B. Hertell, R. Derech Consolidated Edison Company of New York

L. A. Treinish, A. P. Praino, H. Li, J. Cipriani IBM Research, Yorktown Heights, NY

Meteorologists have a number of statistical measures at their disposal to help verify weather forecasts. To those who have limited meteorology or statistical analysis background the results can be difficult to interpret. In an attempt to make the performance of forecasting tools better understood we developed an easy to understand forecast index or score.

There are numerous forecasts in various formats provided by a multitude of service providers to businesses. Determining their relative usefulness and value is difficult given the temporal and spacial differences among the forecasts. This presentation will discuss a methodology that has been developed and utilized by Con Edison to relate various forecasts to each other so that they can be compared against the observations using statistical methods like RMSE, bias, etc. The methodology then goes one step further to create a “forecast score” that combines the statistical measures and provides a one number result based on a 0-100 scale.

Weather forecasts from the IBM Deep Thunder high resolution WRF model customized for the ConEdison service territory are compared to other numerical weather models, as well as various subscription and free forecast services utilized by the company using this method.

The pros and cons to developing and employing such a methodology will be discussed, as well as ways to evolve, improve, and adapt the scoring methodology for various needs.