Abstract Indian summer monsoon rainfall (ISMR) provides 80% of total annual precipitation and thus has a significant impact on water resource management, agricultural yield, and consequently on India’s gross domestic product. The monthly rainfall quantity of the monsoon is more useful than the total monsoon rainfall quantity when it comes to reservoir operations, crop planning, water distribution to different users, etc. In the present study, an assessment of hydro-climatic teleconnection between monthly ISMR and 19 large-scale atmospheric/oceanic circulation indices is performed by using multivariate linear regression (MLR) and a machine-learning technique named support vector regression (SVR), which has not been performed by any reviewed study. Monthly composite indices (MCIs) are developed between monthly ISMR and final selected significant indices (FSSIs) by using the MLR technique for development phase periods of 1951–1985 and 1951–1988, and these MCIs are tested during testing phase periods of 1986–2014 and 1989–2014, respectively. SVR models also have the same training and testing periods as that of MLR models. Correlation coefficient (CC) is evaluated between observed and simulated monthly ISMR corresponding to both MLR and SVR models for development/training and testing phases. The study revealed that, the CC obtained from SVR models is better than that obtained from MLR models for the testing phases. Research highlights <UnorderedList Mark="Bullet"> <ItemContent> Assessing the hydro-climatic teleconnections (HCTs) between monthly Indian summer monsoon rainfall (ISMR) and large-scale atmospheric/oceanic circulation indices (each index has four lags) by using two techniques, namely the multivariate linear regression (MLR) technique and the support vector regression (SVR) technique with a linear kernel function. </ItemContent> <ItemContent> By assessing input significance and input independence for input variable selection by using correlation at a 5% significance level. </ItemContent> <ItemContent> HCTs of monthly ISMR is found to be changing with respect to time. </ItemContent> <ItemContent> SVR performed well in the testing phase. </ItemContent> </UnorderedList>

Assessment of hydro-climatic teleconnection and prediction of monthly Indian summer monsoon rainfall by using multivariate linear regression and support vector regression

期刊论文
发表日期：

2025年11月14日

查看全文

Rahul Verma,
Ganesh D Kale

摘要

Abstract

Indian summer monsoon rainfall (ISMR) provides 80% of total annual precipitation and thus has a significant impact on water resource management, agricultural yield, and consequently on India’s gross domestic product. The monthly rainfall quantity of the monsoon is more useful than the total monsoon rainfall quantity when it comes to reservoir operations, crop planning, water distribution to different users, etc. In the present study, an assessment of hydro-climatic teleconnection between monthly ISMR and 19 large-scale atmospheric/oceanic circulation indices is performed by using multivariate linear regression (MLR) and a machine-learning technique named support vector regression (SVR), which has not been performed by any reviewed study. Monthly composite indices (MCIs) are developed between monthly ISMR and final selected significant indices (FSSIs) by using the MLR technique for development phase periods of 1951–1985 and 1951–1988, and these MCIs are tested during testing phase periods of 1986–2014 and 1989–2014, respectively. SVR models also have the same training and testing periods as that of MLR models. Correlation coefficient (CC) is evaluated between observed and simulated monthly ISMR corresponding to both MLR and SVR models for development/training and testing phases. The study revealed that, the CC obtained from SVR models is better than that obtained from MLR models for the testing phases.

Research highlights

Assessing the hydro-climatic teleconnections (HCTs) between monthly Indian summer monsoon rainfall (ISMR) and large-scale atmospheric/oceanic circulation indices (each index has four lags) by using two techniques, namely the multivariate linear regression (MLR) technique and the support vector regression (SVR) technique with a linear kernel function.

By assessing input significance and input independence for input variable selection by using correlation at a 5% significance level.

HCTs of monthly ISMR is found to be changing with respect to time.

SVR performed well in the testing phase.