Day-Day Forecasts updated through December 2003.
NWS forecast operations have changed considerably since 2000, as the transition to producing a National Digital Forecast Database (NDFD) has been completed. Offices now produce detailed 5 km gridded forecasts out to 168 hours (7 days) for over a dozen parameters, which populate the national database. Our suite of text and tabular products are now generated from each local office gridded database, including the traditional Zone and Coastal Waters Forecasts (ZFP and CWF), as well as the new Point and Area Forecast Matrix Products (AFM and PFM). This database also allows each office to produce a series of graphical forecast products available on the internet.
While verification during the past 10 to 15 years has focused on temperature and probability of precipitation (POP), verification efforts as a result of NDFD will allow for an expansion of verification to more locations in each service area, and to more elements, such as wind and dew point forecasts. Results from these efforts may be available by early 2005.
To a small degree, some forecast error may be related to acclimatization with new operational systems (porting of model data into grids), and the smoothing that occurs as fields are manipulated at 5 kilometers (km). For instance, at night the Boston area frequently experiences at least 1 degree per mile temperature gradients (BOSTON vs. NORWOOD), with a typical west to northwest light surface flow and clear sky (nearly ideal radiation). These very tight temperature gradients between the city and suburban centers will be dampened in a 5km database.
We are showing forecaster issued Mean Absolute Error (MAE) of the high and low temperature at 6 sites through seven days (out to day 7 of the forecast). This is 14 successive, twelve hour forecast periods, or a total 168 hour forecast.
Temperatures are the 12 hour maximum and minimum for the appropriate period, from two separate forecast cycles...the early morning (~4 am zone release) based on the 0000 UTC model guidance and the late afternoon (~4 pm zone release) based on the 1200 UTC model guidance
The following is a comparison for the periods Jan 1, 2002 through December 31, 2002 followed by 2003 for each selected representative location. Each location has listed 2002 (blue) top and 2003 (green) just below. You’ll note significant improvements generally across the board in 2003 except where negative is in parentheses ( ).Boston has especially large improvements on the mid shift issuance - 4 am. All sites on the day shift issuance - 4pm, show significant improvements with no 12 hour period showing a degradation from the previous year!
As you can see from the data, there is a gradual decrease in forecast temperature skill as we move further into the future.
We continue to expect to make slow but steady improvement as seen when looking at the raw mean absolute temperature errors (MAE) from 2002 to 2003 above, and should be able to demonstrate reduced MAE errors to you over the next few years.
This is only one method to measure performance and by no means is to be considered the only tool for assessing forecaster skill.
If you’ve any questions, please write firstname.lastname@example.org