Developing and validating a model for predicting gypsy moth (Lepidoptera: Lymantriidae) defoliation in Connecticut

A logistic regression model was developed that predicts the probability of forest defoliation by the gypsy moth, Lymantria dispar L., in Connecticut. Data used to develop the model were obtained from sketch maps of defoliation in Connecticut from 1982 to 1994, interpolations (using ordinary point kriging) of fixed area egg mass counts from permanent plots in Connecticut from 1986 to 1994, digital soil maps, and digital elevation models. Subjective comparisons of predicted and actual defoliation maps over years used to fit model parameters were similar. The model was validated with data from years not used in developing the model (1975 to 1985 and 1995). When defoliation was averaged over all locales for validation years, average predicted defoliation tracked actual average defoliation, although the association between them was not significant. Correlations between predicted and actual defoliation, when pairs being compared were at the same locations, were below 0.2 for years in which <2% of the state was defoliated. However, for years in which >2% of the state was defoliated, correlations ranged up to 0.9. The model developed here should be useful in providing probabilities of defoliation that can be used in population suppression programs.