Image Processing Reference
In-Depth Information
Table 12.2
Accuracy assessment results (table published in Stefanov and Netzband 2005, copyright
Elsevier)
Accuracy (%)
Reference
totals
Classified
totals
No.
correct
Class name
Producer
User
k
Asphalt
50
50
45
90.00
90.00
0.890
Soil/bedrock
94
77
68
72.34
88.31
0.860
Agricultural soil
50
50
44
88.00
88.00
0.869
Fluvial sediments
43
50
42
97.67
84.00
0.827
Water
43
43
42
97.67
97.67
0.975
Agricultural vegetation
52
50
44
84.62
88.00
0.868
Undifferentiated vegetation
45
50
41
91.11
82.00
0.805
Grass and shrubs
61
69
56
91.80
81.16
0.789
Mesic built
45
45
38
84.44
84.44
0.831
Xeric built
51
50
49
96.08
98.00
0.978
Reflective built surfaces
44
44
40
90.91
90.91
0.902
Total
578
578
509
Overall classification accuracy = 88.06%
0.868
cover data. A grid with 1 km
2
elements was created from the reprojected MODIS
data pixels to allow for direct comparison with the aggregated land cover classifica-
tion. The full extent of the Phoenix urban area is not captured by the ASTER data,
but we selected the location and extent of the analysis grid to capture a representative
portion of both the urban and peri-urban regions of Phoenix (Fig.
12.3
). Four metrics
were selected for analysis: Class Area, Mean Patch Size, Edge Density and
Interspersion/Juxtaposition Index. These metrics were computed using the
FRAGSTATS software package (McGarigal and Marks
1995
). This suite of metrics
was selected as best capturing the key spatial aspects of urban landscape structure
that can be easily interpreted in terms of biophysical variables obtained from 1 km
MODIS data.
Class Area (CA), as an essential descriptive statistical number, equals the area
(m
2
) of the given land cover type divided by 10,000 (to convert to hectares). Mean
Patch Size (MPS), as a measure for the grain of the investigated landscape, indi-
cates the mean land cover pixel size in hectares and is a function of the total area
of the landscape and the number of land cover types. Smaller values indicate a
higher fragmentation of the landscape. Edge Density (ED) for the evaluation of
edge complexity and the density of a given landcover class equals the sum of the
lengths of all edge segments involving the corresponding land cover type divided
by the total landscape area (converted to hectares). The Interspersion/Juxtaposition
Index (IJI), as an indicator for the extent of interspersion of landscape patches,
describes the observed interspersion over the maximum possible interspersion for
a given number of patch types within the landscape. IJI approaches 0 when the cor-
responding land cover type is adjacent to only 1 other land cover type. IJI equals
100 when the corresponding land cover type is equally adjacent to all other land
cover types (i.e., maximally interspersed and juxtaposed to other land cover types)
within the landscape.
