Energy industry to benefit from improved temperature forecasts

by David Stensrud and Nusrat Yussouf

Energy companies rely heavily on temperature forecasts to allocate power in the most economical way. A two-degree error in a temperature forecast, especially in hot weather, can have a substantial impact on energy demand. Improved forecasts of near-surface conditions could result in better and more efficient delivery of electric power and lead to lower costs for consumers. NSSL is one of the partners addressing this issue as part of the NOAA Temperature Forecasting Pilot Project that took place over the New England region the past two summers.

One of the goals of this program is to quantify the improvements in temperature forecasting that result from new and augmented observations and modeling. In collaboration with the National Center for Environmental Prediction/Environmental Modeling Center (NCEP/EMC) and the Forecast Systems Laboratory (FSL), a shortrange ensemble forecasting system was constructed using over 20 different model forecasts. Scientists wanted to see if an ensemble approach could provide improved 2-m temperature and dewpoint temperature predictions when compared against model output statistics (MOS), the statistical post-processing available from present operational forecast models.

In our first attempt to improve upon MOS, we developed a simple bias-corrected ensemble mean. This method used the past seven complete days of forecasts and observations to bias- correct both the 2-m temperature and dewpoint temperature predictions for each individual model at each forecast output time. Results from 48 days during the summer of 2002 indicate that this bias-corrected ensemble is competitive with, or better than, MOS from the Nested Grid Model. In addition, the biascorrected ensemble provides information on the probabilities of temperatures exceeding selected threshold values. This additional probability information provided by the ensemble can be quite valuable to many end users of weather forecasts when used in a simple cost-loss model. In particular, the ensemble adds the most value above that provided by MOS for the more unlikely, and often the most important, events. An additional benefit of the ensemble technique is that it can be developed for any observing station location and needs only a week of forecast and observational data to produce the bias-corrected forecasts. MOS, in comparison, requires many years of data before forecasts can be provided. Forecasts from the summer of 2003 are currently being examined.