Classification of shoreline vegetation in the Western Basin of Lake Erie using airborne hyperspectral imager HSI2, Pleiades and UAV data

ABSTRACT

Mapping land and aquatic vegetation of coastal areas using remote sensing for better management and conservation has been a long-standing interest in many parts of the world. Due to natural complexity and heterogeneity of vegetation cover, various remote sensing sensors and techniques are utilized for monitoring coastal ecosystems. In this study, two unsupervised and two supervised standard pixel-based classifiers were tested to evaluate the mapping performance of the second-generation airborne NASA Glenn Hyperspectral Imager (HSI2) over the narrow coastal area along the Western Lake Erie’s shoreline. Furthermore, the classification results of HSI2 (using the whole Visible-Near Infrared (VIS+ NIR) hyperspectral dataset, and also the spectral subset of Visible (VIS) spectral bands) were compared to multispectral Pleiades (VIS+ NIR) and Unmanned Aerial Vehicle (UAV) VIS classified images. The goal was to explore how different spectral ranges, and spatial and spectral resolutions impact the unsupervised and supervised classifiers. While the unsupervised classifiers depended more on the spectral range, spectral or spatial resolutions were important for the supervised classifiers. The Support Vector Machine (SVM) was found to perform better than other classification methods for the HSI2 images over all twenty-two study sites with the overall accuracy (OA) ranging from 82.6%–97.5% for VIS, and 81.5%–95.6 % for VIS + NIR. Considerably better performance of the supervised classifiers for the HSI2 VIS data over the Pleiades data (OA = 74.8–83.4%) suggested the importance of spectral resolution over spectral range (VIS vs. VIS+ NIR) for the supervised methods. The unsupervised classifiers exhibited low accuracy for both HSI2 VIS and UAV VIS imagery (OA< 30.0%) while the overall accuracy for the HSI2 VIS+ NIR and Pleiades data ranged from 60.4%–78.4 % and 42.1%–66.4%, respectively, suggesting the importance of spectral range for the unsupervised classifiers.

Acknowledgments

This project was funded by the National Oceanic and Atmospheric Administration (NOAA), Ohio Division of Natural Resources (ODNR) and University of Toledo. Authors also gratefully acknowledge National Aeronautics and Space Administration (NASA) for providing data. We thank Roger P. Tokars, Robert C. Anderson, John D. Lekki, Mr. Larry C. Liou (NASA Glenn Research Center) and Ms. Terri Benko for establishing the collaboration.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the National Oceanic and Atmospheric Administration U.S Department of Commerce through the Ohio Coastal Management Program, administered by the Department of Natural Resources, Office of Coastal Management [NA13NOS4190057]; Ohio Division of Natural Resources; University of Toledo.