In this paper, we briefly overview the basic characteristics of design of agricultural disaster evaluation system (ADES) and highlight its importance for modern agricultural production. We further propose a new approach to simple design and fast implementation of ADES. The proposed approach differs from earlier approaches in three major ways: it is based on a layered architecture model to support fast design and implementation of ADES; it can reasonably integrate some advanced information processing techniques such as GIS, simulation and expert system in a simpler and powerful framework; and it uses a commercial GIS package as a tool for the integration. The conceptual approach presented in this paper offers a step forward to designing and implementing a more advanced ADES for providing a number of powerful decision supports for senior managers of agricultural production.
In this paper, we present this new method of building ADES model or ADES. This method includes three key information processing techniques such as geographic information system (GIS), simulation and expert system (ES) techniques. Really, they are also the most important techniques for implementing excellent performance ADES. The ADES based on this method not only makes use of advantages of these techniques as more as possible, but also has much better performance and more users than existent ADES.
Disaster Evaluation: Analytical techniques are important tools for disaster evaluation. However, the analytical techniques often require unrealistic assumptions and trend to be mathematically untraceable as the structure of the disaster measure becomes complex. Furthermore, it is very difficult to estimate disaster statistics, which are required as inputs to the analytical model. On the other hand, discrete-event simulation (DES) (Law and Kelton, 1991) is a viable alternative to analytical techniques. A major advantage of DES over any analytical technique is that a disaster evaluation system can be modeled with much less stringent assumptions, and more complex disaster level measures can be handled with relatively simple case. However, DES usually suffers from significant computation burden. In our ADES, this two techniques have been widely used for disaster evaluation.
Decision Making: This task requires decision parameter optimization, and can be accomplished by numeric analysis method, which utilizes random measurements over a finite period of time to estimate the finite difference quotient of the performance measure with respect to decision variables.
Elements of agricultural system can fail for any number of reasons, including architecture defects, design defects, and inadequate maintenance procedures. Intrusions can come from acts of earthquake, flood, hurricane, accidents. Survivability discussion of agricultural system involves analyzing availability of agricultural system, computing quality of agricultural system and evaluating disaster- based survivability. Most of the research on survivability of agricultural system has concentrated on simple analysis mechanisms related to single simple agricultural system. However, in order to implement the survivability evaluation of very large agricultural systems, it is necessary to be able to build an effective model for this evaluation. The objectives related to the evaluation are typically set out in the form of general policies of evaluation into a number of more specific policies of evaluation to form a policy hierarchy in which each policy represents its plans to meet its objectives. The integrity of agricultural system is defined as: the ability of an agricultural system provided to deliver high quality, continuous service while gracefully absorbing, with litter or no customer impact, failures of or intrusions into the hardware or software of elements of agricultural system.
Traditionally, agricultural production system aim at satisfying some specified performance objectives under normal conditions without explicit consideration of agricultural production system survivability. Performance under agricultural disaster can be unpredictable for agricultural production system designed with these methods. A major benefit of setting survivability performance objectives will be to ensure that, under given disaster scenarios, agricultural production system performance will not degrade below predetermined levels. Thus, such a set of performance objectives should be translated to realize the design and management goals of agricultural production system.
How to find survivability function is an important work to implement agricultural disaster evaluation. The general procedure for finding survivability function of agricultural disaster is as the followings:
(1)Define agricultural production system survivability S;
(2)Specify disaster type to be studied;
(3)Define "normal operation" of agricultural production system, which is related to S;
(4)List the sample points {e}, or all combinations of events that may happen under the disaster type being considered;
(5)Determine the survivability Se;
(6)Determine or assign probability Pe of each event e;
(7)Calculate survivability function, P[S=s]=Sum Pe.
Before calculating the survivability function, one should first specify the type of disaster and the definition of the "normal operation" of agricultural production system. This is important since different disaster types may have different effects on agricultural production. Moreover, we may also obtain different results depending on the feature of the agricultural production for which we are calculating survivability. The next step is to list all possible combinations of events that could happen under the given disaster type. However, this may be a very complex work because listing all the sample points may be difficult under giving a more general agricultural production system and a different definition for "normal operation". In addition, the sample space may be too large, and we may want to eliminate sample points which are very unlikely to happen. Further, the work to calculate the survivability measure for each sample point will depend on our definition of survivability. If the definition is the number of fields connected to a particular agricultural disaster, then we would need an efficient method that can determine whether there a relationship to the disaster in the fields. For each sample point, we can assign a probability measure representing the likelihood of its occurrence. The assignment of probabilities to sample points should be based on historical observations or experience. If the disaster under consideration happens so rarely that historical observations are not available, then one will need to use subjective judgment when assigning probabilities. In this case, a study of the sensitivity of results to variations in probability assignment is also necessary in order to establish the confidence level of obtained results.
Inability (I) is defined in terms of a unit of agricultural production. For an agricultural production system (APS), the most common function of the system is the ability to provide agricultural product. The unit of production is a basic planting area or a special agricultural production zone. In this instance, inability is the percentage of units that meet disaster. For a grain production system, the unit of production is a unit area on which a certain grain yield can be obtained. The inability is defined as the percentage of areas that meet disaster.
Duration (D) define the time during which the inability condition exists in a agricultural production system. It is measured by determing the beginning and ending points of a disaster, based on the inability being above a given threshold.
Geographic area(G) includes the context of geographic area condition in which the inability exceeds a given threshold.
Crop(C) includes the context of crop condition in which the inability exceeds a given threshold.
Water(R) includes the context of water condition in which the inability exceeds a given threshold.
Meteorological factor(M) includes the context of meteorological condition in which the inability exceeds a given threshold.
Weight (W) includes the context of all other factors in which the inability exceeds a given threshold.
The triple (I, D, G, C, R, M, W) is the basis for measuring and quantifying agricultural disasters and their impact on agricultural production. Depending on the value of I, D, G, C, R, M, and W, agricultural disasters may be classified as catastrophic, major, and minor. APS Survivability can be analyzed through its performance models and measures. There are two basic approaches to survivability analysis defined here. The first approach uses probability of disasters and, possibly rates of restorable normal production to calculate various probabilistic measure of APS ability or inability. The second is a conditional approach, defining measures of a APS after given disaster events have occurred. This approach may either use probabilistic weight of the resulting states of the APS and resulting APS restorable production after the disaster or use deterministic analysis of these states. Both approaches can be used to evaluate different restorable and preventive methods depending on different applications in our system. There are many ways to describe APS survivability and define "measurements".
(1)ARCHITECTURE MODEL: One of the most important tasks to design ADES is carefully to determine its various function parts and to build the logical relationship between these parts, so that these parts and relationships are used to construct ADES architecture model. However, in practical applications, ADES involves decision-making based on complex interactions between people, pests, lands, crops, natural disasters, and other natural resources relevant to agricultural disasters. Modeling these interactions and representing them in ADES for supporting decision have presented difficulties in its design and implementation when traditional design methods were used. Thus, during ADES design, a layered architecture model that includes both seven layers and relevant functions is used to guide the ADES design and implementation. Further, the ADES described in this paper utilizes a layered model of architecture which involves the seven different layers.
The model is in practice a widely accepted structuring technique. The functions of ADES are partitioned into a vertical set of layers. Each layer performs a related subset of the functions required to exchange information with another similar system which has the architecture. On the other hand, the ability to exchange information can flexibly support the large size of distributed ADES design and implementation on a network environment. A layer relies on the next lower layer to perform more primitive functions and to conceal details of those functions. It provides services to the next higher layer. Ideally, the layers should be defined so that changes in one layer do not require changes in the other layers. Thus, we have decomposed one complex problem about how to construct ADES into a number of more manageable small problems. The task of our research team was to define a set of layers and the services performed by each layer in order to make the design and implementation of ADES become simpler and easier. Further, the partitioning should group functions logically, should have enough layers to make each layer manageably small, but should not have so many layers that the processing overhead imposed by the collection of layer is burdensome. Thus, we define carefully all functions of ADES according to the practical application requirements of ADES. Further, we discuss briefly each of the layer and the architecture model implementation as follows.
(2)ARCHITECTURE IMPLEMENTATION: The environmental disaster layer (bottom or first layer) provides the mechanism for managing the data about various environmental disasters, which include mainly a number of data related to natural disasters such as flood, land, pollution and other natural ones about geographic environment. The second layer (weather disaster layer) attempts to make use of the services from the first layer and provides one means to mange the data about various weather disasters. The basic service of the biological disaster layer (the third layer) is to provide the management data for biological disasters. The purpose of layer 4 (decision data layer) is to provide a mechanism to generate various data to support the decision and evaluation about agricultural disasters. The layer complexity depends on the type of service it can get from layer 3. The decision layer provides a mechanism for making various decisions to evaluate agricultural disaster with the help of a number of decision and evaluation models. The evaluation layer is concerned with the integrated evaluation decisions for a number of special groups of agricultural disasters. Its purpose is mainly to define various standard survivability-oriented evaluation decisions. Finally, the application layer (top layer or layer 7) which is relevant to ADES application provides a means for various users or application processes to access ADES. This layer contains management functions about the applications and some useful mechanisms to support local and remote applications. According to the description of various layers of the architecture model, ADES architecture model implementation can be divided into three subsystems. The first is basic evaluation subsystem, which encompasses the model's layers 1, 2 and 3. The subsystem is primarily intended to provide the "raw" evaluation service of agricultural disasters which is directly used by an end user and does not support further decisions based on raw evaluation. Two main components such as GIS support environment and Data base management system support its major operations.
The second is advanced evaluation subsystem, which is used to perform the analysis of basic evaluation data from the first subsystem and to recommend the best basic evaluation decisions or strategies for disaster evaluation application subsystem. The decision and evaluation models, expert system (Davis and Clark, 1989), and simulation models are used to support its major operation. The simulation models (Law and Kelton, 1991) generate a complete data for each basic evaluation decision of the expert system. In general, the expert system is capable of integrating the knowledge of several disciplines about ADES evaluation into a single knowledge base system to support decisions about the evaluation (Coulson et al., 1987). The properly developed expert system (Edwards, 1991) is a powerful tool for providing managers or users of ADES with the day-to-day decision support need to evaluate various agricultural disasters and to make the relevant decisions. The evaluation models in the subsystem are some object-oriented programs (Booch, 1991) designed to analyze alternatives over both numeric and non-numeric evaluation criteria. According to various different application requirement, these programs form the final recommendation for the best disaster evaluation decision with the user's preferences and perceptions about the set of evaluation decisions. The third is disaster evaluation application subsystem shown in Fig. 4, which provides a means for application processes or users to access ADES. It is composed of three classes of models: local application models to handle various services for local users, remote application models to provide various services for remote users, and maintenance models, which provide for the testing of ADES components and assist in fault isolation and identification, for testing system functions.
(3)GIS INTEGRATION: The above subsystems can be regarded as collections of tools or methods that serve a special role in ADES. Thus, we can use a real world model based on GIS (Esri, 1989, 1990) as the joining tool in ADES integration. However, our research shown that Esri ARCVIEW (Esri, 1994a) provides a very satisfactory GIS framework for the integration. The object-oriented model of ADES based on the above architecture has been implemented in ARCVIEW. The O-O design does not allow subsystems to communicate directly (Folse et al., 1990). All communication is governed by the real world model or ARCVIEW kernel. The advantage of this approach was that ADES would not be committed to or built around any specific type of subsystem, making ADES compatible with more application environments (Henderson and Edwards, 1990). In practical ADES implementation, the Microsoft's VISUAL BASIC 4.00 and VISUAL C++ 2.00 as well as Esri's AVENUE (Esri, 1994b) were used to implement the object-oriented model in ARCVIEW environment. Especially, a number of scripts were developed by AVENUE as a tool to complete complex integration, which make the design and implementation of some complex user interfaces become very simple. This integration process is easily completed in ArcView. A typical example of ADES application is to evaluate the effect of disaster weather on stored grain, so as to apply suitable means to improve dangerous status of stored grain to safety one. Our experiences shown that the practical application of ADES reduces largely stored grain loss due to some disaster weather, which is equal to increase the grain yield in field. Further, the evaluation results from ADES are used to make or design various effective control strategies to maintain a very good ambient environment of a number of warehouses of stored grain distributed over a very large geographic area with the help of the remote models of ADES. For arriving at the objective, ADES uses a number of weather data to evaluate various possible influence of current disaster weather on stored grain and to provide satisfactory evaluation services for a number of managers of stored grain.
(a)Disaster survivability provides an effective method for analyzing effect of agricultural disasters on agricultural production system.
(b)Disaster survivability evaluation function (I, D, G, C, R, M, W) provides an effective method for evaluating the effect of agricultural disaster on agricultural production system.
(c)The above layered architecture model is an effective method for implementing ADES.
(d)GIS is the best platform for implementing ADES.
These findings suggest that for ADES, disaster evaluation framework can potentially be developed with further studies. A possible better approach would be to allow application of some powerful analysis tools such as application of PETRI NET (Reisig, 1985) in the framework. However, this requires analysis of the more complex disaster evaluation framework to support ADES development.
Coulson, R. N., Folse, L. J. And Loh, D. K. (1987). Artificial intelligence and natural resource management. Science, 237: 262-267.
Davis, D. L. and Clark, J. L. (1989). A selective bibliography of expert systems in natural resource management. AI Appl. Nat. Resour. Manage., 3(3): 1-18.
Edwards, J. S. (1991). Building Knowledge-Based System-Towards a Methodology. Pitman Publishing, London, 248 p.
Esri. (1989). ArcInfo Vol. 1: Users guide. Environmental systems Research Institute Ltd., Redlands, CA, 377 p.
Esri. (1990). AML users guide. Environmental Systems Research Institute Ltd., Redlands, CA, 510 p.
Esri. (1994a). Introducing ARCVIEW. Environmental Systems Research Institute Ltd., Redlands, CA, 98 p.
Esri. (1994b). Introducing Avenue. Environmental Systems Research Institute Ltd., Redlands, CA, 120 p.
Folse, L. J., Mueller, H. E. and Whittaker, A. D. (1990). Object-oriented simulation and geographic information systems. Artif. Intell. Appl. Natur. Resour. Manage., 4(2): 41-47.
Henderson, S. B. and Edwards, J. M. (1990). The object-oriented systems life cycle. Commun. Assoc. Comput. Mach., 33: 142-159.
Law, A. M. and Kelton, W. D. (1991). Simulation Modeling and Analysis. McGraw-Hill, Inc., New York.
Liang, T. and Khan, M. A. (1986). A natural resource information system for agriculture. Agric. Syst., 21: 81-105.
Reisig, W. (1985). PETRI NETS- An Introduction. Springer-Verlay, Berlin, Germany, 161p.