Thin-cirrus detection from Artificial Neural Network and IASI-NG

This study proposes an Artificial Neural Network approach for the detection of optically thin cirrus using observations from the Infrared Atmospheric Sounding Interferometer New Generation (IASI-NG) and from its predecessor, IASI. The Thin Cirrus Detection Algorithm applies a Feed-forward Neural Network (NN) to IASI/IASI-NG samples previously declared as clear by a cloud detection algorithm. The NN training, test and validation datasets are generated from a set of ECMWF 5-generation reanalysis (ERA5) processed with the sigma-IASI radiative transfer model to simulate IASI/IASING radiances. The IASI and IASI-NG Thin Cirrus detection algorithms were validated against an independent dataset showing better performances for the IASI-NG thin-cirrus-detection algorithm. Moreover, IASI thin-cirrus-detection algorithm outputs were compared against Cloudsat/CPR and SEVIRI cloud products, showing good probability of detection: 0.84 for SEVIRI and 0.77 for CPR/Cloudsat.