How well do distribution data reflect physiological tolerance limits?

Background/Question/Methods

Correlative models are frequently used to project the spatial distribution of suitable climate space for species.  A major assumption of correlative modeling is that species’ location data are a reasonable approximation of tolerable climatic conditions.  However, this assumption is difficult to test because most species’ physiological tolerance limits are unknown.  We hypothesized that climatic tolerance inferred by a species’ occurrence records would be less extreme than its physiological limits and, therefore, projections of suitable climate would underestimate the available space.  We also hypothesized that physiological tolerance is better approximated by species with herbarium records covering a larger spatial extent.  We compiled physiological tolerance data for 1790 plant species from the USDA PLANTS database including tolerable minimum temperature and minimum and maximum annual precipitation.  We compared these tolerance data to climatic limits derived from occurrence records, to which we extracted minimum and maximum annual precipitation and minimum January temperature.  We used Mann-Whitney Utests to determine if the physiological tolerance range differs from values extracted to herbarium records and quantified the average difference across all species.  Finally, we compared distribution models based on measured vs. inferred physiological tolerance to quantify how much distribution data cause models to underestimate potential range.

Results/Conclusions

At large geographic scales, climatic suitability based on occurrence data is commonly used to define potential habitat.  While distribution models are assumed to underestimate the ‘fundamental niche’ of climatic tolerance, the magnitude of this problem has never been measured.  We found that the range of minimum temperature, minimum precipitation and maximum precipitation differed significantly between the physiological tolerance values and those extracted to the herbarium point locations (Mann Whitney U tests, p < 0.05 for all climate variables), with physiological values consistently suggesting broader tolerance than distribution data.  Similarly, distribution models based on occurrence data showed consistently smaller predictions of suitable habitat.  Underestimation of a species’ potential range can influence how we interpret projected responses to climate change.  For example, larger potentially suitable areas might allow species to persist in situ with climate change longer than previously expected, leading to less dramatic extinction risk.  Over large spatial extents, correlative models based on occurrence data are often the only tools available for projecting species’ responses to climate change.  Therefore, it is critical to understand the degree to which these models underestimate potential habitat if they are to be useful for conservation planning.