Modeling potential habitat distribution for freshwater diatom Didymosphenia geminata in the continental United States

Background/Question/Methods The diatom Didymosphenia geminata (Lyngbye) Schmidt, or didymo, is a single-celled freshwater alga. Nuisance blooms of D. geminata impact the diversity, abundance, and productivity of other aquatic organisms. Didymosphenia geminata is transported by humans and therefore, accurate spatial prediction of habitats that are likely to be invaded by D. geminata is urgently required to guide field crews and policy makers, identify priority areas for early detection and rapid response, and evaluate the cost-effectiveness of monitoring and control programs. We evaluated four different modeling methods for predicting potential habitat distribution for D. geminata including two presence-absence and two presence-only methods. The two presence-absence methods included stepwise multiple logistic regression and classification and regression trees (CART). Presence-only methods included a fairly recently introduced Maxent (maximum entropy modeling), and the widely used GARP (Genetic Algorithm for Rule-set Prediction). We evaluated the model performances using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, Cohen’s maximized Kappa, and correct classification rate.

Results/Conclusions The Maxent model provided the most accurate predictions with an AUC value of 0.92 and Kappa value of 0.74, followed by logistic regression (AUC = 0.91; Kappa = 0.70), CART (AUC = 0.86; Kappa = 0.59), and GARP (AUC = 0.82; Kappa = 0.32). The greatest number of suitable habitats for D. geminata was predicted in the western United States, in relatively cooler sites at higher elevations and high base flow index. This work is the first to model the distribution of a diatom based on climatic variables. The results provide insights into factors that affect the distribution of D. geminata and a spatial basis for the appearance of future nuisance blooms. Our results also show that presence-only modeling methods can be as good as presence-absence methods for modeling aquatic species distributions.