Application of Neural Networks for Avalanche Forecasting

Abstract

Application of neural networks is investigated for the prediction of avalanches on Chowkibal-
Tangdhar 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.