Show simple item record

dc.contributorJames Gibeaut
dc.contributorJohn W. Tunnell
dc.creatorWood, John S.
dc.description"A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Coastal and Marine Systems Science."
dc.descriptionRemote sensing using multi- and hyperspectral imaging and analysis has been used in resource management for quite some time, and for a variety of purposes. In the studies to follow, hyperspectral imagery of Redfish Bay is used to discriminate between species of seagrasses found below the water surface. Water attenuates and reflects light and energy from the electromagnetic spectrum, and as a result, subsurface analysis can be more complex than that performed in the terrestrial world. In the following studies, an iterative process is developed, using ENVI image processing software and ArcGIS software. Band selection was based on recommendations developed empirically in conjunction with ongoing research into depth corrections, which were applied to the imagery bands (a default depth of 65 cm was used). Polygons generated, classified and aggregated within ENVI are reclassified in ArcGIS using field site data that was randomly selected for that purpose. After the first iteration, polygons that remain classified as `Mixed' are subjected to another iteration of classification in ENVI, then brought into ArcGIS and reclassified. Finally, when that classification scheme is exhausted, a supervised classification is performed, using a `Maximum Likelihood' classification technique, which assigned the remaining polygons to the classification that was most like the training polygons, by digital number value. Producer's Accuracy by classification ranged from 23.33 % for the `MixedMono' class to 66.67% for the `Bare' class; User's Accuracy by classification ranged from 22.58% for the `MixedMono' class to 69.57% for the `Bare' classification. An overall accuracy of 37.93% was achieved. Producers and Users Accuracies for Halodule were 29% and 39%, respectively; for Thalassia, they were 46% and 40%. Cohen's Kappa Coefficient was calculated at .2988. We then returned to the field and collected spectral signatures of monotypic stands of seagrass at varying depths and at three sensor levels: above the water surface, just below the air/water interface, and at the canopy position, when it differed from the subsurface position. Analysis of plots of these spectral curves, after applying depth corrections and Multiplicative Scatter Correction, indicates that there are detectable spectral differences between Halodule and Thalassia species at all three positions. Further analysis indicated that only above-surface spectral signals could reliably be used to discriminate between species, because there was an overlap of the standard deviations in the other two positions. A recommendation for wavelengths that would produce increased accuracy in hyperspectral image analysis was made, based on areas where there is a significant amount of difference between the mean spectral signatures, and no overlap of the standard deviations in our samples. The original hyperspectral imagery was reprocessed, using the bands recommended from the research above (approximately 535, 600, 620, 638, and 656 nm). A depth raster was developed from various available sources, which was resampled and reclassified to reflect values for water absorption and water scattering, which were then applied to each band using the depth correction algorithm. Processing followed the iterative classification methods described above. Accuracy for this round of processing improved; overall accuracy increased from 38% to 57%. Improvements were noted in Producer's Accuracy, with the `Bare' vi classification increasing from 67% to 73%, Halodule increasing from 29% to 63%, Thalassia increasing slightly, from 46% to 50%, and `MixedMono' improving from 23% to 42%. User's Accuracy also improved, with the `Bare' class increasing from 69% to 70%, Halodule increasing from 39% to 67%, Thalassia increasing from 40% to 7%, and `MixedMono' increasing from 22.5% to 35%. A very recent report shows the mean percent cover of seagrasses in Redfish Bay and Corpus Christi Bay combined for all species at 68.6%, and individually by species: Halodule 39.8%, Thalassia 23.7%, Syringodium 4%, Ruppia 1% and Halophila 0.1%. Our study classifies 15% as `Bare', 23% Halodule, 18% Thalassia, and 2% Ruppia. In addition, we classify 5% as `Mixed', 22% as `MixedMono', 12% as `Bare/Halodule Mix', and 3% `Bare/Thalassia Mix'. Aggregating the `Bare' and `Bare/species' classes would equate to approximately 30%, very close to what this new study produces. Other classes are quite similar, when considering that their study includes no `Mixed' classifications. This series of research studies illustrates the application and utility of hyperspectral imagery and associated processing to mapping shallow benthic habitats. It also demonstrates that the technology is rapidly changing and adapting, which will lead to even further increases in accuracy. Future studies with hyperspectral imaging should include extensive spectral field collection, and the application of a depth correction.
dc.descriptionPhysical and Environmental Sciences
dc.descriptionCollege of Science and Engineering
dc.rightsThis material is made available for use in research, teaching, and private study, pursuant to U.S. Copyright law. The user assumes full responsibility for any use of the materials, including but not limited to, infringement of copyright and publication rights of reproduced materials. Any materials used should be fully credited with its source. All rights are reserved and retained regardless of current or future development or laws that may apply to fair use standards. Permission for publication of this material, in part or in full, must be secured with the author and/or publisher.
dc.subjectremote sensing
dc.titleHyperspectral analysis of seagrass in Redfish Bay, Texas

Files in this item


There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record