Ensemble Forecasts for Weather and Seasonal Climate I
Presiding: T Krishnamurti, Florida State University; B Rajagopalan, University of Colorado
A13B-01 INVITED 13:30h
Recent Developments with the NCEP Ensemble Forecast Systems
The National Centers for Environmental Prediction (NCEP) of the National Weather Service produces operational ensemble forecasts on both weather and seasonal climate time scales. Three systems are used for generating these forecasts: an ensemble of Global Forecast System (GFS) forecasts for weather out to 16 days lead time; a limited area ensemble forecast system, embedded into the global ensemble (Short-Range Ensemble Forecasts, SREF); and a coupled ocean-atmosphere, time-lagged ensemble system. This presentation will review recent research and development results related to the configuration, as well as the initial and model perturbations of the ensemble systems used at NCEP.
http://wwwt.emc.ncep.noaa.gov/gmb/ens/index.html
A13B-02 13:45h
Warm Season Mesoscale Superensemble Precipitation Forecasts
With current computational limitations, the accuracy of high resolution precipitation forecasts has limited temporal and spatial resolutions. Forecast accuracy drops dramatically after a 24 hour forecast. Current operational mesoscale models run only to 48-72 hours. However, with the recent development of the superensemble technique, the potential to improve precipitation forecasts at the regional resolution exists. The purpose of this study is to apply the superensemble technique to regional precipitation forecasts to generate more accurate forecasts pinpointing exact locations and intensities of strong precipitation systems. This study will determine the skill and predictability of a regional superensemble forecast out to 60 hours. Precipitation results were stratified by time of day to allow detections of the diurnal cycle. As expected, warm season daytime precipitation is commonly forced by convection which is difficult to accurately model. Results were also stratified by lead time which reveals how quickly the forecasts degrade in time. Currently, mesoscale models such as those utilized in the ensemble are approaching the limits of precipitation predictability. Major synoptic regimes, including subtropical high, mid-latitude trough/front, and tropical cyclone, were examined to determine the skill of the superensemble under various synoptic conditions. Finally, different rainfall intensities were examined which revealed the superensemble forecast significantly improved the forecast at significant rainfall amounts. The regional superensemble consists of 12 to 60-hour daily quantitative precipitation forecasts from 6 models. Five are independent operational models, and one comes from the physical-initialized FSU regional spectral model. The superensemble forecasts are verified during the summer 2003 season over the southeastern US using a merged RFC Stage IV radar/gauge and satellite analyses. Precipitation forecasts were skillful in outperforming the operational models at all model times. Skill measurements that were examined include ETS, Bias, FAR, and POD.
A13B-03 14:00h
Very High Resolution Multi-Model Ensemble Simulations of the Dryline/Pacific frontal merger during STORM-FEST IOP 17
During the Stormscale Operational and Research Meteorology-Fronts Experiment Systems Tests (STORM-FEST) intensive observing period (IOP) 17, which ran from 8 to 10 March 1992, an extratropical cyclone produced widespread severe weather throughout the southern United States. This severe weather produced 13 tornadoes and flash flooding that accrued over $20 million in damage. One of the goals of STORM-FEST was to test the limits of mesoscale predictability. Since the STORM-FEST program in 1992, advances have been made in mesoscale prediction. Increased computational power, improved physical parameterizations, and ensembling methods have all contributed to enhanced mesoscale prediction capabilities. Previous studies by Locatelli et al (1995), Martin et al (1995), and Wang et al (1995), used the PSU/NCAR Mesoscale Model 4th Generation (MM4) at 45 km resolution to simulation the structure and progression of the upper level trough and dryline that moved through the south-central U.S. These studies were able to resolve larger-scale (meso-alpha and synoptic) features of the systems but not some of the finer scale details described in Neiman et al (1998). In this study, we will utilize a more intensive modeling approach through very high resolution (1 km) ensembling as well as probabilistic methods, focusing on a 24-hour period beginning at 1200 UTC 8 March 1992. Two mesoscale models--the PSU/NCAR (MM5) and the Weather Research and Forecasting (WRF) model will be utilized in the ensemble. By creating a multi-model ensemble with perturbations in physical parameterizations at 1 km resolution, we test and verify the predictability of meso-beta scale phenomena that developed during the STORM-FEST IOP-17 as well as potentially elucidate meso-gamma structures which were not well observed.
A13B-04 14:15h
Real Time Numerical Weather Prediction by The Florida State University Superensemble
The Florida State University (FSU) Superensemble technique as applied to real-time numerical weather prediction will be described. An evaluation of the skill of the Superensemble forecasts will be presented in comparison to the skills of the seven global numerical weather prediction models that comprise the Superensemble. Forecast variables that will be examined include lower and upper tropospheric wind fields, mean sea level pressure, mid-tropospheric geopotential height, and precipitation. Forecast skill will be evaluated globally, as well as for a number of sub-regions, such as the Indian monsoon region, North and South America, and the tropical North Atlantic Ocean. Statistical measures of forecast skill will include root mean square error, anomaly correlation, and systematic error for most variables. Forecast precipitation will also be evaluated by use of correlation, bias, and equitable threat scores. The skill scores will be presented the years 2000, 2001, and 2004. The FSU Superensemble technique uses multiple linear regression to derive coefficients from a comparison of member model forecasts to a benchmark analysis during a training period of 120 days. This procedure removes the bias of each individual forecast model and allows for an optimal linear combination of the individual model forecasts, which takes into account the relative skill of each model. The result is a forecast that has greater skill than the individual model forecasts and the ensemble mean forecast. The real-time FSU Superensemble forecasts are available on a website that shows the forecasts for the entire globe, as well as for ten sub-regions of the world. The website has links to the skill scores that are routinely updated, as well as to a number of journal articles that describe the FSU Superensemble technique in detail. Overall, the FSU Superensemble has been shown to be a valuable tool for significantly improving upon the numerical model forecasts emanating from the world's operational forecast centers. Such forecasts are used daily as importance guidance in issuing weather forecasts around the world. These forecasts have important implications for the world's economy, and the life and safety of its people.
A13B-05 INVITED 14:30h
Ensemble Particle Filter with Posterior Gaussian Resampling
An ensemble particle filter was recently developed as a fully nonlinear filter of Bayesian conditional probability estimation, along with the well known ensemble Kalman filter. A Gaussian resampling method is proposed here to generate the posterior analysis ensemble in an effective and efficient way. As a result the ensemble particle filter has good stability and potential applicability to large-scale problems. The Lorenz model is used here to test the proposed method. Multi-modal probability distributions can appear either with state dependent stochastic model errors or nonlinear observations. Ensemble Kalman filter (EnKF)is known to have a difficulty in tracking state transitions accurately. Current implementations of EnKF have not taken non-Gaussian contributions into account. With the posterior Gaussian resampling method the ensemble particle filter can track state transitions more accurately. Moreover, it is applicable to systems with typical multi-modal behavior, provided that certain prior knowledge becomes available about the general structure of posterior probability distribution. A simple scenario is considered to illustrate this point based on Lorenz model attractors. The present work demonstrates that the proposed ensemble particle filter can provide an accurate estimation of multi-modal distribution and is potentially applicable to large-scale data assimilation problems.
http://www.csit.fsu.edu/~navon
A13B-06 14:45h
Breeding in Coupled Ocean-Atmosphere Models for Ensemble Prediction
We have performed breeding in the Cane-Zebiak model (Cai et al, J. of Clim., 2003) and in the NASA and NCEP models (Yang et al, J. of Clim., subm.) with similar results that suggest bred vectors are coupled perturbations useful for ensemble perturbations and data assimilation. We will present results with the operational NASA coupled model that indicate the bred perturbations are sensitive to the ENSO cycle and that bred vectors with one-month rescaling are similar to the analysis increments.