Lab 8 Geospatial Certification

Introduction

For this Project, I am earning a geospatial certification where I will be calculating impervious surfaces from Spectral Imagery. According to crd education, impervious surfaces are land surfaces that repel rainwater and do not permit it to infiltrate (soak into) the ground. In the context of this lesson, I am given a neighborhood geodatabase shown in Figure 1. To see the source of the lesson click here

Figure 1: Neighborhood Geodatabase 

Why impervious surfaces?

Aside from being a liability issue to a land owner, having accurate data about impervious surfaces can provide specific charges to individual land owners rather than having blanket charges. In other words, having data that can show more accurate impervious surfaces enables fairer taxes for homeowners. Currently, the image has bands and natural color combinations. In Figure 2: the neighborhood is adjusted with an extracted band and no yellow parcels. 

Figure 2: Neighborhood Extracted With Hidden Yellow Parcels

After using the Image Classification wizard and following its associated steps in the tutorial, I then created training samples. This was perhaps the hardest part of the tutorial. Something that I had trouble with was finding the correct way to sample the samples. In other words, if i made too little samples, the data would produce errors, if I made too many samples, the data would produce errors and if I didn't cover a variety of areas for certain samples, the data would produce errors. After many trials and errors, I selected areas around the map shown in figure 3.

Figure 3: Creating Training Samples

Once my training samples were processed, I now had assigned colors per image that I classified shown in Figure 4. Although not perfect, I am pleased to see how the program was able to identify and re-assign values throughout this area within minutes.  

Figure 4 Training Sample with Assigned Values

Next, I had again use the image classification wizard to edit, snip and merge subclasses into their parent classes shown in Figure 5.

Figure 5: Assigning Subclasses to Parent Classes with Accuracy Points

Below in Figure 6, Is the Confusion Matrix as a result of the accuracy points. Shown in the bottom right hand corner, the confusion matrix has an over 92% accuracy.

Figure 6 Confusion Matrix

Conclusion

Below in Figure 7, is a map I created showing the impervious surfaces throughout the Louisville Neighborhood. Although I am currently trying to figure out a good way to label the colors, the darker the area, the more impervious the area is. Compared to satellite imagery, UAS data has the ability to better accurately map this site at a more affordable price. As you can see, areas that are water resistant such as roads and structures that are designed to contain water are the most impervious.  

Figure 7: Map of Impervious Surfaces