• Bayesian;
  • eastern North America;
  • historical validation;
  • palaeoecology;
  • spatial random effect



The distributions of many organisms are spatially autocorrelated, but it is unclear whether including spatial terms in species distribution models (SDMs) improves projections of species distributions under climate change. We provide one of the first comparative evaluations of the ability of a purely spatial SDM, a purely non-spatial SDM and a SDM that combines spatial and environmental information to project species distributions across eight millennia of climate change.


Eastern North America.


To distinguish between the importance of climatic versus spatial explanatory variables we fit three Bayesian SDMs to modern occurrence data for Fagus and Tsuga, two tree genera whose distributions can be reliably inferred from fossil pollen: a spatially varying intercept model, a non-spatial model with climatic variables and a spatially varying intercept plus climate model. Using palaeoclimate data with a high temporal resolution, we hindcasted the SDMs in 1000-year time steps for 8000 years, and compared model projections with palynological data for the same periods.


For both genera, spatial SDMs provided better fits to the calibration data, more accurate predictions of a hold-out validation dataset of modern trees and higher variance in current predictions and hindcasted projections than non-spatial SDMs. Performance of non-spatial and spatial SDMs according to the area under the receiver operating curve varied by genus. For both genera, false negative rates between non-spatial and spatial models were similar, but spatial models had lower false positive rates than non-spatial models.

Main conclusions

The inclusion of computationally demanding spatial random effects in SDMs may be warranted when ecological or evolutionary processes prevent taxa from shifting their distributions or when the cost of false positives is high.