Changes in precipitation and temperature will affect the location of suitable habitat for tree species in many forest ecosystems. Predicting the location of suitable habitat in the future will likely demand the consideration of species’ response to novel climatic conditions. Such predictions come with uncertainty, not only about future climate, but also with uncertainty born from measurement error in the predictor variables used to develop predictive models.
While error is often present in such data, it is not always accounted for in subsequently developed models. However, the severity of consequences from failing to address variable measurement error explicitly in models remains unclear. A simulation study was undertaken to assess the impact of different types of variable measurement error on predictions of suitable habitat for tree species in the Pacific Northwest states (OR, WA, and CA). We simulated marginal and joint species response curves to precipitation, temperature, and a composite climate moisture index, and sampled with varying types and magnitudes of error. Simulated model predictions were then compared with a real data example.
Results/Conclusions
In general, predictions had greater bias when measurement error in predictor variables was unaccounted for in models. However, the severity of bias varied depending on the relative influence of the predictor variable on that species’ distribution, the modeling approach, and the physiological characteristics of the individual species. Failing to account for measurement error in climatic variables tended to weaken the climatic signal in the location of suitable habitat for a species. These results highlight the importance of considering variable measurement error, particularly when predictions of suitable habitat under future climatic conditions are of interest. This research lends support for growing calls to consider variable measurement error explicitly when modeling suitable habitat for species.