Post on 11-Oct-2020
Russell Large April 28, 2015 Remote Sensing and Air. Interp. 374
Final ProjectReport Summary
The purpose of this project is to select a moderate resolution multispectral dataset of our choosing, create unsupervised/supervised classifications, and conduct an Error Assesment. Conducting an unsupervised classification can be used as a tool via ArcMap called “Iso Cluster Unsupervised Classification”. This feature allows the user to choose the amount of output classes he/she wishes to use for the area of interest. This unsupervised classification, however, may not be quite as accurate as the supervised classification. This is why the unsupervised classification is used manly as giving an estimate of how many classifications to use for a given area.
The supervised classification is then used as a more precise tool to more accurately create polygonal shapes of each classification. This method is called “Training Samples”. For each classification, I zoomed to areas around my area of interest and created polygonal training samples. After I had completed this task at least 15 times, I then “merged” the training samples into one classification. I did this for each of my classifications (Forest, Water, Residential Housing, Bare Soil, Commercial, Sparse vegetation, and Roads) as well as change the color and label. The next step was to run the “Maximum Likelihood Classification” and allow the tool to use my training samples. This supervised classification was then used to conduct the Project error Assesment with Random Points.
While creating an Error Assesment with the random points, I noticed my classification had issues misinterpreting sparse vegetation for forest areas. This did effect my Error Assesment quite a bit, having my sparse vegetation count for 18 where it should have been forest. Although I did have 24 sparse vegetations correctly classified. This example shows why my Khat Coefficient of Agreement was about 56%. This is an indication that an improvement can be made in my supervised classification. The largest needed room for improvement within my classification was the sparse vegetation, although residential housing and roads could also be improved. What may have happeend is I had not completed a thorough enough training sample for those classifications. What is important when doing so is making sure the samples done are from all over the map, not just in certain locations. Overall, I believe my Project Error Assesment could have been improved knowing exactly where improvements can be made.
The NDVI image showed in white indicate areas of healthy vegetation (reflectiveness) and dark areas indicate where red and nearinfrared are reflected poorly. An example of low value is the lakes/riviers/streams because they show darker colors in the NDVI image. Any area where higher values are indicate Where healthy vegetation, such as the river form th North and flows down towards the South East.
Table 1. M.L. Classification and Reference Sample Point Data.
Sample Point
M.L. Classification
Reference Data
Sample Point
M.L. Classification
Reference Data
1 Residential Housing
Forest 50 Sparse vegetation
Forest
2 Sparse vegetation
Sparse vegetation
51 Sparse vegetation
Sparse vegetation
3 Sparse vegetation
Sparse vegetation
52 Sparse vegetation
Sparse vegetation
4 Sparse vegetation
Forest 53 Forest Forest
5 Forest Forest 54 Sparse vegetation
Sparse vegetation
6 Forest Forest 55 Sparse vegetation
Sparse vegetation
7 Forest Forest 56 Residential Housing
Residential housing
8 Bare soil Bare soil 57 Forest Forest
9 Residential Housing
Roads 58 Forest Forest
10 Residential Housing
Residential housing
59 Sparse vegetation
Forest
11 Sparse vegetation
Sparse vegetation
60 Residential Housing
Residential housing
12 Forest Forest 61 Sparse vegetation
Sparse vegetation
13 Sparse vegetation
Sparse vegetation
62 Sparse vegetation
Forest
14 Roads Forest 63 Sparse vegetation
Sparse vegetation
15 Bare soil Bare soil 64 Sparse vegetation
Forest
16 Sparse vegetation
Forest 65 Residential Housing
Residential housing
17 Sparse vegetation
Sparse vegetation
66 Residential Housing
Forest
18 Sparse vegetation
Sparse vegetation
67 Sparse vegetation
Sparse vegetation
19 Forest Forest 68 Forest Forest
20 Forest Forest 69 Forest Forest
21 Forest Forest 70 Bare soil Bare soil
22 Sparse vegetation
Forest 71 Forest Forest
23 Sparse vegetation
Forest 72 Commercial Commercial
24 Forest Forest 73 Roads Residential housing
25 Sparse vegetation
Sparse vegetation
74 Sparse vegetation
Sparse vegetation
26 Roads Roads 75 Forest Forest
27 Forest Forest 76 Water Water
28 Sparse vegetation
Forest 77 Sparse vegetation
Forest
29 Sparse vegetation
Sparse vegetation
78 Sparse vegetation
Sparse vegetation
30 Sparse vegetation
Forest 79 Forest Forest
31 Water Forest 80 Sparse vegetation
Forest
32 Forest Forest 81 Sparse vegetation
Sparse vegetation
33 Forest Forest 82 Forest Forest
34 Forest Forest 83 Sparse Forest
vegetation35 Roads Roads 84 Sparse
vegetationSparse vegetation
36 Sparse vegetation
Sparse vegetation
85 Residential Housing
Bare soil
37 Sparse vegetation
Forest 86 Residential Housing
Forest
38 Bare soil Bare soil 87 Sparse vegetation
Sparse vegetation
39 Forest Forest 88 Sparse vegetation
Sparse vegetation
40 Sparse vegetation
Sparse vegetation
89 Sparse vegetation
Sparse vegetation
41 Forest Forest 90 Sparse vegetation
Residential housing
42 Sparse vegetation
Forest 91 Sparse vegetation
Forest
43 Forest Forest 92 Commercial Forest
44 Forest Forest 93 Residential Housing
Forest
45 Sparse vegetation
Sparse vegetation
94 Roads Bare soil
46 Residential Housing
Forest 95 Forest Forest
47 Sparse vegetation
Forest 96 Residential Housing
Residential housing
48 Residential Housing
Sparse vegetation
97 Forest Forest
49 Sparse vegetation
Forest 98 Sparse vegetation
Forest
99 Sparse vegetation
Sparse vegetation
100 Bare soil Bare soil
Table 2. M.L. Classification and Reference Data Matrix
Reference data
ML Classification Water Roads Res. Hous. Comm. Bare soil
Forest Sp. veg.
Total
Water 3 0 0 0 0 1 0 4Roads 0 2 1 0 1 1 0 5Residential Housing
0 1 5 0 1 5 1 13
Commercial 0 0 0 2 0 1 0 3Bare soil 0 0 0 0 5 0 0 5Forest 0 0 0 0 0 27 0 27Sparse vegetation 0 0 1 0 0 18 24 43Total 3 3 7 2 7 53 25 100
Khat Coefficient of Agreement 0.56391
Producer's AccuracyWater 3/3=0Roads 2/3=.33Residential Housing 5/7=.29Commercial 2/2=0Bare soil 5/7=.29Forest 27/53=50Sparse vegetation 24/25=.04
User's Accuracy Water 3/4=.25Roads 2/5=0.6Residential Housing
5/13=0.62
Commercial 2/3=.33Bare soil 5/5=0Forest 27/27=0Sparse vegetation 24/43=.45