Improving the MJO Forecast of S2S Operation Models by Correcting Their Biases in Linear Dynamics

• Published in: Geophysical research letters 2021-03, Vol.48 (6), p.n/a
• Main Authors: Wu, Jie, Jin, Fei‐Fei
• Format: Article
• Language: English
• Subjects: Climate; Climate variability; Decay rate; Dynamic models; Dynamical systems; Dynamics; El Nino; El Nino phenomena; El Nino-Southern Oscillation event; Errors; Extreme weather; Geology; GEOSCIENCES; Growth rate; Inclusions; Madden-Julian oscillation; Modelling; Nonlinear dynamics; Nonlinear systems; Operators; Phase velocity; Predictions; Propagation; Seasonal variations; Skills; Southern Oscillation; Systematic errors; Tropical environments; Variability
• ISSN: 0094-8276; 1944-8007
• DOI: 10.1029/2020GL091930

The operational dynamic subseasonal to seasonal (S2S) models for Madden‐Julian oscillation (MJO) forecasting mostly still suffer from systematic errors in capturing the MJO's key dynamic features, such as its growth rate and propagation speed. By deriving the linear dynamic operators using the linear inverse modeling (LIM) approach, we propose a method to partly correct the errors in MJO linear dynamic operators to improve the MJO predictions of three operational dynamic S2S models. Correcting the deficiencies of the too‐fast decay rates and the unrealistic propagating phase speeds lead to MJO prediction skills being extended by approximately 2–4 days. The improvements are more significant for the models with larger biases in MJO amplitude and propagation. This approach in principle may be extendable to predictions of other types of climate variability such as ENSO on one hand, and possible inclusions of nonlinear dynamics effects on the other hand. Plain Language Summary The Madden‐Julian oscillation (MJO) is the dominant mode of intraseasonal variability in the tropics and can exert a significant influence on global circulation and extreme weather events. Most of the current generations of state‐of‐the‐art operational dynamic models used in MJO predictions have yet to fully capture the essential features of the MJO. These deficiencies hinder the models’ MJO prediction skills. By taking advantage of mass hindcast data, a method is proposed in this study to correct the effect of errors of the linear dynamics of each model on its MJO predictions by carrying out a dynamics‐based postprocessing on the model predictions. This proposed approach is applied to three different operational dynamic models and shows improvements in MJO prediction skills. This kind of approach may be extended to improve the predictions of other types of climate variability such as ENSO and may be expanded to include nonlinear dynamics effects as well. Key Points The operational dynamic forecast models mostly suffer errors in capturing Madden‐Julian oscillation's (MJO) fundamental dynamic properties A dynamics‐based postprocessing correction method may extend model's MJO forecast skill by approximately 2–4 days The gain in skill comes from reducing errors in the decay rate and propagating phase speed in the model's MJO predictions