Hierarchical tree classifiers to find suitable sites for sandplain grasslands and heathlands on Martha’s Vineyard Island, Massachusetts

Abstract

The grasslands and heathlands of Martha’s Vineyard Island, Massachusetts provide habitat for unusual, rare, and endangered species. Currently, these globally rare ecosystems exist as fragments on the southern coast of the island within a matrix of wooded, agricultural, and developed land. We used existing land cover patterns to train a hierarchical tree classifier model according to 10 biogeoclimatic and positional variables. This model was then used to predict suitable sites where grasslands and heathlands do not presently occur but could be efficiently established. The tree classifier model indicates that many sites compatible with sandplain grasslands currently contain agricultural lands, residential development, mowed grassland, and commercial development. Efficient establishment could be undertaken at locations buffering existing grasslands to increase their extent by 67% relative to current conditions and join isolated fragments. Sites compatible with heathlands also currently contain agriculture and residential development as well as later successional vegetation including maritime forests and pitch pine. Conversion of these sites to heathlands, as indicated by the model, would increase the current extent by 25% and increase average patch size. An adaptive management approach to grassland and heathland supplementation could be employed to validate the causal mechanisms between the predictor variables and the vegetative communities. Testing of the predictive accuracy of the Martha’s Vineyard Island tree classifier models at other coastal sites would also provide information on the true mechanisms responsible for distribution patterns.