Module 5: Unsupervised & Supervised Classification

 In the first exercise, the UWF campus was used for our data. We used a MrSID file (multiresolution seamless image database), specifically a high-resolution aerial photograph of the UWF campus. In ERDAS for this part, I completed an unsupervised classification of the UWF image. I loaded the image with the correct parameters. Once loaded, under the Raster tab, the details mattered for the setup in the analysis. Name the Input and Output rasters, Output Signature unchecked, Number of Classes to 50. I had to change the color schemes to R3, G2, and B1. Setting the skip factors to 2 helped expedite the processing time because of how every other pixel gets analyzed. Once the analysis is complete, I reclassified the 50 classes in the attribute table. The reclassification meant changing the colors and establishing categories for the colors to represent. After the reclass and setting the classifications, I recoded the Class Names. This is an extension, my words, of the reclass process. It is necessary to establish a connection with the data for the final product (file). In Recode the values entered are to combine the reclass values into one numbered value, i.e., 1-4 for Grass then becomes 1 for all four of the Grass features in the image. Once these are Merged it is easier to conduct analysis, I calculated the percentage difference between permeable and impermeable surfaces.

In the second exercise, we used Grays Harbor, Washington imagery. In this part, I conducted a Signature Collection for Supervised Classification. Once the image is loaded and the Signature Editor tool is opened, I need the Drawing tool to draw polygons around areas for coordinates that are given. This is one way of gathering the data I wanted for an Area of Interest (AOI). The next way is by Creating Signatures from the AOI Seed tool by growing a region around an area where land cover is known. Two signatures that are used with this tool are Spectral Euclidean Distance and Neighborhood. When using this tool, is very similar to the previous method but this one is from the Inquire Legacy box where I set the At Inquire and the distance value, 11 for the Spectral Euclidean Distance, but it could be anywhere from 0-255 for pixel value. I captured the areas of interest as polygons, saved them and now they are ready for analysis. 

Final map

Module 4 Lab: Spatial Enhancement, Multispectral Data, and Band Indices

 

Image with darker pixel values

Image with higher pixel values

Image with different levels of reflectivity

In this week's lab, we were asked to perform exercises with included tasks to increase our understanding of spatial enhancement, multispectral data, and band indices. These exercises and tasks were to be conducted in ERDAS Imagine and ArcGIS Pro. In the first part of the lab, we analyzed images using different methods of filtering for high pass, low pass, and sharpen filter. The high pass filters offer advantages when looking at edges using the Range statistic creating edge detect. The low pass filter generalizes the images because the filter is being run on images that have already been filtered. Lastly, the sharpen filter is similar to the high pass filter only slightly sharpens details. In the last part of the lab, we examined histograms to locate three areas in an image based on pixel values. Grouping of pixel values towards one end over another was how we were able to determine these locations. One where the pixel value was too low, showing darker colors, another showing lighter colors, and the final one showed different levels of reflectivity of one color. 

Module 3a and 3b Lab: Intro to ERDAS Imagine

This week’s lab introduced us to ERDAS Imagine, software that is raster-based and provides tools to extract information from images while being able to change the bandwidth of the images to study them more in-depth. It was a fun lab and the biggest challenges were making sure to go “by the numbers” in lab exercise steps and working with formulas calculating the frequency, wavelength, and energy of photons for the process. We added an image provided to us and manipulated it to get an output that would be added to ArcGIS. Once we added a random selection from the image to ArcGIS we calculated the hectares of an area and created a layout of map. 

Image from ERDAS loaded into ArcGIS

The intention of the lab was to familiarize ourselves with ERDAS, after all this is Remote Sensing. The software is somewhat similar to ArcGIS in the sense that there are tools, content panes, ways of adding data, and creating data as an output to be exported. One of the differences was how ERDAS handled the raster images allowing different bandwidths to be changed in the display. ERDAS was easy to work with as long as you are patient and have a fast computer when working on the server. 

Module 2a Lab: Land Use / Land Cover Classification and Module 2b Lab: Ground Truthing and Accuracy Assessment

 Module 2a Lab: Land Use / Land Cover Classification 

For this week’s lab, we were introduced to Land Use / Land Cover Classification and Ground Truthing and Accuracy Assessment. The task is to create a map classifying land in Pascagoula, MS and afterward to ground truth the points. This area has an assortment of land types and makes for learning how to classify, although not hard it is detail-oriented and a lot of fun. In the first part of the lab, I created and added a new polygon feature class called LULC. In the new feature, I created polygons for each of the classifications in Pascagoula. After creating the polygons, labels were added, and the symbology was changed to unique values. Overall I created eight level-one codes with level-two classification, see table below. 

Module 2b Lab: Ground Truthing and Accuracy Assessment

The second part of the lab involved creating 30 points and then ground truthing those locations in Google Maps, I used Google Earth. After creating a new feature class for ground truthing I used the Create Random Points tool to generate 30 points. I used the Coordinate Conversion tool to export the random point coordinates as a kmz file. Once the file was exported I loaded it into Google Earth and was able to search each coordinate to “truth” the points compared to the random points in ArcGIS. When compared to the random points 6 of the 30 were inaccurate (wrong) from Google Earth. The accuracy assessment was 80%: 24 with “YES” correlation and 6 with “NO” correlation.

The map has an underlying image I used to create polygons of each of the classifications. The codes are based on USGS Level II. There was more that could be classified in the image but due to time constraints these were the ones that summed up the area of Pascagoula quite well.

Module 1 Lab: Visual Interpretation

The objective of exercise 1 was to identify Tone and Texture in an image. I created polygons that identified areas of tone ranging from very light to very dark. For texture, I created polygons for areas that ranged from very fine to very coarse. The objective was accomplished. Even though the image is older there are still many identifiable features representing both of these remote sensing characteristics.

The objective of exercise 2 was to identify features in the image using these factors shape/size, shadow, pattern, and association. These features were created as points within each of the feature classes when added to the image. I was able to accomplish the objective. This particular task was interesting in that it made me take a deeper look into the image. It was not easy to find something relevant to each of the criteria but I did succeed. 

Lab 6 Topic 3 Scale Effect and Spatial Data Aggregation

 Part 1b Scale Effects on Vector and Raster Data

This week’s lab was determining the effect of scale and resolution on vector and raster data. Another lab part was analyzing boundaries with Modifiable Area Unit Problem (MAUP), this involved looking at Gerrymandering in U.S. Congressional Districts.

For the vector data, the scale of data was 1:1200, 1:24000, and 1:100000. Because maps have different scales, a greater emphasis should be put on ensuring spatial accuracy is adhered to as much as possible. Understanding the effect of scale and resolution on vector data differs from observing raster data. 

In the first part of the lab, we used the Clip tool for our hydrography datasets with the county as the “clip to” feature. After clipping all the data to the county, we added fields and calculated geometry to get length, area, and total count.

As resolution decreases, the accuracy and details diminish. Scale expresses the amount of detail for vector data; the hydrographic features are polylines and vector data. Because the large scale map has more detail and the small scale has less detail, these show how the relationship between scale and these hydrography data are affected.

Map Scale 1:1500 Scale and resolution effects

Map Scale 1:20,000

Part 2b Gerrymandering

The Merriam-Webster Dictionary defines gerrymandering as “dividing or arranging a territorial unit into election districts in a way that gives one political party an unfair advantage in elections.” Its history dates back to the early 1800s when it became official and later defined but was known prior to this time. The Modifiable Areal Unit Problem (MAUP) is an issue with boundaries and scale in spatial analysis. It highlights potential issues of delineation, creating bias within voting areas, i.e., congressional districts. In this final part of the lab, the feature class consisted of the continental U.S. I used the Dissolve tool to amalgamate the districts and in doing so I was able to find out the number of polygons each Congressional District (CD) consisted of. The below picture is of CD 01, the compactness score from the Polsby-Popper test was the lowest of all the districts we looked at in this lab. It is the "worst offender" of having bizarre-shaped legislative districts.

Congressional District 01

Lab 5 M2.2 Surface Interpolation

 This week’s lab focused on water quality in Tampa Bay, officially Surface Interpolation. It is always interesting to learn how there are different ways of studying data and interpreting results. We worked with different ways of interpolating data, specifically Thiessen, Inverse Distance Weighted (IDW), and Spline (Regularized and Tension). The data (BOD_MGL) for the study used BOD (Biochemical Oxygen Demand) in MGL (Milligrams Per Liter ) to measure data points for water quality in Tampa Bay (the body of water). We needed to determine areas with low and high water quality based on the results using different interpolation techniques. 

The techniques we used to interpolate gave somewhat similar results. The Thiessen offered the same results as the non-spatial information. The IDW was very similar to Thiessen, only offering a difference in standard deviation. Spline was the interpolation technique that offered the greatest variation from the others. Interpolation offers a way to study the spatial distribution of phenomena across a wide range of points. These are a few of those options.

Thiessen-This interpolation technique contains only a single point having any location within the output polygon closer than any other point, it defines an area around a point. It divides areas into proximal zones or polygons. Thiessen polygons are also called Voronoi polygons or Voronoi diagrams.

Inverse Distance Weighted (IDW)-As the name suggests it relies on inverse distance from points with emphasis placed on the nearest ones. The mapped variables have decreased influence as distance increases from the sampled location.

Spline-This technique has two types: Regularized and Tension. Regularized offers a smooth changing surface and has values that may be outside of its range. Tension offers a less smooth surface but has data that adheres closer to sample data ranges. Both can be altered in the number of points and the weight when running the tool.

Thiessen Polygons

Spline Regularized

Spline Tension

Inverse Distance Weighted

Compare these different interpolation techniques. They are similar but do offer different levels of insight to this study area.