LocalAgricultural disaster warning system (LADWS) has a very important influence on the modern agricultural production in local area, because the decrease of influence of agricultural disasters is really one effective method of increasing crop production. In general, GIS and AI are two key techniques for implementing LADWS. However, how to make use of them for building LADWS has been being an important research direction. In this paper, we discuss a technique based on PC ArcInfo and AI for building LADWS, Which is also called flexible reasoning framework. This technique integrates reasonably the data acquisition, the spatial analysis and the decision generation processes in LADWS for warning various possible agricultural disasters so as to increase crop production. This framework is really a flexible interface between PC ARC/INFO and AI, which makes PC ARC/INFO support the AI's reasoning, the knowledge base operation and the visualization computation of LADWS.
1.INTRODUCTION
In the last few years, agricultural disaster warning system
(LADWS) has been being an important agricultural production management
system during agricultural modernization, which provides agricultural
managers with various decision information about agricultural
disasters. For high-output and low-input modern agricultural
production, the information is very important(Keulen, 1986). In
general, LADWS's major tasks are: according to the historical and
current information about agricultural production, to predict various
possible agricultural disasters and to provide management of these
disasters with necessary decision information. Thus, LADWS is also an
important tool for managing agricultural disasters. This management
includes usually: to determine pattern of the disasters, to analyze
their influence and to provide related decision information and method
for decreasing loss of these disasters as much as possible. In the
ten past years, a lot of research has been executed in LADWS,
especially information technology has been widely applied in this
field. These technology includes the mathematical modelling, the
statistical analysis, AI (Rich, 1991), the expert system ( Brownston,
1985), the database system and GIS(Tomlin, 1990). Further, the
development of LADWS can be divided into three phases as the following:
(1) Early LADWS is mainly based on the mathematical modelling, the
statistical analysis and the database system. In LADWS, a lot of
historical and current data of agricultural disasters are stored in
database system, the mathematical modelling and the statistical
techniques are respectively used to build the mathematical model
of agricultural disasters and to predict their occurrence. There are
some obvious disadvantages in the LADWS, including the lack of some
intelligence about the above prediction, the complex architecture, the
very difficulty of implementation and management. And, due to the
limited precision of fixed model for predicting agricultural
disasters, prediction correctness of this LADWS is usually
unsatisfactory.
(2) Then, AI technology is introduced into LADWS, so that both
reasoning and knowledge base methods are used to analyze a number
of the statistical model and data for predicting agricultural
disasters so as to improve the performance of LADWS in very large
extent. However, the supports of the interface and visualization
computation of user in this LADWS are still unsatisfactory.
Especially, both organization and architecture of this LADWS are very
complex, so that its application, management and maintenance are very
difficult. On the other hand, GIS is also introduced into LADWS, but
it is only used as one kind of query and statistical analysis tool and
has little ability to provide the reasoning analysis based on AI with
various possible agricultural disasters.
(3) Recently,one begins to find that the most important character of
LADWS is to manage a number of entities which have geographic or
spatial information feature and interact. Thus, an ideal LADWS must
include GIS and AI at the same time, in which GIS is used as a key
framework for integrating AI and database systems so as to provide a
high effective method of building LADWS with agricultural managers.
Further, some obvious advantages of this method are the more effective
management of spatial data of all entities, the easier implementation
and management of LADWS, the excellent user interface of visualization
computation and the very good decision support for past-disaster
action.
This paper is divided into five parts. In the second, we discuss
LADWS's architecture and show the basic method for designing LADWS.
In the third, PC ARC/INFO function and how to apply PC ARC/INFO in
LADWS are discussed respectively. In the fourth, AI technique and how
to apply it in LADWS are discussed. In the fifth, an effective
technique for integrating PC ARC/INFO and AI is discussed in detail.
Finally, an example of LADWS's prototype is analyzed.
2.LADWS'S ARCHITECTURE
(1) Principle: LADWS's architecture is a logical function design for
implementing reasonably LADWS's function. This design describes
usually the various function modules of LADWS and their logical
relationship, and is suited to the application of layer-division
design principle. The division and definition of function modules are
usually object-oriented or implementation-oriented in order to make
the implementation and adjustment of each module easier and the
logical relationship among all modules clear. In general, this
method of designing the architecture is really the step-by-step
precision of the architecture until there is a clear relationship
between each module and each special object. In our LADWS, some
important, alone-function and high practical value entities are
selected as the objects so as to make the description and
implementation of architecture easier.
(2) Design method:In order to describe reasonably the architecture of
LADWS, we apply a step-by-step precision method, in which PC ARC/INFO
is thought as a logical centre of LADWS. First, total architecture of
LADWS is divided into three main modules that are PC ARC/INFO, AI and
database modules about the information processing. Second, PC
ArcInfo's supports required by AI and database modules are determined
so that a logical relationship between PC ARC/INFO module and these
modules can be built. Finally, an interface or framework supported by
PC ARC/INFO can be also built with the help of this relationship, so
as to make the operation of final LADWS more efficient.
(3) Reasonableness:To make LADWS's architecture reasonable is a very
important work. In general, the major goal of this work is to
simplify and adjust some parts of the architecture in order to make
the architecture more reasonable, which is called the architecture
optimization. In this process, four principles should be followed:
a)Optimization should not reduce the user function of LADWS;
b)Optimization should reduce the complexity of implementation of
LADWS's architecture as much as possible;
c)Optimization should reduce the intensity of interaction among
various functional entities as much as possible;
d)Optimization should make the management and maintenance of LADWS
simpler.
4)Typical disasters: Function to warn the typical agricultural
disasters is described and implemented in the modules of application
entities of LADWS's architecture. The agricultural disasters
considered in our LADWS are the meteorological disasters, the pest
disasters and the weed disasters. In general, the meteorological
disaster is the most important disaster and influences on occurrence
of the other disasters. The meteorological disasters include usually
drought, flood, freezing disasters and other disastrous weather.
Further, the intensity of the pest and weed disasters relates not only
to the meteorological condition but also to the crop and special
geographic environment. Thus, the analysis of meteorological
information about special geographic environment, including the
historical and current information,is used as the fundamental of
prediction of agriculture disasters in our LADWS. Because the basic
period of crop growth is usually one year, we can determine that the
basic period of our LADWS's prediction and decision is usually also
one year. The major method of predicting agricultural disasters is to
recognize the meteorological-geographic-crop pattern of possible
occurrence of the disaster, to analyze and reason on the current
agricultural management data or pattern by existing patterns of the
knowledge base and to predict the various possible agricultural
disasters. On the other hand, various decision patterns about these
disasters are also stored in the knowledge base and are analyzed or
reasoned with the help of the prediction information in order to
provide the suitable decision information with various disasters. All
modules of typical disasters are placed in the application layer.
5) Layer design:Further,LADWS's architecture may be divided into four
layers, i.e., application layer, PC ARC/INFO layer, AI layer and
database layer. The application layer describes the various typical
agricultural disasters and the relevant form of decision pattern, and
analyzes the various pattern or data of meteorological-geographic-crop
of input in order to construct a form of middle data and support the
other module operation of LADWS. Both PC ARC/INFO layer and AI layer
may be combined together and are called basic support layer, and are
discussed respectively as the following:
(1)PC ARC/INFO layer: Its major function are:
a)to support the knowledge base management of AI reasoning;
b)to support AI reasoning operation;
c)to support the visualization computation of LADWS.
In general, all information of agricultural disasters almost has
obvious geographic feature. Thus, the evaluation of influence of the
disaster requires that the various information of geographic feature
must be analyzed, so that PC ARC/INFO layer becomes a basic tool for
implementing the above works.
(2)AI layer: This layer describes mainly the reasoning mechanism based
on the knowledge base. Its major function is discussed as the
following:
a)Manage the various existing patterns in the knowledge base;
b)Implement the pairing process between the various data pattern of
inputandthe existing relevant patterns for optimum prediction by
reasoning on the current input data of agricultural management;
c)Update the contents or patterns of the knowledge base;
d)Select the different reasoning method according to the reasoning
requirement.
In fact, AI layer is one of key modules of LADWS and determines
the main performance of LADWS, but the function of AI layer is
perfectly implemented on condition that the suitable supports of PC
ArcInfo must be provided.
6)Architecture description: a typical LADWS's architecture is designed
by the division-layer method. It consists of the typical application
layer, the PC ARC/INFO layer, the AI layer and the database
layer. Next, we describe general function of this architecture by a
typical application of LADWS. Suppose a typical user pattern WGC1 of
meteorological-geographic-crop which accesses the application
layer. After it is processed in the application layer, the WGC1
accesses the PC ARC/INFO layer. Then, a graphic coverage (WGC1, D1)
is generated in this PC ARC/INFO layer, so that this coverage is
feeded to the AI layer and is analyzed or processed in it. As the
final results of AI's reasoning analysis on the coverage (WGC1,D1),
the AI layer and PC ARC/INFO layer generate together three coverages
(WGC1,D1, R1), (WGC1, D1, R2), and (WGC1,D1,R3), respectively. These
coverages describes respectively the agricultural disasters of
possible occurrence, the influence of these disasters and the
suggested decision for reducing the influence of these disasters. The
reasoning analysis is completed by a process of pattern comparison in
the knowledge base. Further, the knowledge base consists of two sub-
knowledge bases that are the sub-knowledge base of the warning
disaster pattern and the sub-knowledge base of the disaster decision
pattern. Obviously,the coverage (WGC1, D1, R3) is only generated on
condition that the coverages (WGC1,D1,R1) and (WGC1,D1,R2) have been
generated.
3.PC ARC/INFO MODULE
In our LADWS, the function of PC ARC/INFO module is divided into
both logical function and physical function in order to make LADWS'
implementation as simple as possible. The logical function is really
a abstract description of the PC ARC/INFO ability to process spatial
information (Cressie, 1991) and also a framework for supporting the
implementation of PC ARC/INFO physical function. On the other hand,
the physical function describes the function related to the
application of LADWS or the object function. Further, this division
of function of PC ARC/INFO module make the design and implementation
of LADWS easier. Some obvious advantages of this division are given
as the following:
(1)Make the internal design and implementation of the PC ARC/INFO
module easier;
(2)Make the extension of function of PC ARC/INFO module easier;
(3)Make the management of internal entities in the PC ARC/INFO module
more convenience;
(4)Make the implementation of the above flexible framework for
supporting LADWS's application and AI reasoning easier.
In our current design of PC ARC/INFO module, we define the logical
function of PC ARC/INFO module as the function of spatial information
processing, which can be discussed further as the following:
(1)function to filter the data information: This function can be used
to filter the input data information of PC ARC/INFO feature so as to
delete those unnecessary information and to obtain the required
information, with the PC ARC/INFO ability to analyze the information
of geographic and spatial feature. This function is usually
implemented with the help of the function of complex graphic
processing of PC ARC/INFO. For example, the information of input WGC
pattern may include some useless information for AI reasoning, so that
this information must be filtered out.
(2)Added-value function: This function is used to add the application
or use value of input information in order to increase the efficiency
of information processing. In general, the process of added-value
information consists of the process of information input, the process
of searching the related information and the process of revising the
information. For example, this added-value function in our LADWS can
be used to update the knowledge base. When the information for
updating the knowledge base is given, the added-value function is used
to find the sub-knowledge bases related to this revision of
information with the help of PC ARC/INFO ability to process the
spatial information. Then, all information of revision which the
different sub-knowledge bases require is generated respectively and is
used to update these sub-knowledge bases.
(3)Information synthesis: This function of information synthesis
describes the PC ARC/INFO ability to synthesize the geographic or
spatial information and is used to synthesize these related
information before the different information is processed. Then, the
synthesis information is again processed in order to increase the
efficiency of information processing. For example, this function in
our LADWS is used not only to update the knowledge base and but also
to synthesize all related output information for warning agricultural
disasters into one graphic coverage.
(4)Visualization computation: In fact, the PC ARC/INFO itself is an
excellent tool for visualization computation. This function in our
LADWS is used to describe the visualization computation process of
LADWS by a series of graphic coverages, so that the module and final
results of computation of LADWS have more obvious physical meaning and
make the result of LADWS application better.
The physical function of PC ARC/INFO represents mainly the
application feature of PC ARC/INFO layer and is usually implemented by
the suitable combination of the logical function of PC ARC/INFO and
the related data. Some typical physical function is described as
following:
(1)historical data processing of agricultural disaster prediction:
First, this processing function implements the filtering-processing
and added-value processing of input historical data information.
Then, the processed information is used to update the warning disaster
patterns in the knowledge base. The updated contents includes both
the prediction and the decision patterns of the knowledge base.
(2)AI data processing: This processing provides mainly the suitable
form of information with the reasoning analysis of the current input
information of LADWS. Further, the operation of this processing
includes the filtering information, added-value information and
information synthesis.
(3)Other data processing: This processing includes the data management
in the knowledge base and the database of PC ARC/INFO. The relevant
operation are still the filtering information, the added-value
information and the information synthesis.
(4)Visualization data processing: This processing includes the
organization of various visualization graphic coverages of the
possible disasters and the disaster influence. The relevant operation
includes still the filtering information, the added-value information
and the information synthesis in various visualization graphic
coverages.
4.AI MODULE
There are three issues relevant to the design of AI module, which
are discussed as the following:
(1)Understanding the patterns of agricultural disasters: This
understanding makes really use of the historical information about
the occurrence of past agricultural disasters to build the various
suitable patterns of prediction for predict the possible occurrence of
future agricultural disasters. One of important problems in
understanding the patterns of agricultural disasters is how to
represent the information of agricultural disasters. Further, this
problem is also how to represent or store the current/historical
information and the discovered patterns of agricultural disasters in
computer.
(2)Knowledge acquisition: This issue includes the learning or
knowledge acquisition. Further, this issue is really how to discover
the patterns from the stored information trace(Sleeman, 1992).
(3)Inference or knowledge use: According to the input information of
agricultural disasters and the captured patterns, the AI module must
provide the information of future possible occurrence of agricultural
disasters with the users of LADWS.
In order to introduce the learning and reasoning ability into our
LADWS(Pearl, 1988), we apply some method, including as the following:
(1)construct global views: These views are really a set of global
views. A global view is a virtual object class defined from all
agricultural disasters or relevant decisions via logical rules. These
global views can be implemented and serve as windows though which
LADWS applications can access entities of LADWS. A logical rule in
our LADWS has the generic form: IF X THEN Y, where X is its body part
and Y is its head part. A body or head part has one, or more than
one, formula which can represent the status of a disaster object or an
action to update an object's status. All logical rules are implemented
in AI module.
(2)Equip the LADWS with adaptive learning and reasoning ability: The
agricultural disaster patterns for supporting this ability are learned
from a historical database which contains a chronological measurement
and statistic analysis trace(Ullman, 1988). These discovered patterns,
represented in the form of logical rule, describes the relationship
between the objects of agricultural disasters.
(3)Inference implementation: Based on these disaster patterns and
prespecified domain knowledge of agricultural disasters, forward and
backward inference can be triggered to access global view, predict
possible disasters, fire the relevant decision, and analyze the
relevant influence. Unlike a general expert system with only
prespecified domain knowledge, AI module of our LADWS has learning
ability to augment its knowledge related to the specific agricultural
disasters and the relevant decision.
Learning and reasoning ability is used as a key framework of AI
module function of LADWS. In general, learning goal is to discover
the patterns of agricultural disaster, and reasoning goal is to
predict the possible occurrence of agricultural disasters and to
search the decision regarding agricultural disasters though reasoning
on the discovered patterns of agricultural disasters and the
prespecified domain knowledge. The AI database for supporting the
reasoning on agricultural disaster contains usually the pattern of
agricultural disasters, the abstract description and definition of
these disasters and the prespecified domain knowledge regarding the
agricultural disaster and relevant decision, which all are represented
by a set of logical rules.
To make the operation of AI module more efficiency, we apply a
reasoning system based on logical rules for the prediction of the
agricultural disasters. Some obvious advantages of the application of
logical rules are discussed as the following:
(1)Knowledge for solving the problem of the prediction of agricultural
disasters varies as the time;
(2)Pattern of the prediction of agricultural disasters also varies as
the time;
(3)Requirement of supporting the user interface based on the PC
ArcInfo module;
(4)Requirement of pairing the patterns in complex situation;
(5)Requirement of natural representation of the solving method and
pattern based on IF-THEN relationship.
In fact, the updating knowledge for solving the problem of LADWS
must be continuously created because the new service and decision
about the agricultural disasters is being introduced. Further, there
are the different patterns of prediction of agricultural disasters in
the different agricultural system, which varies as the time. Thus,
the application of IF-THEN rule here is reasonable.
In our LADWS, there are a number of logical rules to be used for
reasoning on the patterns. Three typical logical rules are described
as the following:
(1) This rule is written in the form of Horn clause, which are
statements of the form: if A1, A2, ... , and An are true, the B is
true. Following the prolog syntax, it is written as:
B:-A1, A2, ... , An
where the formulas B and Ai are predicates with a list of argument,
e.g. P(X1,..., Xk).
(2)This rule is written in Horn clause with certainty factors, which
may be written as:
Confidence Factor=P% B-A1, A2, ..., An
This expression shows that, if A1, A2, ..., and An true, B is
concluded to be true with probability P. A formula Ai or B is a
condition that represents the status of an object of agriculture
disaster.
(3)This rule is written in form of a production rule. It is written
in the form of syntax as:
(A1, A2, ... An - B1, B2, ..., Bm)
which represents that, if A1, A2, ..., and An are true, then B1, B2,
..., and Bn will be executed. Here, Ai is a condition and Bj is
action about the condition. These rules can be used to invoke the
control decision by forward inference. They can also emulate backward
inference to predict the agricultural disasters.
5.INTEGRATION OF PC ARC/INFO AND AI
One of keys to implement LADWS is the integration of PC ARC/INFO
and AI modules in the architecture of LADWS. This process of
integration can be represented as:
(1)PC ARC/INFO module is how to support the operation of AI module;
(2)AI module is how to interact the agricultural system though PC
ArcInfo module.
In general, the operation of AI module is based on the logical
rules. However, the historical information of measurement must be
processed only to form the logical rules for reasoning on the
agricultural disasters in AI module. Further, the techniques of
statistical analysis, e.g. regression analysis, are usually used to
construct these logical rules. The Implementation of this analysis
and the creation of logical rules require a lot of data computation
relevant to regression analysis. In general, the regression analysis
is mainly used to analyze the relationship between the various
meteorological factor, geographic factors and crop factors and the
agricultural disasters, which is usually implemented by PC ARC/INFO
module. This regression analysis based on PC ARC/INFO, which is also
called the spatial regression analysis due to the spatial operation of
PC ARC/INFO, can be used to describe the result of regression analysis
of some graphic information though a graphic coverage. Otherwise,
some special function, e.g. the filtering information, the added-
value information and the information synthesis, cab be also used in
the regression analysis. Thus, the regression analysis based on PC
ArcInfo spatial operation is one important advantage of our LADWS.
Obviously, PC ARC/INFO module becomes not only an interaction
interface between LADWS and its user, but also an interface for
supporting the operation or reasoning of AI module. Because the PC
ArcInfo module is really an access interface of external user and its
character is also affected by the operation of AI module, there is a
mechanism based on the PC ARC/INFO and AI modules for supporting this
function, which is called flexible reasoning framework of LADWS, whose
major characters are discussed as following:
(1)Integrate effectively the data acquisition, the spatial analysis
and the process of decision support of LADWS;
(2)Make learning function of AI module support many different
application of LADWS;
(3)The data acquisition and spatial analysis based on geographic-
oriented feature simplify the process of creation and recognition
of the patterns of agricultural disasters;
(4)Reasoning based on geographic-oriented feature simplifies the
operation of AI module;
(5)object-oriented techniques are easily used.
However, how to implement the above flexible reasoning framework
is a key to integrate PC ARC/INFO and AI techniques. Here, an
effective technique, called data acquisition-spatial analysis-decision
support technique(DSPT), is used to implement this goal. Further,
each query to LADWS is defined as an action of LADWS, the output of
LADWS to the action is defined as a response of LADWS. Thus, we think
that LADWS consist of a number of pairs of action-response. Clearly,
for any action, there is a certainty response to it. In the other
word, there is a relationship between the action and the response,
called A-R relationship, and all A-R relationships consist of the
solving space of LADWS. After the ability of spatial analysis of PC
ArcInfo is introduced, the spatial operation of PC ARC/INFO generates
really a division for this solving space, and all A-R relationships of
this space are divided into several geographic-feature classes, which
is also called relationship class. Further, the introduction of
relationship class simplifies really the organization of A-R
relationships. In fact, a number of the A-R relationships are used to
build the knowledge base based on the disaster patterns and the
prespecified domain knowledge, and the programming method of
geographic-oriented feature can be used to implement effectively the
integration of PC ARC/INFO and AI techniques.
6.SOME EXPERIMENTAL RESULTS
In the current time, we are developing a prototype of LADWS for
warning the meteorological disasters, the pest disasters and weed
disasters. The experience obtained from this prototype development
will be used to guide the development of practical LADWS. The
development environment of PC ARC/INFO selected for this prototype is
Esri's PC ARC/INFO, the development environment of AI is a programming
expert system and a machine learning system, and the various
historical data of the past 40 years are used to construct the
patterns of prediction of agricultural disasters. All the above
environments are used to construct the prototype of local LADWS. The
precision of this prototype for warning the agricultural disasters is
very satisfactory. This LADWS has successfully predicted some typical
agricultural disasters above 82% confidence degree, and provides a
number of decision information for reducing the influence of these
disasters.
Our selected PC ARC/INFO has powerful function, and the
performance of AI system consisted of both the programing expert
system and the machine learning system is very satisfactory. Thus, we
use their Marco or programming languages to implement the above LADWS.
Further, we find that the performance of testing this prototype is
also satisfactory. Especially, the introduction of PC ARC/INFO
provides the geographic-oriented feature with a lot of information so
that the construction and implementation of LADWS are very simple. On
the other hand, the application of future probability reasoning will
improve largely the performance of LADWS. Form the results of current
experiment on the prototype of LADWS, the application results of LADWS
not only are very satisfactory but also have very obvious economic
gain. For example, a typical warning result and possible decision are
shown in a geographic coverage.
7.CONCLUSION
The reasonable integration of PC ARC/INFO and AI techniques for
building LADWS is a very effective and practical method. However, the
key which influences the precision of LADWS is mainly AI system
design, including various patterns of warning agricultural disasters.
But, the agricultural system itself is very complex, and a number of
problems require to be solved in future. We will continuously perfect
the prototype of our LADWS in order to increasing the precision of the
prediction of agricultural disasters.
8.ACKNOWLEDGE
Authors would thank Agricultural Information System Research
Center and Institute of Agricultural Modernization of JiangSu Academy
of Agricultural Sciences for their help and warm research environment.
9.REFERENCES
Keulen, H. andWolf, J. 1986, Modelling of Agricultural Production:
Weather, Soils and Crops, Pudoe Wageningen.
Rich,E.and Knight,K.1991, Artificial Intelligence,
Second Edition,New York: McGraw-Hill.
Brownston, L., Farrell, R., Kant, E. and Martin, N. 1985, Programming
Expert System in OPS5: An Introduction to Rule-Based Programming,
New York: Addison-Wesley.
Tomlin, D. 1990, Geographic Information Systems and Cartographic
Modelling, Prentice Hall.
Cressie, N. 1991, Statistics for Spatial Data, John Wiley & Sons.
Sleeman, D. and Edwards, P. 1992, Machine Learning: Proceedings of
the Ninth International Workshop, New York:Morgan Kaufmann.
Pearl, J. 1988, Probabilistic Reasoning in Intelligent Systems:
Networksof Plausible Inference, NewYork: Morgan Kaufmann.
Ullman, J. D. 1988, Principles of Database and Knowledge Base Systems,
Vol. I, Computer Science Press.
Sun Ling
A. Reseacher, JiangSu Academy of Agricultural Sciences
JiangSu Academy of Agricultural Sciences
Nanjing, JiangSu, 210014, P. R. China
Telephone:86-025-4431481-325
Fax:86-025-4432691