Skip to main content
U.S. flag

An official website of the United States government

Dot Gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.


The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Bringing It All Together: A Prototype For A Probabilistic National Blend of Models


It is well understood that Numerical Weather Prediction (NWP) output (a) suffers from lead-time dependent systematic errors, (b) lacks fine scale User Specific Variables (USV), and (c) proliferates due to a diverse set of data sources. Traditionally, each guidance is statistically calibrated independent of the others. The National Weather Service (NWS) recognized the unsustainability of such an approach and launched the National Blend of Models (NBM) project. The present, NGGPS (Next Generation Global Prediction System) supported study of the Bayesian Processor of Ensemble (BPE, Krzysztofowicz and Toth 2008) aims at prototyping a future probabilistic version of NBM. Recognizing that lead-time dependent systematic forecast errors manifest in Model Prognostic Variables (MPV) and that the generation of fine scale USVs is altogether an independent problem, we design a 2-stage Statistical Post-Processing System (SPPS). Stage-1 (SPPS-1) brings and fuses all predictive information (i.e., unperturbed high resolution forecasts, lower resolution ensembles from different centers) together to produce calibrated (i.e., reliable, consistent with reanalysis) and informative (i.e., skillful, with high statistical resolution) probabilistic guidance for all MPVs on the model grid. BPE uses prior climatological distributions based on a large sample of reanalysis to impose reliability. To assess the relative performance of, and combine predictive information from various NWP forecasts, a smaller joint sample of forecasts and verifying analyses are used. At its core, BPE uses multiple linear regression, a staple of Model Output Statistics. Due to the novel use of distributional and Bayesian techniques, BPE produces reliable forecasts, including extremes at a significantly reduced need for computer-intensive hindcasts. With a meta-Gaussian approach BPE is configured to process all continuous prognostic variables, regardless of their distribution type. For each variable, output is provided in a full array of formats: cumulative and probability distribution functions, quantiles, and ensemble forecasts consistent with the posterior distributions, all derived from just two parameters used in the context of the prior distribution function, only doubling the storage requirement compared to current practices. Stage-2 (SPPS-2) uses the calibrated MPVs from Stage-1 that are statistically consistent with reanalysis, for the derivation of fine scale USVs. Unlike Stage-1 that corrects lead-time dependent systematic forecast errors, Stage-2 needs no forecast data whatsoever as it builds instantaneous (non-predictive) relationships between calibrated MPVs (i.e., reanalysis) and USVs (i.e., observationally based fine scale analysis). The MPV - USV relationships can either be physically based (readily usable components of the Unified Post-Processing package) or statistical (in design phase, leveraging existing Stage-1 BPE components). Stage-1 BPE algorithms and software have been developed by the University of Virginia and the Global Systems Division of NOAA Research, respectively in a modular form for ease of implementation, maintenance, and further development and testing of alternative techniques. The software has been transitioned to the Meteorological Development Laboratory of the NWS where it undergoes further testing, from which initial results of a comparison with the operational Ensemble Kernel Density Model Output Statistics (EKDMOS) system will be presented at the conference. With the BPE project we attempt to bring it all together: (i) Theoretically based and time-proven scientific techniques and software engineering solutions; (ii) academic, NOAA research and operational communities and organizations; (iii) elements of a well-structured end-to-end workflow; (iv) the fusion of climatology with all informative NWP forecast guidance; (v) processing of all MPVs and USVs into (vi) calibrated and informative guidance in a (vii) comprehensive array of output formats (viii) serving diverse applications and the functionalities of a future probabilistic NBM. Krzysztofowicz, R. and Z. Toth, 2008: Bayesian Processor of Ensemble (BPE): Concept and Implementation, 4th NCEP/NWS Ensemble User Workshop, Laurel, Maryland, May, see at:

Article / Publication Data
Available Metadata
Fiscal Year
Published On
January 01, 2017

This publication was presented at the following:

2017 - 97th AMS Annual Meeting
Amer. Meteor. Soc.
Conference presentation


Not available


Authors who have authored or contributed to this publication.

Not available