Spatial differences in predicted <i>Phalaris arundinacea</i> (reed canarygrass) occurrence in floodplain forest understories

Abstract Reed canarygrass ( Phalaris arundinacea L.) is one of the most common invaders of floodplains and wetlands in North America. In the Upper Mississippi River floodplain, invasion by reed canarygrass in forest understories can inhibit forest regeneration when gaps form in the overstory. Understanding the distribution of reed canarygrass in forest understories is essential for effective management and control. We used an ensemble of species distribution models including Bayesian additive regression trees, boosted trees, and random forest algorithms to predict habitat suitability for reed canarygrass in forest understories across the Upper Mississippi River floodplain (~41,000 ha). Data from forest inventory study plots with reed canarygrass presence and absence were combined with 10 hypothesized environmental predictors of reed canarygrass invasion. We applied three approaches to better understand and incorporate the influence of spatial autocorrelation among our predictor variables, including random cross‐validation, spatial cross‐validation, and spatial cross‐validation with Euclidean distance fields. Flood frequency, distance to contiguous floodplain, distance to forest edge, and distance to invaded wet meadow were among the most important environmental predictors across the three algorithms. Generally, the mean probability of reed canarygrass presence decreased with increasing flood depth, distance to contiguous floodplain, distance to invaded wet meadow, forest cover, and forest height, while relationships with other predictors were more variable. The ensemble of the three models (i.e., the average prediction) was used to map and summarize potential reed canary grass habitat suitability across the landscape. The maps generated quantified the habitat suitability for reed canarygrass and areas of agreement among the models in forest understories across the floodplain. This information can be used to better understand the extent of invasion, prioritize restoration efforts, and develop further research.