Investigating forced transient chaos in monsoon using Echo State Networks

Abstract

Forecasting Indian Summer Monsoon Rainfall (ISMR) is a formidable task due to its intricate variability. This study harnesses the power of machine learning (ML) to decipher the chaotic trajectory within ISMR, drawing inspiration from ML's success in predicting analogous systems. By utilizing ERA-interim data, the method dissects ISMR's chaotic nature through correlation dimension-based techniques. Employing the Lorenz-96 model on daily rainfall data, trained with an Echo State Network (ESN), the technique discerns patterns within a span of 1 model time slightly trailing its performance in other systems. This discrepancy could stem from the intricacies of observational data and the training process involving 500 initial conditions. Notably, this method achieves accuracy in slightly over 50% of cases. Despite its current limitations, this approach exhibits promise in shedding light on the chaotic behaviour enforced in ISMR. As a result, it contributes to the advancement of monsoon forecasting technique