Multi-scale sampling of native and non-native plant diversity: Examples of data analyses and applications

Multi-scale vegetation data were collected in Rocky Mountain National Park, Colorado, USA. Analyses were conducted on subsets of data depending on specific study objectives, the area of the Park in question, and the appropriate statistical approach.; Species-area curves and measures of heterogeneity combined with information on the area covered by each vegetation type showed that the types making the greatest contributions to biodiversity at the landscape-scale covered the smallest areas. This approach may provide an accurate and relatively rapid way to rank landscape-scale hotspots of plant diversity within regions of interest.; Butterfly diversity was sampled in conjunction with vegetation in the Beaver Meadows area of the Park. Over 180 vascular plant species were unique to the aspen vegetation type (N = 32 plots). The slope of the mean species-area curve for the aspen vegetation type was the steepest of the 10 types sampled, and aspen plots contained the greatest number of native butterfly species. However, aspen plots were the most heavily invaded by non-native plant species, which could have negative effects on native plant and butterfly species diversity.; I used stepwise multiple regression and modified residual kriging to estimate the numbers of native and exotic species, and the probability of observing a non-native species in 30 m x 30 m cells over a 54,000 ha area. These models can efficiently direct resource managers to areas in need of further inventory, monitoring, and non-native species control efforts.; I compared the abilities of trend surface analysis alone and in combination with regression tree analysis to develop full-coverage surfaces of the cover and richness of total, native, and non-native plant species in an 80,000 ha portion of the Park. The combination of approaches consistently outperformed the trend surface analysis alone with an increase in average R2 from 0.17 to 0.54.; Future applications of this research approach could be useful for developing the concept of “scalable models” where the area modeled is adjusted to produce the most accurate models given sparse data in complex terrain. These models could then be used to estimate variables of interest in the unsampled area to direct future work.