Application of Neural Networks for Avalanche Forecasting

Abstract

Application of neural networks is investigated for the prediction of avalanches on ChowkibalTangdhar road axis in Jammu and Kashmir. The networks are developed and trained monthwise using the past snow and weather parameters recorded at the Stage-II Observatory on the axis to generate an assessment of avalanche and non-avalanche activities. Two approaches have been considered for training the network. In the first approach, only avalanche activities observed in the axis were taken for training, and in the second approach, along with the observed activities, the opinion of expert forecasters were also considered. The performance of the networks varies from 67 to 82 per cent for correct predictions. Winter data for [2001][2002] has been used to validate the network performance.