Geoscience Reference
In-Depth Information
Table 6.7
Error Matrix for the Land-Cover Map in Study Area, Rondonia 1999
Reference Data
Sec.
Forest
Bare
Soil
User's
Accuracy
Classified Data
Forest
Transition
Crops
Pasture
Water
Total
Forest
2370
73
0
3
0
0
6
2452
96.7%
Secondary forest
44
233
0
12
8
4
0
301
77.4%
Transition
0
0
54
19
5
16
0
94
57.5%
Crops
0
58
0
206
14
21
6
305
67.5%
Pasture
0
7
0
5
198
9
2
221
89.6%
Bare soil
0
0
0
64
7
29
1
101
28.7%
Water
0
2
0
1
0
0
44
47
93.6%
Total
2414
373
54
310
232
79
59
3521
Producer's accuracy
98.2%
62.5%
100.0%
66.5%
85.3%
36.7%
74.6%
Note:
Overall classification accuracy = 89.0%. Kappa statistic = 0.78.
The pattern of misclassification and confusion between LC classes is similar for both the 1994
and 1999 error matrices (Table 6.6 and Table 6.7), although different image sensors (TM and ETM
)
were used. Confusion between primary and secondary forest was expected because our classification
scheme did not separate secondary forest for different successional stages. Some of the polygons
delineated in the field as secondary forest exceeded 12 years of regrowth and closely resembled
semideciduous primary forest. Accordingly, there was probably some spectral overlap between “old”
secondary forest and the semideciduous primary forest. Confusion between secondary forest and
crops occurred because many coffee areas were shaded with native species such as rubber tree
+
(Hevea
brasiliensis)
, freijó
(Cordia goeldiana)
, and Spanish cedar
(Cedrela odorata,)
or included pioneer
species such as embaúba
spp.). Therefore, shaded crops appeared as partially forested areas.
Despite the lack of homogeneity within the “transition” LC class, confusion with other cover
types (crops, pasture, and bare soil) was minimal. Confusion most likely occurred because the
transition cover type was not particularly unique (Table 6.6 and Table 6.7). The confusion seen in
the error matrices between pasture and bare soil and the confusion between bare soil and recently
planted coffee areas were expected. Overgrazed pasture had little vegetative matter, allowing these
areas easily to be misclassified as bare soil. Also, it was unlikely that spectral reflectance by coffee
plants less than 0.5 m tall planted in a 3-
(Cecropia
3-m spacing was detected and discriminated from the
surrounding soil background, resulting in confusion between young coffee and bare soil. Water,
although spectrally distinct, was easily biased along edge pixels. This was particularly true in the
case of small and circuitous watercourses in mixed systems.
¥
6.3.3
Bringing Users into the Map
Initially, the local farmers expressed substantial distrust and skepticism about the mapping
project; however, trust was established throughout the mapping process and a good working rela-
tionship was established. To best present our findings, we organized community meetings in the
areas of the farmer associations involved earlier in the process. Participation in these meetings ranged
from as few as six individuals to packed rooms with more than 30 people. These meetings inten-
tionally included the broader community and farmers who had not taken part in the data collection.
Each farmer that had provided input during the data collection phase of the study received a color
copy of the 1999 LC mapping results. Additional meetings were arranged with agricultural extension
agents, leaders of the local rural labor union, municipal officials, and middle school students.
Upon examination, farmers provided verbal confirmation of our estimates and errors. Specific
concerns closely resembled the classification errors shown in the accuracy assessment matrices
(Table 6.6 and Table 6.7). More than 30 farmers who did not participate in the data collection
process compared their estimates of LC for their individual parcels to the statistics generated from
