The Subseasonal Prediction Experiment

The sub-seasonal experiment (SubX) is a multi-model sub-seasonal prediction experiment with both a research and real-time component. Seven global models have produced seventeen years of retrospective (re-) forecasts and one year of weekly real-time forecasts. Both the re-forecasts and forecasts are archived at the IRI Data Library for research on sub-seasonal predictability and predictions. The real-time forecasts have been produced weekly, beginning in July 7, 2017 to provide guidance to the NOAA/NCEP Climate Prediction week 3-4 outlooks. The real-time forecasts are provided to CPC by 5pm every Wednesday of every week. This talk describes detailed results from the analysis of the retrospective forecast quality for the individual models and the multi-model ensemble including the bias, and a suite of deterministic and probabilistic measures. The seasonality of forecast quality is assessed, and how forecast quality varies with well-known low frequency variability (e.g., ENSO). The emphasis is on T2m and precipitation, but some phenomenological (e.g., MJO, tropical storms) measure of performance are also provided. Specific case studies of success (and failure) are also noted.