The plots on this web site are generated using the raw (unadjusted) output from the NCEP Global Forecasting System (GFS) Ensemble. Four times each day, NCEP produces this ensemble using a single computer model that is run multiple times with slightly different initial conditions. These various initial states approximate the uncertainties inherent in the initial atmospheric state. As the forecasts progress in time, the slight discrepancies in the initial conditions will, in general, grow until the forecasts in the ensemble become quite different. As the differences in the ensemble members evolve, they provide a measure of the forecast uncertainty. The more disagreement there is among the ensemble members for a particular region, the less certain the forecast is for that region. Note that this uncertainty arises through the use of a single model.

Additionally, the ensemble itself has been shown to have deficiencies in describing the forecast uncertainty. Studies of the ensemble's performance indicate that the ensemble tends to underestimate the extremes. That is, a forecast for a little rain is often wetter than it should be, and a forecast for a lot of rain should often be for even more rain.

Research in this lab and elsewhere is being conducted to develop post-processing techniques to reduce the error of individual forecasts within the ensemble and to adjust the statistical distribution represented by the ensemble to represent the forecast probabilities more accurately. Even in their present form, the plots on this web site can be used to provide an idea of both the forecast and its uncertainty along with the range of possibilities that exist. Remember, though, the ensemble is typically a little too wet for small amounts of precipitation and not wet enough for large amounts of precipitation.