Geoscience Reference
In-Depth Information
where
E
= relative errors (in percentage) of a given landscape index for a given thematic map,
index
I
= landscape index value derived from a thematic map, and
I
= landscape index value derived
map
ref
from a reference map.
Thematic maps were assigned to three accuracy groups based on the overall accuracy maps at
site A (
= 23). Landscape metrics were computed for each map with the FRAGSTATS for site A
(McGarigal and Marks, 1994) and with patch analyst (PA) for site B (Elkie et al., 1999). Nine
landscape indices were used for site A: largest patch index (LPI), patch density (PD), mean patch
size (MPS), edge density (ED), area-weighted mean shape index (AWMSI), mean nearest neighbor
distance (MNN), Shannon's diversity index (SHDI), Simpson's diversity index (SDI), and contagion
index (CONTAG). Thirteen landscape indices were used for site B: PD, MPS, patch size coefficient
of variance (PSCOV), patch site standard deviation (PSSD), ED, mean shape index (MSI), AWMSI,
mean patch fractal dimension (MPFD), area-weighted mean patch fractal dimension (AWMPFD),
MNN, mean proximity index (MPI), SDI, and%LAND. These landscape indices had broad repre-
sentation within the different cover categories (McGarigal and Marks, 1994).
n
15.2.1
Relative Errors of Area (REA)
If a thematic map
contains
n
classes or types, its accuracy can be assessed with an error matrix
(Table 15.1).
For a given patch type
k
(1
£
k
£
n
), the reference value of %LAND (LR
) is computed as:
k
n
Â
n
f
+
f
Â
ik
kk
f
ik
i
ik
=
π
1
f
N
+
k
i
=
1
LR
==
=
(15.2)
k
N
N
The classification value of %LAND (LC
) is derived as:
k
n
Â
n
f
+
f
Â
f
kj
kk
kj
j
jk
=
π
1
f
N
j
=
1
k
+
LC
==
-
(15.3)
k
N
N
Table 15.1 A General Presentation of an Error Matrix
Adapted from Congalton and Green (1999)
Reference Data
Classified
Cover Type
1
j
n
Total
1
f
f
f
f
11
1
j
1
n
1+
i
f
f
f
f
i
1
ij
in
i
+
n
f
f
f
f
n
1
nj
nn
n
+
Total
f
f
f
N
+1
+
j
+
n
Note:
n
= the total number of land cover types;
N
= the total
number of sampling points;
) =
the joint frequency of observations assigned to type
f
(
i
and
j
= 1, 2, …,
n
ij
i
by classification and to type
j
by reference data;
f
=
i
+
the total frequency of type
i
as derived from the clas-
sification; and
f
= the total frequency of type
j
as
+
j
derived from the reference data.
