POTATO LATE BLIGHT FORECASTING MODELS FOR THE SEMIARID ENVIRONMENT OFSOUTH-CENTRAL WASHINGTON

Relationships between weather and outbreaks of potato late blight in t he semiarid environment of south-central Washington for a 25-year peri od were examined with linear discriminant and logistic regression anal yses. The response variable was a year either with or without a late b light outbreak. A linear discriminant function with an indicator varia ble 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 da ys of rain during July and August (R(ja)) correctly classified the dis ease status for 92% of the years. The percentage of years with disease outbreaks correctly classified (sensitivity) and years without diseas e correctly classified (specificity) were each 92%. A second linear di scriminant function with Y-p, R(am), and total precipitation during Ma y when daily minimum temperature was greater than or equal to 5 degree s C (P-m) correctly classified the disease status for 88% of the years , with a sensitivity of 92% and a specificity of 85%. Logistic regress ion, which unlike discriminant analysis does not assume multivariate n ormality, led to similar results. These results suggest that the relat ive disease status of a given year can be predicted before the first o f 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 sufficie nt time to thoroughly monitor individual fields and initiate fungicide sprays.