sed as
) less
(perts
based
alised
TES
ystem
, for
ise of
'S the
cally
ide a
1 and
ware
ed to
jased
blem
using
ledge
antic
and
from
e. By
ledge
well
, The
ence
Y CIN
iseful
2).
osion
states
Department of Agriculture Soil Conservation Service(USDA-
SCS). The system, which is based on the EXSYS shell, uses a
knowledge base of farm equipment and soil characteristics to
assist farmers in selecting the best combination of farm
equipment and conservation planting technology to minimize
soil erosion(Morrison et al, 1989).
Domain independent expert systems which contain only search
control and general inference procedures are referred to as
expert shells. Expert shells can be used to create expert systems
for specific domains of applications. Depending on the
knowledge representation method used expert shells may be
characterized as rule based or frame based. Based on the search
control or inference strategy used expert shells may also be
characterized as forward chaining, backward chaining, or
forward and backward chaining(Barr and Feigenbaum, 1981,
1982).
2. THE SLEMS CONCEPT.
SLEMS is essentially a rule based and semantic network based
expert system. Knowledge is therefore captured and stored in
the SLEMS knowledge base in the form of either a semantic
network of facts or rules. Fuzzy or vague information is stored
directly in the form of fuzzy expressions which can be queried
by a fuzzy query processor as explained later.
The capture and representation of domain knowledge is,
generally, an iterative two step process. First a knowledge
engineer supported by the domain expert analyses and abstracts
the structure of the domain knowledge. The knowledge engineer
then stores the abstracted knowledge using appropriate
knowledge representation structures such as frames, semantic
networks or rules. After knowledge acquisition the domain
expert performs validation and verification tests on the system.
The two processes are repeated as necessary to refine the
representation of the domain knowledge(Ebrahimi, 1987 and
Green and Keyes, 1987).
The SLEMS uses both the rule based representation and
semantic network structures to capture domain knowledge. The
rules are designed as simple English-like "IF ... THEN ..."
rules. These are compiled interactively by the SLEMS rule
editor in the form of premises and conclusions. However in one
of the SLEMS sub-system, LEARN, real world objects and
facts are represented by a semantic network of object triplets.
The SLEMS knowledge base is queried by a layer of
specialized inference operators which exploit the hierarchical
structure of the semantic network to infer facts not directly
stored in the knowledge base. Alternatively a backward
chaining strategy is used to search for knowledge base objects
satisfying the premises of a rule. When a rule is satisfied the
rule conclusion is asserted as a new fact and stored into the
knowledge base.
HAS
USLE | L——————————be| PARAMETERS
The design of the SLEMS enables the capture and storage of
three kinds of knowledge: knowledge directly needed for the
solution of soil erosion problems, auxiliary knowledge required
for the extraction of soil erosion information from external
Sources, such as, aerial photointerpretation and remote sensing,
and knowledge about the organization and use of the SLEMS
knowledge base.
The soil erosion domain knowledge stored in the SLEMS is
based on Wischmeier's empirical formulae called the Universal
Soil Loss Equation(USLE). It includes the six parameters, S, L,
K,C, R, P, which represent the erosive effects of terrain
morphology, soil characteristics, vegetation cover, hydrological
and meteorological factors, and conservation
practice(Wischmeier, 1984; Meyer, 1984). The USLE is widely
used by soil conservation and agricultural experts as a design
tool for soil erosion control.
Several variations of the USLE exist, each designed to handle
different geographic, topographic, hydrologic, soil and ground
cover conditions. The knowledge required by a soil erosion
domain expert to select specific USLE models and compute
values for the model parameters can be analyzed and organised
by semantic network representation. For example Figure 1
shows a semantic network representation of the USLE concept.
In this scheme the top of the hierarchy contains a general USLE
model characterized by a default parameter set (PARAMETERS)
whose instances are the six USLE parameters K,S, C, R, L, P.
Specific USLE models, such as for example the USLE_1 which
might represent the USLE model for Eastern Canada, appear
lower in the semantic network hierarchy. Specific models are
characterized by parameter sets containing both specific and
default parameter descriptions. Each parameter instance is
characterized by a value (VAL.). The semantic network
organization thus enables missing parameters to be inherited
from models defined higher up in the hierarchy.
Soil erosion is a complex problem. The study and the solution
of soil erosion problems requires complex multi-variable
information and a multi-discplinary approach to information
management and analysis. Certain aspects of the problem such
as data capture and processing, information extraction, and
knowledge acquisition and extraction are amenable to solution
by computerized knowledge intensive methods. Figure 2 is a
functional model of the soil loss estimation and modelling
problem. Thick arrows in the figure indicate the forward flow of
information, from the data capture stage to the knowledge
extraction and application stage. Thin arrows indicate the
interplay between the various components of the problem.
IS_A
USLE À USLE_n
HAS HAS
PARMETERS | … | PARMETERS | HAS HAS HAS HAS HAS HAS
Figure 1: A Semantic Network Representation of USLE Concepts
(Source: E.G. Mtalo, 1990, page 65).
