Monday, 7 January 2019
Hall 4 (Phoenix Convention Center - West and North Buildings)
Accurate subseasonal-to-seasonal forecasts (3 weeks to 3 months) of precipitation along the U.S. West Coast are of great interest to stakeholders and emergency managers. Largely driven by atmospheric rivers, precipitation extremes can lead to periods of drought or flooding that pose significant challenges. Recently, we developed an empirical prediction model to forecast atmospheric river activity along the U.S. West Coast. It uses the current state of the Madden-Julian Oscillation (MJO) and the Quasi-biennial Oscillation as predictors and has significantly positive skill scores that extend to subseasonal lead times of 3–5 weeks. In contrast, state-of-the-art dynamical models from NCEP and ECMWF only have skillful lead times of approximately 2 weeks. However, these same dynamical models do reasonably well at predicting the MJO, with skillful lead times extending out to 4–5 weeks. Here, we reconfigure our empirical prediction model by conducting two tasks. First, rather than forecasting for atmospheric river activity, we train the empirical model to forecast for actual precipitation. Second, instead of using the current state of the MJO as a predictor, we ingest dynamical model predictions of the future state of the MJO into the empirical model. Thus, we create a hybrid dynamical-empirical model to forecast precipitation along the U.S. West Coast. We train and test the model using reforecast data acquired from the S2S Project and SubX databases. Our testing demonstrates that the model has improved skill scores, and, moreover, these positive skill scores extend deeper into subseasonal lead times – offering important precipitation guidance for concerned interests along the U.S. West Coast.
![](images/flag.png)
![](images/blue_ribbon.png)