Background
DICast is a completely automated consensus forecast system that generates forecasts of sensible weather elements at specified sites. Statistical and fuzzy logic techniques are used to enhance forecast data obtained from a variety of sources. DICast forecasts statistically outperform the forecasts from the ingredient data sources. Since monthly output statistics (MOS) from the National Weather Service's (NWS) forecast products can be included in the ingredient mix, DICast can improve upon the NWS’s best guidance.
This forecast technology was developed at NCAR/RAL in 1999 for The Weather Channel (TWC) and it continues to serve as the backbone for TWC’s on- air and online operations. In recent years, it has been adopted for operational forecast generation by other large and small U.S. commercial weather providers, including Peak Weather, Inc. DICast forecasts have also been used in several RAL projects such as the FHWA Winter Maintenance Decision Support System (MDSS), FAA Ceiling and Visibility, Fire Weather, and the Army Test and Evaluation Command (ATEC) projects.
Current Activities and Accomplishments
Research and development continued in 2005 as did support and maintenance of existing operational systems. Improvements to the system were made in several areas.
- NWP Model-based Forecasts: A new Dynamic Model Output Statistics (DMOS) system was developed to improve forecasts of unusual (large departures from average) weather phenomena. The new approach, Model Error Correction (MEC), provides equally skillful forecasts for common events (compared to the old DMOS approach), but does significantly better on rapidly changing “extreme” events.
- Improved Interpolation Techniques: Cubic spline techniques were implemented to better capture the diurnal temperature cycles for temporally interpolated forecasts.
- New Weather Model Input: New weather models were ingested into the system. The NOAA/FSL Rapid Update Cycle (RUC) model was made operational for the MDSS project. The NWS Downscaled GFS with Eta Extension (DGEX) model and Environment Canada Global Environmental Multiscale (GEM) model have been ingested and software was developed to derive DMOS (MEC) predictors.
- GRIB Evolution: Software has been implemented to convert data in the GRIB-2 format into the older GRIB-1 format. This allows use of higher resolution data sets. Unfortunately, many of the higher resolution data sets are too large to be used once in GRIB-1 format.
- Daytime Maximum and Nighttime Minimum Temperatures: Software was developed to generate location-specific day/night specific maximum/minimum temperature forecasts. The previous max/min temperature forecasts were for the UTC day (0000Z-2359Z). These were sometimes unsatisfactory for operational usage.
Plans for 2006
DICast will continue to provide operational forecasts to a variety of projects and commercial weather providers. RAL DICast staff will continue to provide the required support. Many of the improvements listed above will require further verification and tuning before installation in operational systems.
Other areas of expected development include:
- The inclusion of radar-based precipitation extrapolations. Precipitation forecasts generated by these techniques outperform model-based precipitation forecasts for the first hours of the forecast.
- The development of gridded DICast forecasts. Algorithms will be developed to produce forecasts on a 5-km grid across the CONUS. This will most likely be done by combining DICast point forecasts and climatological data.