This paper reports on recent activities on geometric calibration and validation of the Copernicus Sentinel-2, performed jointly by the Sentinel-2 Mission Performance Center (MPC), ESA and CNES.

The current geometric calibration strategy relies on the optimization of geometric calibration coefficients used by the geolocation processing on-ground. More precisely, the processing corrects for biases in Roll, Pitch and Yaw between the nominal and real orientation of the instrument reference line-of-sight. The process is as follows:
• Continuous monitoring of the performance by the MPC geometric validation team at TAS
• In case of degrading performance, new calibration processed by the MPC geometric calibration team at Airbus
• independent assessment performed at regular intervals by the CNES geometry team

Since the launch of Sentinel-2A in 2015, some lessons have been learnt from this process. The alignment biases of Sentinel-2A evolved slowly until after launch and reached stabilization after one year. The evolution spanned several tens of meters on ground, which required relatively frequent corrections. For Sentinel-2B, the stabilization period seems to be even longer, as significant changes are still observed more than one year after launch. In addition to this post-launch evolution, some oscillations can be observed with different time periods:
• Seasonal oscillations visible on Sentinel-2A (not yet on Sentinel-2B): approximately 2 meters on ground
• Orbital oscillations: approximately 10 meters from North to South, correlated with the local Sun declination
• Short term oscillations discovered more recently, about 3 m peak to peak

In the near future, another calibration approach will be introduced, using the Sentinel-2 Global Reference Image (GRI). In this approach, all images are co-registered on-ground to the reference image. This reference image is constructed by the MPC geometric calibration team at IGN, and independently validated by CNES, the MPC validation team at TAS for performance aspects, and by the MPC operation team regarding format aspects. During the last year, progress on the elaboration of the GRI concerned:
• Extension of the area to Northern latitudes and isolated islands
• Correction of minor format errors
• Completion of the performance validation

In parallel, improvement and validation of the refinement algorithm has progressed on the following topics:
• Handling of overlapping GRI images
• Filtering of homologous points based on texture content
• Optimization of refinement parameters
• Consolidation of the multi-temporal performance using the Ground Prototype Processor (GPP)

Another axis of improvement of the geometric performance is the use of more accurate Digital Elevation Model (DEM). First, a set of requirements for the procurement of a new DEM suitable for Sentinel-2 processing has been established. A global DEM with 30 m horizontal sampling distance (vs. 90 m currently) is expected to significantly improve the performance in “risky” areas (terrain with high slopes on the edges of the swath). In order to limit the impact on processing time, an adaptive resolution approach is currently under study.

Thanks to the geometric refining and the new DEM, significant improvements are expected both for the absolute geolocation performance and the relative (multi-temporal) performance.

The land surface reflectance is a fundamental climate data record at the basis of the derivation of other climate data records (Albedo, LAI/Fpar, Vegetation indices) and has been recognized as a key parameter in the understanding of the land-surface-climate processes. Here, we present the inversion and the validation of the Land surface reflectance for Landsat-8 and Sentinel-2 data.
This work relies on a generic approach developed to derive surface reflectance over land from a variety of sensors. The approach is based on the inversion of the radiative transfer equation in the Lambertian case, with no adjacency effects, that account for a simplified coupling of the absorption by atmospheric gases and scattering by molecules and aerosols as implemented in the 6SV radiative transfer code. The processing code relies on look-up tables generated by 6SV, for which the accuracy (~1%) has been well documented in several papers. The LaSRC code uses ancillary data such as pressure and gas concentrations but relies on a per pixel inversion of the aerosol properties to assure the best possible accuracy for the surface reflectance, as aerosols can be highly variable both in space and time. This new aerosol inversion builds on the extensive dataset acquired by the Terra platform, combining MODIS and MISR to derive an explicit and dynamic map of band ratio’s between blue and red channels and is a refinement of the operational approach used for MODIS and LANDSAT over the past 15 years. The aerosol inversion is generic and applicable to a variety of sensors. We use this basic approach to derive Landsat 8 and Sentinel 2 surface reflectance products.
Then, we present different schemes of validation. The first uses AERONET data with an error budget associated. The second presents performance of LaSRC inter-compared with other operational approaches in the framework of the CEOS-ACIX (Atmospheric Correction Intercomparison eXercise) Calibration Validation Task co-Lead by ESA and NASA.
Finally, we present the validation of the Sentinel 2 surface reflectance during the summer-fall of 2018 over a corn field at the United States Department of Agriculture Beltsville Agricultural Research Station. Measurements of the corn field reflectances for the coincident sentinel 2 data are taken from a boom lift 80 feet above the ground using four cameras (Red, Green, Blue and Near Infrared). The results of this first measurement campaign are presented as well as recommendations for future automated implementation of this system over a larger number of sites.

The Ocean and Land Colour Instruments (OLCI) are flying on-board the Sentinel3-A and Sentinel 3-B platforms. The Sentinel3-A satellite has been launch in February 2016 while Sentinel3-B was launch in April 2018. The two instruments have exactly the same design and their data are processed by the same processing chains, in continuity with the ENVISAT/MERIS mission with numerous improvements.
The OLCI radiometric calibration, as for MERIS, is based on in-flight measurements of sun-lit diffusers. Calibration relies on the diffusers on-ground characterisation, acting as on-board secondary reflectance standards: known diffuser BRDF and viewing/illumination geometry during acquisitions allow predicting spectral radiance at instrument entrance and, together with measured numerical counts appropriately corrected from instrumental effects, deriving calibration coefficients. Spectral calibration is a key input of this process as it directly impacts the computation of the diffuser radiance. OLCI, as MERIS, allows in-flight spectral calibration thanks to its ability of redefining the spectral channels.
Radiometric performance is then validated by indirect methods using observation of natural targets.

An overview of the instrument principle will be presented; as well as a sketch of the calibration processing chain. Results of in-flight radiometric performance monitoring will be presented for both instruments: radiometric diffuser stability monitoring and modelling, sensor sensitivity evolution monitoring and modelling, spectral absolute characterization and stability monitoring, Signal to Noise ratio assessment and stability monitoring.

Finally, results of the radiometric validation will be presented, as the only independent way to assess the quality of the diffuser-based radiometric calibration. The verification of the in-flight calibration is performed over natural targets through different indirect approaches such as absolute calibration over Rayleigh scattering, cross-calibration and trending over desert and inter-band calibration over sunglint. Corresponding calibration/validation sites have been selected for their appropriate optical properties to validate the radiometry of optical sensors: e.g. spatial and temporal radiometric stability for desert sites, maximization of the Rayleigh scattering over clear and table oceanic surface and under clear atmospheric conditions, etc.

Sentinel-2 is an optical imaging mission devoted to the operational monitoring of land and coastal areas. The Sentinel-2 mission is based on a satellite constellation deployed in polar sun-synchronous orbit. It offers a unique combination of global coverage with a wide field of view (290km), a high revisit frequency (5 days with two satellites), a high resolution (10m, 20m and 60m) and multi-spectral imagery (13 spectral bands in visible and shortwave infra-red domains).

The main objective of the Ssentinel-3 mission is to measure sea surface topography, sea and land surface temperature, and ocean and land surface colour with high accuracy and reliability to support ocean forecasting systems, environmental monitoring and climate monitoring, through acquisitions by multiple sensing instruments. Two of them, OLCI (Ocean and Land Colour Imager) and SLSTR (Sea and Land Surface Temperature Radiometer), are optical sensors designed to provide a continuity with Envisat's MERIS and AATSR instruments. OLCI is based on the opto-mechanical and imaging design of MERIS instrument, i.e. a push-broom imaging spectrometer with five cameras for a 1270 km total FOV. The instrument flyes 21 spectral bands from 0.4 to 1.02 microns with an IFOV of 300 m over land and ocean. SLSTR is based on the design of AATSR instrument, i.e. a conical scanning imaging radiometer employing the along track scanning dual view technique for a total swath of 1420km nadir (and 750km backwards). The acquisition is made for a spatial sampling of 500m (VIS-SWIR) and 1000m (MWIR, TIR) and in 9 spectral bands from 0.55 to 12 microns.

Sentinel-2 and Sentinel-3 sensors are routinely calibrated in-flight using dedicated on-board devices: sun diffusers (OLCI, SLSTR, MSI) and a black body (SLSTR). Afterward, vicarious calibration methods and instrumented sites are used in order to validate the radiometry for spectral bands in the reflective range.
Indeed, the calibration can be checked using dedicated natural targets such as Rayleigh scattering, sunglint, desert sites, Antarctica, and tentatively deep convective clouds, as well as instrumented sites (RadCalNet network). Based on measurements on these sites, it is possible to provide an accurate checking of many radiometric aspects such as absolute and cross-band calibrations, the trending correction, the calibration consistency within the field-of-view, and more generally it to provide an assessment of the radiometric consistency for various types of targets.

The Centre National d’Etudes Spatiales (CNES) supports ESA on radiometric validation activities for Sentinel-2 and Sentinel-3 satellite constellations - now fully deployed. This paper will present radiometric performances results for Sentinel2&3 optical sensors, including their radiometric consistency with regard to many other instruments like MERIS AATSR MODIS, validation with RadCalNet instrumented sites and the early radiometric validation results of recently-launched Sentinel3-B.

The VENµS mission

VENµS (Vegetation and Environment on a New Micro Satellite) is an Earth observation demonstration mission developed in cooperation between FRANCE and ISRAEL, launched on August 2nd, 2017.

The main scientific mission is to acquire data over land in order to improve the understanding and modelling of land surface and vegetation processes, and to develop new applications (water balance, crop yield, carbon fluxes assessment). VENµS provides 10m resolution images in 12 shortwave spectral bands every two days over a set of scientific sites, with constant view angle and overpass time.

The calibration and performance assessment activities of the commissioning phase took place in the second semester of 2017. This phase consisted in checking the whole system, including the satellite, camera, download of the images and data, ground processing chains, as well as the geometric and radiometric calibrations, and products validation.

Radiometric calibration and performance assessment

The radiometric calibration includes fine tuning of the dynamic, equalization of the detectors, identification of defective pixels, absolute calibration, viewing parameters tuning. A special focus is made on specificities of the mission concerning level-1 radiometric calibration:

Equalization and radiometric spikes: An unusual behavior has been observed on numerous pixels, called “radiometric spikes”, on VENµS images in-flight as well as on ground measurements. The cause of this behavior is probably an electronic effect at detector level. The behavior of these rogue pixels cannot be taken into account in the standard linear radiometric model, since the response of the pixel depends on the incoming radiance up to a threshold, and is a constant offset for high radiances. An improvement of the VENµS radiometric model has been implemented to remove these spikes. The model and the final equalization performances will be presented.

Absolute calibration using the moon: The moon calibration consists in comparing the global irradiance of the VENµS moon image with the reference from ROLO radiometric model. The results in the 12 spectral bands of VENµS will be presented and compared to the results from other vicarious calibration methods on terrestrial targets.

Absolute calibration using simultaneous nadir observations with Sentinel-2: Since Sentinel-2A and Sentinel-2B are currently in orbit with very similar spectral bands, it is possible to use Simultaneous Nadir Observations (SNOs) of VENµS and Sentinel-2. Sentinel-2 image is re-sampled to VENµS resolution and the same location is extracted on both images. The cross-calibration results will be presented.

Straylight modelling and correction: The optical straylight level exceeds the requirements by far. The probable causes of this defect are discussed. Two types of straylight are observed: local ghost, caused by scattering of optical surfaces of the instrument and multiple reflexions between detector and filter or filter window, or lenses. Cross-talk ghost is constituted of light going through the filter of one band, and impacting another band. Local straylight model is computed as a global system MTF, whereas the estimated cross-talk contribution is subtracted from the image. The correction model will be presented, as well as the in-flight performances estimated over specific targets (moon, sand/water transitions).

Geometric calibration and performance assessment

The geometric calibration and performance assessement consisted in assessing and correcting the pointing bias, assessing the geolocation performance, calibrating the line of sight (LOS) of each detector, assessing the multi-spectral as well as multi-temporal registration performances, and assessing and setting up reference images. Since the performance of scientific algorithms depends strongly on registration performances, a focus is made on the image registration methodology and assessment, and on the LOS calibration methodology:

Automatic image registration: After radiometric corrections in raw geometry, Venµs images are registered with a subpixel accuracy on a Venµs cloud free reference image by means of tie points automatically extracted between the slave and the reference image. The registration is virtually done by an adjustment of the parameters of the camera geometric model. The attitude and LOS biases are jointly optimized in order to minimize tie points location errors in a least square reduction.

LOS calibration: LOS absolute calibration is based on the ground control points matching between a reference spectral band image and absolute references, which are in our case Sentinel-2 GRI raw images. LOS relative calibration is then performed between the reference spectral band and all the other bands, thanks to dense tie point matching. The separation of the attitude and LOS errors contributions is a prerequisite to high quality calibration. When the attitude errors do not have a zero-mean over acquisition time, this separation is a major issue. The errors can then be jointly estimated using global optimization on several images.

Multi-spectral and multi-temporal registration quality: The multi-spectral registration requirement is 2m rms. This registration is constrained by LOS calibration quality, the DTM quality, and attitude restitution quality. Considering the maximum viewing angle disparity, the SRTM DTM quality yields a negligible contribution, except on mountainous areas where the DTM error is typically more than 40m. The LOS residuals and attitude errors are assessed independently since LOS errors affect the registration column-wise and the attitude errors affect the registration line-wise. The multi-temporal registration requirement is 3m rms. Since the viewing conditions (roll, pitch, yaw) are always the same, there is no DTM contribution to the error budget. This requirement is mostly constrained by the registration quality on the reference image. This registration quality is mostly depending on the attitude refinement quality during the registration processing, and on the Venµs reference image geometric quality. The multi-spectral (resp. multi-temporal) registration performances have been assessed on all spectral band couples (resp. on one reference band and several dates) thanks to dense tie points matching on several scientific sites.

Long term climate data records derived from satellite observations provide vital
information for climate monitoring and research. However, most satellites only operate
over a limited time after which they are replaced. Thus, providing a long term climate data
record requires the combination of measured values from several different satellite
sensors.

A simple combination of calibrated measurements from several sensors will produce an
inconsistent record. For historical sensors, the behavior in space can be different
from its behavior during pre-launch calibration in the laboratory
and more scientific value can be derived from considering the series as a whole.

Here we consider harmonisation as a process that obtains new calibration coefficients
and/or a new calibration model for the sensor by comparing the output of one satellite to
a more radiometrically accurate sensor using appropriate match-ups, such as
simultaneous overpasses. When we perform a comparison of two sensors using match-ups
we must take into account the fact that those sensors are not observing exactly the same radiance.
This is in part due to uncertainties of the radiation due to the match-up process itself and due
to differences in the spectral response functions of the two instruments, even when
nominally observing the same 'band'. We do not aim to correct for spectral response
function differences, but to reconcile the calibration of different sensors given their
estimated spectral response function differences.

We present a harmonisation framework that establishes calibration coefficients
for several sensors simultaneously and rigorously with respect to their uncertainties and
covariances. We discuss this problem, and its mathematical formulation as a large
structured non-linear least squares problem called Marginalised Error In Variables.
The mathematical problem is challenging to solve because it involves large numbers of match-ups
and has significant error correlation in the measured data.
The posterior uncertainties of the optimized parameters are evaluated as the inverse of the Hessian matrix at the minimum.
The algorithm is matrix free since it uses matrix vector products
which are computed efficiently by code generated by Automatic Differentiation.
Results of the harmonisation of AVHRR sensors will be presented and discussed.
The posterior statistic of the remaining misfit is consistent with assumptions and uncertainties.

This work was undertaken within the project “Fidelity and Uncertainty in Climate data
records from Earth Observation (FIDUCEO). FIDUCEO has received funding from the
European Union’s H2020 Research and Innovation programme, under Grant Agreement
638822.