Deep learning for bias correction of MJO prediction

H Kim,Y G Ham,S W Son,Y S Joo

doi:10.1038/s41467-021-23406-3

Abstract

Producing accurate weather prediction beyond two weeks is an urgent challenge due to its ever-increasing socioeconomic value. The Madden-Julian Oscillation (MJO), a planetary-scale tropical convective system, serves as a primary source of global subseasonal (i.e., targeting three to four weeks) predictability. During the past decades, operational forecasting systems have improved substantially, while the MJO prediction skill has not yet reached its potential predictability, partly due to the systematic errors caused by imperfect numerical models. Here, to improve the MJO prediction skill, we blend the state-of-the-art dynamical forecasts and observations with a Deep Learning bias correction method. With Deep Learning bias correction, multi-model forecast errors in MJO amplitude and phase averaged over four weeks are significantly reduced by about 90% and 77%, respectively. Most models show the greatest improvement for MJO events starting from the Indian Ocean and crossing the Maritime Continent.

Highlights

Producing accurate weather prediction beyond two weeks is an urgent challenge due to its ever-increasing socioeconomic value
We demonstrate that the Deep learning (DL) postprocessing substantially reduces the Madden-Julian Oscillation (MJO) forecast errors from the state-of-the-art dynamical forecasting systems, making strides towards improving global extended range forecasts
It shows the multi-model mean of predicted Real-time Multivariate MJO indices (RMMs) composite on a phase-space diagram[45] as a function of initial MJO phases and forecast lead days from day 1 to day 28 (4 weeks)

Summary

Introduction

Producing accurate weather prediction beyond two weeks is an urgent challenge due to its ever-increasing socioeconomic value. Due to errors originating from imperfect numerical models, the MJO prediction skill has not reached its theoretical predictability, which is known to be ~7 weeks[26] This indicates that there is considerable room for improvement[23,25,26,27,28]. One of the greatest challenges in current dynamical forecast systems is the fast damping of the MJO signal as the forecast lead time increases, which results in a rapid decrease of forecast skill[25,29,30] This systematic damping of the MJO convection signal is apparent when the MJO starts over the Indian Ocean and is expected to propagate through the Maritime Continent and move further into the western Pacific. Given that the MJO prediction alone presents considerable systematic biases, the global weather forecast beyond

Methods

Results

Conclusion