Assessing forecast uncertainty in the National Digital Forecast Database

NOAA's National Weather Service (NWS) has implemented a National Digital Forecast Database (NDFD) to provide its customers and partners access to gridded forecasts of sensible weather elements. The NDFD contains a seamless mosaic of digital forecasts from NWS field offices, working in collaboration with the National Centers for Environmental Prediction (NCEP). Most of the NDFD weather elements represent single-value forecasts at a point in space, valid for a either a point in time or over a span of time. The NWS' Meteorological Development Laboratory (MDL) is developing a set of techniques that quantify the uncertainty associated with these single-valued forecasts. These techniques use recent NDFD performance and related guidance derived from Numerical Weather Prediction (NWP) models to infer the expected distribution of observations for that weather element.

Matched pairs of forecasts and observations are amassed to form a set of developlmental data. These developmental data are used to form a model from which the joint distribution of forecasts and observations is inferred. It has proved useful to transform both the forecasts and observations from their native values to climatological percentiles since it allows aggregation of data from disparate regions. Additional diagnostic data can be added to further refine the modeled distribution. Once the distribution has been modeled, one can infer a conditional distribution of the observations given the current forecast and diagnostic data.

Efforts to date have been focused on modeling the daily maximum temperature. Early results and prototype products are explored.