Calculate Impervious Surfaces from Spectral Imagery

Introduction


In this lab, we made use of an online tutorial on ArcGIS Online.  The tutorial (Calculate Impervious Surfaces Imagery)  walked us through classifying an aerial image to determine the surface types.  This can be helpful, as ground surfaces that are impenetrable to water are dangerous and can cause serious environmental problems, including flooding and contaminated runoff.  With the classification of the surface types, this can determine where the impervious surfaces are on the properties of homeowners in which the government can actually fine the owners.  In this tutorial, the data is supplied by the local government of Louisville, Kentucky in which this very action is looking to be done.

Methods and Results


The goal of this lab as mentioned in the introduction was to classify the surfaces of an aerial image of a Louisville neighborhood in which the government would then use to issue fines.  To start, the data was downloaded, and the project file was opened. This can be seen in Figure 1: Project Opened.

Figure 1: Project Opened

The project’s folder has already been linked to the project, so this step will not need to be done.  Along with this, there is a premade shapefile and accuracy points feature class that will be used later.  The spectral bands can now be extracted. This is done by clicking on Raster Functions. The pane appears in which the raster file needs to be selected for raster with the method being Bands IDs, combination being “4 1 3” and the missing band action being Best Match.  This can be seen in Figure 2: Raster Functions Pane.

Figure 2: Raster Functions Pane  

The new layer can now be created. This new layer has the bands extracted in which the vegetation appears as red, the roads appear as gray, and the roofs appear as shades of gray or blue. This can be seen in Figure 3: Bands Extracted.

Figure 3: Bands Extracted

After making sure the Extracted Bands layer is selected, the Classification Wizard can be selected in the Imagery tab.  This brings up the Image Classification Wizard in which the classification method is set to Supervised, the type is Object Based, the Classification Schema is set to Default (NLCD2011), and the output location is added to the database.  This can be seen in Figure 4: Image Classification Wizard. 

Figure 4: Image Classification Wizard

The Next button is then clicked.  In order to make it easier to classify, adjacent pixels are grouped with similar spectral characteristics into segments.  This is done by adjusting three parameters. The first is the Spectral detail which is set to 8. The second is the Spatial detail which is set to 2.  The third is the Minimum Segment Size in Pixels which is set to 20. This can be seen in Figure 5: Parameters to Segment the Image.

Figure 5: Parameters to Segment the Image

This can be seen in the Contents pane with the name of Preview_Segmented and can be seen in Figure 6: Preview_Segmented Layer.

Figure 6: Preview_Segmented Layer

Next, a supervised classification of the segments, or indicating what types of pixels or segments should be classified in what way, will be performed.  To do this, a new class has to be added. This is done on the Training Samples Manager page of the wizard in which the default classes are right clicked and the Remove Class is selected.  The NLCD2011 is then right clicked and the Add New Class is chosen. This can be seen in Figure 7: Add New Class.  

Figure 7: Add New Class

In the Add New Class window, the class was named Impervious with the Value of 20 and the color Gray 30%.  This can be seen in Figure 8: New Class - Impervious.

Figure 8: New Class - Impervious

This is done again, but this time, the class was named Pervious with a  value of 40 and the color of Quetzal Green. This can be seen in Figure 9: New Class - Pervious.

Figure 9: New Class - Pervious

Subclasses were added for both of these classes.  This is done by right clicking the parent class and choosing Add New Class.  The first was the gray roof surfaces which were named Gray Roofs with a value of 21 and the color of Gray 50%.  This subclass is now created, and samples need to be added now for references. The samples can be added by selecting the subclass and clicking the Polygon button.  This can be seen in Figure 10: Gray Roofs Subclass Created.

Figure 10: Gray Roofs Subclass Created

The roofs were zoomed in on and were outlined with the polygon feature.  This creates a new training sample which can be seen in Figure 11: New Training Sample.

Figure 11: New Training Sample

More training samples were drawn and added.  This can be seen in Figure 12: More Training Samples Added. 

Figure 12: More Training Samples Added

These training samples can now be collapsed into one, using the Collapse button.  This can be seen in Figure 13: Samples Collapsed.

Figure 13: Samples Collapsed

Two more impervious subclasses were created based upon the following table in Figure 14: Two Additional Impervious Subclasses.

Figure 14: Two Additional Impervious Subclasses


The subclasses for the pervious parent class were then created based on the table in Figure 15: Four Pervious Subclasses.

Figure 15: Four Pervious Subclasses

The training samples were then drawn throughout the image for these new subclasses.  This can be seen in Figure 16: Training Samples Drawn.

Figure 16: Training Samples Drawn

These training samples can now be collapsed into the perspective subclasses.  This can be seen below in Figure 17: Training Samples Collapsed.  The Save button can now be clicked and the Next button can be selected.  

Figure 17: Training Samples Collapsed

The classification method can now be selected.  For this project, the Support Vector Machine classifier was selected with the Maximum Number of Samples per Class being 0.  This can be seen in Figure 18: Classification Method.

Figure 18: Classification Method

The Run button can now be clicked.  Multiple processes run, so it may take a long time.  After the process has been finished, a preview of the classification is displayed on the map which can be seen in Figure 19: Preview of Classification.

Figure 19: Preview of Classification

If the preview looks good, the Next button can be clicked which will lead to the Classify page.  This page is used to run the actual classification which will be saved to the geodatabase. The parameters were left alone, and the classification ran which the classified raster was added to the map.  The next page is the Merge Classes page where the subclasses must be merged with their parent classes. Each subclass was merged with the parent class of Impervious or Pervious and can be seen in Figure 20: Subclasses Merged with Parent Classes.  The next button was then clicked.

Figure 20: Subclasses Merged with Parent Classes

The Reclassifier page is the final page of the wizard which includes tools for reclassifying small errors in the raster dataset.  For this, all layers were unchecked except the Preview_Reclass and LouisvilleNeighborhood.tif layers. While the Preview layer was selected, the Swipe tool was used under the Appearance tab.  This can be seen in Figure 21: Swipe Tool Used.

Figure 21: Swipe Tool Used

One of the errors noticed was the muddy pond area.  This area was classified as Impervious, or man made, which is not the case.  This can be seen below in Figure 22: Muddy Pond Area Error.

Figure 22: Muddy Pond Area Error

This error is corrected by clicking Reclassify within a region.  This tool allows a polygon to be drawn on the map and everything within it to be reclassified.  In the Remap Classes section, the Current Class is set to Any with the New Class selected as Pervious.  These settings can be seen below in Figure 23: Remap Classes Settings.

Figure 23: Remap Classes Settings

The polygon can now be drawn around the area.  This can be seen in Figure 24: Polygon Drawn.

Figure 24: Polygon Drawn

This area has now been reclassified as a pervious surface and can be seen in Figure 25: Reclassification to Pervious Surface.

Figure 25: Reclassification to Pervious Surface

This classified dataset can now be saved and run.  The final product of this layer can be seen in Figure 26: Reclassified Raster.

Figure 26: Reclassified Raster

For this particular lab, it was not required to complete the final lesson (I did however just for the experience, but I will not include it so if interested, please let me know).  A final map was created to show what was done in the lab and can be seen below in Figure 27: Final Map.

Figure 27: Final Map

Conclusion


As seen in this final map, the data has been classified. The terrain was not only classified by the type of terrain (grass, roads, roofs, water, etc.), but it has also been classified with the bigger picture of is it pervious or impervious. This is crucial information that the government can use to issue fines for too much impervious land use which can be dangerous for the environment. With this example in particular, the pattern looks to be that neighborhood still has quite a bit of pervious land, and the only impervious, is the intended use of houses and roads. This is a good sign, and one that will hopefully continue.

Comments

Popular posts from this blog

Processing Pix4D Imagery with GCPs

Using Arc Collector

Building Maps with UAS Data