Date of Award
Level of Access Assigned by Author
Master of Science (MS)
Wildlife Ecology and Wildlife Conservation
William B. Krohn
Second Committee Member
Daniel J. Harrison
Third Committee Member
Steven A. Sader
Habitat association models designed to predict species occurrence are often tested by comparing predictions to field observations. Two types of error are then reported, omission (Yo of species not predicted but present on a site) and commission (% of species predicted but not present on a site). The purpose of this research was to assess the Maine Gap Analysis vertebrate predictions using the traditional site-specific approach and to determine what factors influence the amounts of error reported. I also developed a species-specific approach for testing the accuracy of the vertebrate predictions and compared these results to the site-specific method. When tested with the site-specific approach, the Maine Gap habitat models were found to have low omission errors (medians across all sites: 0.0% for both amphibians and birds, 10.0% for reptiles, and 5.4% for mammals) and higher commission errors (medians across all sites: 0.0% for amphibians, 5.0% for reptiles, 18.9% for mammals, and 91.9% for birds). Error rates were influenced by factors such as test site size and survey length, how species are defined as present, and how likely a species is to be observed during a field survey. Using a liberal definition of avian occurrence on a site increased omission error with a corresponding decreased in commission error. Test site size and inventory length also influenced commission error. As test site size q d field survey length increased, the commission error decreased p > 0.003). How likely a species is to be observed during a field survey also influenced commission error. Using an a priori ranking system called Likelihood Of Occurrence Ranks (LOORs) the commission error for birds decreased as the species' LOOR increased (ρ = -0.87 to -1.0). To date, testing of multiple-species predictions has focused on calculating site-specific error rates. Omission and commission errors are reported by taxonomic class for each site and across the entire state. An alternative approach would be to use the same data to look at the discrepancies for each species across all of the test sites. This approach would compare the predictions to field records of presence or absence for each species on sites within their range limit. Assumptions of data completeness were used to calculate error ranges that indicated model performance and variability of the error for each species. Commission error range was significantly correlated with species distribution (ρ = -0.583, P < 0.000), as well as with the likelihood of detecting a species in the field (ρ = -0.657, P < 0.000). In cases where high error range is reported for a species with a high LOOR the most likely cause for the over prediction is in the model. However, if a species has a low LOOR and a high error range, the over prediction error is likely caused by incompleteness in the test data. Site-specific and species-specific approaches to testing predicted occurrences are calculated from the same data, but provide different information. Therefore, I recommend that both approaches be used when testing predicted occurrences of multiple vertebrate species.
Schaefer, Sandra M., "An Assessment of Methods for Testing the Reliability of Wildlife Occurrence Models Used in Gap Analysis" (2002). Electronic Theses and Dissertations. 646.