Digitizing Benthic Habitats
Individual georeferenced mosaics were loaded into ArcGIS with the Habitat Digitizer and Effects toolbars activated. The ArcGIS Effects toolbar allows each image to be easily manipulated so as to optimally adjust contrast, brightness, and color. The Minimum Mapping Unit (MMU) restriction was set to 1 acre in the Habitat Digitizer extension. One acre was selected based on the scale of the imagery and the objectives of the mapping project. As a result, some features visible in the imagery such as small isolated patch reefs and sea walls that, while important features, are quite small and beyond the scope of this mapping project.
Digitizing scale was set to 1:6000 in the Habitat Digitizer. Experimentation indicated that digitizing at this scale optimizes the trade-off between positional accuracy of lines and time spent digitizing. In general, line placement conducted while zoomed in at large scales results in excellent line accuracy and detail but can be quite time consuming. Conversely, while zoomed out, lines can be drawn quickly but lack both detail and positional accuracy.
Determining the Optimum Digitizing Scale
Results of an experiment conducted during benthic habitat mapping of the Caribbean were used to determine the optimum digitizing scale to maximize accuracy and minimize map production time. In the Caribbean digitizing experiment a 25 acre area composed of a variety of habitat types was mapped at 1:1500, 1:3000, 1:6000, and 1:12,000 on-screen scale (scale that the image appears on the computer monitor). Five replicates were conducted at each scale. Each trial was timed so we could evaluate the influence of mapping scale on production time. Resulting maps were evaluated for deviations in polygon detail relative to the map digitized at 1:1500 scale. At 1:1500, individual pixels are clearly discernible allowing highly detailed and accurate maps to be created by closely following the contours of even the most convoluted habitat boundary. Additional increases in zoom do not result in an increase in map detail and accuracy since individual pixels are already visible at 1:1500. Therefore, the map created at 1:1500 scale was used as a reference against which to compare maps digitized at scales of 1:3000, 1:6000, and 1:12,000.
The results of this experiment indicated that there is no appreciable loss in polygon detail and accuracy by digitizing at 1:6000 while mapping time was dramatically reduced. Therefore all polygons were digitized at this scale except when subtle habitat boundaries were not easily discernible at 1:6000 and zooming out to a more broad scale was required to place boundaries correctly. In this case, digitizing generally took place at a scale of approximately 1:10,000.
Using the Habitat Digitizer, habitat boundaries were delineated around signatures (e.g., areas with specific color and texture patterns) in the orthorectified mosaic corresponding to habitat types in the Classification Scheme. This was often accomplished by first digitizing a large boundary polygon such as the habitats that compose the shoreline and then appending new polygons to the initial polygon or splitting out smaller polygons within. Each new polygon was attributed with the appropriate habitat designation according to the classification scheme. It is believed that the positional accuracy of polygon boundaries is similar to that of the mosaics since delineation is performed directly on the digital imagery. Brightness, contrast, and occasionally color balance of the mosaic were manipulated to enhance the interpretability of some subtle features and boundaries. This was particularly helpful in deeper water where differences in color and texture between adjacent features tend to be more subtle and boundaries more difficult to detect. Particular caution was used when interpretation was performed from altered images, since results from color and brightness manipulations can sometimes be misleading.
The visual interpreter was provided a series of imagery files to aid in delineating and attributing polygons. These included the unmodified multispectral scenes (4m), normalized reflectance scenes (4m), and pansharpened multispectral scenes (1m).