Browsing College of Natural Science and Mathematics (CNSM) by Subject "wetland mapping"
Now showing items 1-1 of 1
Airborne hyperspectral imaging for wetland mapping in the Yukon Flats, AlaskaThis study involved commissioning HySpex, a hyperspectral imaging system, on a single-engine Bush Hawk aircraft; using it to acquire images over selected regions of the Yukon Flats National Wildlife Refuge; establishing a complete processing flow to convert raw data to radiometrically and geometrically corrected hypercubes, and further processing the data to classify wetlands. Commissioning involved designing a customized mount to simultaneously install two-camera systems, one operating in the visible and near infrared region, and the other operating in the shortwave infrared region. Flight planning incorporated special considerations in choosing the flight direction, speed, and time windows to minimize effects of the Bidirectional Reflection Distribution Function (BRDF) that are more dominant in high latitudes. BRDF effects were further minimized through a special processing step, that was added to the established hyperspectral data processing chain developed by the German Space Agency (DLR). Instrument commissioning included a test flight over the University of Alaska Fairbanks for a bore-sight calibration between the HySpex system's two cameras, and to ensure the radiometric and geometric fidelity of the acquired images. Calibration resulted in a root mean square error of 0.5 pixels or less for images acquired from both cameras at 1-meter spatial resolution for each geometrically corrected flight line. Imagery was radiometrically corrected using the ATCOR-4 software package. No field spectra of the study areas were collected due to logistics constraints. However, a visual comparison between current spectral libraries and acquired hyperspectral image spectra was used to ensure spectral quality. For wetlands mapping, a 6-category legend was established based on previous United States Geological Survey and United States Fish and Wildlife Service information and maps, and three different classification methods are used in two selected areas: hybrid classification, spectral angle mapper, and maximum likelihood. Final maps were successfully classified using a maximum likelihood method with high Kappa values and user's and producer's accuracy are more than 90% for nearly all categories. The maximum likelihood classifier generated the best wetland classification results, with a Kappa index of about 0.90. This was followed by the SAM classifier with a Kappa index of about 0.57 and lastly by the hybrid classifier that achieved a Kappa index of only 0.42. Recommendations for future work include using higher-accuracy GPS measurements to improve georectification, building a spectral library for Alaskan vegetation, collection of ground spectral measurements concurrently with flight image acquisition, and acquisition of LiDAR or RGB-photo derived digital surface models to improve classification efforts.