HR: 14:40h
AN: A53E-07    [Abstracts]
TI: Forecast Performance of the New Local Three Month Temperature Outlook
AU: * Timofeyeva, M
EM: Marina.Timofeyeva@noaa.gov
AF: University Corporation for Atmospheric Research, SSMC2 room 13370 1325 Eastwest Hwy, Silver Spring, MD 20910, United States
AU: Bair, A
EM: Andrea.Bair@noaa.gov
AF: National Weather Service, 1325 Eastwest Hwy, Silver Spring, MD 20910, United States
AU: Hollingshead, A
EM: Annette.Hollingshead@noaa.gov
AF: RS Information Systems, 2525 Correa Rd, Ste 250, Honolulu, HI 96822, United States
AU: Unger, D
EM: David.Unger@noaa.gov
AF: National Weather Service, 1325 Eastwest Hwy, Silver Spring, MD 20910, United States
AU: Livezey, R
EM: Robert.e.livezey@noaa.gov
AF: National Weather Service, 1325 Eastwest Hwy, Silver Spring, MD 20910, United States
AB: NOAA's National Weather Service (NWS) introduced a new operational Local 3-Month Temperature Outlook (L3MTO) in January 2007. The product is available for 1170 locations nationwide and can be accessed via any NWS Weather Forecast Office (WFO) climate website (under the "Climate Prediction" tab, or the NWS Climate website http://www.nws.noaa.gov/climate/l3mto.php ). L3MTO methodology (1) applies linear regression to identify the statistical relationship between a station parameter and the corresponding forecast region and (2) adjusts the regression parameters to the most recent temperature trends at the station. Long-term forecast performance evaluation plays an important role in the product development process because it is essential to guide the ongoing improvement of the forecasting procedures. The CPC 3-month temperature forecasts from 1994 to 2005 were used to create a L3MTO hind cast for the same time period. To avoid possible contamination of the verification results by dependent observations, the L3MTO hindcast computations train each year of 1994 to 2005 using the present forecast methodology and data for regression and trend adjustment of appropriate time period. For example, for the 1994 (2005) forecast, the regression is based on 1961-1990 (1971-2000) data, and the 1984 -1993 (1995-2004) 10-year trend is adjusted for significant changes in the difference between the station and CPC forecast region temperature. As the L3MTO presentation includes different formats (e.g., three categorical forecast, probability of exceedance, etc.) the verification uses different verification statistics: modified Heidke Skill Scores, Continuous Ranked Probability Skill Scores (CRPSS), and reliability diagrams. To avoid possible issues of small sampling, a combination of all stations is used to analyze the forecast skill versus the lead relationships. This analysis concludes that the 1994 -2005 forecasts overall did not indicate a large difference between skills for short and long leads. Therefore, to identify each individual station's performance, all leads have been used. This study identifies locations and 3-month periods with satisfactory forecast performance, and their spatial and temporal variability. The most skillful L3MTO performance is during late fall and early winter seasons in the Southwestern US.
DE: 0399 General or miscellaneous
DE: 0520 Data analysis: algorithms and implementation
DE: 0545 Modeling (4255)
DE: 0550 Model verification and validation
SC: Atmospheric Sciences [A]
MN: 2007 Fall Meeting