AbstractsGeography &GIS

Oak savannas are valuable and complex ecosystems that provide multiple ecosystem goods and services, including grazing for livestock, watershed regulation, and recreation. These ecosystems of the woodland-prairie ecoregion of the Midwestern United States are, however, in danger of disappearing. The Sheyenne National Grassland, North Dakota, a protected Prairie grassland-savanna, is a representative of such rare habitats, where oak savanna is found at the landscape scale. In this research, I map the distribution patterns of oak savanna in the Sheyenne using a combination of remote sensing and geospatial datasets, including landscape topography, soils, and fire disturbance. Further, I interpret the performance of a suite of advanced Species Distribution Modeling approaches including Maximum Entropy, Random Forest, Generalized Boosted Model, and Classification Tree to analyze the primary environmental and management factors influencing oak distributions at landscape scales. Woody canopy cover was estimated with high classification accuracy (80-95%) for two study areas of the Sheyenne National Grassland. Among the four species distribution modeling approaches tested, the Random Forest (RF) approach provided the best predictive model. RF model parameters indicate that oak trees favor gently sloping locations, on well-drained upland and sandy soils, with north-facing aspect. While no direct data on water relationships were possible in this research, the importance of the topographic and soil variables in the SDM presumably reflect oak preference for locations and soils that are not prone to water saturation, with milder summer temperatures (i.e. northern aspects), providing conditions suitable for seedling establishment and growth. This research increases our understanding of the biogeography of Midwestern tall-grass oak savannas and provides a decision-support tool for oak savanna management. Advisors/Committee Members: Mandira Sigdelphuyal.