Applying logistic regression to determine regeneration risk to stand replacement fire on the Kootenai National Forest, Montana

In 1994, fire managers on the Kootenai National Forest observed that wildfires had produced regeneration loss in some stands but not in others. They questioned what site characteristics and management activities were related to this loss. To address this question and to establish guidelines to "triage" stands and prioritize management efforts, we applied a logistic regression model to data from a set of regeneration stands (n = 135) located on the Libby, Rexford, and Three Rivers Ranger Districts. The occurrence of a stand replacement fire was modeled as a logistic function of Aspect, Habitat Type, Fuel Treatment, and logarithm of trees/ac (log_TPA), with R-2 = 0.523 (P < 0.05). Odds ratios derived from logistic regression identified the descriptor characterizing regeneration stands "most at risk" for a stand replacement fire and provided a means to triage stands. Southwest and south aspects had the highest odds ratios (22 and 9) and largest coefficients Of variation (3.07 and 2.22) for the Aspect variable. Western hemlock/queencup beadlilly (Tsuga heterophylla/Clintonia uniflora) and western red-cedar/queencup beadlilly (Thuja plicata/Clintonia uniflora), with respective odds ratios of 30 and 17, had the largest coefficients of variation (3.40 and 2.83) for the Habitat Type variable. For the Fuel Treatment variable, the "no fuel treatment" category had the highest odds ratio (11) and coefficient of variation (2.38). Stands with stand replacement fire had a mean log_TPA significantly lower than that of non-stand replacement fire stands (P < 0.001). Competition from understory, vegetation may explain these findings. Bracken fern (Pteridium aquilinum), commonly found in cedar-hemlock stands and on southerly aspects, may outcompete tree seedlings and provide a fine-fuel hazard.