Multi-task machine learning improves multi-seasonal prediction of the Indian Ocean Dipole

Area of Science:

• Climate Science
• Artificial Intelligence
• Data Science

Background:

• The Indian Ocean Dipole (IOD) is a major driver of interannual climate variability with significant global socio-economic impacts.
• Current prediction skills for IOD events are limited, typically extending only three months ahead.

Purpose of the Study:

• To develop an advanced prediction model for the Indian Ocean Dipole (IOD).
• To enhance the lead time and accuracy of IOD event forecasting.

Main Methods:

• Utilized a multi-task deep learning model, named MTL-NET.
• Trained and validated the model using four decades of historical IOD data.
• Compared MTL-NET performance against established dynamical models.

Main Results:

• MTL-NET demonstrated accurate IOD prediction up to seven months in advance.
• The model outperformed several world-class dynamical prediction models.
• MTL-NET effectively identified key predictors and captured complex, nonlinear relationships.

Conclusions:

• The MTL-NET model offers a significant advancement in IOD prediction capabilities.
• This deep learning approach provides a more efficient and accurate tool for IOD forecasting.
• Improved IOD prediction has the potential for better mitigation of its socio-economic impacts.