Seasonal Precipitation Forecasting with a 6–7 Month Lead Time in the Pacific Northwest Using an Information Theoretic Model

Abstract

An entropy minimax analysis for the forecast of seasonal precipitation with a 6?7 month lead time was performed for two regions in the Pacific Northwest. A model for the forecast of winter precipitation in the Willamette Valley, Oregon was developed using the 92 year period 1854?1945 for model building. Predictions made for the reserved test years 1946?82 had a 74% accuracy. Statistical tests establish a 99% confidence that this success is not due to chance. Models for the forecast of winter and spring precipitation in eastern Washington were developed using the 73 year period 1873?1945. The winter model had 69% predictive accuracy on the test years, with a 97% confidence against chance. The spring model had an accuracy of 54%, which is statistically indistinguishable from chance. Further analysis reveals that the spring model had an accuracy of 67% for the period 1946?63 and 43% for the period 1964?82, with an 88% confidence against the accuracy decrease arising from chance. This suggests a marginal validity period for the spring model in the order of 15 years following the model building period. The accuracies of the winter model for both western Oregon and eastern Washington show fair uniformity throughout the 37 year test period, with a slight downward trend in the later portion.