Models that relate species presence/absence data to environmental variables are commonly used to predict species distributions. However, few attempts have been made to evaluate the accuracy of such models to predict suitable, yet presently unoccupied, habitat. Unoccupied areas identified by the model as being suitable may represent errors of commission (false positives). Alternatively, these “errors” of commission may indicate that the study species is unable to colonize otherwise suitable areas as a result of biotic interactions or dispersal limitations.
We have run GLM and maximum entropy models for eastern hemlock (Tsuga canadensis) that indicate a large number of unoccupied, apparently suitable, habitat patches at local scales in southern Indiana where hemlock has a sporadic distribution. We hypothesize that these potential errors of commission actually represent suitable habitat patches that are not occupied. We tested this hypothesis with a pilot field transplant study. We planted 240 hemlock seedlings in April 2011 in a full factorial design with two variables: presence or absence of adult conspecifics, and high or low quality habitat as determined by the models. Survivorship of seedlings was recorded at two week intervals over the growing season.
Results/Conclusions
In the first six weeks of the experiment, seedlings planted in areas identified by the model as high quality habitat survived at significantly greater rates than those in low quality habitat (25.6% average increase in survivorship, p values <0.01), regardless of the presence or absence of conspecifics. In subsequent weeks, seedlings in high quality habitat maintained a higher survival rate, though not at significant levels (p values ranging from 0.12 to 0.35), largely because of the low number of treatment replicates (n=3) in our pilot study. Different causes of seedling mortality were observed in high quality habitats (desiccation) and low quality habitats (animal disturbance and smothering by leaf litter). Preliminary results from a larger field study (800 seedlings, n=10) currently in progress will also be presented.
Our pilot study indicates our species distribution models are accurately predicting areas of high quality habitat that are presently unoccupied by the species. Three potential causes for this result will be discussed: the models are predicting, to some degree, the fundamental species niche; the distribution of eastern hemlock is not in equilibrium at this scale; and the combined effect of dispersal limitations and local population extinctions due to rare stochastic events.