98 
To summarize, the presented method has the following 
advantages: 
1. It is relatively fast. 
2. It enables a flexible combination of automated 
learning with the existing background knowledge. 
3. It exploits information contained both in the 
continuous and discrete GIS attributes while no 
assumption regarding the distribution of the GIS data 
is made. 
4. Hypotheses in the form of classification rules can be 
quickly examined and their errors visualized. 
5. Once the decision trees are defined, the periodical 
updating of the map should be easier. 
The presented method has the following drawbacks: 
1. The classification accuracy depends also on the 
analyst's knowledge of the study area. 
2. Many explanatory GIS data layers are necessary. 
3. Although being simple in principle, it is technically 
complicated to implement because of the wide 
spectrum of necessary tools. 
4. The method does not provide means for accounting 
for the uncertainty in the input data. 
The first step decision tree: 
UNSUPERVISED_RESULT = Forest: 
. .PROX_LAKE = 0 : 
:..TM5 <= 43: Water 
: TM5 > 43 : Marsh 
PROX_LAKE > 0 : 
:..SLOPE > 0: Forest 
SLOPE = 0 : 
:..TM5 <= 50: Forest 
TM5 > 50: Marsh 
UNSUPERVISED_RESULT = Farmland: 
.FOREST81 = Forest: 
:..NDVI <= 211: Farmland 
: NDVI > 211: Forest 
FOREST81 = Non-forest: 
:..NDVI > 156: Farmland 
NDVI <= 156: 
:..PROX_SET_HW <= 51: Unvegetated 
PROX_SET_HW > 51: Farmland 
UNSUPERVISED_RESULT = Marsh: 
. SLOPE = 0 : 
:..TM5 <= 50: Forest 
: TM5 > 50: Marsh 
SLOPE > 0 : 
: . .PROX_LAKE = 0 : 
PROX_LAKE 
:..TM5 <= 
TM5 
Marsh 
0 : 
69 : Forest 
6 9 : Shrub 
UNSUPERVISED_RESULT = Unvegetated: 
..PROX_SET_HW > 195: Farmland 
PROX_SET_HW <= 195: 
:..PROX_SET_HW <= 25: Unvegetated 
PROX_SET_HW > 25: 
:..NDVI <= 105: Unvegetated 
NDVI > 105: 
:..PROX_LAKE = 0: Water 
PROX_LAKE > 0: Farmland 
UNSUPERVISED_RESULT = Shrub: 
.-..FOREST81 = Forest: Forest 
:..TM5 <= 75: Forest 
: TM5 > 75: Shrub 
FOREST81 = Non-forest: 
:..SLOPE = 0: 
..PROX_WATER = 0: Marsh 
PROX_WATER > 0: 
:..NDVI <= 204: Marsh 
NDVI > 204: Shrub 
SLOPE > 0: 
:..PROX_SET_HW <= 430: Farmland 
PROX_SET_HW > 430: 
:..POP_DENSITY <= 64: Shrub 
POP_DENSITY > 64: Farmland 
The second step decision tree: 
TREEl_RESULT = Forest: 
:..PROX_LAKE > 0: Forest 
: PROX_LAKE = 0 : Water 
TREEl_RESULT = Farmland: 
.NDVI <= 190: Farmland 
NDVI > 190: 
:..TM5 <= 81: Abandoned_pasture 
TM5 > 81: Farmland 
TREE1_RESULT = Water: 
:..FOREST81 = Non-forest: Water 
: FOREST81 = Forest: Forest 
TREEl_RESULT = Marsh: 
:..FOREST81 = Forest: Forest 
: FOREST81 = Non-forest: Marsh 
TREE1_RESULT = Shrub: 
:..TM5 <= 80: Shrub 
TM5 > 80: Abandoned_pasture 
Table 5: The two successive decision trees for reclassification of the (intermediate) unsupervised classification results