Synergistic methods to generate predictive models at large spatial extents and fine resolution

Abstract

Non-native plant species must be mapped and contained because they have negative economic impacts and degrade native ecosystems and wildlife habitats. By combining data from many different sources, known as data synergy, knowledge of their locations is greatly improved. Different data types (two species lists, quarter quad survey, weed mapping, and vegetation plots from 45 datasets) added different information, and the number of species recorded per county in Colorado increased on average by 30% even in the most intensively surveyed areas. These combined data were then used to create predictive models for Colorado for non-native and native species richness and a probability of occurrence model for Euphorbia esula, leafy spurge, a poisonous, non-native weed. The non-native species richness model had an r2 of 0.36, suggesting that significant general patterns could be identified. The most invaded areas in the Colorado riparian areas and the least invaded areas were high elevation vegetation types. The generalized linear model (GLM) for E. esula explained 62% of the variance in occurrence. The final model, including spatial autocorrelation, did not discriminate well when compared to a quarter quad level statewide survey of county weed managers for estimates of E. esula presence, but this result occurred, at least in part, from differences in the datasets (estimated current distribution from survey versus model of potential distribution of a spreading species) rather than actual model performance. Overall, data synergy holds promise to increase knowledge of non-native species current locations and their predicted distributions and richness over large geographic areas.