Skip to main content
ARS Home » Research » Publications at this Location » Publication #159645
Title: A HEURISTIC METHOD FOR TIME DISAGGREGATION OF SEASONAL CLIMATE FORECASTS.

Author
item Schneider, Jeanne
item Garbrecht, Jurgen
item UNGER, DAVID - NOAA/NWS/NCEP/CPC

Submitted to: Weather and Forcasting
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/1/2004
Publication Date: 9/24/2004
Citation: Schneider, J.M., Garbrecht, J.D., Unger, D.A. 2004. A heuristic method for time disaggregation of seasonal climate forecasts. Weather and Forcasting. 20:212-221.

Interpretive Summary: Seasonal climate forecasts are issued for overlapping 3-month periods covering the next year. To be useful in practical applications that employ daily weather generators, these forecasts need to be disaggregated into a sequence of 1-month forecasts. Direct linear algebraic approaches to disaggregation produce unrealistic sequences of monthly forecasts; this failure is an artifact of the forecast generation process. A heuristic method has been developed to disaggregate seasonal precipitation forecasts, and tested on regional precipitation data for 1971-2000 covering the contiguous United States. This heuristic method produces monthly values that reasonably replicate the forecast 3-month variations without any unrealistic behavior. It performs equally well across widely different precipitation climates. Differences between the disaggregated values and the test data are small enough to be acceptable.

Technical Abstract: To be immediately useful in practical applications that employ daily weather generators, seasonal climate forecasts issued for overlapping 3-month periods need to be disaggregated into a sequence of 1-month forecasts. Direct linear algebraic approaches to disaggregation produce unrealistic sequences of monthly forecasts, an artifact of the methods currently used to generate the 3-month forecasts. A heuristic method has been developed to disaggregate the NOAA/CPC Probability of Exceedance seasonal precipitation forecasts, and tested on precipitation data for 1971-2000 for the 102 forecast divisions covering the contiguous United States. This simple method produces monthly values that replicate the direction and amplitude of variations on the 3-month time scale, and approach the amplitude of variations on the 1-month scale, without any unrealistic behavior. It performs equally well across widely different precipitation regimes, and does a reasonable job reproducing the sudden onset of strong seasonal variations such as the southwest U.S. monsoon. Root-mean-square-errors between the disaggregated values and actual precipitation over the 30 year test period and all forecast divisions averaged 0.94 inches, which is 39% of the mean monthly precipitation, and roughly half the monthly standard deviation.