Synergies between the EPIC OCRA/ROCINN algorithm tandem and the operational Sentinel-4 UVN cloud product

Abstract

With the launch of the Meteosat Third Generation - Sounder (MTG-S) in summer 2025 in a geostationary orbit (GEO), the awaited Sentinel-4/Ultraviolet-Visible-Near-infrared spectrometer (S4/UVN) will provide hourly information of trace gas species, aerosols and clouds over Europe and parts of Northern Africa, which are essential for air quality monitoring. This mission completes the Geo-Ring constellation, which aims to monitor the atmospheric composition and the air quality over the Northern Hemisphere with hourly resolution during daytime, and which already includes the South Korean mission GEMS (Geostationary Environment Monitoring Spectrometer, launched in 2020) and the US-American mission TEMPO (Tropospheric Emissions: Monitoring of POllution, launched in 2023).

The knowledge about basic cloud properties like cloud cover and cloud height is needed to monitor the Earth's radiation budget, and it is also essential for trace gas- and greenhouse gas retrieval algorithms, since these require a description of the cloud contamination of the remote sensing scenes as algorithm inputs. As part of the ESA Sentinel-4 Level-2 Operational Processors (S4 L2OP), the German Aerospace Center (DLR) develops the operational cloud product for S4/UVN by means of the OCRA/ROCINN algorithm tandem. OCRA (Optical Cloud Recognition Algorithm) determines the scene cloud fraction using image analysis. ROCINN (Retrieval Of Cloud Information using Neural Networks) determines two additional cloud properties by means of an optimal estimation approach in which the forward model consists of two fast neural networks that emulate with high precision the outputs of a time-consuming radiative transfer model. The OCRA/ROCINN tandem has proven to work operationally in other low-Earth orbit (LEO) missions, including GOME/ERS-2, SCIAMACHY/ENVISAT, GOME-2 on MetOp-A/B/C and TROPOMI/S5P. Furthermore, OCRA/ROCINN has also been implemented successfully for the EPIC/DSCOVR mission, located at the Lagrangian point L1.

In this work, we show the synergies between the OCRA/ROCINN implementation for EPIC/DSCOVR and the operational S4 L2OP cloud processor. S4/OCRA benefits from EPIC/OCRA through the usage of adapted EPIC/OCRA monthly cloud-free maps during the start of S4/UVN operations, since the generation of these maps with S4/UVN measurements cannot be performed until at least one year of instrument measurements is available. S4/ROCINN benefits from EPIC/ROCINN through the simplification of the neural network configuration thanks to the usage of dimensionality reduction techniques, which reduce the number of neural network output dimensions.