Enhancing Subseasonal Temperature Prediction by Bridging a Statistical Model With Dynamical Arctic Oscillation Forecasting

Abstract

Subseasonal predictions, which fall between medium-range forecasts of up to 10 days and seasonal predictions of up to 3 months, have gained increasing scientific interest due to a rise in demand for accurate and reliable outlooks. In recent decades, the prediction skill of dynamical forecast systems has greatly improved via the use of ensemble forecasting techniques and improved initializations, couplings, and physical parameterizations (e.g., Saha et al., 2014;Vitart, 2014). Despite this improvement, it remains challenging to produce skillful predictions for surface variables across the globe at lead times of longer than a few weeks (Xiang et al., 2019). This can be attributed to the rapidly worsening prediction skill of dynamical models as initial errors are compounded over time. Nonetheless, dynamical forecast systems generate reliable predictions for major climate modes, such as the North Atlantic Oscillation (NAO) and the Madden-Julian Oscillation (MJO) beyond a subseasonal time scale (e.g., Lin, 2018;Kim et al., 2018).To overcome the shortcomings of dynamical models, statistical models have been developed by employing slowly varying climate modes as a potential source of prediction skill (Baggett et al., 2018;Black et al., 2017;Mundhenk et al., 2018). Johnson et al. (2014) constructed a statistical model, known as a phase model, based on the probability distributions of predictands (e.g., surface air temperature) contingent upon the phases of chosen tropical climate modes, such as the MJO and El Niño-Southern Oscillation (ENSO). For weeks 3-4, the phase model demonstrated marginally higher skill for temperature predictions over the

Improved subseasonal prediction of South Asian monsoon rainfall using data-driven forecasts of oscillatory modes

“…The idea of combining full-field dynamical forecasts with datadriven forecasts of intraseasonal oscillations was previously suggested by various studies (15,21,22,26,27). While there have been a number of works that correct dynamical forecasts of an oscillatory mode using either data-driven forecasts (28)(29)(30) or post-processing (31)(32)(33), these did not attempt to correct the full-field forecasts. In this work, we correct a full-field dynamical forecast using data-driven forecasts of specific modes.…”

Extended-Range Forecast of Regional Persistent Extreme Cold Events Based on Deep Learning

“…The result was substituted into the trained model to obtain the predicted precipitation field. In addition, many scholars have proposed methods using dynamical statistics to predict meteorological elements for an extended range [18][19][20].…”

Adaptive Bias Correction for Improved Subseasonal Forecasting

“…To overcome observed systematic errors of physics-based models on the subseasonal timescale, there have been parallel efforts in recent years to demonstrate the value of machine learning and deep learning methods in improving subseasonal forecasting [13,14,15,16,17,18,19,20,21,22,23]. While these works demonstrate the promise of statistical models for subseasonal forecasting, they also highlight the complementary strengths of physics-and learning-based approaches and the opportunity to combine those strengths to improve forecasting skill [15,20].…”