Potato late blight forecasting models for the semiarid environment of south-central Washington

Relationships between weather and outbreaks of potato late blight in the semiarid environment of south-central Washington for a 25-year period were examined with linear discriminant and logistic regression analyses. The response variable was a year either with or without a late blight outbreak. A linear discriminant function with an indicator variable for the occurrence of an outbreak during the preceding year (Y-p), number of days of rain during April and May (R(am)), and number of days of rain during July and August (R(ja)) correctly classified the disease status for 92% of the years. The percentage of years with disease outbreaks correctly classified (sensitivity) and years without disease correctly classified (specificity) were each 92%. A second linear discriminant function with Y-p, R(am), and total precipitation during May when daily minimum temperature was greater than or equal to 5 degrees C (P-m) correctly classified the disease status for 88% of the years, with a sensitivity of 92% and a specificity of 85%. Logistic regression, which unlike discriminant analysis does not assume multivariate normality, led to similar results. These results suggest that the relative disease status of a given year can be predicted before the first of June, which is 4 to 10 weeks after planting and 14 days before late blight has been observed in this region in any year. With this system, growers can be alerted relatively early in the growing season of the likelihood of a late blight outbreak in the region, providing sufficient time to thoroughly monitor individual fields and initiate fungicide sprays.