Using spatiotemporal correlative niche models for evaluating the effects of climate change on mountain pine beetle

Over the last decade, western North America has experienced the largest mountain pine beetle (Dendroctonus ponderosas Hopkins) outbreak in recorded history, and Rocky Mountain forests have been severely impacted. Although bark beetles are indigenous to North American forests, climate change has facilitated the beetle's expansion into previously unsuitable habitats. We used three correlative niche models (maximum entropy [MaxEnt], boosted regression trees, and generalized linear models) to estimate (1) the current potential distribution of the beetle in the U.S. Rocky Mountain region, (2) how this distribution has changed since historical outbreaks in the 1960s and 1970s, and (3) how the distribution may be expected to change under future climate scenarios. Additionally, we evaluated the temporal transferability of the niche models by forecasting historical models and testing the model predictions using temporally independent outbreak data from the current outbreak. Our results indicated that there has been a significant expansion of climatically suitable habitat over the past 50 yr and that much of this expansion corresponds with an upward shift in elevation across the study area. Furthermore, our models indicated that drought was a more prominent driver of current outbreak than temperature, which suggests a change in the climatic signature between historical and current outbreaks. Projections under future conditions suggest that there will be a large reduction in climatically suitable habitat for the beetle and that high-elevation forests will continue to become more susceptible to outbreak. While all three models generated reasonable predictions, the generalized linear model correctly predicted a higher percentage of current outbreak localities when trained on historical data. Our findings suggest that researchers aiming to reduce omission error in estimates of future species responses may have greater predictive success with simpler, generalized models.