Testing, refinement and demonstration of probabilistic multi-model, calibrated sub-seasonal global forecast products

Sub-seasonal to seasonal (S2S) forecasting (lead times between 2 weeks and 2 months) is a new area of climate prediction that occupies the time range between medium range weather forecasts and seasonal climate prediction. The S2S range is currently the focus of intense research effort through the NOAA MAPP program and WWRP/WCRP S2S project, whose aims include improving forecast skill and understanding, and developing well-calibrated probabilistic forecast products in the sub-seasonal range. Since 2015 NOAA has been issuing experimental sub-seasonal probabilistic week 3–4 outlooks of precipitation and temperature every Friday, and there is a need and opportunity to demonstrate the potential of ongoing research results for improving the guidance used in these NOAA Climate Prediction Center (CPC) forecasts.

Under NWS NGGPS funding we have recently developed and tested a multi-model, calibrated, probabilistic sub-seasonal forecasting methodology for weekly-averaged precipitation (weeks 1–4), as well as week 3–4 two-week averages. The method has been evaluated using hindcasts from three models from the S2S database (CFSv2, ECMWF and CMA) over the U.S. (Vigaud et al. 2017a), and over boreal summer monsoon regions (Vigaud et al. 2017b). In these results, both the calibration and subsequent multi-model ensembling (MME) of the forecast probabilities were found to achieve probabilistically reliable forecasts with skill in some cases. It is this methodology that we propose to refine, further test and to transition to CPC operations in this CTB project.

NOAA MAPP’s S2S activities include the “SubX” sub-seasonal forecasting experiment, that combines multiple global models from NOAA, NASA, Environment and Climate Change Canada (ECCC), the Navy, and National Center for Atmospheric Research to produce once-a-week real- time experimental forecasts of at least 32 days in length, together with 17 years of reforecasts (1999–2015). SubX began a one-year pilot period of real-time predictions in July 2017, as additional guidance to CPC for their week 3–4 outlooks. The SubX project will test the skill of individual prediction systems as well as multi-model combinations. While SubX is limited to a 1-year real- time pilot period, it includes models from both NCEP and ECCC which are operational centers and will continue to issue real-time forecasts after SubX ends. The SubX forecast and reforecast data
is being archived in the IRI Data Library4