Application of Statistical and Artificial Neural Network(s) on Selected Features of RSRW Data to Nowcast Severe Thunderstorm

Abstract Severe thunderstorm is one of the extreme natural calamities. In the present study, different weather parameters such as- moisture difference, adiabatic lapse rate and wind shear have been chosen to forecast severe thunderstorm. Here, moisture difference at 5 different geo-potential heights of atmosphere, dry adiabatic lapse rate at 5 different geo-potential heights of atmosphere, and vertical wind shear at 3different geo-potential heights of atmosphere that is, total 13 weather parameters have been chosen for prediction purpose. Here, the Naïve Bayes method, Multi Layer Perceptron, and K-Nearest Neighbor method have been considered. Radial Basis Function Network (RBFN) has been introduced here which produce far better result in comparison with other methodologies. Applying Principal Component Analysis (PCA) on these weather parameters 3variables of adiabatic lapse rates and 3 variables of vertical wind shear have been selected as determinant factor. Again Naïve Bayes method, Multi Layer Perceptron and K-Nearest Neighbor methodologies have been applied on the chosen predicators. This gives better outcome than before, but application of Radial Basis Function Network gives most accurate results among other methodologies. The RBFN gives more than 96% accurate forecast for both squall days and no squall days. Moreover this study has a lead time of 10–12 hours which is very much important to take necessary precaution to save life and property.