Forecasting Late Blight in Potato Crops of Southern Idaho Using Logistic Regression Analysis.

Abstract

Previously published late blight forecasts which predict the threat of disease based on the presence or absence of favorable weather have not been effective in semi-arid potato-producing areas of the Pacific Northwest (Idaho, Oregon, and Washington). Research was conducted to identify weather variables useful for forecasting late blight in southern Idaho. The objectives of this research were to (i) determine if regional weather variables could be related to the occurrence of late blight in southern Idaho, (ii) determine if disease severity (scale of 0 to 4) could be predicted using variables found to be correlated with the annual occurrence of late blight, and (iii) validate the efficacy of this model in predicting disease incidence in regions of the Columbia Basin. Weather data were collected from five locations over a 9-year period (1995 to 2003). A binary logistic regression model (0 = no disease and 1 = disease) indicated that the number of hours with favorable conditions (10°C ≤ temperature ≤ 27°C, relative humidity ≥ 80%) in April and May (HF80m) was a significant disease predictor. Logistic regression analysis using an ordinal disease scale (0 = no disease and 4 = severe disease) indicated amount of precipitation (APj) and favorable hours (HF80j) with extended periods from April to June as significant disease predictors. The binary model predicted disease occurrence more accurately, with 67.5% accuracy (27/40 years correctly predicted), 75% sensitivity (12/16 late-blight years predicted), and 62.5% specificity (15/24 non-late-blight years predicted) using a leave-1-year-out error estimate. The binary model was validated with data (1995 to 2003) from the semi-arid Columbia Basin regions, predicting disease with 80.8% accuracy (21/26 years predicted), 84% sensitivity (21/25 outbreak years predicted), and 0% specificity (0/1 non-outbreak years predicted).