Lots can be made with multi-spectral data if the right skills are used. This post features the creation of a map of the amount of impervious land contained within each land parcel all through the processing of spectral signatures. Creating profiles for the spectral signatures of different areas of land cover, then further classifying these into the binary categories of pervious and impervious, amount of impervious land was able to be found in different sections of the parcel shapefile. The resulting data shows which areas contribute more to storm surge.
Methods:
After downloading and decompressing the data from ESRI's tutorial site (https://learn.arcgis.com/en/projects/calculate-impervious-surfaces-from-spectral-imagery/), I opened the ArcGIS Pro project file. Bringing the included 6-inch spatial resolution, 4-band image "Louisville_Segmented.tif" into the map, the task, '"Calculate Surface Imperviousness" was then opened. From here the change band combination task was clicked. With the the parameters set to the neighborhood 4-band raster, "Band IDs," the band combination set to "4 1 3", and "Best Match" also selected the layer was created containing the new band combination. From here the new band combination layer was ready to be segmented. For this, the "Group similar pixels into segments" was clicked and then the parameters were set to the specified parameters in the tutorial. Going through the steps of the task I then confirmed that they were correct in creating the segments desired for the differentiation purpose. The resulting raster was then processed and created.
Imperviousness was now classified. To begin this classifications needed to be made. Because ArcGIS Pro does not have this feature yet, this needed to be done in ArcMap. After opening the segmented neighborhood raster, the image classification toolbar was opened. Classifications were made with the shape tools for the specified different land-covers. These were then saved. In ArcGIS Pro, the "Classify Imagery" folder of tasks was now processed. The classifier was trained with the samples created in the last step, and the image was classified, then reclassified, creating a binary raster with impervious surfaces classified as 0 and pervious ones classified as 1.
By clicking on the "Assess Classification Accuracy" task folder, then walking through the "Create accuracy assessment points" task, accuracy points were created, tested, then recorded in the attribute table. Statistics of these correct and incorrect points are then recorded by using the Confusion Matrix. With classification accuracy above 90%, continuation to the "Calculate Impervious Surface Area" task folder was okay. Using the "Tabulate the area" task, area of each land parcel that is pervious or impervious was calculated. This table was now joined to the parcel data attribute table. From here "Impervious Area" field was symbolized and a map made.
Results:
The resulting data (shown in Figure 1) should be very accurate, and is pleasing to look at. When the confusion matrix was performed a high degree of accuracy was shown (between 90 and 100% for both classifications). This data shows that the roads and larger areas (such as the middle area with the pond) have more impervious area however and this could be confusing to a viewer because this is only because they are larger areas. A viewer needs to understand that this map does not need to be normalized because the context of the map is to see just the raw amount of impervious surface for storm surge billing.
 |
| Figure 1 |
Conclusion:
This process can be immensely helpful for working with aerial imagery, and in the specific context of this class for UAS imagery. With the GIS spectral classification, areas of land that have different cover can be specified without manual drawing of areas. This process, while not as entirely accurate as manual drawing, can be very close, and save huge amounts of time when working especially with larger data sets. This process could be used not only for city landcover and imperviousness but in analysis of irrigation, areas where pollution could be draining from, or other applications.
No comments:
Post a Comment