Eradication of biological invasions quickly becomes impractical once species have spread extensively. An important part of invasive species management is thus preventing or slowing of spread into new areas. Typically, human activities are responsible for larger-scale spread, yet they are seldom addressed quantitatively. We present the results of simulation models of the landscape-scale spread of Microstegium vimineum, an invasive grass which has rapidly invaded North American forests. This cellular automaton model is a landscape consisting of cells of different types (roadsides, edge, interior forest, disturbed/logged forest). At the level of a cell, local reproduction and spread processes are parameterized using data from patch-scale experiments, which predict average spread of less than a meter per year in all habitats. Local dispersal only is not adequate to explain observed invasion speeds. At a larger scale, we use data from long-distance dispersal experiments to parameterize the effects of road grading on rural roads. Road grading is a highly stochastic event, and we sample directly from observed distributions ranging from 0 to 270 m in a single grading event.
The resulting landscape-scale spread does mimic the observation that entire roadsides quickly become infested; however, larger-scale spread rates are still slower than observed rates unless we include additional long-distance dispersal by other vehicles. We also describe the impacts on long-distance spread rates of various management practices such as limiting grading in sensitive areas or removing the invasive along sensitive corridor areas. These results highlight the need to explicitly address human-mediated spread, often from multiple vectors, when managing invasive species, as this likely determines the speed of an invasion, rather than the natural dispersal abilities of the species.