A machine learning based prediction system for the Indian Ocean Dipole

Area of Science:

• Climate Science
• Machine Learning
• Artificial Intelligence

Background:

• The Indian Ocean Dipole (IOD) significantly impacts global climate, influencing regions like East Africa, Australia, and India.
• Predicting the IOD is crucial due to its widespread climatic effects.
• Previous IOD prediction studies relied on complex coupled climate models.

Purpose of the Study:

• To introduce and evaluate artificial neural networks (ANNs) for Indian Ocean Dipole (IOD) prediction.
• To assess the forecasting skill of ANNs for the IOD index.
• To compare ANN performance against persistence forecasts and established climate models.

Main Methods:

• Utilized artificial neural networks (ANNs) for IOD index forecasting.
• Derived predictive attributes from sea surface temperature and geopotential height anomalies (850 hPa and 200 hPa) from 1949-2018.
• Generated an ensemble of 500 ANN forecasts using a jackknife resampling approach.

Main Results:

• ANN models demonstrated high skill in forecasting the IOD index well in advance (May-November from February-April conditions).
• Forecast accuracy significantly surpassed persistence-based predictions.
• ANN models outperformed the North American Multi-Model Ensemble (NMME) in terms of correlation coefficients and RMSE.

Conclusions:

• Machine learning, particularly ANNs, offers a powerful new approach for skillful IOD prediction.
• ANNs provide superior forecasting capabilities compared to existing methods, enabling earlier climate impact assessments.
• This study highlights the potential of ANNs to enhance climate variability predictions.