The main objective of the Sentinel-3 mission is to provide accurate and reliable measurements of sea surface topography, sea and land temperature, and ocean and land surface color. The mission foresees the simultaneous operation of a constellation of up to three satellites and it is vital to ensure the delivery of consistent, high quality and well calibrated data to the scientific community.
The deployment of the Sentinel-3 constellation started with the launch of the A platform in 2016, followed after two years by the launch of the B platform, on 25 April 2018. During the commissioning phase of Sentinel-3B, it has been foreseen to operate the satellite in “tandem” configuration with the A module: the “tandem” phase consists in the operations of both satellites flying on the same orbit, with the B module preceding the A module of ~30 seconds with exactly the same viewing geometry. This unique configuration provides a huge number of very refined matching dataset of the Sentinel-3A and -3B observations, in the sense of strongly reducing matching errors (because of the simultaneity and the similar geometry), which allows powerful statistical analysis, for instance the relative non-linearity behavior, the relative temporal evolution, the intercalibration of all bands, and for the whole field of view of the instrument.
In this study, we propose a technique to systematically process and match Sentinel-3 data (OLCI and SLSTR) in tandem configuration by means of an image registration approach using the maximization of the Enhanced Correlation Coefficient (ECC) [1]. The strength of this technique stands in being invariant with
respect to photometric distortions which allowed to process a dataset with potential radiometric mis-calibration, especially in the early phase of the Sentinel-3B mission. We decided to use this image-based registration approach in stead of a more classical collocation method based on the use of geographical information to avoid any potential deformation introduced by projection of the coordinates.
This technique allowed to create tandem registered datasets for OLCI A/B, SLSTR A/B Nadir/Oblique and to perform analyses at pixel level.
The orbital stability achieved during the operation of the two platform in tandem configuration allows to easily reach a very good level of registration (mis-registration within 1 pixel) between products almost everywhere, with the only exception of cloudy targets, which usually showed a mis-registration of 1 to 3 pixels due to clouds’ movement. The considered tandem dataset consists of a time series of one full day every 10 days (23.06, 04.07, 14.07, 28.07, 08.08, 18.08, 30.08, 08.09, 18.09).
The statistical analysis of the tandem couples has been carried out selecting homogeneous areas with very high level of registration and low standard deviation. The selection of the candidates for the intercomparison can be also optionally triggered by the cloud masks or land/sea mask or over particular geographic areas.
The OLCI tandem dataset has been analyzed at camera and band level. The relative evolution of the B instrument with respect to the A one has been also addressed and compared to the relative evolution obtained by the operational monitoring of the on-board diffusers. The relative inter-band calibration and residual non-linearity have been addressed as well.
The same analysis has been repeated systematically for SLSTR couples, for Nadir and Oblique views, evaluating the relative differences and residual non-linear behavior between the two modules for the VIS/NIR and TIR bands.
The results are critically discussed and presented, which reveal specific signatures for both OLCI and SLSTR sensors with a very high accuracy. RTM simulations have been also carried out to better understand the large differences observed between –A and –B for bands for which the instrument spectral response may slightly differs, which could be crucial for some bands operating close to strong atmospheric absorption lines.
[1] Evangelidis and Psarakis, Parametric Image alignment Using Enhanced Correlation Coefficient Maximization, IEEE Transaction on pattern Analysis and Machine Intelligence, vol.30, N10, October 2008
|