Classification of spruce beetle hazard in Lutz and Sitka spruce stands on the Kanai Peninsula, Alaska

Abstract

Stand data from Lutz and Sitka spruce forest types occurring on the Kenai Peninsula were analyzed by tree-based classification and abductive inference to develop decision models for classifying spruce beetle hazard. Model development and validation data sets contained 286 and 88 stand observations, respectively. The final decision-tree structure yielded 13 possible hazard outcomes based on total stand basal area, percentage of total basal area composed of spruce, percentage of spruce basal area composed of trees with diameter > 25 cm, stand elevation, and stand aspect. Four paths in the decision tree led to low-hazard outcomes (spruce basal area loss ≤ 10%); two paths each led to low-medium, medium, and medium-high hazard outcomes (spruce basal area loss of ≤ 40%, 11–40%, and > 10%, respectively); and three led to high-hazard outcomes (spruce basal area loss > 40%). Results of model verification were considered very acceptable; in the worst case, predictions of high hazard were correct for 67% of the observations. Model validation results also were considered acceptable, but predictions of medium and high hazard showed a marked drop from verification results. For comparison, the same analysis was performed using abductive inference to test a modeling method better suited to automatic processing of numerous stands for landscape-level analysis. There was a high degree of correspondence between predictions of the two analytical methods.