3B.4 Multi-model Ensembling and Calibration of S2S Precipitation Forecasts over North America

Friday, 28 July 2017: 2:15 PM
Constellation F (Hyatt Regency Baltimore)
Nicolas Vigaud, IRI, Columbia Univ., Palisades, NY; and A. W. Robertson and M. K. Tippett

A Multi-model Ensemble (MME) of Ensemble Prediction System (EPS) forecasts at week 1 to 4 leads is designed using Extended Logistic Regressions (ELR) to produce probabilistic forecasts of weekly and week 3-4 averages of precipitation over the 1999-2010 period for which ECMWF, NCEP and CMA Subseasonal-to-Seasonal (S2S) reforecasts are available. The ELR parameters are fitted separately at each gridpoint and lead time for the three EPS reforecasts with starts in Jan-Mar (JFM) and Jul-Sep (JAS). The ELR used for calibration produces tercile category probabilities for each models that are then averaged together with equal weight. The resulting MME forecasts are characterized by good reliability but low sharpness, while multi-model ensembling is found to largely remove negative skill scores compared to individual forecasts. For forecasts of weekly averages, skill decreases with lead times with steep drops after one and two weeks and is higher in winter than summer. Week 3-4 forecasts have more skill along the US east coast and the southwest US in winter, as well as over west/central US regions and the Intra-American Seas/east Pacific during summer. Skill is also enhanced when the ELR parameters are fit using spatially smoothed observations and forecasts. The skill of week 3-4 precipitation outlooks has a modest, but statistically significant, relation with ENSO and MJO activity, particularly in winter over the southwest US.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner