Predicting habitat use from opportunistic observations: a case study of the Virgin Islands tree boa (Epicrates granti)

Wildlife managers must often make conservation decisions based on uncertain and incomplete information. The challenge is to make the most robust predictions of species' requirements given these limitations. This is particularly the case when the species is rare and difficult to locate and baseline data are virtually non-existent. In the absence of other data, we used 143 opportunistic observations collected over 25 years and geographical information systems to predict the habitat of the endangered Virgin Islands tree boa (Epicrates granti) on St Thomas, United States Virgin Islands. We compared the habitat characteristics surrounding observations to the rest of the island using logistic habitat models with varying spatial resolution. Models formed with smaller-scale presence definitions were better able to discriminate areas of occurrence from the rest of the island but were more biased towards developed areas. To investigate habitat associations below the resolution of the models, we compared microhabitat near high-certainty observations with microhabitat at nearby, random locations. Snakes were disproportionately found in low elevation (<150 m) areas with non-stony soils. Vegetation near snakes consisted of woody plants 5–10 m tall with a high degree of vegetation continuity (e.g. mangroves, drought deciduous forests, thicket/scrub). This habitat occurs primarily along the southeastern coast of St Thomas. Our multi-scale approach allowed a more informed prediction of the snakes' requirements than any single-scale approach, particularly in light of the variable resolution of the observations.