A novel machine learning retrieval for the detection of ice crystal icing conditions based on geostationary satellite imagery

Abstract. High ice water content (HIWC) conditions are a concern for aviation as the ingestion of ice particles in the jet engines can induce ice crystal icing (ICI), which results in performance loss and damage. To constantly monitor these conditions, retrievals for the detection of ICI were recently developed based on geostationary satellite imagery, but their calibration is limited to targeted flight campaigns or scattered samplings from ICI events databases. In this work, we close this gap, using exclusively remote sensing data to develop and assess a new retrieval for potential ICI conditions. Cloud IWC measurements are provided from the synergy of radar and lidar (DARDAR) on board the polar-orbiting satellites CloudSat and CALIPSO. HIWC conditions (IWC≥0.5 g m−3) at typical cruise altitudes are used as the proxy for areas with potential ICI formation. The HIWC conditions predictors are taken from a combination of observations and retrievals of the geostationary satellite Meteosat Second Generation (MSG). A random forest is trained and tested based on the collocated dataset of active and passive measurements during the summer months of 2013 and 2015, covering the European domain. The input predictors are the brightness temperature difference between the MSG channels at 6.2 and 10.8 µm wavelengths, the visible channel at 0.6 µm wavelength, the cloud optical thickness at 0.6 µm wavelength, and four convection metrics related to the distance to the closest convective cell, area extent of the convective cells, and convection density in the pixel surroundings. Over Europe, 83 % of HIWC conditions measured in the DARDAR dataset are correctly detected. The associated false alarm rate is 51 %. The retrieval is further tested with the ICI events database reported by Lufthansa. Four out of seven events are correctly detected. In conclusion, the retrieval achieves performances comparable to previously developed retrievals. An operational application would enable aircraft rerouting around areas with high ICI probability.