Optimal Estimation Retrieval Framework for Daytime Clear-Sky Total Column Water Vapour from MTG-FCI Near-Infrared Measurements

Abstract. A retrieval of total column water vapour (TCWV) from the new daytime, clear-sky near-infrared measurements of the Flexible Combined Imager (FCI) on-board the geostationary satellite Meteosat Third Generation Imager (MTG) is presented. The retrieval algorithm is based on the differential absorption technique, relating TCWV amounts to the radiance ratio of a non-absorbing band at 0.865 µm and a nearby water vapour (WV) absorbing band at 0.914 µm. The sensitivity of the band ratio to WV amount increases towards the surface, which means the whole atmospheric column down to the boundary layer moisture variability can be observed well.

The retrieval framework is based on an Optimal Estimation (OE) method providing pixel-based uncertainty estimates. It builds on well-established algorithms successfully applied to other passive imagers with similar spectral band settings. Transferring knowledge gained in their development onto FCI required some new approaches. The absence of additional, adjacent window bands to estimate the surface reflectance within FCI's absorbing channel were mitigated using a Principle Component Regression (PCR) from the bands at 0.51, 0.64, 0.865, 1.61 and 2.25 µm.

Since a long-term calibrated FCI dataset is not available yet, we build a second forward model for two equivalent NIR bands (0.865 and 0.9 µm) on the Sentinel-3 Ocean and Land Colour Instrument (OLCI). A long-term validation of OLCI against a single Atmospheric Radiation Measurement (ARM) reference site without the PCR resulted in a bias of 1.85 kg/m², centered root mean square deviation (cRMSD) of 1.26 kg/m² and r² of 0.995. In order to test the PCR which uses FCI bands in the visible to short-wave infrared, we replaced the bands missing in OLCI with bands from the Sea and Land Surface Temperature Radiometer (SLSTR). A spectrally similar dataset was created from SLSTR and OLCI data on Sentinel-3A/B during June 2021. This dataset is used to test the retrieval with regards to robustness and global performance of the PCR. A first verification of this OLCI/SLSTR "FCI-alike" TCWV against well-established ground-based TCWV products concludes with a wet bias between 1.23–3.12 kg/m², a cRMSD between 1.88–2.35 kg/m² and r² between 0.95–0.97. In this set of comparison, only land pixels were considered. Furthermore, a dataset of FCI Level 1c observations with a preliminary calibration was processed. The TCWV processed from FCI data aligns well with reanalysis TCWV and collocated OLCI/SLSTR TCWV but show a dry bias. A more rigorous validation and assessment will be done, once a longer record of FCI data is available.

The PCR may be extended to include more diverse water-bodies. In future iterations, more bands in the visible spectral range may be added to further increase performance in presence of aerosol over dark surfaces.

This novel TCWV dataset derived from geostationary satellite observations enhances monitoring of WV distributions and associated meteorological phenomena from synoptic scales down to local scales. Such observations are of special interest for the advancement of nowcasting techniques and Numerical Weather Prediction (NWP) accuracy as well as process-studies.