Document: MAT-3-68-11

The role of topography in determining re-burned areas in two large Rocky Mountain wilderness areas .

ROLLINS, M.G.* ¹, T.W.SWETNAM ¹ and P.MORGAN ²

University of Arizona, Tucson, AZ 85721 ¹
University of Idaho, Moscow, ID 84833 ²

Abstract:
Little is known about the factors that determine spatial patterns of fire frequency at landscape scales (1,000 to 1,000,000 ha). Frequency of fire determines the distribution of vegetation structure, composition, and successional trajectory across landscapes. Explicit information on the spatial patterning of fire frequency is critical for successful efforts focused on re-introducing fire as a keystone disturbance process on Rocky Mountain landscapes. We present empirical models describing relationships between topography and re-burned areas in two, regionally distinct wilderness areas: the 226,455-ha Gila Wilderness Area, New Mexico and the 546,627-ha Selway-Bitterroot Wilderness Area, Idaho/Montana. We used continuous surface models for elevation (digital elevation models) and derived slope and aspect data to predict areas burned two or more times during the 20th century. We used likelihood ratio tests, logistic regression, and classification and regression trees (CART) to investigate associations between topography and area burned multiple times in each wilderness area. Our emphasis was to develop models based on easily obtained (or derived) topographic information, along with spatially and temporally continuous fire perimeter data for the 20th century to develop probability surfaces for use in future ecological and operational wilderness management planning. Results indicated that slope and aspect subordinated to elevation in our models. Coefficients for elevation, slope, and aspect were obtained using logistic regression, but standard logistic regression models failed to identify non-monotonic relationships between fire frequency and topography. CART models were simple, with bifurcations defined by elevation. While CART models correctly classified non-monotonic relationships between elevation and fire frequency, slope, and aspect had no overall effect until cluster membership was too low to affect the overall model. While statistical significance of our logistic regression models and CART models was limited, the overall trends and patterns represented by the models indicate that important inferences about fire frequency may be determined using simple derivatives of digital elevation models.

Keywords: Fire Frequency, Wilderness, GIS

This abstract is being presented at: 3:30 PM in session:
Oral Session #35: Fire Ecology.