Statistical species distribution models (SDMs), which correlate the observed presence/absence or abundance of a species with environmental variables, are increasingly being used to forecast the distributions of species in response to changing climate. Although SDMs have high predictive accuracy when modeling contemporary distributions, many assumptions of the models may be violated when predictions are projected to future climate conditions. In particular, SDMs assume that species are in equilibrium with the environmental conditions that limit their distributions and that their observed distribution adequately represents a species’ fundamental niche. However, even in the absence of biotic limitations to a species’ fundamental niche, its observed distribution will not adequately represent its response to novel climates, i.e. climate conditions beyond the range of our contemporary climate conditions. Late-Quaternary fossil pollen records across North America show that associations of plant taxa with no current analog have occurred. No-analog plant communities were most common between 17000 and 12000 years ago, during periods with higher-than-current annual ranges of temperature and insolation, suggesting that these taxa grew in climates with no contemporary analog. Understanding how species actually respond to novel climates as well as how SDM predictions are affected by novel climates is an important step in more accurately projecting future distributions. Fossil records of species occurrence and abundance from across North America from the past 21000 years and paleoclimate simulations from global climate models were used to examine whether the distribution of taxa along climatic axes varied as climate changed from the late Pleistocene through the Holocene. We built calibration datasets using multiple time slices from the late Pleistocene as a way of building a fuller realization of the fundamental niche for species, and test the SDM predictions  by comparing them with observed contemporary species distributions.  

Results/Conclusions

We highlight the differences in prediction accuracy among taxa in which the correlations with environmental conditions are more or less conserved to evaluate the effect of novel climates on predictive accuracy. We compare the projections to future climate simulations from SDMs calibrated with contemporary data to SDMs calibrated with a composite of data spanning the entire time period of 21000 years ago to examine the sensitivity of future predictions to the range of climate conditions sampled by the calibration data. We show that predictions from models calibrated from contemporary distributions have higher uncertainty than models calibrated from multiple time slices due to poor calibration for novel future climates.