GEOSTATISTICAL MODEL FOR FORECASTING SPATIAL DYNAMICS OF DEFOLIATION CAUSED BY THE GYPSY-MOTH (LEPIDOPTERA, LYMANTRIIDAE)

Outbreaks of the gypsy moth, Lymantria dispar (L.), typically occur ov er large areas but are difficult to predict. Previously developed mode ls forecast defoliation from preseason counts of egg masses in a given stand. In this Study, we take a different approach to defoliation pre diction: forecasts are based upon the statistical autocorrelation of d efoliation through space and time. Spatial and temporal autocorrelatio n of defoliation in historical data was quantified at a variety of sca les using variograms. We used a 30-yr time series of aerial sketch map s of gypsy moth defoliation in Massachusetts to calculate these variog rams. The variograms were then used to parameterize a geostatistical e stimation technique: three-dimensional simple kriging. Kriged estimate s are weighed averages of values from nearby locations and are typical ly used to interpolate two-dimensional data. In this study, we used kr iging to extrapolate future defoliation maps into a third dimension, t ime. Kriged estimates were expressed as probabilities of detectable de foliation. Predicted probabilities were estimated for each year of the time series and were compared with actual defoliation maps for that y ear. The kriging procedure usually performed well in predicting the sp atial distribution of outbreaks in a given year, but the magnitude of regionwide outbreaks generally lagged a year behind actual values. Tho ugh this approach is not currently suitable for operational use, it re presents a novel approach to landscape-level forecasting of insect out breaks. These models may ultimately outperform current forecasting sys tems.