LAUSR

Abstract Starry stonewort (Nitellopsis obtusa; Characeae) is a freshwater macroalga that is considered an invasive species in North America and has only recently been identified in the upper Midwest states of Minnesota and Wisconsin (USA). While the current known extent of N. obtusa invasion in the Midwestern U.S. is limited, there is significant potential habitat for continued expansion and thus a pressing need to target surveillance and response efforts to limit further spread. Here we use data on N. obtusa presence and lake-level environmental conditions from locations in New York state to train a set of ecological niche models using three separate algorithms: random forests, boosted regression trees, and ecological niche factor analysis. These models were then used to predict habitat suitability and potential invasion risk for a set of ∼900 lakes in the upper Midwest using publicly available lake-level water quality data. Based on a cross-validation study we found that the random forest method provided the most accurate predictions, though only marginally better than boosted regression trees. Ecological niche factor analysis, while offering better than random predictions, had the highest cross-validation error rates. Though there were some inconsistencies between modeling approaches, all three tended to agree on regions of relatively high risk in central Minnesota and eastern Wisconsin and relatively low risk in north-central Wisconsin. However, there are inherent limitations to developing ecological niche models with data from one geographical region and predicting into a different region, thus there is a need for additional efforts to validate model predictions.