Automatic day time cloud detection over land and sea from MSG SEVIRI images using three features and two artificial intelligence approaches

Abstract

A fuzzy logic and neural network approaches are proposed to generate a cloud mask for Meteosat Second Generation Spinning Enhanced Visible and Infrared Imager (MSG SEVIRI) images. MSG images are classified into land/sea clear sky or cloudy. A spatial and temporal properties of three solar channels (0.6, 0.8 and 1.6 $$\upmu $$ m), and three thermal infrared channels (3.9, 6.2 and 10.8 $$\upmu $$ m) are used for this aim. The proposed methods were tested and evaluated using seventy-two MSG images taken at different times and dates. The fuzzy logic method leads to cloud detection accuracy of 84.41%, and the neural network achieved an average accuracy of 99.69%. Our proposed methods detected not only thick clouds but also thin and the less bright clouds. To give more support to our results, we made a comparison between the proposed approaches and the cloud mask product which is one of the applications integrate into software package Satellite Application Facility to NoWCasting and Very Short Range Forecasting (SAFNWC/MSG) by the European Organisation for the Exploitation of Meteorological Satellites EUMETSAT applied on the same MSG data. The high average accuracy achieved by our neural network proposed method (and more less for fuzzy logic) demonstrates its effectiveness and robustness and also the utility to benefit of using the artificial intelligent techniques in remote sensing imagery applications.