Breathing new life into an old friend:
Objective analog seasonal forecasting using NCEP reanalyses

Bob Hart, Jeremy Ross, Charlie Hosler, and Mike Fritsch

Department of Meteorology The Pennsylvania State University

Rich Grumm

National Weather Service State College, PA


Using the 1948-2001 NCEP/NCAR reanalysis dataset, an objective ensemble analog system is employed to produce 3-yr forecasts of tropical 1000-500hPa thickness anomalies for each day in a 51-yr period (1948-1998). When averaged over the entire 51-yr period, the analog system is more skillful that either climatology or persistence out to 25 months. Examination of the analog systems' skill within the 51-yr period reveals that the analog system is capable of predicting the onset, evolution, and end of major thickness anomalies even when the initial conditions are near climatology. For example, the analog system forecasts both the formation and decay of the major 1982-1983 thickness anomaly (associated with El Nino) when starting from early-mid 1982 initial conditions, i.e., 5-12 months prior to the onset and end, respectively, of the El Nino event. The forecasting system provided similar 5-12 month lead times on the formation and decay of major thickness anomalies for the 1987, 1990 and 1998 El Nino events. The development and maintenance of major La Nina events (e.g. 1955-1956) were also predicted. Moreover, the system exhibited periods of considerable forecast skill during neutral ENSO events (e.g., 1979-1980), suggesting that the analog method is successful at anticipating significant tropical thickness anomalies other than those driven by ENSO.

Once an optimal ensemble of analog matching dates is determined, analog forecasts can be created for any variable (e.g., temperature, precipitation, wind speed, visibility, cloud cover, etc.) for which there is a historical record for the same period from which the analogs are extracted. A preliminary test to extract temperature and precipitation forecasts for San Diego (an ENSO-sensitive city) was conducted. Analog forecast skill was found for moderate ENSO cycles, with the best skill observed during the 1983 and 1998 El Nino events.

It is emphasized that the analog system designed for these preliminary experiments is rudimentary at best. Therefore, there may be great potential for improvement. In particular, it is expected that further enhancements and customization of the analog matching technique in the mid and high latitudes will lead to increased forecast skill over climatology in locations outside the tropics and during non-ENSO related events.

Finally, in addition to presenting analog forecasts for historical events, forecasts for the next six months (based on analogs of December 2002 and January 2003) will be presented.