SESSION: INTERCALIBRATION

• Will CPF look at PICS- Yes but scale and number to be dependent on operations not a priority of mission so will depend on lifetime etc. 
• For angular correction: LUT built with simulated spectra for 10000 scenes with various combinations of solZen, satZen and relativeAzim. identify in a Data base which type of scene observed, then select the corresponding spectrum for the target sensor geometry. Knowing that solZen will be very close as acquisitions are within 10min and sat view Zen are within 2-2.5°. Relative Azim is tolerable with up to a few degrees. 
• Is there a sensitivity to clouds moving within nominal 10 minutes matching window? Effect is not strong within a large intercal FOV. But there is a scene homogeneity criterion used in defining a match-up, which will account for cloud. 

Talk 2: Ali Mousivand (Eumetsat),“Lunar and vicarious calibrations: PICS, BRDF characterisation, DCC and more at EUMETSAT”

• Eumetsat Micmics monitoring and calibration tool of Eumetsat (in prototype) described – provides operational use of PICS as references, PICS as a transfer from a reference sensor, DCC, and the moon. For a range of sensors primarily targeting operational sensors at present but not limited to.
• Both PICS methods focus on six priority sites, and utilise a 4 parameter BRDF model and RTM based on 6S with custom aerosol model but scaled (US standard atmosphere) atmospheric parameters derived from ECMWF and AOT from CAMS. 
• Methods achieve around 2-3% Reference sensor is VIIRS (N20) – primary sources of uncertainty are surface BRDF model and polarization, atmospheric parametrisation, cloud masking and site homogeneity and target instrument spectral response function knowlege. BRDF parameters the most important uncertainty. 
• DCC method uses DCC as a reference target to transfer cal from reference sensor VIIRS – 11-13LST data used takes rolling average of upto 30 days exact criteria for what is enough still subject of research. Spectal band adjustment factors from NASA to adjust between reference and target instrument. 
• Moon uses either of two models GSICS ROLO (GIRO) or LESSR and only used to monitor trends due to absolute uncertainty of models- Main error sources are Over sampling factor and pixel solid angle (which can vary as a function of pixel) and observation. 
• Results show that methods (particularly PICS illustrated) when used over a long period to remove seasonality effects due to BRDF residuals etc can be <0.5% in relative terms and for some spectral bands better than this. Results show a very consistent bias between S3 OLCI A and B for example 
• Important to note that different methods are better for different things and sensors and a strategy to optimally combine methods is still work in progress. 
• MICMICS and Vicarious cal methods on it shown to be highly valuable for FCI following on-board cal facility failure. 
  

• Asked about uncertainty for DCC 1% assessed (relative to the reference sensor) 
• Can TRUTHS do angular over DCC to assess BRDF? Probably only needed to confirm Lambertian nature which should be the case. DCC only to be observed 11:00 to 13:00 and can only provide high reflectance values (easy to saturate pixels) 
• How to identify DCCs can we use Oxygen A band in absence of TIR channel? Suggested to also make use of Geo satellites. 
• Ali Noted how FCI difficulties highlight the value of Missions like TRUTHS and also in particular the value to missions without on-board capabilities like commercial satellites. 
• It was noted that the RTM code and its parameterisation in MICMICS is out of date and could benefit from more bespoke solutions for higher accuracy. 
• Question raised that the value of sites like RadCalNet were limited due to the relatively large variation of BRDF as a function of time – It was noted that this was primarily for the standard RadCalNet product which only gives Nadir values however, time corrected BRDF corrections are available but only from the site owners and so such sites are highly valuable for higher resolution sensors as change from a nominal TRUTHS calibration at a point in time can be extended by the local instrumentation to greater than 10 minute windows. 
• Facilties like Micmics would benefit from a TRUTHS reference calibration and also from BRDF calibrations of the PICS. 

• Long term vision of TRUTHS as a central support function for CEOS/GSICS calibration infrastructure presented i.e. PICS, RadCalNet, moon, DCC, Rayleigh, Sunglint, Ocean Col Buoys. 
• Sensitivity analysis on choice of spectral characteristics of TRUTHS- bandwidth for S2S calibration for different sensors and scene types – impact on residual uncertainty due to spectral matching - showed 4 nm allows 0.1 uncertainty, Similar for spectral sampling interval (equal to bandwidth is ok) wavelength knowledge of 0.1 nm also resulted in <0.1 uncertainty 
• Wavelength and FWHM can be assessed in orbit via Fraunhofer lines and spectral features of ions in doped glass certainty. 
• Processing chain for comparing sensors viewing same target at nominally same time presented showing impact of spatial, spectral matching and resultant uncertainties for different scene types using S2 as a sensor under test and real observation data. 
• Illustrated uncertainty reporting via a Fiduceo like ‘tree’ 
• Showed for some bands near the absorption features and sharp spectral features e.g. red edge wavelength knowledge can be critical - However note as TRUTHS in matching a spectral response profile of sensor under test wavelength knowledge uncertainty is from sensor under test. But also this is true for the sensor in its use for applications! 
• Critical issue to note is that TRUTHS can only really provide a calibration to a sensor if the spectral response function of the Sensor under test is known or assumed – its knowledge in orbit will limit achievable uncertainties.

• Goal to develop E2E global intercalibration simulator (intercal digital twin) 
• Evaluate S2S on a global scale for all conditions where defined match up criteria are met – latitude band (avoiding high polar regions) only land, time limited to values e.g. <1 minutes. 
• Orbit simulated at 5s resolution and a match up defined as at least 30 km - for each match up BRDF corrected using MODIS data and atmosphere AOT using CAMS. 
• Directional effects (temporal changes to sun angle and view angle matching) found to be most simulating. Simulation with S2 shows that if no limit to VzA can have a 1% bias but if limit to <5 degrees can have no (<0.01%) bias. The latter still having a reasonable number of match ups for realistic conditions e.g. cloud etc for a global scale but less so for individual specific sites. 
• For specific relatively homogenous sites e.g. PICS, RadCalNet for even 15 degree Vza can achieve <0.3 % error without any BRDF correction. With ~20 manouevers per year can have ~30 cloud free match ups for a specific site of <15 degrees. 
  

• It was noted that number of manoeuvres will be limited to avoid impact on a long time base global benchmark 
• Recommendation to sample BRDF where possible for key sites e.g. Libya 4 and also to sample in the principle plane where most of the key info on BRDF can be derived. 
• Sampling BRDF from space provides complementary information to that measured at surface.