Sequence Model
based Cloudburst Prediction for the Indian State of Uttarakhand
Sivagami M., Radha P. and Balasundaram A.
Disaster Advances; Vol. 14(7); 1-9; doi: https://doi.org/10.25303/f2512105; (2021)
Abstract
Predicting the phenomenon of cloudburst has been a larger than life challenge to
many weather and rain scientists. The very nature of cloudburst occurrence itself
complicates the prediction of cloudburst. Since, cloudburst downpour occurs over
a short span of time and is confined to very narrow geographic location, it is highly
difficult for weather scientists to make any cloudburst predictions.
In this work, the authors propose a cloudburst prediction model that leverages deep
learning techniques to predict the occurrence of cloudburst in a location. The authors
have collected the data pertaining to the cloudburst events that have occurred in
the Indian State of Uttarakhand over the past decade and developed the model. Experiments
were conducted using time series sequence models namely Long Short Term Memory (LSTM)
and Gated Recurrent Unit (GRU). Predictive Power Score (PPS) has been used to extract
the essential features that are fed as input to these sequence models. The performance
of sequence models has been discussed in terms of loss function and accuracy and
the results are promising for GRU based model in comparison with other sequence
models.
