Application of machine learning technique and multivariate linear regression for the assessment and prediction of monthly Indian summer monsoon rainfall by using eighteen large-scale circulation indices

Abstract

The monthly Indian Summer Monsoon Rainfall (ISMR) quantity is more useful than the total ISMR quantity. Therefore, assessment of hydro-climatic teleconnection (HCT) of monthly ISMR is essential. None of the reviewed studies have assessed HCTs between monthly ISMR and more than eleven circulation indices, along with more than five lags. Thus, in the present study, HCTs between monthly ISMR and eighteen circulation indices (each index having eight lags) is assessed by employing the multivariate linear regression (MLR) technique and machine learning technique named support vector regression (SVR) through the formulation of two models. In the present study, two development/training phase periods considered are 1951-1985 and 1951-1988 and two testing phase periods considered are 1986-2014 and 1989-2014 corresponding to models 1 and 2, respectively, which are used in each technique. Initial significant lagged circulation indices (ISLCIs) are derived based on a significant correlation between lagged circulation indices and monthly ISMR. Multi-collinearity amongst ISLCIs is removed, if present, to obtain the significant and independent lagged circulation indices (SILCIs). Then, the monthly composite index (MCI) for every month of the ISMR is developed using SILCIs and corresponding monthly ISMR data. Each MCI is used to forecast rainfall during the testing phase. Similarly, the SVR technique has used the SILCIs corresponding to the two development and testing phases. The Study found some common SILCIs in both the models along with the effect of circulation indices other than El Nino and Southern Oscillation (ENSO) and Equatorial Indian Ocean Oscillation (EQUINOO) on monthly ISMR. It can be concluded that the predictive power of the SVR technique is more than the MLR technique.