For purposes of conservation, we require methods to predict where species occur currently, and where they are likely to move in the future. Species distribution models can be useful tools, by predicting where species are in areas that cannot be surveyed but where management decisions must occur. However, they are rarely tested using independent field surveys to determine their accuracy. These models are also generally developed over broad spatial scales; in contrast, distributions must often be predicted over local to regional scales for conservation purposes. Using butterfly surveys along transects in eastern Ontario, we tested whether independent regional observations of butterflies could be predicted from distribution models developed on Canada-wide species ranges. We also developed a set of regional models using Maximum entropy software to compare with the independent field surveys. Our results show that regional field occurrences are not accurately predicted by the national models, which explain between 0 and 31% (median 4.2%) of the deviance in species presence/absence. Even when species distribution models developed using regional variables have high apparent accuracy (AUC values exceed 0.8), they do not appear to accurately predict occurrence at local spatial extents, explaining between 0 and 46% (median 3.4%) of the deviance in species presence/absence. Species distribution models' inaccuracy in predicting species' presence within their ranges indicates that field-based observations remain indispensable when making conservation decisions. Misusing models can be costly if they lead to errors in predicting the presence of species of concern or their habitat and can result in poor management decisions if used improperly.
| Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/28612 |
| Date | January 2010 |
| Creators | Nadeau, Julie L |
| Publisher | University of Ottawa (Canada) |
| Source Sets | Université d’Ottawa |
| Language | English |
| Detected Language | English |
| Type | Thesis |
| Format | 52 p. |
Page generated in 0.0096 seconds