Species distribution modeling is commonly used to forecast species ranges under climate change, and is a valuable tool for conservation managers and ecologists planning for climate change impacts. However, uncertainty in forecasts, stemming from modeling decisions and inherent climate change uncertainty, is rarely sufficiently investigated, limiting usefulness of range predictions.
Using a large data set of observations from the Botanical Information and Ecology Network (BIEN), we modeled present and future ranges using MaxEnt of 592 species of cacti (Cactaceae), representing over a third of the family. Given that cacti are one of the most endangered clades on the planet, and little is known about their vulnerability to climate change, there is a pressing need for range forecasts for this family. Furthermore, they are an excellent model family given their morphological diversity and presence in every ecosystem except tundra.
Maps were validated against expert maps provided by the IUCN SSC Cactus and Succulent Plants Specialist Group. Future projections were produced for 2050 and 2070 climate averages under two carbon emission pathways and eight climate models and their average. Maps were built under three dispersal limitation assumptions as well as the presence or absence of impacts of human land use. Environmental variables were chosen in four ways: randomly, using the same set of variables for every species, and with two approaches in which variables were chosen based on climate variability in geographic space, allowing for model flexibility as projections are scaled to many species. Finally, we built sets of models with a tendency to overfit as well as underfit occurrence patterns in order to better bracket uncertainty. Overall, 3,456 future maps were produced per species. The effects of the different sources of uncertainty were disentangled using ANOVA at the level of individual species as well as resulting richness maps.
Results/Conclusions
Both present species and diversity maps matched expert maps well. Median predicted range size changes across species varied from a decrease of 40% to over 90% depending on climate scenario and modeling decisions. Richness maps predict a decrease in intensity of current cactus diversity hotspots as well as geographic shifts of their locations. Preliminary analyses indicate that variable selection is the strongest driver of uncertainty in richness forecasts, followed by choice of climate model. Our results highlight the negative impact of climate change on a group of plants commonly believed to be one of the most resilient plant families on the planet.