Data-Driven Machine Learning Models for Predicting Monthly Rainfall Using Lagged Climate Indices

Abstract

Rainfall prediction is vital for several economic activities, including agriculture. The present study is focused on developing machine learning models to predict rainfall using different climate indices. Four machine learning techniques - RF, M5P tree, REP Tree algorithm, and ANN were used for predicting rainfall from climate indices. The practicality of the approach presented herein was demonstrated through application to rainfall data at Safdarjung meteorological station in New Delhi, India. The machine learning models use climate indices—Indian Ocean Dipole, El Niño–Southern Oscillation, Pacific Decadal Oscillation, Atlantic Multidecadal Oscillation and North Atlantic Oscillation as feature variables and rainfall as the target variable. The dataset was divided into a training set and a testing set. The ratio of the split was eighty to twenty. Among the four machine learning models evaluated, random forest model demonstrates best performance, with coefficients of correlation of 0.9853 and 0.6674, mean absolute errors of 15.1624 and 33.7026, and root mean square errors of 25.1173 and 44.0744 for training and testing phases, respectively. The Taylor diagram also showed that the RF-based model performed better at predicting rainfall.