Invasions by non-natives contribute to the loss of ecosystem biodiversity and productivity,
modification of biogeochemical cycles, and inhibit natural regeneration of native species. Japanese
stiltgrass (Microstegium vimineum) is one of the most prevalent invasive grasses in the forestlands of
Tennessee. Hence, we aim to identify potential determinants of invasion and quantify the relative
importance of each factor. We analyzed extensive field data collected Forest Inventory and Analysis
Program of the U.S. Forest Service to quantify the range expansion of Japanese stiltgrass from 2000 to
2011. We then conducted the analysis with a process of machine learning that uses boosted regression
trees and fitted the model in R containing at least 1000 trees. We calculated the response variance
explained, the area under the receiver operator characteristic curve (AUC), and the overall accuracy based
on the aggregated cross-validation results. We evaluated the reliability and validity of the optimal model.
We then used the gbm library to derive the relative influence of each potential explanatory variable in the
optimal model and constructed partial dependence plots and fitted values plots for the most influential
variables.
Results/Conclusions
As indicated by the FIA records from 2011, Japanese stiltgrass spread extensively throughout
the forestlands of Tennessee during an eleven-year period. The range expansion has been
particularly dramatic in northern latitudes. The presence of Japanese stiltgrass almost doubled
from 269 plots (7.5%) in 2000 to 404 plots (11.3%) in 2011. Of the four most influential
variables, one was a landscape feature, two were forest features, and one was a disturbance
factor. Elevation was the most influential variable, contributing 24.12%. Basal area, road
distance, and stand age were the second, third, and fourth most important variables, contributing
20.69%, 16.17%, and 15.09%, respectively. Distance to the nearest road was the only variable
selected from the disturbance factors while no forest management activities were chosen.
Landscape features, forest features, and disturbance factors had total contributions of 38.59%,
45.24%, and 16.17%, respectively. In conclusion, our analyses suggest that the range of Japanese
stilgrass will continue to expand in forested lands of Tennessee. While predicting vulnerable
habits to invasion remains difficult, our model may provide information that guides effective
control strategies and facilitate detection of newly established invasions of Japanese stiltgrass.