Rainfall forecasting in the Barak river basin, India using a LSTM network based on various climate indices

The proposed study employs a long short-term memory (LSTM) neural network (NN) to forecast monthly rainfall in the Barak river basin northeastern region of India for prediction horizon up 4 months advance. Out nine significant climate variables, sea surface temperature (SST), level pressure (SLP), Nino 3.4 index, Indian summer monsoon (ISMR) anomalies and dipole mode index (DMI) were identified be best-suited predictors introduced as inputs NN. LSTM is special kind recurrent (RNN) which specializes feature extraction storing its cell state cumulatively. model results display strong correlations between potential predictor sets distribution across basin. obtained scrutinized terms various statistical measures predictions found at par with real time observations (correlations greater than 0.90 hit score 85%). testing phase produced root mean square errors range 12.45% 15.65% highlighting satisfactory performance. method incorporating different indices form novel approach may lead timely effective management water resources.