S100 Using Model Climatology to Develop a Confidence Metric for Operational Forecasting

Sunday, 22 January 2017
4E (Washington State Convention Center )
Taylor S Mandelbaum, Stony Brook University, Stony Brook, NY; and B. A. Colle and T. Alcott

Handout (2.3 MB)

Probabilistic forecasting is an important tool for both public and private sectors. The use of ensemble models increases the awareness of uncertainty and errors in the output of a model. Although the use of ensembles has increased, there exist many opportunities for better visualization of ensemble model output, which is a major objective of the Stony Brook University CSTAR collaborative project with the National Weather Service (NWS). The Ensemble Situational Awareness Table (ESAT), managed by the NWS, compares forecasts from the North American Ensemble Forecast System (NAEFS) and Global Ensemble Forecast System (GEFS) to reanalysis (R-Climate) and model reforecast (M-Climate) climatologies. Standardized anomalies, percentiles and return intervals are calculated to assist in identifying potentially significant weather events.  While M-Climate output from the GEFS reforecast can place the current ensemble mean forecast in context, it does not assess the ensemble spread relative to similar anomalous events. We have attempted to take the M-Climate diagnostic a step further by assessing whether confidence in the developing anomaly is unusually high or low. Our goal is to output an operational spread anomaly product that will complement the existing ESAT. We utilized the GEFS Reforecast between 21 November 1985 and 10 March 2015. To test the approach, we chose cases restricted to the winter (DJF) timeframe over the contiguous United States. In these cases, midlatitude synoptic cyclones are the most prevalent high impact events, especially along the U.S. East Coast. Our initial test variables include mean sea-level pressure and surface 2-m temperature. A future goal is to include other variables such as wind (magnitude and direction) and precipitation. The ensemble mean is used to determine standardized anomalies, at every point on the forecast grid. The current forecast for a given variable is compared to the DJF M-Climate distribution to account for seasonal variability. Reforecast values in a 5x5 grid, centered about each gridpoint, within a threshold of similar standardized anomalies to the forecast ensemble mean are utilized to construct a new spread M-Climate. Using this method, a spread anomaly can be calculated for each point on the domain. The resulting visualization displays a novel metric of quasi-confidence in the GEFS forecast. We will demonstrate the technique for a few notable East Coast storms in recent years.

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