GIS Blog: Melanie A. Sutton

Locating Potential Beachrock Shelves in Dry Tortugas National Park Using Unsupervised and Supervised Classification Techniques on Aerial Imagery This final project specifically examined how unsupervised and supervised classification methods available in ERDAS Imagine could be used to locate potential beachrock shelf areas near Loggerhead Key in Dry Tortugas National Park, Florida. Near-shore beachrock shelves in this region serve as important habitats for juveniles of some fish species tolerant to the extreme conditions in these formations compared to adjacent coral reefs (Rummer et al., 2009). Monitoring of these sites is therefore relevant to understanding their importance as nurseries within reef ecosystems (Speaks et al., 2012). Automatic identification of these areas from satellite or aerial imagery avoids disturbing these sensitive habitats and can also be used for path planning the acquisition of lower cost imagery useful for continued management ( Casella et al., 2017)....

This week's lab focused on using the tools within ERDAS Imagine to perform unsupervised and supervised classification of pixels representing various features in an image. In the map above, supervised (maximum likelihood) classification was used to classify pixels based on eight different classes of land use. The results above demonstrate many urban/residential areas were misclassified within the roads class. The results can be improved by selecting more signature samples for these two classes and then also evaluating these signatures to determine the optimal 3 bands that help with differentiation prior to running the classification algorithm. Two helpful tools in ERDAS Imagine for signature evaluation include examining histogram plots and mean plots of the image bands of the features you are seeking to differentiate.

Map 1 Map 2 Map 3 The subimages above were derived from imagery from the Landsat 5 satellite. To locate the features most prominent in each of the noted areas, analysis of the histograms of the various bands was performed to locate specific peaks. Following this, each subimage was colored using a color band combination that most effectively highlighted the discovered feature with the specific histogram characteristics.  Map 1 was colored so that bodies of water would appear dark against contrasting land and urban areas. Map 2 was colored so that snow in mountainous areas would be distinghishable from surrounding areas of vegetation. Finally, Map 3 was colored to enhance shades of blue within waterways where sediment was present.

This week's lab introduced features within the ERDAS Imagine tool that can be utilized to understand and process satellite data. In the map above, the full-size Landsat Thematic Mapper image was loaded into ERDAS Imagine for preprocessing to crop a select region in Northwest Washington State. The attribute table within ERDAS Imagine was then supplemented with a field to determine area values for each of the land classifications in that area. After preprocessing, the image was then loaded into ArcGIS Pro to create the final layout. A focus of this week's skill development was cleaning up the formatting in the legend area to highlight only those classes relevant to the displayed image. In addition, we learned how to format the legend to include the area values that were imported into ArcGIS Pro from ERDAS Imagine.

This map demonstrates practice creating polygons representative of land use / land cover by examining the underlying features in the aerial photograph for details relevant to classify at Level II of the USGS Standard Land Use / Land Cover Classification System. This lab further demonstrated ways to utilize Google maps as  part of ground truthing a random set of 30 points drawn throughout the image to measure overall accuracy (shown in the map above at 70%).

This map demonstrates how to identify areas of an image based on tone (brightness/ darkness) and texture (smoothness/roughness, as measured by how much the tone changes in a small area).  This map demonstrates how to use image characteristics/criteria to identify features in the image (shape and size, shadowing, patterns, and association).  This week's lab focused on various ways to visually interpret aerial photographs. By using a 5-point scale to understand tone and texture, I was able to identify areas of an image that differed by the height of the imaged features. In the next part of the lab, I learned how to include characteristics such as shape, size, shadows, pattern, and association to identify broad categories of features (e.g., neighborhoods) and specific types of features associated with this  category (e.g., residential housing).