ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensional GIS”, Bangkok, May 23-25, 2001 
401 
priori information existed indicating the influence of other 
environmental factors, e.g. geology, vegetation, landuse, etc., 
the above landscape metrics were considered of primary 
importance in the Medina study area by default. GIS data 
layers on these five environmental variables were derived from 
a 3 by 3 meter resolution DEM using conventional digital 
terrain analysis (Zevenbergen and Thorne, 1987). 
3.2.2 Identifying unique environmental combinations using 
FCM. An unsupervised FCM classification described in Section 
2.2.2 was carried out on the environmental dataset compiled in 
Section 3.2.1. The FCM were applied on the environmental 
data across three different weighting exponents (m = 1.5, 2.0, 
and 2.5, after Bezdek (1981)). For each run (per m) the 
number of clusters examined ranges from 2 to 15. By 
examining the improvements in partition coefficient (F) and 
entropy (H) for all three sets, we observed the consistent 
improvement in both entropy and partition coefficient across all 
three runs at ten clusters. No other cluster values improved 
consistently across all three run, we argue that there are ten 
clusters (ten unique combinations of environmental conditions) 
within the data. Figure 1 shows the membership distributions 
of Class 7 and Class 9. 
Figure 1: Membership distributions, (a): Class 7; (b): Class 9 
3.2.3 Investigating the class centroids in the field. Over 20 field 
observations (2 to 3 per class) were made to determine the 
soil classes associated with each of the environmental 
classes. The field observations were guided by the 
membership values. For each class, the observation sites 
were at locations where the membership values for the class 
are very high. 
The association between the environmental classes and soil 
types observed in the field is summarized in Table 1. It is clear 
that there is a good association between the environmental 
classes and the soil types in the area. However, the 
association is not one-to-one. A soil type can occur under 
different unique environmental conditions. It is important to 
point out that we did not observed different soil types at a 
single environmental class. This validates the utility of fuzzy 
membership maps in allocating field investigation efforts since 
the membership distribution allowed us to avoid sampling in 
the transitional areas. 
Table 1: Association between environmental clusters (for 
m= 2.0) and soil types. 
Cluster ID 
Observed Soil Class 
7 
Kidder 
9 
Kidder 
1 
McHenry 
5 
Transitional 
6 
St. Charles 
4 
St. Charles 
2 
St. Charles 
3 
Mayville 
8 
Virgil 
10 
Sable 
3.2.4 Distilling the soil-landscape model for the area. By 
interpreting the cluster centroids in comparison with field 
observations and examining the membership distribution of 
each environmental cluster, the local soil scientist created an 
environment description for each of soil class. Examples of 
these descriptions are shown in Table 2. These environmental 
descriptions constitute the essence of the soil-landscape 
model of the areas. Local soil scientists can map the 
distribution of each soil series using the environmental 
description for a given soil series. 
Table 2: Descriptions of environmental conditions of Soil 
Series Kidder and McHenry in the study area. 
For Soil Series Kidder 
Environmental Variable 
Environmental Conditions 
Elevation 
850-1050 feet 
Gradient 
15-35% 
Profile Curvature 
Slightly concave to convex 
Planform Curvature 
Slightly concave to convex 
Upstream Drainage Area 
Low 
Landform Position 
Drumlin tops and upslopes 
For Soil Series McHenry 
Environmental Variable 
Environmental Conditions 
Elevation 
850 - 940 feet 
Gradient 
6-15% 
Profile Curvature 
Slightly concave to convex 
Planform Curvature 
Slightly concave to slightly 
convex 
Upstream Drainage Area 
Moderately low 
Landform Position 
Strongly sloping back-slopes and 
foot-slopes 
3.3 Evaluating the so-derived soil-landscape model 
3.3.1 Applying the soil-landscape model using SoLIM. To 
assess the validity of the soil-landscape model constructed 
using the FCM approach, the soil-landscape model was used 
under the SoLIM approach to generate soil maps for the area. 
The SoLIM approach is a knowledge-based approach for soil 
mapping. It combines the knowledge of soil-environmental 
relationships (soil-landscape model) with the environmental 
conditions characterized in a GIS to infer the spatial 
distribution of soils (Zhu, 1997; Zhu 1999; Zhu et al., 2001). 
Case studies have demonstrated that the SoLIM approach to 
soil mapping is successful (Zhu et al, 2001). However, the 
quality of the soil maps from SoLIM largely depends on the 
quality of the soil-landscape model. Thus, the SoLIM approach 
provides us with the opportunity to examine the quality of the 
soil-landscape model constructed with the use of FCM. 
3.3.2 Evaluating the quality of the soil-landscape model. The 
soil map derived from SoLIM with the use of the soil-landscape 
model above is shown in Figure 2. The soil map shows a 
catenary sequence of the soils in the area: soil series Kidder at 
the ridge (drumlin) tops and upper part of the slopes; McHenry 
at the back-slopes; St. Charles at the foot-slopes; Mayville at 
the toe-slope; Virgil on the flat area of valleys and Sable at the 
wet areas besides the streams. This pattern matches field 
observation of catenary sequence in the area well. 
To validate this soil map, observations of soils at 50 field sites 
were made. Soil type at each field site was identified at the 
series level. The field observed soil series at these sites were 
then compared with the soil series obtained from the inferred 
soil map at these locations. Soil series from the inferred soil 
map match field observed soil series at 38 of the 50 sites,