Machine-learning algorithm for 24h Detection of Fog and Low Stratus over Europe based on MSG-SEVIRI infrared bands

In this communication we present a pixel-based algorithm for detection of fog and low stratus (FLS) during the 24H day cycle over land and across Europe, based on geostationary satellite observations.

Fog and low Stratus are both a persistent aggregation of water particles in liquid and/or solid phases (cloud) close to the Earth surface. As the cloud-base-altitude is the only real difference between the two (fog: touching the ground; low stratus: above ground), they are frequently treated together as a single category from satellite perspective (FLS). This study presents a pixel-based method for detection of FLS over land across Europe based on Meteosat-11 SEVIRI (Spinning Enhanced Visible and InfraRed Imager) infrared observations. The method is based on a gradient boosting machine learning model that is trained with the observations from Meteorological Aviation Routine Weather Reports (METAR) and German Weather Service (DWD) stations. An intensive validation of the product over 356 METAR stations across Europe over five years of daytime winter data revealed that the method proposed is well capable of detecting FLS over land.  Specifically, the algorithm is found to detect FLS with probabilities of detection (POD) ranging from 0.83 to 0.88 (for different inter-comparison approaches), and false alarm ratios (FAR) between 0.34 and 0.36. As the algorithm operates based on the SEVIRI infrared observations only, it can be applied over day and night, making it feasible to continuously monitor the FLS status over large areas over the 24H day cycle.