Geography Reference
In-Depth Information
Table 3.10 Comparison of different classification among the IGBPDIS, SCSLC, WESTDC and
UMD products (km 2 )
Class name
WESTDC
SCSLC
GLC
UMD
Evergreen needleleaved forest
13,174
11,532
19,250
9,075
Deciduous needleleaved forest
312
4,309
25,702
10,819
Deciduous broadleaved forest
36,330
32,120
33,971
67,687
Mixed forests
6,680
3,434
123
3,529
Closed shrublands
7,219
4,878
16,212
24,049
Open shrublands
4,198
4,945
184,674
Grasslands
267,639
234,562
325,539
265,344
Permanent wetlands
18,731
18,284
10,055
Croplands
195,069
161,682
163,214
7,038
Urban and built-up
10,274
9,059
5,182
Cropland/natural vegetation mosaic
7,419
84,276
Barren or sparsely vegetated
47,482
46,324
29,255
48,860
Water bodies
11,739
10,861
2,945
9
Table 3.11 The confusion matrix for the vegetation classification from land use type to land
covers scheme
Class
EN
DN
DB
MF
CS
OS
Classified
total
Number
correct
Accuracy
Producer's
User's
EN
87,920
452
121
389
9,754
8,792
90.14
90.86
DN
990
5,346
327
213
5,985
5,346
89.32
84.62
DB
720
72
4,783
412
5,339
4,783
89.59
87.70
MF
7,130
448
223
4,956
6,340
4,956
78.17
83.02
CS
3,268
123
3,391
3,268
96.37
89.95
OS
365
4,222
4,587
4,222
92.04
97.17
Reference
total
96,760
6,318
5,454
5,970
3,633
4,345
35,396
31,367
Overall classification accuracy = 88.62 % Overall kappa statistics = 0.86
EN evergreen nedleleaved forest, DN deciduous needleleaved forest, DB deciduous broadleaved forest,
MF mixed forests, CS closed shrublands, OS open shrublands
Throughout the classification process, the accuracy of the classification maps is
assessed by a set of 35,396 sample points selected with the stratified random
sampling method; these sampling points were randomly selected for each of the
classes in the first generated classification map in this research. For each map, a
confusion matrix is created and the accuracy is measured. The use of measure-
ments such as the overall accuracy, Kappa statistics, producer's accuracy and
user's accuracy have been quite common and have been explained in detail in
numerous publications. The confusion matrix is constructed with the land cover
data using the decision rule and the large scale land cover mapping with the
integration of multi-source information, which is recognized as the real data. The
result indicates that an overall accuracy of 88.69 % is achieved, which suggests
that it gained about a 17.69 % increase in accuracy in comparison to the WESTDC
map (Table 3.11 ).
Search WWH ::




Custom Search