Predicting the distributions of regional endemic dragonflies using a combined model approach

Climate warming is predicted to have large effects on insects, yet several data shortfalls, including distributional information, impede effective conservation strategies. Knowledge of species distributions is a critical component for assessing conservation need but is often lacking for endemic or rare taxa, especially invertebrates. One approach to better inform this gap is by using species distribution modelling (SDM) to predict suitable habitat and guide field surveys. Here, we combine the predictions of two machine learning algorithms, maximum entropy and Random Forest, to estimate the current and future distributions of two endemic dragonflies of the Ozark-Ouachita Interior Highlands region in the southcentral United States. Current suitable areas predicted by both algorithms largely overlapped for each species, but different environmental variables were most important for predicting their distributions. Field validation of these models resulted in new detections for both species showing their utility in guiding subsequent field surveys. Future projections under two climate change scenarios support maintaining current suitable areas as these are predicted to be strongholds for these species. Our results suggest that combining outputs of multiple species distribution models is a useful tool for better informing the distributions of geographically limited or rare species.