sun light on a dense cloud is «white» in VNIR inter-band calibration Use of «white» clouds in VNIR In sub-tropical convective systems, above the ocean High altitude clouds, very thick and large No contamination by tropospheric aerosols, surfaces, cirrus or non-cloudy neighbourhood But not pure white clouds reduced radiative effects of the upper atmospheric layer (low contribution of molecules, absorbing gases and stratospheric aerosols) Reflectance of the cloud is lower in NIR and MWIR (higher absorption by ice crystals)
Inter-band calibration on Deep Convective Clouds Assets Validation of other calibration methods Multi-temporal calibration (~0.5%) Limitations and drawbacks Absolute reference needed Not operational for SWIR bands Interband calibration (~1%) The whole FOV can be observed Not very sensitive to spectral instrument issues Polarization calibration All bands in VNIR => Very reliable method
Interband calibration over sunglint Principle of the method Accurate computation of the 2 main contributors of TOA signal: Rayleigh scattering Sunglint Minor contributions: Ocean surface ( climatology) Aerosols ( threshold using another viewing direction) Gaseous absorption (O 3, NO 2, H 2 O) Reference band: often red band Assets Can be used for all VNIR bands including absorption bands Precision: about 1 to 2% in cross-calibration Limitations and drawbacks It is an interband method: a reference band is needed Limited area in the FOV Precision: about 2% for temporal monitoring Not very sensitive to spectral instrument issues or misknowledge
atmosphere Radiometric calibration : in situ calibration using LACRAU robotic station (ROSAS) 380-1650nm (9=>12 bands) Ground + atmosphere caracterisation Proc. SPIE 8153, Earth Observing Systems XVI, 815311, 2011 Measured ground reflectance Modelled ground reflectance Simulated TOA reflectance Simulated TOA reflectance Sensor TOA reflectance Absolute calibration (all bands except B9, B10 & B12) 10 days revisit RadCalNet ( Gobabeb )
Calibration with instrumented sites ROSAS Robotic station RADCALNET network of instrumented sites (recommended by CEOS) LaCrau (South East of France) Flat plain of a few kilometers, covered with white pebbles and grass Used since 1987 (SPOT), automated since 1997 Gobabeb desert, Namibia Arid desert, tens of kilometers Very low cloud coverage Instrumented in 2017
MUSCLE / SADE List of projects and missions Rayleigh Deserts Snow Clouds Sunglint The moon Instr. sites POLDER 1, 2, 3 X X X X X VGT 1, 2 X X X X X SPOT 1,2 X X X SPOT 4,5 X X X X SPOT 6,7 X X X X PLEIADES 1A, 1B X X X X X MERIS X X X X X FORMOSAT X THEOS X SeaWIFS X X MODIS X X X Sentinel-2 A & B X X X X LANDSAT-8 X X Sentinel-3 X X X X X AATSR Soon Venµs X X X X X 3MI Soon Soon Soon Soon Soon (Soon)
Sentinel-2 absolute calibration monitoring Radiometric S2A Calibration RRQI, 28 January Workshop, 2016, 30-31/08/2017 CNES
Desert (PICS) vicarious calibration Principle Cross-calibration wrt. reference sensor (MERIS, MODIS, LANDSAT8 ) 19 desert sites Geometric matching : viewing and solar angles no temporal matching Temporal evolution + diffuser monitoring S2A S2B IOC Result Good agreement with MERIS (spectral reference) ESRIN Aug 30-31 2017
ESRIN Aug 30-31 2017 Desert (PICS) vicarious calibration S2A S2B IOC Result Spectral interpolation Good consistency with MODIS / L8
ESRIN Aug 30-31 2017 Desert (PICS) vicarious calibration S2B/S2A intercalibration S2B/S2A Both instruments are consistent within 1-2 %
ESRIN Aug 30-31 2017 Desert (PICS) vicarious calibration Sentinel-2A Temporal evolution Temporal evolution within +/-1% Same behavior for all spectral bands Calibration on the diffuser seems to accurately correct the temporal evolution of Sentinel 2A
ESRIN Aug 30-31 2017 + atmosphere in situ calibration using LACRAU robotic station (ROSAS) Principle Simultaneous in-situ measurements at Lacrau (South-Eastern France) Radiative transfer Not many acquisitions esp. for S2B Simulated TOA reflectance S2A All spectral bands within 3% - except for B8 (5%) S2B IOC Result
ESRIN Aug 30-31 2017 Principle Inter-band calibration on Deep Convective Clouds Diffusion of sun light on a DCC is «white» in VNIR inter-band calibration All bands within 3% (B8A 4%) spectral shape is representative of the fact that we don t have proper DCCs! (not core method) but similar for both satellites All VNIR spectral bands within 3%
ESRIN Aug 30-31 2017 Vicarious Calibration over molecular scattering (OCEANS) Principle Statistical approach based on molecular scattering (Rayleigh) over a dark surface (ocean) Molecular scattering constitutes most of TOA signal Calibration from blue to red-nir spectral bands (440nm to 740nm) Surface Reflectance Change of S2A/B2 spectral shape S2A All bands within 2-3% S2B IOC Result
ESRIN Aug 30-31 2017 S2B S2A intercalibration What if we used each method as a transfer method (double difference)? This allows us to get rid of spectral issues for each method since they are the same both satellites S2B/S2A S2B/S2A Results are consistent with the direct S2B/S2A intercalibration : no more spectral features this should be investigated later when more data are available.
Conclusion Sentinel-2 Overall quality of S2 intercalibration is very good (solid results for S2A, to be confirmed for S2B) Rayleigh : absolute calibration B1=>B6 < 2-3% Deserts Cross-calibration with MERIS, LANDSAT8 & MODIS shows a very good agreement with the diffuser (<2%). Data continuity with other missions is assured. Temporal monitoring confirms no degradation of the diffuser so far for S2A. S2A/S2B very good intercalibration (within 1-2%) Cloud Interband B1=>B8A within 3%, but very good S2A/S2B intercalibration! Lacrau Good consistency with diffuser, better than 3% (5% for B8) See Revel et al. 2017 EJRS paper (submitted)
Discussion Sentinel-2 Properly take into account S2A/B2 in MUSCLE-SADE DCC Change S2 DCC site based on S3A experience? Density map of usefull S3 measures Desert sites selection : To facilitate the methods intercomparaison, CNES can present results on a subset of sites for S2, (ie. 6 sites from DIMITRI) Dome S2 campaign? Instrumented sites Gobabed first results to come during fall!!
BACKUP SLIDES 29
S2A / B2 S2A/B2 spectral response at intrument seems not valid according to ADS (problem with test setup), S2B/B2 will be used instead. Is a significant improvement observed with the new response (not yet included in MUSCLE-SADE)? What do we do if not? The shape of the detector sensitity (less in the blue) is not visible in S2B? Would it be possible to model the spectral response using the tests that have been performed at FPA level? What about B1?
Variation of the spectral response in the field of view
Computing ERS for different landscapes, and different Sentinel-2 SUMMARY S2A S2B ERS statistics S2A ~ S2B, S2B seems a bit more affected by the problem Worst case is B5 (red-edge band) over vegetation, ~7-8% peak to peak A spatial signature is generally located around detector edges, but also we also observe steps between adjacent detectors or low spatial frequency evolution within a detector All the figures presented here are fully dependent on ground characterization processing that may be biased (test setup and/or processing)
VNIR variation~0.5% 1.05 1.03 1.01 0.99 0.97 0.95 1.05 Launch Solar diffuser Absolute calibration S2A 04/2015 07/2015 10/2015 01/2016 05/2016 08/2016 11/2016 03/2017 06/2017 09/2017 B1 B2 B3 B4 B5 B6 B7 B8 B8A B9 SWIR variation~1-2% Very high stability of the instrument 1.03 1.01 0.99 0.97 0.95 04/2015 ESRIN 07/2015 Aug 10/2015 30-31 2017 01/2016 05/2016 08/2016 11/2016 03/2017 06/2017 09/2017 B10 B11 B12 SWIR focal plane decontamination
Solar diffuser Absolute calibration S2B Launch VNIR variation~0.1% ESRIN Aug 30-31 2017 SWIR variation~2-3% SWIR focal plane decontamination 34
ESRIN Aug 30-31 2017 Sentinel-2 Comparison with other missions Landsat SPOT Sentinel-2 16 days 8-11 bands (inc TIR) Nadir viewing geometry Global Coverage 30m / 190 km Free data <26 jours 3-4XS + PAN Varying geometry On-demand coverage 10m / 60 km Commercial satellite 5 (10) days 13 bands Nadir viewing geometry Global Coverage 10m / 290 km Free data
ESRIN Aug 30-31 2017 Radiometric Absolute Calibration Specifications The absolute radiometric accuracy shall be 3% (goal) / 5% (threshold). The inter-band calibration accuracy shall be 3%. The intercalibration (S2A/S2B) shall be 3% Nominal method: onboard solar diffuser Validation methods: vicarious calibration using SADE/MUSCLE Molecular scattering over the ocean Crosscalibration over PICS Interband calibration over DCC Instrumented Site over La Crau Results are presented for Sentinel 2A (2 years in orbit) and 2B (4 months in orbit)
BRIGHTNESS Moderate Intermediate Very bright Cross-calibration over desert sites Profiles of the desert sites (homogeneity, brightness) Arabia 3 Libya 2 Niger 1 Egypt 1 Libya 4 Arabia 2 Sudan 1 Mali 1 Algeria 4 Niger 3 Mauritania 1 Libya 3 Algeria 2 Algeria 3 Libya 1 Mauritania 2 Algeria 1 Algeria 5 Arabia 1 Niger 2 Heterogeneous Inhomogeneous Acceptable Homogeneous Very homogeneous HOMOGENEITY
Surface reflectance Cross-calibration over desert sites Profiles of the desert sites Wavelength (nm)