Predicting rare plant occurrence in Great Smoky Mountains National Park, USA

Abstract

We investigated the applicability of biometric habitat modeling to rare plant inventory and conservation by developing and field testing a geographically explicit model for Cardamine clematitis Shuttleworth ex A. Gray (mountain bittercress), an endemic plant of the southern Blue Ridge Mountains, USA. For each of 187 confirmed coordinates for C. clematitis in Great Smoky Mountains National Park, 13 habitat variables were measured with a geographic information system. These data were used to calculate Mahalanobis distances for each 30-m x 30-m pixel within the study area; small values of Mahalanobis distance represented site conditions similar to those of known locations of C. clematitis , whereas larger distance values represented dissimilar conditions. Following model development, we tested model performance by sampling 120 randomly distributed plots for C. clematitis presence. Logistic regression showed that Mahalanobis distance values were strongly related to C. clematitis occurrence (P = 0.039). Overall, 75% of all known occurrences of C. clematitis had associated Mahalanobis distance values below 17.7, and 95% of all occurrences were below 33.8; the median Mahalanobis distance value for the study area as a whole was 40.0. A habitat suitability cutoff value was defined which identified roughly 23,640 ha (19.5% of the study area) as suitable habitat. Although the model successfully predicted species absence in test plots with high Mahalanobis distance values, many sites with low values did not contain C. clematitis . Only 16.2% of test plots below the habitat suitability cutoff contained C. clematitis . The absence of C. clematitis from sites with low Mahalanobis distance values (low specificity) is not necessarily indicative of a poor model; metapopulation processes (e.g., recolonizations, local extinctions) have been shown to play a major role in presence or absence of many plant species. That may be partially the case with our model as evidenced by a relationship between C. clematitis presence and habitat patch size.