Data Capture
_ Info Requirements
Data Processing
Information Extraction
Y
Knowledge Base
[rman }
Y
S
oil Properties
Ground Cover
Hydrologic info
Meteorologic info
Topography
Erosion State
Human Expert
Rules |
| Expert System |
Field & Laboratory EEE
GIS & DBMS )
Computer
IAS & Topo Maps al
qe Parameters
Conservation Practice
Soil Loss Estimates
ifm
Crop Estimates Facts
(ER J reet |
Knowledge Utilization ul — — —]
Knowledge Extraction
Figure 2: A Conceptual Model of the Soil Loss Estimation and Modelling Process(After
E.G. Mtalo, 1990, pp.23).
3. SYSTEM DESIGN AND COMPONENTS OF SLEMS.
SLEMS is an experimental knowledge based system prototype
designed and developed for application in soil loss estimation
and modelling as part of an Msc. thesis research at the UNB
CanLab-INSPIRE in 1990 (Mtalo, 1990). At the current level of
development the SLEMS can only manipulate attributive data
and information. SLEMS consists of the following sub-
systems(Figure 3):
- The EXPERT,
The LEARN,
The FUZZ,
The Data Base Management System.
The first three sub-systems perform knowledge based
functions. The EXPERT and the LEARN sub-systems are
based on algorithms and source code published by Schildt
(1987), modified and augmented for SLEMS implementation
(Mtalo, 1990). The FUZZ sub-system was developed to
process fuzzy knowledge. The Data Base Management System
based on the CDATA (Stevens, 1987) performs conventional
database management functions for the system. The inference
layer consists of operators which exploit the semantic links
between knowledge base objects to facilitate extraction of facts
not directly stored in the knowledge base. The low level
operators are basic search operators.
All the system modules were written in the C programming
language and implemented on a SUN 4 Work Station. Since
then a demonstration version of the SLEMS EXPERT was
implemented for DOS based PC. Currently a new DOS version
based on object oriented programming is being developed using
BORLAND C++. In the new version the CDATA interface will
be replaced by a new one based on the Paradox Engine. As
mentioned in the introduction the primary objective is to produce
a system which is affordable by developing countries.
200
3.1 The Rule Based EXPERT
The EXPERT is a rule based expert shell whose main function
is to capture procedural and classification knowledge in the form
of English-like "IF ... THEN ..." rules. Given certain facts the
EXPERT uses a backward chaining search strategy to locate a
rule satisfied by the given facts. If no rule is directly satisfied by
the given facts the inference mechanism tries to find if there is a
rule whose conclusion would provide the additional facts
needed to satisfy another rule. Newly deduced facts are added
into the knowledge base and used in subsequent inference.
At a preliminary stage the knowledge required to solve specific
tasks must first be analyzed and organised into a decision tree
where branches indicate alternate solution paths or related
components of the knowledge. There are three objectives to the
preliminary analysis. The first and most important objective is to
enable the domain knowledge to be broken down into smaller
manageable chunks without losing the semantic relationships
inherent within the body of knowledge. Secondly the resulting
structure(Figures 4) facilitates easy search of the resulting
knowledge base. Thirdly the organization of the domain
knowledge into a semantic network structure enables the direct
translation of the structure into "IF... THEN..." rules and
facilitates checks against errors in the generation of rules.
The structure of the knowledge required to calculate soil loss by
the USLE model is shown in Figure 4. Based on this the first
stage in the estimation of soil loss involves the selection of a
specific model. Using the decision tree model USLE models
may be classified by geographic area (e.g. Eastern Canada), by
inventor (e.g. Wischmeier or Holy) or by method, such as,
MUSLE (modified USLE), SLEMSA (modified soil estimation
model for Africa), Revised Slope Factor, or Varying K (varying
soil erodibility factor) as shown in Figure 4.
If, for example, the modified USLE (MUSLE) is selected then