Machine learning prediction of the Madden-Julian oscillation

Abstract

The socioeconomic impact of weather extremes draws the attention of researchers to the development of novel methodologies to make more accurate weather predictions. The Madden–Julian oscillation (MJO) is the dominant mode of variability in the tropical atmosphere on sub-seasonal time scales, and can promote or enhance extreme events in both, the tropics and the extratropics. Forecasting extreme events on the sub-seasonal time scale (from 10 days to about 3 months) is very challenging due to a poor understanding of the phenomena that can increase predictability on this time scale. Here we show that two artificial neural networks (ANNs), a feed-forward neural network and a recurrent neural network, allow a very competitive MJO prediction. While our average prediction skill is about 26–27 days (which competes with that obtained with most computationally demanding state-of-the-art climate models), for some initial phases and seasons the ANNs have a prediction skill of 60 days or longer. Furthermore, we show that the ANNs have a good ability to predict the MJO phase, but the amplitude is underestimated.

Ensemble forecast of an index of the Madden Julian Oscillation using a stochastic weather generator based on circulation analogs

“…One of the reasons is that it is often used for MJO forecast (e.g. Kim et al, 2018;Rashid et al, 2011;Silini et al, 2021).…”

“…Finally, we evaluate the ensemble mean forecast of RM M 1 and RM M 2 using the usual scalar metrics for MJO forecast (Rashid et al, 2011;Silini et al, 2021;Kim et al, 2018). We computed the bivariate anomaly correlation coefficient (COR)…”

Deep Learning for Multi‐Timescales Pacific Decadal Oscillation Forecasting

“…Owing to this advantage, RNN has been widely used in time series forecasting tasks, such as North Atlantic SSTs prediction (Nadiga, 2021), continuous forecasting of chlorophyll a (Du et al., 2018) and hurricane trajectories (Alemany et al., 2018). Gated recurrent unit (GRU) (Cho et al., 2014) is a special kind of RNN capable of learning long‐term dependencies, and it has been successfully applied to Madden‐Julian oscillation prediction (Silini et al., 2021), wind power forecasting (Niu et al., 2020). As deep learning methods are data‐driven, they can directly learn from the historical data and achieve accurate prediction of complex dynamical systems.…”

Improving the prediction of the Madden-Julian Oscillation of the ECMWF model by post-processing

“…Machine Learning (ML) algorithms are being extensively used in many fields, and they are gaining a foothold in weather and climate forecasts (O'Gorman and Dwyer, 2018;Nooteboom et al, 2018;Dijkstra et al, 2019;Ham et al, 2019;Dasgupta et al, 2020;Tseng et al, 2020;Gagne II et al, 2020;Silini et al, 2021) among many others. Although MJO predictions obtained using ML models do not outperform dynamical models (Silini et al, 2021;Martin et al, 2021a), a hybrid approach, combining dynamical models and ML techniques, may improve the results.…”

“…Machine Learning (ML) algorithms are being extensively used in many fields, and they are gaining a foothold in weather and climate forecasts (O'Gorman and Dwyer, 2018;Nooteboom et al, 2018;Dijkstra et al, 2019;Ham et al, 2019;Dasgupta et al, 2020;Tseng et al, 2020;Gagne II et al, 2020;Silini et al, 2021) among many others. Although MJO predictions obtained using ML models do not outperform dynamical models (Silini et al, 2021;Martin et al, 2021a), a hybrid approach, combining dynamical models and ML techniques, may improve the results. In this way, it is possible to use dynamical models that have been developed across decades, based on physical phenomena, in combination with data-driven ML techniques, an approach that has shown its ability to reduce the gap between observations and dynamical models' forecasts (Rasp and Lerch, 2018;McGovern et al, 2019;Scheuerer et al, 2020;Haupt et al, 2021;Vannitsem et al, 2021).…”