GIS5027 Module 2: LULC Classification & Ground Truthing and Accuracy

 

Lab 2 this week in Photo Interpretation and Remote Sensing involved using the provided aerial raster image of Pascagoula, Mississippi area to utilize what we’ve learned so far to determine Land Use and Land Cover (LULC), classifying with the Anderson Classification System (USGS). We started by creating a polygon shapefile (“LULC”), ensuring to add two text fields for the Code and Code Description. Once that was created, we identified areas that would fit into Level I Classification, and then further determined the Level II Classification. A lot of polygons were digitized to indicate the different land use and land cover areas as shown below in my map (Figure 1).  I also ended up making two Feature Class files (“WaterBodies”, “Water2”) to use the Trace tool in the Create Features pane to mark the larger bodies of water that I considered Lakes, as well as to mark the Streams and Canals.

Figure 1: Land Use and Land Cover of the Pascagoula, MS area

Quick note here – while white is generally used to denote Perennial Snow or Ice, I elected to use it on the Transportation, Communications and Utilities as I felt that was the best indicator at the time.

Ultimately, the point of this section of the lab was to practice determining land use and land cover up to Level II Classification. I did throw in a Level III Classification from Urban or Built-up Land (Level I) by marking Mobile Home Parks (115) due to the clustering, size, and shape of the objects in one location (see Figure 1).

The second section of this lab involved Ground Truthing. While we didn’t go into the field for this one, we did utilize Google Maps to help us determine if 30 points were “true” to the LULC we designated for that area. I tried to make a Stratified Random sampling of the aerial image, but I’m not sure my points are proportional to the land cover types, so I think my sampling location would be more accurate to just Random sampling.

Out of my samples, 25 were “true” (correct LULC Classification was chosen) and 5 were “false” (were not my original LULC Classification designation). Using this information, I calculated for Overall Accuracy: 25 (total of “yes” accurate points) / 30 (total points) * 100 = 83.33% Accuracy.

 

A few notable points were:

N	11	Residential	72	This one would have been Transitional Area (76) in the provided aerial image after comparing to the Google Earth current image. This location now has a home/pool built on it, so was possibly in transition to a Residential (11) site at the time the aerial image was taken.
N	12	Commercial and Services	76	This one looks like I got half right, as half of the property now has a home where previously seems it was sand/other after comparing to Google Earth location. The other half is a cemetery, and that would be more appropriate in the Commercial and Services (12) Level II code.