Remote Sensing of the Environment : Rio Grande Delta Marshes
Classification of PaloAlto CASI Data
The Palo Alto Battlefield is of interest due to its historical importance. Approximately 3300 acres are under negotiation for purchase by the National Parks Service in an effort to restore the land back to conditions during the time of the Mexican-American War. The Palo Alto Battlefield was significant as U.S. General Zachary Taylor defeated the Mexican Army under the command of General Arista at this location.
Field reconnaissance yielded the following classes for the Palo Alto Battlefield:
- Spartina spartinae (dense, tall clumps)
- Low Spartina spartinae (dense cover, but smaller plant heights)
- Bare Soil
- Spartina spartinae mixed with Borrichia frutescens
- Open Spartina spartinae (sparse cover)
- Low Salt Marsh (a mixture of Monanthacloe littoralis and Salicornia)
- Water
- Trees
- Borrichia frutescens
CASI data is collected in 17 bands with wavelengths ranging from 0.448µm to 0.805µm. Figures 1 and 2 show the classified images obtained from the Best Basis and Hierarchical Tree algorithms respectively with the class colors given below. The overall importance of each band is given in Figure 3, Tables 1 and 2 show the classification accuracies with a confusion matrix:
Fig 1. Palo Alto (CASI) - Hierarchical Tree Classifier |
Fig 2. Palo Alto (CASI) - Best Basis Classifier |
Table 1: Confusion matrix(%), Best Basis Classification
Class |
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
1 |
89.294 |
2.4934 |
0 |
0.2595 |
1.3646 |
0.3934 |
0 |
13.261 |
0 |
2 |
4.5755 |
92.467 |
0 |
0.0865 |
3.6725 |
0 |
1.5625 |
1.297 |
0 |
3 |
0 |
0 |
100 |
0.173 |
0 |
0.3147 |
0 |
0.0265 |
0 |
4 |
0.1609 |
0.6897 |
0 |
80.234 |
0.7412 |
15.4996 |
0.1042 |
0.0794 |
7.9225 |
5 |
0.3396 |
2.2812 |
0 |
0.4758 |
88.6792 |
2.3603 |
2.1875 |
0 |
2.8169 |
6 |
0.286 |
0.5305 |
0 |
10.467 |
0.6907 |
80.1731 |
2.2917 |
0.1588 |
0 |
7 |
0.0715 |
0.3714 |
0 |
0 |
3.5546 |
0.7081 |
93.854 |
0.0265 |
0 |
8 |
5.1117 |
1.1671 |
0 |
0 |
0.0168 |
0 |
0 |
85.0715 |
0 |
9 |
0.1609 |
0 |
0 |
8.3045 |
1.2803 |
0.5507 |
0 |
0.0794 |
89.261 |
Overall accuracy: 88.78%
Table 2: Confusion Matrix(%), Hierarchical Classification
Class |
1 |
2 |
3 |
4 |
5 |
6 |
7 |
8 |
9 |
1 |
90.7775 |
2.8117 |
0.2415 |
0.346 |
1.0108 |
7.3958 |
0.4167 |
8.8671 |
0 |
2 |
5.5228 |
84.615 |
0.0805 |
0.3028 |
3.5714 |
0.5507 |
1.25 |
1.9852 |
0 |
3 |
0 |
0 |
99.436 |
0.173 |
0 |
0.0787 |
0 |
0 |
0 |
4 |
0.1966 |
4.8806 |
0 |
74.611 |
0.5896 |
15.0275 |
0 |
0 |
5.2817 |
5 |
0.0357 |
2.122 |
0 |
0.1298 |
89.7406 |
2.439 |
1.5625 |
0 |
16.021 |
6 |
0.0536 |
0.7427 |
0 |
10.424 |
0.1685 |
71.3611 |
1.5625 |
0 |
0 |
7 |
0 |
3.3422 |
0 |
0.0433 |
1.7857 |
2.3603 |
94.792 |
0 |
0 |
8 |
3.3244 |
0.1061 |
0.2415 |
0 |
0.0168 |
0.0787 |
0 |
89.0948 |
0 |
9 |
0.0894 |
1.3793 |
0 |
13.971 |
3.1166 |
0.7081 |
0.4167 |
0.0529 |
78.697 |
Overall accuracy: 85.9%
Notice that the Best Basis algorithm gave a higher overall accuracy. The most important band for the Best Basis technique is found as band 13. We can expect to obtain higher accuracies by using HYMAP data since it has a wider wavelength range than CASI hence more bands that can be use to better discriminate the classes.
Comparing the two classified images we observed that the Best Basis algorithm increases the number of
pure borrichia and low salt
marsh, the confusion matrices show that the accuracies for these classes are significantly improved when we use the Best Basis instead of Hierarchical Tree classifier.
Fig 3. Best Basis Weighted bands for Palo Alto CASI data.
Last Modified: Wed Apr 14, 1999
CSR/TSGC Team Web
|