JP6.14
Calibrating probabilistic multimodel ensemble pentad and weekly precipitation forecasts: a comparison of techniques
Dan C. Collins, NOAA/NWS/NCEP/CPC, Camp Springs, MD; and T. M. Hopson and D. A. Unger
Model ensembles provide a sample of possible outcomes of a particular forecast given the initial conditions. However, because models are imperfect, errors are not simply due to the spread of the distribution of possible outcomes and ensemble sampling error. Distributions derived from model ensembles will not be accurate representations of the probability of any particular outcome due to model and representativeness errors. Calibration of ensemble probabilistic forecasts modifies the ensemble distribution to represent the true distribution. This study examines methods for improving the reliability of pentad and weekly precipitation forecasts using the NCEP Global Ensemble Forecast System and Canadian Modeling Centre Ensemble models. A new CPC unified global precipitation dataset is used as observations. The skill and reliability of probabilistic ensemble forecasts of accumulated precipitation is calculated both with and without calibration and compared. The strengths of various calibration methods are examined for several end-user products, such as quantile, extreme and probability distribution forecasts. Techniques considered include quantile regression of the ensemble probability density function, ensemble regression of the individual members and ensemble mean, and reliability of ensemble member counts as a linear function of the resolution or conditional likelihood of an event. Reliable ensemble probabilities contribute to the improvement of the operational extended range and US hazards forecasts issued daily by the NCEP Climate Prediction Center.
Joint Poster Session 6, Improvements to NWP and Short-term Forecasting
Wednesday, 14 January 2009, 2:30 PM-4:00 PM, Hall 5
Previous paper Next paper