Geography Reference
In-Depth Information
111355
120000
93861
96072
89667
100000
80000
48151
60000
34563
42072
27106
25002
21454
2210
8545
19703
25673
12012
40000
12129
22113
7262
15648
17259
2007
16219
11132
7303
5003
20000
5545
8718
1028
0
1768
2572
-2806
4460
1955
8210 1127
2701
1847
-3510
-
5033
-5332
-1372
-1225
-20000
Multi-Stage-Classification-Approach-Statistics/Hectare
Statistic Records of the Agriculture Ministry/Hectare
The Variance
Fig. 5.45 The comparison between the areas of various LULC-classes that generated from the
supervised classification of remotely sensed data with the statistical records in the Menbij Region
in 2007
The second method (Table
5.6
, Fig.
5.46
) is quantitative, more automated, and
used either non-remotely sensed data (e.g., GPS-measurements) or remotely
sensed data as truth-reference/s, based on suitable founded mathematical equations
(see
Sect. 5.13
).
This evaluation showed that: (1) the accuracy values range from 49.56 to
99.02 %; (2) after comparison of each of the three used algorithms (MLC, NN, and
SVM) with the three different spatial resolutions of remotely sensed data (ASTER-
15 m, LANDSAT-TM-30 m, and LANDSAT-MSS-60 m) and various spectral
resolution (ASTER-3-bands, TM-6-bands, and MSS-4-bands), for the 12 indi-
vidual classification levels, it can be concluded that the MLC-algorithm had the
highest accuracies in general, followed by SVM and finally, NN. Generally, the
accuracy decreased horizontally with the reduction of the spatial resolution at
almost each classification level, with the exception of ASTER-data at the more
detailed levels. Although these data had the best spatial resolution, there was no
corresponding increase in accuracy. Therefore, the higher spectral resolution by
LANDSAT-data with coarser spatial resolution was more important than the
higher spatial resolution by ASTER-data with coarser spectral resolution. In
addition, accuracy decreased vertically with the increase in the information
extracted at individual level; (3) after comparison of the final overall accuracy of
classification using the multi stage classification approach and the MLC, NN and
SVM algorithms (with accuracy values resulting from using one classification
approach and the same three algorithms), it was evident that the first approach
always showed a higher accuracy among the three classification algorithms. Also
here, MLC harvested the higher accuracy in both approaches. The higher accuracy
was found by using LANDSAT-MSS-data, while the offered classified classes
were too little and wide than those generated from other used remote sensing data;
and (4) therefore, the optimized results for the used remote sensing data, the
classification approach and classification algorithm were found to be LANDSAT-
TM-data (ASTER-data had insufficient spectral resolution, while LANDSAT-
Search WWH ::
Custom Search