A predictive spatial model based on the integration of spatial information (field data, remotely sensed data, GIS) and spatial statistics is important for modeling large-scale and small-scale variability of landscapes and for predicting the distribution and patterns of native and exotic plant species and soil characteristics. We studied an area of 9252 ha within the eastern region of Rocky Mountain National Park, Colorado, USA. To predict the probability of exotic species richness, we used a new model using trend surface analysis and stepwise regression. This process is based on the ordinary least squares (OLS) estimates. Landsat TM bands 1, 2, 5, 6 and 7, elevation, slope, and aspect were found to be significant predictors with R~2 = 0.2727, residual standard error = 0.4090, F-statistic = 43.64, and p-value < 0. The probability exotic species predictive model was selected for lowest values of AIC and AICC statistics. To model the spatial continuity of small-scale variability based on cokriging,a gaussian semi-variogram model was selected for the lowest values of AIC and AICC statistics. Large-scale and small-scale variability models may be improved by recording certain vegetation types that are associated with weed invasion and native plants (indicator variables) as present or absent. Furthermore, we are refining this approach for studying single plant species (e.g., noxious weeds) to examine the association between spatial patterns and environmental variables.
展开▼