Hail detection from Meteosat satellite imagery using a deep learning neural network and a new remote sensing index

Abstract

Severe cloud storms that produce hail (hailstorms) are common across Europe and pose major episodes of meteorological risk. The study presents all the methodological steps that were followed to develop a Deep machine learning Neural Network (DNN) model for hail detection (in terms of hail probability of occurrence), using the Meteosat multispectral infrared (IR) imagery, exclusively. The DNN model was trained using numerous cases of hail events as they were recorded from the European Severe Weather Database (ESWD). In each pixel with a recorded hail event, a set of parameters was calculated which were used to train the DNN model. Among them, a new remote sensing index named Hail Potential Index (HPI) was used to achieve optimal accuracy of the proposed DNN model in hail detection. The accuracy assessment of the DNN model was found satisfactory, with a Mean Absolute Error (MAE) of 1.16%. Also, the DNN model was applied in two different case studies (hail episodes) with satisfactory efficiency. The accuracy achieved allows its operational mode for hail detection using multispectral IR information from modern meteorological satellites worldwide.