Logistic regression models to predict stripe rust infections on wheat and yield response to foliar fungicide application on wheat in Kansas

Abstract

Stripe rust, caused by Puccinia striiformis f. sp. tritici, historically has been a minor problem in the Great Plains. However, Kansas had significant losses due to stripe rust in 2001, 2003, and 2005. Recent research on the population of P. striiformis suggests changes in the fungal population may have been responsible for these epidemics. The objectives of this research were to determine conditions that are favorable for the infection of P. striiformis f. sp. tritici isolates from the current population and develop models to predict infection events. Two week old potted seedlings were inoculated with an isolate of P. striiformis and exposed to ambient weather conditions for 16 hours. Results of this bioassay were used to develop logistic regression models of infection. Models using hours at relative humidity >87%, leaf wetness, and mean relative humidity predicted infection with 93%, 80%, and 76% accuracy. Future research will use these results to determine weather patterns that influence the probability of stripe rust epidemics and to facilitate the development of regional prediction models for stripe rust. Foliar diseases of wheat result in an average yield loss of 7.8% in Kansas. Although it is possible to reduce these losses with foliar fungicides, the yield increases resulting from these applications may not justify the additional costs. The objective of this research was to develop models that help producers identify factors associated with disease-related yield loss and the profitable use of foliar fungicides. Data were collected for two years at three locations in central Kansas to determine yield response to fungicide application on eight varieties with varying degrees of resistance. Logistic regression was used to model the probability of a yield response >4 bushels per acre based on disease resistance of a variety, historical disease risk, and in-season disease risk. The accuracy of the resulting prediction models ranged from 84% to 71%. A model combining in-season disease risk and variety resistance was most accurate. The prediction accuracy of the model was 79% when tested with an independent validation dataset. In the future, these models will serve as educational tools to help producers maximize profit and productivity.