Prediction of number of rainy days over different monsoon regions in India

Article Type

Research Article

Publication Title

Journal of Data Information and Management

Abstract

Indian monsoon rainfall is an extremely important affair in socio-economic well-being of the country. Recently, climatic changes have introduced significant uncertainty and irregularity in the Indian Summer Monsoon (ISM) cycle. Such unpredictability has been evidenced in all the elements of monsoon, e.g., onset, intensity, and regularity. Rain amount has been the eye of all predictions for obvious reasons. But, in the recent scenario, other components of the monsoon process have also become extremely crucial to be forecasted. The current work has regionalized Indian subcontinent using a rough-fuzzy c-means algorithm and proposed five updated monsoon zones. Four deep learning networks using Bi-LSTM architecture for four of the resulting regions have been developed for prediction of the number of rainy days one month ahead of time with a spatial resolution of 10 × 10. The model accuracy is found to be 67.48%, 92.80%, 70.72% and 88.75%. A stronger model (architecture) has then been developed by ensembling the said four models whose validity is tested on the fifth region (monsoon zone) with an accuracy of 79.80%. The spatial matching between the actual and predicted number of monsoon rainy days has been found to be good. The certainty values of prediction are determined by rough set-based decision rules. Such studies on prediction of ISM elements using deep learning methods can be of compelling interest in tropical meteorology in near future.

First Page

1

Last Page

14

DOI

10.1007/s42488-023-00106-9

Publication Date

3-1-2024

Link to Full Text

Share

COinS