Design and validation of a neural network system for forecasting and monitoring fog events in the UAE using SEVIRI-MSG data
Satellite remote sensing is an important tool in the detection and nowcasting (short range forecasting) of fog events. Fog over land develops primarily during the late-night and pre-dawn hours, infrared remote sensing is indispensable in observing fog formation, while visible imagery helps to monitor the extent and density of fog after sunrise. Satellite remote sensing is widely used in the United States in detecting and nowcasting fog events. The temperature difference between two infrared bands (11 ým and 4 ým) forms the basis for fog detection and classification.
Several fog detection and monitoring algorithms from remote sensing data have been developed during the last two decades. Retrieval of fog-related information from geostationary platforms has matured into applications at operational level in many national weather service programs in the USA, Europe and more recently in China. In this study, a neural network based tool for fog detection and monitoring in the UAE will be presented. Data collected by SEVIRI instrument onboard of the European Satellite Meteosat Second Generation (MSG) were the main source of spatial data. Fog forecasts are first generated before sunrise using thermal imagery collected at night. The predicted maps will be then corrected and updated with the first visible images received and processed as soon as the sun rises.
The development approaches and the preliminary results will be presented and discussed in this paper. The preliminary results have shown that the use of thermal difference between the two MSG channels (T04 and T09) give a good indication of fog presence. The analysis of thermal data measured at night has shown that the thermal difference T04-T09 ranges between -5 and -8 degree Kelvin under dense fog conditions, while it ranges between -2 and -5 degree Kelvin under clear sky conditions.