The 239th R-CCS Cafe - part 2

The Madden–Julian Oscillation (MJO) is a massive cluster of convective activities in the tropics that spans thousands of kilometers and travels slowly eastward from the Indian Ocean to the central Pacific in approximately 20 to 60 days. It has vast influence on the global weather and is recognized to be one of the most important sources of predictability in extended-range weather forecast longer than weeks. Prediction of the MJO has been recognized to be notoriously difficult with physics-based weather prediction models. To address this problem, we employ the reservoir computing technique to construct a machine learning model that forecasts the real-time multivariate MJO index (RMM), a macroscopic variable that represents the state of the MJO. The training data was refined by developing a novel filter that improved the recurrency of MJO associated signals in the raw atmospheric data and selecting a suitable time-delay coordinates of the RMM. The model demonstrated the skill to forecast the state of MJO events for a month from pre-developmental stages.