Improving infrared MSG satellite images classification, application on rainy areas detection

Abstract

The investigation of the positive impact of using artificial intelligence systems in meteorological satellite images classification is presented in this study. An application for classifying clouds observed northern Algeria by the MSG (Meteosat Second Generation) satellite is developed in this work. The proposed identification system is based on the spectral information derived from the infrared channels of MSG satellite images. The classification aim is to identify pixels as rainy or not, and to delineate precipitating areas for northern Algeria in real time. An SVM (Support Vector Machine) classifier is adopted in this work as the intelligent system used for the identification. Models are trained and validated against in situ rainfall measurements, collected during the rainy period of the year 2012, by automatic rain-gauge stations distributed over the study area (North of Algeria). Five statistical assessors are used to evaluate performances. Results of the classification by the SVM are compared to a statistical method, based on thresholding in histograms representing classes' features. The classification error has been significantly reduced when using the SVM classifier, allowing a more accurate estimation of precipitations by infrared MSG satellite images.