A connectionist approach to thunderstorm forecasting

Thunderstorms are considered to be global phenomena, as they may occur anywhere in the world, at any instant. Lightning, damaging straight-line wind, large sized hails, heavy precipitation and flooding are some damage-prone factors associated with these storms. Depending on their types (specially the supercell thunderstorm, having the tendency to form tornadoes), some of them may possess great potentiality to produce serious damages to human life and property. Now-a-days, due to rapid technical development, many sophisticated instruments (such as Doppler radar, Satellite, Radiosonde, etc.) are available to record weather data. Efforts are being made to use these data by designing models based on statistical, mathematical, and soft computing techniques, in order to forecast damaging weather conditions with greater reliability. In this article, we have used an artificial neural network (ANN) based model, with backpropagation learning, for classifying the occurrence and non-occurrence of seasonal thunderstorms over the eastern coastal region of India. Rules are extracted from the trained network. These enable the system to provide advance prediction of oncoming thunderstorms, based on relevant weather parameters, in human-understandable form. It is concluded that the application of such connectionist approach may be useful in the area of weather forecasting, and can facilitate the research and development in several fields, including civil-aviation, radiocommumcation, and meteorology.