Comparative Machine Learning approach to seasonal ENSO Forecasting with Time-Series Performance Evaluation

Deeksha Sudheer, Charuvi Suresh, Abhilash M Bairy, Arjit Praveen

The El Niño-Southern Oscillation (ENSO) is one of the most important climate phenomena that affects weather conditions worldwide. It influences monsoon seasons, droughts, and crop productivity. Forecasting of ENSO processes is rather difficult since the system is inherently nonlinear. Moreover, the Spring Predictability Barrier limits the predictive capacity of forecasters during the boreal spring period.

In this paper, four machine learning algorithms – Ridge Regression, Support Vector Regression (SVR), Random Forest, and Long Short-Term Memory (LSTM) are considered. Training was performed on the historical data of the Niño 3.4 and Oceanic Niño Index indices collected over decades. The quality of predictions was evaluated using several lead times from 1 to 15 months with RMSE, MAE, and Pearson correlation coefficient being utilized as metrics.

This paper evaluates and compares the abilities of four machine learning methods in predicting ENSO processes. Although simpler methods give relatively good forecasts on a short term, LSTM shows better results with a longer horizon due to its ability to temporal dependencies. Besides, the influence of the Spring Predictability Barrier on the predictability of ENSO events is evaluated.