Accuracy Assessment of High Resolution Multispectral Satellite Imagery for Remote Sensing Identification of Wetlands and Classification of Vernal Pools in Eastern Sacramento County, California

I certify that this student has met the requirements for format contained in the University format manual, and that this thesis is suitable for shelving in the Library and credit is to be awarded for the thesis. Conservation of wetlands and their functions is important for maintaining biodiversity. California's vernal pool wetlands are unique habitats that support a suite of species with local, regional, and global conservation significance. However, due to development and agricultural conversion, vernal pool habitat has been significantly reduced throughout California and particularly in Sacramento County. Conservation of vernal pool wetlands depends on accurate identification and classification of vernal pools. Although remote sensing has been extensively used to detect wetlands, few studies have been conducted that examine the accuracy of satellite remote sensing methods to identify and classify wetlands as small as vernal pools. Consequently, this study addresses the hypothesis that remote sensing classification of high resolution satellite imagery can accurately identify wetland plant communities and classify vernal pool deep and shallow communities to a 95% level of accuracy. Three study sites were used to assess accuracy by statistically comparing high-resolution multispectral satellite imagery classification with reference areas classified through ground-surveys. Reference areas were classified into 6 land cover classes using a classification key that incorporated the USACE Wetland Delineation Manual (1987) for wetland identification and Barbour et al. (2003) for classification of vernal pool deep and shallow communities. Wetland land cover classes were correctly identified between 74 and 92% of the time across study sites. Overall site classification accuracies for all 6 land cover classes ranged from 50 to 62% and did not differ significantly among sites. Mean accuracies of land cover classes ranged from 26 to 94% and differed significantly across site. Only the upland cover class accuracy significantly differed among sites. Results show that high resolution multispectral satellite imagery can accurately identify open water wetlands, but do not accurately identify or classify other wetland types, including vernal pools and vernal pool sub-communities. This study demonstrates that remote sensing identification and classification of vernal pools using high resolution multispectral imagery is a potentially valuable method of identifying open water wetlands, such as inundated vernal pools, but suggests that limitations still exist to achieving a high level of accuracy for other wetland classes. If methods are developed to address the limitations identified in this study, future studies should be able to accurately identify wetlands and classify …