Estimating habitat value using forest inventory data: The fisher (Martes pennanti) in northwestern California

Abstract

Managing forests for multiple objectives requires balancing timber and vegetation management objectives with needs of sensitive species. Especially challenging is how to retain the habitat elements for species that are typically associated with late-seral forests. We develop a regionally specific, multivariate model describing habitat selection that can be used – when linked to an institutional forest inventory program – to assess, monitor and forecast habitat conditions for a key wildlife species. We use the fisher (Martes pennanti) in northwestern California as our example and develop a predictive model for resting habitat that is created using data from the specific region where it will be used. We explore how this resting habitat model differs from a similar model developed for the Sierra Nevada and consider the implications for forest management. We developed the model using MaxEnt by comparing vegetation data at 99 randomly selected fisher resting structures on public and tribal lands in northwestern California with 883 Forest Inventory and Analysis (FIA) plots within the same ecoregion. A total of 58 alternative vegetation models were specified and the top 10 were nearly identical in their performance (Gain>1.08; Area Under the Curve [AUC]>0.89). We chose a five-variable model (canopy closure, tree age, total basal area, volume of “large” wood and basal area of hardwoods) because it included the fewest variables and included only those that could be affected by management. This model was similar to the Sierra Nevada model, but did not include topographic features (e.g., slope) nor did it include a variable representing the density of small trees. The absence of variables related to topography may make it easier for managers to affect positive change in resting habitat suitability since all variables can be influenced by management actions. Moreover, the model indicates that small trees appear to be less important (compared to southern Sierra Nevada) and therefore the probability of producing high-value resting habitat without higher fire risk is greater. We also created a spreadsheet that simplifies the process of generating habitat predictions from new data. Since metrics of stand structure and wildlife habitat are sensitive to sample design, collecting new data with FIA protocols will provide the most accurate estimates of predicted habitat with this model. Together, the Sierra Nevada and northwest California models provide managers in California a quantitative means to assess and monitor resting habitat suitability using current and future data that are part of an institutionally supported program to inventory forest vegetation.