Development of Regional Models that Use Meteorological Variables for Predicting Stripe Rust Disease on Winter Wheat

Abstract

Meteorological variables can be used to predict stripe rust, a disease of wheat caused by Puccinia striiformis West., at Lind, Pullman, and Walla Walla, Washington and Pendleton, Oregon in the Pacific Northwest of the United States. Regional models developed using different methodologies are described and evaluated for accuracy. Disease intensity data, collected from 1968 to 1981, were converted to a 0?9 disease index (DI) and were used as the dependent variable in regression analysis. Meteorological data were expressed as standardized negative degree days (NDDZ) accumulated during December and January, the Julian date of spring (JDS) [defined as the date when 40 or more positive degree days (PDD) accumulated during the subsequent 14 days] and PDD for the 80-day period after the JDS. In one of the regional models, NDDZ was accumulated for adjusted time periods at sites other than Pullman. Mallow's Cp criterion was used to evaluate the regression equations with different numbers of independent variables. The most accurate model uses NDDZ and JDS as the independent variables. The models were cross-validated by randomly removing 2 years' data and reformulating the model based on the remaining data; the new model was then used to compare actual and predicted DI. Predicted DI was within one standard error of the actual DI 60% of the time. Incorrect predictions occurred during years when spring was unusually favorable or unfavorable for disease development. The methodology described is applicable to developing statistical models relating other pest occurrences to meteorological conditions.