Mapping and modeling leafy spurge spread in Theodore Roosevelt National Park, North Dakota using spatial information and spatial statistics

Leafy spurge, an exotic species, is a durable, troublesome invasive weed with significant economic and social costs. The economic results in the northern Great Plains have made the study and control of leafy spurge a current and important research topic. Leafy spurge was first reported in Theodore Roosevelt National Park (THRO) in the late 1960's. In 1970, it was estimated that 13 ha of the park was infested. The latest 1998 estimate was 1,198 ha of the 18,680 ha ‘South (Management) Unit.’; The primary objectives of this research were to define the relationships between various physical factors and the presence of leafy spurge on the landscape at THRO, develop a ranked probability map of occurrence, and to estimate the rate of spread. Spatial statistical analysis was used to develop a binary regression tree classifications for mapping leafy spurge, to evaluate spatial autocorrelation and cross correlation statistics between leafy spurge and physical environmental variables, and to develop the Ordinary Least Squares (OLS) based on stepwise regression analysis and examining residual characteristics using kriging based on semi-variogram models.; The traditional classification for presence and absence of leafy spurge had a 62% overall mapping accuracy, and the binary regression classification trees had a 74% overall mapping accuracy. The criteria for selecting all the spatial statistical models were the lowest values of standard errors, Akaike's Information Criteria (AICC) statistics, and high R2. The trend surface predictive probability model and the presence and absence (suitable vs. non-suitable habitat) predictive model accounted for ∼35% of the variability of leafy spurge on the landscape, using the OLS procedure. All variables used to predict these two models were significant at a <0.05 level. To capture the fine-scale variability for both models, semi-variogram models, based on Exponential (probability model) and Gaussian (presence and absence) models were used to krig the residuals of both surfaces. The R2 values were 42.5% (probability) and 39.9% (presence and absence).; The combined predictive spatial model indicated high probability areas in the drainages and midslope swales in the northwest wilderness area of the South Unit of THRO. Additional moderate probability zones were identified in most of the drainage basins. Spread rates averaged 5 to 10-meters per year at THRO. This study concluded that the OLS and kriged model of leafy spurge probability was superior over the OLS alone in performance, and that the slope, aspect, and zone (slope position) exhibited significant cross-correlation with elevation in the presence of spurge on the landscape. The study approach provided useful technical tools to forecast landscape-scale for invasive species (plants or animals) for improved management activities within ecosystems of the landscape at different scale levels.