Guy Carpenter’s William Stikeleather Publishes Papers on Modeling Biases and Mitigating Forecasting Errors

Guy Carpenter’s Will Stikeleather, Analyst and Meteorologist, North America Peril Advisory, has published 2 academic papers concerning the use of algorithms to address weather modeling errors in long-term forecasts. These papers appear in the American Meteorological Society’s Journal of Weather and Forecasting.

The overall goal of the research is to develop a weather model postprocessing algorithm that uses advanced mathematical techniques (Time-Extended Empirical Orthogonal Functions or EEOFs) to identify and remove forecast error from weather model ensemble forecasts that extend from 2 to 5 weeks into the future.

Research is being funded by the National Oceanic and Atmospheric Administration-Climate Prediction Center (NOAA-CPC), with the ultimate goal for the new algorithm to be implemented into its long-term forecasting efforts.

The first paper provides proof of concept that the algorithm can identify model errors in 200 hPa Geopotential Height, a key forecast variable for longer-term forecasts. The second paper tests the algorithm out to day 16 of historical model forecasts, demonstrating that the algorithm is effective in reducing overall forecast errors.

Key Takeaways

Development of a Two-Step EOF Statistical Postprocessing Algorithm to Identify Patterns of Systematic Error and Variance Within GEFSv12 Reforecasts

• Systematic errors for 200 hPa Geopotential Height weather model forecasts grow rapidly in magnitude beyond day 7.
• The algorithm can successfully identify and reconstruct these errors within the models.
• Subtracting these reconstructed errors from the original forecast can reduce overall error beyond day 7, as shown in paper 2.

Application of a Two-Step Space-Time EOF Statistical Postprocessing Algorithm to Mitigate Sub-Seasonal 200hPa Geopotential Height Forecast Error

• Skill scores show that by day 16 of a 200 hPa Geopotential Height historical forecast, error is reduced by an average of 20% across the mid-latitudes in winter. Some regions see error reduction of higher than 40% on average in winter.
• Research by Will’s co-author, Dr. Paul E. Roundy, of the Department of Atmospheric and Environmental Science at the University at Albany (SUNY), is ongoing to apply this algorithm beyond day 16 and to new weather variables.
• Research is ongoing to apply the algorithm to real-time forecasting. Ultimately, the goal of the research grant is to implement this algorithm into the NOAA-CPC forecasting efforts in real time.