Species distribution models (SDMs) relate known species occurrences to underlying environmental characteristics to develop predictions about habitat requirements and the spatial distribution of suitable habitat. These predictions are useful for a variety of purposes ranging from theoretical study of niche relationships to more applied endeavors such as habitat conservation or predicting potential spread of nonindigenous invasive species. Although SDM data requirements, modeling frameworks, and analysis of outputs have all received considerable attention, one relatively understudied topic is the extent to which the incorporation of systematic, such as subspecies or biotype information, influences predictions. We used the program Maxent to develop SDMs for the aquatic invasive macrophyte hydrilla (Hydrilla verticillata), which occurs in two distinct biotypes (monoecious vs. dioecious) in the United States. The two biotypes occur in different geographic regions and are known to demonstrate optimal growth under different climatic conditions. This lead us to hypothesize that incorporation of biotype into an SDM framework may influence model performance and improve ability to predict future habitat invasions. To test this hypothesis, we compared Maxent models developed with temperature- and precipitation-based environmental data for monoecious-only, dioecious-only, and all available hydrilla occurrence data in the United States.
Results/Conclusions
All models demonstrated strong predictive performance based on the occurrence data with which they were trained (all-data AUC=0.910, monoecious-only AUC=0.970, dioecious-only AUC=0947). Maxent models produced using compiled hydrilla occurrence data regardless of biotype forecasted the species range to spread as far west as California and Oregon, northward into the lower Midwest, and with regions of highly suitable habitats in Texas, Louisiana, Arkansas, and Florida. Monoecious-only models demonstrated an increased northern range for hydrilla, extending further north into the Great Lakes region, and with favorable ranges spanning from New Jersey to North Carolina. Dioecious-only models displayed a predicted distribution more similar to the all-data model, with ranges extending into to Texas and California and slight decreases of habitat suitability in Louisiana and Arkansas. Overall, our results indicate that the incorporation of hydrilla biotype occurrence data can influence and improve SDM predications. Future research should explore whether similar trends occur when other species are considered. If so, incorporation of more species systematic data will benefit predictions of invasive species ranges, conservation of habitat for threatened and endangered species, and other SDM applications.