Models of habitat selection have been developed primarily for mobile animals with well-defined home ranges. The assumptions made by traditional techniques about habitat availability are inappropriate for species with low mobility and large home ranges, such as the wood turtle. We used paired logistic regression, typically used in medical case Ϫ control studies, to model selection of habitat within activity areas in a population of wood turtles in a watershed in western Maine. We also modeled selection of activity areas within the watershed, using nonpaired logistic regression. Within activity areas, wood turtles selected nonforested locations close to water with low canopy cover. Within the watershed, they selected activity areas close to streams and rivers with moderate forest cover and little open water. The difference between selection at these two scales suggests that wood turtles select forest edges to balance thermoregulatory and feeding needs. The model of selection of activity areas within the watershed correctly classified 84% of activity areas and random areas. This model may be useful for identifying wood turtle habitat across the landscape as part of regional conservation efforts. We suggest that paired logistic regression shows promise for analysis of habitat selection of species with movement patterns that violate the assumptions of traditional habitat selection models.

We investigated the habitat use and movements of two turtle species to assess the importance of conserving multiple wetlands and the upland matrix in which they occur. Spotted turtles ( Clemmys guttata) and Blanding's turtles (  Emydoidea blandingii) are considered threatened and endangered, respectively, in Maine where they are near the northeastern periphery of their geographic range. We used resightings of marked individuals (69 spotted, 16 Blanding's) and radiotelemetry (13 spotted, 9 Blanding's, radiotagged for one or two seasons) to investigate the movements and habitat use of both species. Individuals of both species used multiple wetlands throughout the year, including permanent and seasonal pools, forested swamps, and wet meadows. Pools occupied by spotted and Blanding's turtles were small (<0.4 ha), and they were less isolated from other wetlands than pools that did not contain turtles. Both species used uplands extensively for nesting, dormancy, and traveling between wetlands. Turtles traveled 70–570 m (spotted) and 100–1620 m ( Blanding's) to nest, and nests were located 1–120 m (spotted) and 70–410 m ( Blanding's) from the nearest wetland. Spotted and Blanding's turtles entered relatively dormant stages for 15–89 and 3–21 consecutive days, respectively, and upland dormancy sites were 12–80 m (spotted) and 30–110 m ( Blanding's) from the nearest wetland. Total distance traveled overland throughout a season was 0–1680 m and 0–6760 m for radiotagged spotted and Blanding's turtles, respectively. Most spotted turtles followed a seasonal pattern of habitat use: pools for spring activity, upland forest for relative dormancy during part of the summer, and wet meadows or forested swamps for overwintering. A seasonal pattern in the habitat use of Blanding's turtles was not as evident. Our study suggests that protecting small wetlands, maintaining generous terrestrial buffers around individual wetlands, and conserving wetlands in groups are important components of a landscape approach to species conservation.

Models of habitat selection have been developed primarily for mobile animals with well-defined home ranges. The assumptions made by traditional techniques about habitat availability are inappropriate for species with low mobility and large home ranges, such as the wood turtle. We used paired logistic regression, typically used in medical case Ϫ control studies, to model selection of habitat within activity areas in a population of wood turtles in a watershed in western Maine. We also modeled selection of activity areas within the watershed, using nonpaired logistic regression. Within activity areas, wood turtles selected nonforested locations close to water with low canopy cover. Within the watershed, they selected activity areas close to streams and rivers with moderate forest cover and little open water. The difference between selection at these two scales suggests that wood turtles select forest edges to balance thermoregulatory and feeding needs. The model of selection of activity areas within the watershed correctly classified 84% of activity areas and random areas. This model may be useful for identifying wood turtle habitat across the landscape as part of regional conservation efforts. We suggest that paired logistic regression shows promise for analysis of habitat selection of species with movement patterns that violate the assumptions of traditional habitat selection models.