ISPRS, Vol.34, Part 2W2, “Dynamic and Multi-Dimensionai GIS", Bangkok, May 23-25, 2001
399
ASSISTING THE DEVELOPMENT OF KNOWLEDGE FOR PREDICTIVE MAPPING USING
A FUZZY C-MEANS CLASSIFICATION
A-Xing ZHU 12 Edward ENGLISH 1
’Department of Geography
University of Wisconsin
550 North Park Street
Madison, Wisconsin 53706
Tel: (608) 262-0272
Fax: (608) 265-3991
Email: axina@aeoqraphv.vvisc.edu
State Key Lab of Resources and Environmental Information System
Institute of Geography
Chinese Academy of Sciences
Beijing 100101, China
Keywords: Fuzzy Set, Geographic Information System, Soil Mapping, Expert System, Case-Based Reasoning
Abstract
Knowledge of relationships between a given geographic phenomenon and its observable environmental factors is needed for mapping
geographic phenomena/features which cannot be directly observed, for example, soils, habitat potential. The knowledge is often
developed through extensive fieldwork which is not only very labor intensive and slow, but also very costly. Methods are needed to
assist local domain experts to acquire this knowledge efficiently. This paper presents an approach based on fuzzy c-means
classification to assist the development of local domain experts’ knowledge of the relationships. The method is based on the assumption
that the observable environmental factors have dominant impact on the distribution of the given geographic phenomenon and that
unique environmental configurations reflect the unique status or properties of the geographic phenomenon. Under this assumption,
clusters in the environmental space (parameter space) are directly related to different types (status) of the given phenomenon. We
employed a fuzzy c-means classification to identify the natural clusters in the environmental space and use the centroids of these fuzzy
clusters as a guide to allocate field investigation efforts for developing knowledge on relationships between environmental factors and
the phenomenon to be mapped. Through a soil mapping case study, we found the approach is effective in helping local soil scientists to
develop their understanding (knowledge) of soil-environmental relationships in areas the local soil experts are not familiar with the
relationships. The soil map derived using the understandings achieved over 76% accuracy overall, compared to about 60% accuracy
through the extensive fieldwork. For areas with a good environmental gradient, the accuracy is over 90% while the accuracy for area
with little relief is about 63%.
1. Introduction
Predictive mapping, such as soil mapping and habitat potential
mapping, uses observable environmental conditions to predict
the spatial distribution of the geographic phenomenon to be
mapped. Predictive mapping requires the knowledge of the
relationships between the phenomenon and the observable
environmental conditions. This knowledge is often developed
through extensive fieldwork. For example, with conventional
soil mapping, extensive field campaign is required for local soil
scientists to develop the knowledge on how the soils in an
area are related to the observable environmental conditions
(such as elevation, slope gradient). Once the understanding of
relationships has been developed through the campaign, the
local soil scientists will then delineate the spatial distribution of
soils based on the knowledge of the relationships and the
observable environmental conditions.
Field campaigns for natural resource surveys are often very
labor intensive, time consuming, and costly. Although field
scientists may employ statistical sample strategies (such as
random sampling, regular grid sampling, or stratifying
sampling), their field efforts may still not be effectively directed
due to the repeated spatial patterns and the gradation of the
phenomenon/feature. The layout of field investigation is often
dependent on the experience of individual field experts. For
example, some field scientists may be able to avoid transition
areas and distill the major relationships very quickly while
others may spend a lot of time in the transitional areas and
was not able to quickly capture the major relationships. As a
result, the development of knowledge on the relationships is
often very slow and the accumulated knowledge is often very
subjective.
Methods are needed to assist field scientists to quickly identify
the key areas for developing the relationships. This paper
investigates the use of an unsupervised fuzzy classification
technique to identify areas of unique environmental niches
from those areas of environmental transition. Using the results
from the fuzzy classification field investigation efforts are
directed towards areas of unique environmental niches to
discover relationships between these niches and the
geographic phenomena to be mapped.
The next section of this paper describes our method. Section 3
shows a case study using this method. Summaries are given
in Section 4.
2. Methodology
2.1 The Basis:
The underlying assumption behind predictive mapping is that
there is a relationship between the geographic phenomenon to
be mapped and its observable environmental conditions. We
further assume that the unique status of the given geographic
phenomenon (such as different types of soils) is created under
unique combinations of environmental conditions. Under this
assumption, the spatial locations of the geographic
phenomenon with unique status can be approximated by the
locations of unique combination of environmental conditions.
Thus, discovering the relationships between the geographic
phenomenon and its environmental conditions is a matter of
finding which unique combination of environmental conditions
is related to which unique status (types) of the phenomenon.
Relating unique status of a given geographic phenomenon to
unique combination of environmental conditions requires field
investigation. Due to the repetition of spatial patterns and
graduation of geographic phenomenon, locating the unique
combination of environmental conditions is still very
challenging. The efficiency of field investigation can be greatly
improved if we can first identify the spatial locations of these
