Mod 5: Supervised Classification Exercise

This week we learned about Unsupervised and Supervised Classifications. The map above is a Supervised Classification where we were given an image from Germantown, Maryland and tasked with dividing it amongst 8 classes (Agriculture, Deciduous Forest, Road, Urban, Mixed Forest, Grass, Fallow, and Water) to make a land use map for their future planning. Using ERDAS Imagine I was able to create polygons for some signatures or utilize the AOI Seed tool to grow my own. By collecting the signatures, evauluating them, classifying the image, merging the classes and calculating area the process read from lecture has come to reality. Taking into consideration Spectral Euclidian Distances and Neighborhood, then checking the histograms and signature editor columns for clarification I think the output map was an overall success in this lab. The concepts involving the histogram plots, layers and bands are still a little fuzzy in my head, but exercises like this are very helpful in making things clearer.

Mod 4: Spatial Enhancement, Multispectral Data, and Band Indices

This week we dove into the deep end of trying to find features based on feature descriptions or clues in pixel data given to us. By using what I learned in the lectures and lab exercise I was able to locate the 3 descriptions In ERDAS Imagine by either examining the histogram, visually examining the image as a grayscale or multispectral, and using the inquire cursor to find brightness values. Below are 3 maps showcasing the features for each of the 3 descriptions (clues) I was given to find.

This map above shows that water is the feature causing the Layer_4 spike between pixel values of 12 and 18. I found it by looking at the tm_00.img as a grayscale and looking for the darkest features because the spikes between the pixel values of 12 and 18 were on the left side of the graph. As we learned spikes on the left indicate dark in that band, so I concluded it must be the water by visual and the inquire cursor steps. I chose to display the feature in a False Color IR (R=4, G=3, B=2) because the red to black color made it stand out the best.

The map above shows that snow/ice is the feature that a. causes a small spike in layers 1-4 around pixel value 200, and b. a large spike between pixel values 9 and 11 in layer_5 and layer_6. I found snow and ice by looking for something bright because of the pixel value of 200 at the true color tm_00.img and the snow covered mountains were definitely the brightest. Then with the panchromatic image in another pane I could look there to see layers 5 and 6 pixel spike values 9 and 11, and I knew I was looking for something dark. The snow/ice covered areas met both the a. and b. criteria we were looking for. I chose the False Natural Color (R=5, G=4, B=3) to display them because it made the feature stand out in a bright baby blue color.

The map above shows that shallow and deep water is the feature causing certain areas of water in layers 1-3 to be brighter, layer 4 becomes somewhat brighter, and layers 5-6 remain unchanges. This area shows that variation in water. The hint was already there that we were looking for variations in water, so I experimented with different band combinations and looked near land where I knew I might see variations. (The Navy sailor in me was helpful as well, but that’s just my own experience being useful I think) Sure enough I could tell where shallow water was and where the darker deep water was, so those are the only water variations that I could tell and I feel like my hunch was correct here. I ended up using a R=3, G=2, B=1 band combination because that was the clearest to me. Overall, this was a tedious and difficult task. I tried to make it sound simple above, but there was so much confusion and fumbling between the lab and book pages on my end. Hopefully it was a success, but I do have some unease about this topic still.

Mod 3: Intro to ERDAS Imagine

This week we were introduced to a new program called ERDAS Imagine. It appears similar to ArcGIS Pro when you first open it, so nothing to be too intimidated by I hope. The above map image is a smaller image subset from a larger image given to us from a forested area in Washington State. Task included to create the random subset, add an area column to the attribute table, and take the image to ArcGIS Pro to set unique values to the parts of the forest being represented by the pixels and complete the map layout. Not bad for a first go at it from my seat. Looking forward to working with ERDAS Imagine in the future.

Mod 2: Land Use/Land Cover and Ground Truthing

This week we went all the way to Pascagoula, Mississippi in a aerial photo of course for a exercise in Land Use and Land Cover to a Level II classification. In the map above you can see the different uses and covers in polygons, and then randomly choses points to test out my interpretation. With a accuracy of 90%, I'd say I am pretty good at seeing features for what they truly are. With a few changes over time in the town, a little error is to be expected. Things are going well in class so far.