Search

Predictions of defoliation are an important component of planning aerial insect control programs, especially for defoliators such as balsam fir sawfly (Neodiprion abietis (Harris)) that cause severe impacts on forest growth and yield. Currently, defoliation prediction is done manually based on field observations and experience, but it is a good candidate for a Bayesian network (BN), a flexible tool for combining available expert knowledge and empirical data. We created a BN model and linked it to a geographic information system to map predicted defoliation for balsam fir sawfly in western Newfoundland over an area of 5.7 million ha from 2001 to 2008. Based on expert knowledge, probabilistic influence of egg counts, previous defoliation, and stand characteristics (species composition, stand age, and management intervention) on subsequent-year defoliation was quantified. For validation purposes, maps created using the BN model were compared with manual defoliation predictions and with measured aerial defoliation survey maps. BN model defoliation prediction maps were found to be in moderate agreement (mean Kappa value of 0.59) with conventional manual prediction maps. Overall, the BN model showed similar accuracy to manual predictions, but with benefits of automating the process and of providing more spatial detail in predictions. [ABSTRACT FROM AUTHOR]