Date of Completion
Seagrass ecosystems are a valuable resource, but vulnerable to changing conditions in the coastal ocean. Quantification of seagrass density and distribution from aerial imagery can be applied as a tool in resource management and ecosystem health and stability monitoring. This study investigates analytical methods for mapping eelgrass beds in an optically complex shallow, turbid estuary. Hyperspectral imagery (HSI) of Elkhorn Slough, CA was collected by the Spectroscopic Aerial Mapper with Onboard Navigation (SAMSON) instrument. In-situ data of water column and benthic optical properties and Hydrolight Radiative Transfer model were used to create a spectral library describing the reflectance of Elkhorn Slough at different depths with bottom coverage of seagrass or sediment. In the turbid waters of Elkhorn Slough with high levels of suspended particles, very subtle spectral differences between shallow water containing dark gray sediment or eelgrass with Leaf Area Index (LAI) up to 8 could not be accurately modeled. A second set of spectral libraries was created by selecting endmembers from the HSI data with known depth and benthic coverage ranging from sediment to dense eelgrass. Different classification algorithms were tested, and the Spectral Information Divergence algorithm was selected to compare the hyperspectral image pixels to the spectral libraries. This approach produced maps of eelgrass with 61% accuracy using 18 validation points along three transects covering sediment, sparse eelgrass (LAI=1-4), and dense eelgrass (LAI=6-8). Spectra from sediment and optically deep water in the channel were considered indistinguishable and the ancillary bathymetry was used to mask the deep channel. Eelgrass covered 10 ha of the Slough and net primary productivity totaled 1 x108 g C yr-1. These results could not be reproduced with uncalibrated aerial photography (e.g., Google Earth). Atmospherically corrected and calibrated hyperspectral imagery was needed to resolve the subtle spectral differences between sediment and seagrass in this turbid estuary. Better calibration and atmospheric correction of the imagery coupled with improved characterization of the inherent optical properties of the benthos and water column should lead towards the use of more radiative transfer-based approaches to mapping benthic constituents in turbid estuaries.
Bostrom, Kelley J., "Testing the Limits of Hyperspectral Airborne Remote Sensing by Mapping Eelgrass in Elkhorn Slough" (2011). Master's Theses. Paper 208.