Twenty-seven distinct benthic habitat types within eleven zones were mapped directly into a GIS system using visual interpretation of orthorectified aerial photographs and hyperspectral imagery. Benthic features were mapped that covered an area of 790 km^2. In all, 204 km^2 of unconsolidated sediment, 171 km^2 of submerged vegetation, and 415 km^2 of coral reef and colonized hardbottom were mapped.
The three types of imagery were acquired during different days with different weather conditions. The habitat type for the portions of the test area that were not interpretable due to cloud cover, glint or water quality were classified as unknown. The accuracy assessment points that fell within polygons with the habitat type of unknown were not included in the accuracy analysis. As a result, the total number of accuracy assessment points varies between the imagery types within a single area.
Two statistical analyses (Kappa and Tau test and the Z score) were preformed. The Kappa and Tau statistic for the major habitat types showed that the percent overall accuracy of photointerpretation of color aerial photography, IKONOS satellite and hyperspectral imagery is 90.7%, 86.5% and 89% respectively. The Z score showed that at the 90% confidence level there was no significant difference between data gathered from the three imagery sources. At the 95% confidence level there is a significant difference in the quality between aerial photographs and IKONOS satellite imagery.
The accuracy assessments tests showed that the ability to generate benthic habitat maps with an overall accuracy of 90% at the 95% confidence interval is reaching a threshold using imagery with three meter pixel size allowing for spectral enhancement of the imagery with reduced resolution.
GIS topologic quality was established by executing ArcView extension routines that check for: overlapping polygons, multipart polygons, sliver polygons and void polygons. Additionally checks for adjacent polygons with the same habitat attributes were completed. All errors were identified and corrected. This file is believed to be logically consistent.
The minimum mapping unit (MMU) for identifying habitats or features was 1 acre for visual photointerpretation. The software utilized in this project was designed to alert the photointerpreter each time a polygon was drawn smaller than the MMU. When this occurred the photointerpreter has the choice whether to include the polygon in the data set.
All three types of remotely sensed imagery were processed by NOS prior to map production. Individual color aerial photographs were georeferenced and mosaicked. The hyperspectral data composed of 72 ten nm wide bands were subsetted to three band composites that enhanced deep and shallow water features. IKONOS satellite imagery was corrected for atmospheric and water column effects. During the digitizing process, image stretched and manipulating image contrast, brightness and color balance were performed in the ArcView Image Analysis Extension to enhance features in the processed imagery.
A first draft map was completed and features in the imagery where uncertainties existed, due to confusing or difficult to interpret signatures, were identified for future ground validation effort. An ArcView GIS point theme was generated with points positioned on the features of uncertain habitat type or along transects though gradients between habitat types. The GIS points were converted to GPS waypoints using Trimble Pathfinder Software and were navigated to in the field using a Trimble GeoExplorer 3 GPS data logger.
A benthic habitat characterization was conducted at each site by snorkeling, free diving, or via observations from the surface where water depth and clarity permitted. GPS data were collected at each location and site ID, depth, habitat type, zone and the method used to make the assessment were recorded. The ground validation data were incorporated into the second draft of each map.