2.2 Advancing intraseasonal forecasts through the use of a new NOAA global ensemble reforecast data set

Wednesday, 9 January 2013: 11:00 AM
Ballroom F (Austin Convention Center)
Dan C. Collins, NOAA CPC, College Park, MD; and S. Handel, M. L'Heureux, D. Unger, T. M. Hamill, and J. S. Whitaker

Because of chaos, model, and sampling error, statistical post-processing is an essential component in the forecast process. Short training datasets (for instance, the last 90 days) are generally not as useful for the statistical post-processing of forecasts at intraseasonal climate timescales; simply, because the training data will not span enough relevant similar cases.

To aid in the statistical post-processing, NOAA has recently created a multi-decadal global ensemble retrospective forecast (or reforecast) dataset. Every day from late 1984 to present, 11-member global reforecasts were computed using the current (2012) operational version of the Global Ensemble Forecast System (GEFS). Forecasts extend to 16 days lead. As with the operational model, the forecasts were computed at T254L42 (about ½ degree grid spacing) in week 1 and T190L42 (about ¾-degree) in week 2. Data was archived every 3 h to 72 h, and every 6 h thereafter. 99 fields are freely available for fast download from NOAA/ESRL/PSD, including precipitation, soil moisture, and surface fluxes. A full data archive is maintained at the Department of Energy, where the data set was created. Prior to 2012, the Climate Forecast System Reanalysis supplied the control initial condition. Perturbed initial conditions were generated with the operational ensemble transform with rescaling technique.

The talk will describe the application of this reforecast data set to postprocess and improve forecasts at various leads including 6-10 days and 8-14 days. Statistics for postprocessing are calculated using the training data from the first two decades of reforecasts and the skill of independent forecasts from more recent years will be assessed. In general, seasonally and spatially varying statistics are calculated for the forecast and observation variables and for the covariance of forecasts and observations. A variety of methods to postprocess ensemble probabilities will be compared to show the increased value of a reforecast dataset over shorter training datasets and improve operational NOAA intraseasonal forecasts. Deriving forecasts from the NOAA GEFS reforecasts for leads of 6-10 and 8-14 days allows for skill comparisons to the Climate Prediction Center operational forecasts and to forecasts using the NCEP Climate Forecast System reforecasts.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner