Anthropogenic demand for environmental resources such as food, water, timber, and residential development result in unprecedented rates of habitat fragmentation and habitat loss, the leading cause of species extinctions globally. Since the 1950s, exurban landscapes have been the fastest growing land use type in the United States, furthering the development of human dominated landscapes. Exurbanization often results in land cover change; however, in agricultural areas, exurbanization has the potential to increase habitat heterogeneity and bolster local biodiversity. Land cover is an important factor to consider when analyzing species occurrence; however, land cover of exurban landscapes is not well understood as they are often classified as development, due to the spatial resolution of readily available land cover data. The National Land Cover Dataset (NLCD) is commonly used for land cover classification due to accessibility and spatial extent; however, programs like the National Agriculture Imagery Program (NAIP) now provide high resolution imagery that can be classified for land cover at large spatial scales. In this study, we aimed to determine a more accurate methodology of land cover classification using high resolution satellite imagery to further our understanding of how human dominated landscapes are impacting species occurrence.
Results/Conclusions
We used NLCD to classify 80 exurban parcels and select sites as primarily forested or agriculture; however, majority of parcels had a significant amount of development. Preliminary maximum likelihood classification of NAIP imagery for a subset of our study region had an overall accuracy of 92.82% and kappa coefficient of 0.899, suggesting that our classification was accurate compared to the reference land cover dataset. Overall, the categorization of sites in our preliminary analysis did not change across NAIP or NLCD classification. However, through visual assessment the high-resolution imagery classification depicts a more accurate representation of land cover on exurban parcels. Though initial results showed no difference in land cover categorization, only two sites fell within the region of preliminary analysis. In order to perform a full comparison, NAIP imagery has been obtained for Butler and Warren Co. Ohio. Maximum likelihood, support vector machine, neural network, and random forest will be used as classifiers for remotely sensed land cover analysis. Furthermore, previously obtained anuran occurrence data will be used to assess how land cover impacts species occurrence on exurban parcels. This research will further our understanding of how land cover change impacts local species occurrence on human dominated landscapes.