Machine Learning for Fog-and-Low-Stratus Nowcasting from Meteosat SEVIRI Satellite Images

Fog and low stratus (FLS) are meteorological phenomena that have a significant impact on all ways of transportation public safety. Due to their similarity, they often grouped together as single category when viewed from satellite perspective. The early detection these is crucial reduce the negative effects can cause. This paper presents an image-based approach for short-term nighttime forecasting FLS during next 5 h over Morocco, based geostationary observations (Meteosat SEVIRI). To achieve this, dataset hourly night microphysics RGB product was generated native files covering cold season (October April) 5-year period (2016–2020). Two optical flow techniques (sparse dense) three deep learning (CNN, Unet ConvLSTM) were used, performance developed models assessed using mean squared error (MSE) structural similarity index measure (SSIM) metrics. Hourly Meteorological Aviation Routine Weather Reports (METAR) Morocco used qualitatively compare existence in METAR, where it also shown by product. Results analysis show outperform traditional method with SSIM MSE about 0.6 0.3, respectively. Deep promising results first hours. However, highly dependent number filters computing resources, while sparse found be very sensitive mask definition target phenomenon.