INDO-BLIGHTCAST – a model for forecasting late blight across agroecologies

Abstract

ABSTRACTA late blight forecasting model “INDO-BLIGHTCAST” has been developed at the Central Potato Research Institute, Shimla, India using meteorological data and late blight appearance dates at four locations in the Indo-Gangetic plains. The model involves computation of physiological days (P-days) and mean relative humidity (RH) of the night, accrued over seven consecutive days. Late blight was predicted to appear within 15 days if moving cumulative effective temperature (P-days) and RH exceeded 52.5 and 525, respectively for seven consecutive days. The developed model was also tested against independent data sets at three locations in the plains, two in plateau region and three in the hills. Statistical comparisons of observed and predicted dates of late blight appearance showed that the mean absolute error was 10.48, while the residual mean square error was only 13.17 indicating that the errors were quite low. The Willmott D index was 0.84 which is quite close to unity thus indicating high accuracy of model predictions and its applicability across regions and seasons. Receiver operating characteristic analysis also confirmed the superiority of this combination (accuracy of 73.6%, AUC of 0.725) to predict late blight appearance as well as its non-appearance in unfavourable years.