@prefix vivo: . @prefix edm: . @prefix ns0: . @prefix dcterms: . @prefix skos: . vivo:departmentOrSchool "Business, Sauder School of"@en ; edm:dataProvider "DSpace"@en ; ns0:degreeCampus "UBCV"@en ; dcterms:creator "Johnston, Michael Macfarlane"@en ; dcterms:issued "2010-08-11T16:07:48Z"@en, "1989"@en ; vivo:relatedDegree "Master of Science - MSc"@en ; ns0:degreeGrantor "University of British Columbia"@en ; dcterms:description """This thesis reports the results of a project to develop an expert system to predict the ecological effects of prescribed fire. The central goal of this project was to examine the suitability of the prescribed burning domain for the application of expert systems technology and to recommend a strategy for further research in this area. With the sponsorship of the Protection Branch of the BC Forest Service, several experts were contacted in order to ask for their participation in the Fire Effects Expert System project. Five different experts were consulted and several different conceptual models of the domain were developed. The limiting factor concept was used to model the effects of prescribed fire on the direct and indirect growth factors. A prototype of the Fire Effects system was eventually constructed using the VP-EXPERT software package. Many difficulties in using expert systems technology in the prescribed burning domain were identified. These difficulties include: conflicting viewpoints amongst the different experts, uncertainty of knowledge concerning fire effects, and lack of upper management commitment to the project. However, many favourable factors were identified and these include: the scarcity of domain experts, the qualitative nature of the decisions, the large solution space of the problems, and the existence of a large amount of narrow domain-specific knowledge. Several uses for the Fire Effects system were proposed. These uses include: aiding foresters in developing fire prescriptions, educating forest managers, and serving as a basis for group discussions concerning fire effects. A strategy for further research in the prescribed burning domain was also proposed. The principal features of this strategy are to plan and coordinate the development of expert systems at a high management level, to draw extensively on academic work in the area, and to develop specific standards pertaining to all site information data that is collected by the Forest service."""@en ; edm:aggregatedCHO "https://circle.library.ubc.ca/rest/handle/2429/27262?expand=metadata"@en ; skos:note "AN EXPERT SYSTEM TO PREDICT T H E ECOLOGICAL EF F E C T S OF PRESCRIBED FIRE by MICHAEL M A C F A R L A N E JOHNSTON B.Sc, Carleton University, 1979 A THESIS SUBMITTED IN PARTIAL F U L F I L M E N T OF T H E REQUIREMENTS FOR T H E D E G R E E OF MASTER OF SCIENCE in T H E F A C U L T Y OF GRADUATE STUDIES Commerce and Business Administration We accept this thesis as conforming to the required standard T H E UNIVERSITY OF BRITISH COLUMBIA June 1989 © Michael Macfarlane Johnston, 1989 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of The University of British Columbia Vancouver, Canada DE-6 (2/88) ABSTRACT This thesis reports the results of a project to develop an expert system to predict the ecological effects of prescribed fire. The central goal of this project was to examine the suitability of the prescribed burning domain for the application of expert systems technology and to recommend a strategy for further research in this area. With the sponsorship of the Protection Branch of the BC Forest Service, several experts were contacted in order to ask for their participation in the Fire Effects Expert System project. Five different experts were consulted and several different conceptual models of the domain were developed. The limiting factor concept was used to model the effects of prescribed fire on the direct and indirect growth factors. A prototype of the Fire Effects system was eventually constructed using the VP-EXPERT software package. Many difficulties in using expert systems technology in the prescribed burning domain were identified. These difficulties include: conflicting viewpoints amongst the different experts, uncertainty of knowledge concerning fire effects, and lack of upper management commitment to the project. However, many favourable factors were identified and these include: the scarcity of domain experts, the qualitative nature of the decisions, the large solution space of the problems, and the existence of a large amount of narrow domain-specific ii knowledge. Several uses for the Fire Effects system were proposed. These uses include: aiding foresters in developing fire prescriptions, educating forest managers, and serving as a basis for group discussions concerning fire effects. A strategy for further research in the prescribed burning domain was also proposed. The principal features of this strategy are to plan and coordinate the development of expert systems at a high management level, to draw extensively on academic work in the area, and to develop specific standards pertaining to all site information data that is collected by the Forest service. iii TABLE OF CONTENTS ABSTRACT ii T A B L E OF CONTENTS iv LIST OF FIGURES vi A C K N O W L E D G E M E N T . . viii CHAPTER 1. INTRODUCTION . . . 1 1.1 Background 1 1.2 Prescribed Burning 2 1.3 Expert Systems 4 1.4 Prescribed Fire Research in BC 5 1.5 Fire Effects Expert System Project 7 1.6 Structure of the Paper . . . 8 CHAPTER 2. EXPERT SYSTEMS 10 2.1 Overview 10 2.2 Expert System Tasks 13 2.3 Expert System Lifecycle . . . 16 2.4 Selecting and Motivating Experts 17 2.5 Knowledge Engineering 18 2.6 Knowledge 21 2.7 Knowledge Representation 23 2.8 Knowledge Acquisition 26 2.9 Knowledge Validation . 32 2.10 Knowledge Engineering with Multiple Experts 33 2.11 Hypertext 36 2.12 Integration of Expert Systems into the Environment . . . . . . 37 2.13 Expert Systems in the Forestry Domain 39 CHAPTER 3. PRESCRIBED FIRE AND FIRE E F F E C T S 41 3.1 Introduction . . . 41 3.2 Prescribed Fire and Fire Effects Research in B.C 41 3.3 Effects of Fire on Soil . . 48 3.3.1 Physical Changes . . . . . . . . 48 3.3.2 Chemical Changes 50 3.3.3 Biological Changes 51 3.4 Fire and Ecosystem Processes •-. . . . . . . 52 3.5 Fire Severity and Fire Behaviour 54 3.6 Vegetation and Prescribed Fire 57 3.7 Prescribed Burning Procedures in BC 60 3.8 Prescribed Burning Decision Models 61 3.9 The Ecosystem Classification System in B.C 62 3.10 Soil Development and the Forest Floor 64 iv 3.11 Limiting Factor Concept 66 3.12 Prescribed Fire Decision Aids 68 CHAPTER 4. BUILDING T H E EXPERT SYSTEM 71 4.1 Knowledge Engineering 71 4.1.1 Knowledge Engineering Sessions with E . Hamilton . . . 75 4.1.2 Knowledge Engineering Sessions with M . Curran . . . . 77 4.1.3 Knowledge Engineering Sessions with B. Hawkes . . . 83 4.1.4 Knowledge Engineering Sessions with other Experts . . 84 4.2 Prescribed Burning Procedures 85 4.3 Conceptual Model of Fire Effects . . 91 4.4 Motivations of Groups Involved in Prescribed Burning 93 4.5 Structure of the Expert System 95 4.6 Typical Interaction with the System 97 4.7 Comparison of Different Expert System Software 99 CHAPTER 5. DISCUSSION AND CONCLUSIONS 103 5.1 Discussion of Knowledge Engineering 103 5.2 Discussion of Development and Production Environments . . . 108 5.3 Discussion of the Prototype System I l l 5.4 Discussion of Expert Systems in Prescribed Burning 114 5.5 Strategy for Further Expert Systems Research . 119 BIBLIOGRAPHY 123 APPENDLX 1 - PROJECT PROPOSAL 129 APPENDLX 2 - PRE-HARVEST SILVICULTURAL PRESCRIPTION . . 137 APPENDLX 3 - SITE PREPARATION GUIDE 140 APPENDLX 4 - BURNING P L A N 147 APPENDLX 5 - PRESCRIBED BURN ANALYSIS 151 APPENDLX 6 - VEGETATION CONCEPTUAL M O D E L 154 APPENDLX 7 - GROWTH FACTORS 164 APPENDLX 8 - FIRE BEHAVIOUR CONCEPTUAL M O D E L 167 APPENDLX 9 - TYPICAL INTERACTION WITH SYSTEM . . . . . . . . 172 v LIST OF FIGURES FIGURE 1 Expert System Task Types 15 FIGURE 2 The Expert's Transition Curve 19 FIGURE 3 Stages of Knowledge Acquisition 30 FIGURE 4 Criteria for Selecting a Knowledge Engineering Technique . 31 FIGURE 5 A Validation-Based Life Cycle Model of the Knowledge-Based System Development Process 34 FIGURE 6 Flow of Research Needs from the M O F to Forestry Researchers 42 FIGURE 7a Proposed Prescribed Fire Management System . . . . . . . . 45 FIGURE 7b Proposed Prescribed Fire Management System 46 FIGURE 7c Proposed Prescribed Fire Management System 47 FIGURE 8 Factors Determining the Impact of Prescribed Fire on Soils 53 FIGURE 9 Example Demonstrating Limiting Factor Concept 67 FIGURE 10 Key to the Identification of Site Sensitivity to Fire Classes 70 FIGURE 11a Products of Knowledge Engineering Sessions 73 FIGURE l i b Products of Knowledge Engineering Sessions 75 FIGURE 12 Proposed Implementation of Limiting Factor Concept . . . . 81 FIGURE 13 Prescribed Burning Decision Process Model 87 FIGURE 14 Conceptual Model of Prescribed Burning . . . . . . . . . . . . 92 FIGURE 15 Structure of the Fire Effects Expert System 96 FIGURE 16 Comparison of Four Expert System Development Tools . . 101 vi FIGURE 17 Spectrum of Techniques for Analyzing Interaction Between Fire and Site Factors vii ACKNOWLEDGEMENT The author would like to thank several different agencies and individuals for their support of this project. The Protection Branch of the BC Forest Service provided financial support for travelling and other expenses incurred during the course of this project. The Forest Economic and Policy Analysis Project provided financial support and office space for the project. Mike Curran and Evelyn Hamilton of the BC Forest Service gave much of their time and energy to this project. My friend, Chris, encouraged me and proofread draft versions of this thesis. To all of these parties, the author extends his sincere gratitude. viii CHAPTER 1. INTRODUCTION 1.1 Background This thesis is one of the products of a research project involving the BC Forest Service, the Canadian Forest Service (CFS), the Forest Economics and Policy Analysis (FEPA) project, and the Faculty of Commerce and Business Administration at the University of British Columbia. This project involved investigating the feasibility of applying expert systems technology to prescribed burning decisions by developing an expert system to predict the ecological effects of slashburning. Initial contact between the BC Forest Service and The Faculty of Commerce and Business Administration was established through FEPA. The specific project was first suggested following a June meeting between representatives from the Faculty of Commerce and Business Administration, the BC Forest Service, and the Canadian Forest Service (CFS). Since the CFS had already started development work on an expert system for prescribed burn planning, it was suggested any other project should complement the CFS project. Therefore, the Fire Effects Expert System Project was proposed with the goal of developing the Fire Effects module of the CFS-sponsored PB (Prescribed Burn) Planner expert system. However, the Fire Effects Expert System Project was sponsored by the Protection Branch of the BC Forest Service. Appendix 1 contains the proposal for this project. INTRODUCTION / 2 1.2 Prescribed Burning Prescribed burning may be defined as \"the knowledgeable application of fire to a specific land area to accomplish designated land objectives. Implicit in this definition is the studied consideration of the probability and benefit of achieving the objectives, the potential for damage to other resources and the opportunity costs and benefits of alternate land treatments. Fire is not the only land management tool. It is however, an economical and natural tool which offers more potential for manipulation in response to specific site requirements than most other post-logging treatments.\" [Muraro, 1977, page 26] In BC, one of the major uses for prescribed burning is as a site preparation tool prior to replanting of the site. Prescribed burning can improve the planting accessibility, increase the number of planting spots, sanitize the site from insects and diseases, and increase the availability of the nutrients on the site. In 1985, over 45,000 ha of land were burned after timber harvesting [Hawkes and Lawson, 1986]. Cost savings of up to $200/ha can be achieved by using slashburning instead of other site preparation methods [D. Gilbert1, pers. comm.]. 1 Protection Branch, BC Forest Service. INTRODUCTION / 3 Prescribed burning has other uses in addition to its use as a post-logging site treatment. Its most common use throughout the world is to clear land prior to planting crops [Chandler et al., 1983]. Other uses include: the management of grasslands and ranges, the management of wildlife, the management of forest fuels (hazard control), and the maintenance of ecosystems. In 1985, 92,000 ha of land in BC were burned for range and wildlife habitat improvement [Hawkes and Lawson, 1986]. An additional 1,500 ha were burned for hazard abatement. Although fire can be a very useful tool for site preparation, it must be used carefully. Many factors must be considered before fire is used on a site. The impact of a fire on a site varies with the slash loading, the moisture content of the soil and fuels, the weather conditions at the time of burning, the site topography, and the method and pattern in which the fire is ignited. The forest manager can manipulate the impact of fire on a site by defining a fire prescription that states a certain time of year for burning, as well as weather conditions and ignition patterns to burn under. Of course, economic and safety concerns play a crucial role in this decision because the forest manager should only burn when there is a minimal chance of the fire escaping. The effects of a particular fire on the ecology of a site cannot be predicted with much certainty. Fire can affect the site chemistry; the soil structure, the soil temperature as well as the different biochemical processes of the ecosystem. Because of the uncertain knowledge concerning the nature of how fire impacts INTRODUCTION / 4 a site, many efforts have been made to document and evaluate prescribed burns. The goal of these efforts is to gradually improve the prescribed burning decision models that are in use. 1.3 Expert Systems Professor Edward Feigenbaum of Stanford, one of the leading researchers in expert systems, has defined an expert system as \"an intelligent computer program that uses knowledge and inference procedures to solve problems that are difficult enough to require human expertise for their solution. Knowledge necessary to perform at such a level, plus the inference procedures used can be thought of as a model of the expertise of the best practitioners of the field.\" [Harmon and King, 1985, page 5] Expert systems derive much of their power from having the ability to model knowledge about a specific part of a problem domain. However, it is very difficult to incorporate what is generally referred to as \"commonsense\" knowledge into an expert system. Therefore, expert systems technology is only suitable for problems that require extensive knowledge within a narrow domain. Of course, it is also possible to represent knowledge about a narrow domain in a book or manual. However, it is difficult to represent knowledge about strategies that can be used to solve a problem. This is because of the problem of combinatorial explosion of the solution space of a problem as it increases in INTRODUCTION / 5 complexity. Expert systems technology is better able to handle complex problems because it can easily represent all the different types of knowledge that are commonly used to solve a task. The current technology also enables the system to explain the reasoning it used to solve a task in much the same way as a human expert does it. Expert systems have become more popular with the advent of powerful microcomputers and computer languages and programs that allow for symbolic programming. The advent of expert systems have also led to the creation of a new= type of profession. The people who develop expert systems are now commonly referred to as knowledge engineers [Beerel, 1987], Many different expert systems are currently in use in industry [McDermott, 1982; Herrod and Smith, 1986]. 1.4 Prescribed Fire Research in BC There is extensive research presently occurring in BC relating to prescribed burning. The main categories of research are: 1. Operational monitoring of prescribed burns and the development of models to understand fire behaviour and the interaction of fire and site factors. 2. Analyzing old sites that have been burned many years previously and the development of models to predict the long term effects of fire on site productivity. INTRODUCTION / 6 3. Developing techniques to document prescribed burns for use in a prescribed burn computerized database. 4. Developing improved decision aids to be used by the field staff. The final category of research is becoming more and more important as the need to transfer knowledge from the researchers to the people in the field becomes more recognized. One of the original decision aids is the Prescribed Fire Predictor/Planner (PFP) [Muraro, 1977]. The PFP takes as input the desired impact (slash reduction or duff 2 reduction) and produces as output the range of moisture codes that will result in that impact. This set of moisture codes forms a prescription. The user can run this prescription through a historical weather database program to obtain probabilities of the likelihood of a specific prescription occurring at a certain time of year. Decision aids have also been developed to enable the site to be classified in terms of its sensitivity to fire. These decision aids usually take the form of one-page3 keys that use readily observable site characteristics to determine a site's Also r e f e r r e d to as the forest f l o o r . 3 Small flowcharts that take as input several s i t e factors i n order to a r r i v e at a decision, (see Figure 10) INTRODUCTION / 7 sensitivity to fire. The Vancouver Forest Region's Site Sensitivity to Fire key is an example of this type of approach [Klinka et al., 1984]. Recently, the Canadian Forest Service (CFS) has commenced development work on the PB Planner expert system. This expert system is being designed primarily by Bernie Todd of the CFS's Petawawa research centre. The goal of this project is to build a system that will help with all the different prescribed burning decisions. This includes the translation of silvicultural objectives into burn objectives, the scheduling of cutblocks for burning, and the modelling of fire behaviour as it relates to ignition patterns and topography. 1.5 Fire Effects Expert System Project The Fire Effects Expert System project was formally proposed in July of 1988 after a June meeting involving representatives of the CFS, the Protection Branch of the BC Forest Service as well as Michael Johnston and Professor Yair Wand of the University of British Columbia. The general goal of this project was to determine the applicability of expert systems technology to decisions within the prescribed burning field. A more specific goal was to work on a part of the overall PB Planner Expert System and to develop a prototype expert system to predict the ecological effects of slashburning. The objectives were: 1. To study different development environments with the goal of recommending a particular environment for further systems development. INTRODUCTION / 8 2. To examine the integration of expert systems applications into the operating environment. 3. To recommend a strategy for future research into expert systems. The activities of the project included meeting with different experts as well as meeting with potential users. Considerable effort was also devoted to developing a prototype system using an expert system shell. The outcomes of this project include a prototype system to predict the ecological effects of slashburning. This paper also defines a strategy for further research in expert systems in the prescribed burning domain. 1.6 Structure of the Paper In Chapter Two of this paper, the many issues involved in developing expert systems are described. Included in this chapter are sections describing the development lifecycle of an expert system, the process of knowledge engineering, types of knowledge representation, techniques for knowledge acquisition and issues involving knowledge engineering with multiple experts. INTRODUCTION / 9 Chapter Three of this paper describes prescribed fire and how it interacts with the ecology of a site. Specific sections include discussions of the effect of fire on soil as well as its effect on vegetation. The procedures involved in conducting a prescribed burn are also given. Chapter Four describes the process that was involved in building the expert system. Included in this chapter is a section describing the conceptual structure of the domain as well as a section comparing different expert systems software. The final chapter of this paper discusses various aspects of this project including the viability of expert systems technology in this domain. A strategy for further expert systems research is also proposed. CHAPTER 2. EXPERT SYSTEMS 2.1 Overview Expert systems(ES) is an application of what is generally referred to as the field of artificial intelligence. Branches of this field include reasoning, natural language understanding, knowledge representation, and vision. Expert systems are also commonly referred to as knowledge-based systems [Walters and Nielsen, 1988]. This is because expert systems maintain their knowledge in the form of a separate knowledge base. There are three basic components to an expert system: the knowledge base, the inference engine and the user interface [McGraw and Harbison-Briggs, 1989]. The knowledge base contains knowledge in the form of rules or some other type of representation formalism. The inference engine consists of rules that control how the knowledge in the knowledge base is used. The user interface controls the communication between the system and the user. ES are different from conventional systems in that they can be used for problems that have no easy algorithmic solution [Beerel, 1987]. They simply try to model the reasoning of a human expert on a particular problem. Because of this approach to problem solving, they are not expected to give a correct solution for all situations/problems. 10 EXPERT SYSTEMS / 11 The concept of expertise is crucial to the expert systems field. Turban (1988) defines expertise as facts and theories about the problem area, hard and fast rules and procedures regarding the problem area, rules (heuristics) of what to do in a given problem situation (i.e. rules regarding problem solving), global strategies for solving these types of problems, and meta-knowledge (knowledge about knowledge). The many functions that experts provide is an important consideration for researchers in the expert systems field and expert system designers have to be aware of these different functions. Experts do more than simply just provide answers to problems. Experts perform a range of tasks including recognizing and formulating the problem, explaining the solution, learning from experience, and determining inconsistencies in logic. Expert systems have only been able to model some of these functions. Despite the inability of expert systems to perform all the functions of an expert there are many advantages to using this type of technology [Hart, 1986]. These advantages include availability, consistency and comprehensiveness. Availability refers to the fact that experts are busy people and there is considerable expense involved in training new experts. Expert systems technology has the potential to increase the availability of expertise. This is because a computerized system can be made available at all times to many different users. EXPERT SYSTEMS / 12 Consistency is another advantage of expert systems technology. Experts do not always reason the same way when presented with a similar problem. The current emotional state of the expert can also have adverse affects on the reasoning process of an expert. Comprehensiveness refers to the ability for expert systems technology to encapsulate knowledge from many different sources. Therefore, the user of an expert system can have immediate access to all the knowledge that is necessary to solve a problem. This can be done by linking the system to external databases and systems or by consulting different knowledge bases. The main disadvantages of expert systems are summarized by Hart (1986) as follows: choice of domain, acceptability, uncertainty, updating and testing. Choice of domain refers to the unsuitability of many domains for expert systems technology. For many domains, the problems are too complex and the experts disagree quite vehemently. Acceptability refers to whether the users in the domain will want to use a computer to give them advice on problems. The problem of uncertainty refers to the fact that most of the data that is handled by experts is vague and uncertain. EXPERT SYSTEMS / 13 Updating refers to the fact that knowledge is changing quite frequently in many different domains. Therefore, the knowledge base of an expert system has to be updated very frequently which can present many difficulties. Testing an expert system can be difficult to do as well. There are many different solution paths and it is usually not possible to test each of these paths individually. Expert systems are most useful for problems with a large number of possible combinations, and problems that involve interpreting a large amount of signal data, and problems where there are time constraints on the decision and there is a large amount of information to interpret. In order to construct an expert system, it is first necessary to j)btain knowledge from an expert. This process is called knowledge acquisition (KA). There are many methods to do this as is discussed later in this chapter. 2.2 Expert System Tasks Not all problems are suitable for expert systems technology. The knowledge domain and the task type will affect the suitability of the problem for expert systems technology, the potential success of the system, and the knowledge acquisition techniques that could be used effectively. Kidd (1987) classifies EXPERT SYSTEMS / 14 knowledge into four different domains based on how formalized is the language for reasoning and whether there is any clear, coherent theory underlying the domain. She proposes that expert systems development is much more problematic in the domains where there is no such theory. The first domain class comprises domains where humans have developed a strong formal reasoning language and where there is a clear, coherent, underlying theory. This class include mathematics, geometry and the programming languages. The second class of domains includes chemistry and medicine. There is no formal reasoning language for these domains but there are some underlying theories. The third class includes applications software, management and marketing. These domains lack formal reasoning and representation languages as well as a clear, coherent, underlying theory. The fourth domain is spatial reasoning tasks where experts have developed a formal and powerful language for reasoning but are not able to support this language on a machine. Much work is presently being done on matching expert system techniques to tasks [Chandrasekaran, 1984]. Therefore, it is important to understand the nature of the different types of tasks. Boose (1988) provides a fairly comprehensive list of different types of tasks (see Figure 1). These task types are also classified into two broad categories: analysis and synthesis. Analysis tasks are ones where there are a limited number of solutions to the problem and the task is to map one of the pre-defined solutions to the problem at hand. EXPERT SYSTEMS / 15 ANALYSIS TASXS C l a s s i f i c a t i o n - c a t e g o r i z i n g based on ob s e r v a b l e s Debugging - p r e s c r i b i n g remedies f a r m a l f u n c t i o n s D i a g n o s i s - i n f e r r i n g s y s t e i n a l f u n c t i a n s from o b s e r v a b l e s I n t e r p r e t a t i o n - i n f e r r i n g s i t u a t i o n d e s c r i p t i o n s from sensor data SYNTHESIS TASKS C o n f i g u r a t i o n - c o n f i g u r i n g c o l l e c t i o n s of o b j e c t s under c o n s t r a i n t s i n r e l a t i v e l y s m a l l search spaces Design - c o n f i g u r i n g c o l l e c t i o n s of o b j e c t s under c o n s t r a i n t s i n r e l a t i v e l y l a r g e search spaces P l a n n i n g - d e s i g n i n g a c t i o n s Scheduling -planning v i t h s t r o n g time and/or space c o n s t r a i n t s TASXS COHBINIKG ANALYSIS AND SYNTHESIS Comnand and c o n t r o l - o r d e r i n g and governing o v e r a l l system c o n t r o l I n s t r u c t i o n - d i a g n o s i n g , debugging, and r e p a i r i n g student behaviour M o n i t o r i n g -couparing o b s e r v a t i o n s to expected outcomes P r e d i c t i o n - i n f e r r i n g l i k e l y conseguences of g i v e n s i t u a t i o n s Repair -executing plans to a d i i n i s t e r p r e s c r i b e d remedies __zz-FIGURE 1 Expert System Task Types (adapted from Boose, 1988) Synthesis tasks involve constructing the solution from the subproblem solutions. 2.3 Expert System Lifecycle EXPERT SYSTEMS / 16 There is much confusion presently occurring over what should be the standard expert system lifecycle. Much of this controversy concerns whether rapid prototyping is the best approach to use for building expert systems or whether a more formal methodology is preferable. The term \"prototyping\" refers to a program development methodology where the system is built in an iterative fashion. The features of this methodology include user involvement, learning between iterations and evolving requirements as experience is gained [Mathieson, 1988]. This technique contrasts with other, more structured system development methodologies such as the traditional Systems Development Life Cycle where there is no iterative component to the methodology. Parsaye and Chignell (1988) describe a typical lifecycle for ES as: 1. feasibility analysis 2. conceptual design 3. knowledge acquisition 4. knowledge representation 5. knowledge validation 6. technology transfer and maintenance Important features to note about this lifecycle are the distinction between a EXPERT SYSTEMS / 17 conceptual design phase and a knowledge acquisition phase and the inclusion of a knowledge validation phase. 2.4 Selecting and Motivating Experts When building an expert system, it is important to select the right experts to use as sources of knowledge for the system. There are many characteristics that make some experts more desirable than others. The designer should also be aware of the phenomenon usually termed \"paradox of expertise\" whereby an expert may not be able to explain in detail the reasoning he/she used to solve a problem [Beerel, 1987]. The first desirable characteristic that an expert should possess is the ability to articulate his/her own knowledge. The expert must have time to devote to the knowledge acquisition process and the expert must be committed to the goals of the project. The credibility of the expert is also an issue because potential users may want to know that he/she is getting advice from a system based on the knowledge of a respected expert. An additional concern is that the experts may feel threatened by the \"deskilling\" process of the new technology [Beerel, 1987]. For some experts, relinquishing control of knowledge may mean loss of power. Therefore, the concept of deskilling is very important to the motivation of the expert. This is a term that describes the mechanization or automation of a task that is EXPERT SYSTEMS / 18 normally done by a human. A person who holds a special skill is likely to unreasonably defend against this deskilling process. Beerel (1988) has studied the motivation levels of experts during the course of development of an expert system. She describes the motivation level of an expert in terms of a transition curve (see Figure 2). It is claimed that the motivation and enthusiasm for the project will change quite dramatically through the course of the project as is illustrated by the curve in Figure 2. The shape of the curve is affected by the way in which the expert was introduced to the technology, the support of top management, the confidence and abilities of the expert, the skills of the knowledge engineer, the understanding of the objectives set, and the manner in which these are to be achieved. By carefully considering these various factors, it should be possible to maintain a moderately high motivation level in the expert throughout the course of the project. 2.5 Knowledge Engineering Knowledge engineering can be defined as the \"process of synthesizing knowledge into a computer system so that the problems are electronically solved through symbolic manipulation and reasoning of the knowledge base\" [Beerel, 1987, page 127]. The person who obtains the knowledge is referred to as the knowledge engineer. Some people do not like using this term because it seems EXPERT SYSTEMS / 19 EXPERT SYSTEMS / 20 to imply that knowledge engineering is a science when it is in reality closer to an art. In fact, some authors give this occupation a different name such as \"knowledge crafter\" [Walters and Nielsen, 1988]. There is also some controversy as to whether the expert should be building the system himself/herself and whether it is really necessary and beneficial to employ the services of an intermediary (i.e. knowledge engineer). If a knowledge engineer is employed then there also is a debate as to whether that person should be from the expert's domain or not. In addition to knowledge acquisition, the knowledge engineer is usually responsible for many different tasks. In particular, the K E is responsible for: 1. The overall management of the project. 2. The identification of the project. This usually involves reading existing documentation on the problem, doing extensive background reading, and locating the experts. Many researchers have attempted to document what are the most important characteristics for a knowledge engineer to have [Davies and Hakiel, 1988; Beerel, 1987; Hart, 1986]. These characteristics include good communication skills, proficiency in using expert system software, persistence, intelligence and domain knowledge. Good communication skills and domain knowledge are particularly important because the K E must have the ability to enter into the EXPERT SYSTEMS / 21 \"expert's way of thinking\" [Beerel, 1987]. This means that the knowledge engineer must be able to grasp new concepts very easily. There are many phases to the knowledge engineering process. The central phase is the knowledge acquisition phase. However, the validation phase is becoming more and more important as more systems are being implemented and the need for system maintenance is recognized. These phases are described in more detail in subsequent sections of this chapter. 2.6 Knowledge \"Knowledge is not synonymous with information. Rather knowledge is information that has been interpreted, categorized, applied and revised.\" [McGraw and Harbison-Briggs, 1989, page 13]. Knowledge is what gives expert systems their power. This section discusses the concept of knowledge is and leads into the next section which describes different methods to represent knowledge. It is related to expertise in that expertise is a demonstration of the application of knowledge [Mcgraw and Harbison-Briggs, 1989]. Many authors have proposed many different categories of knowledge. Hayes-Roth (1984) classifies knowledge across three dimensions: scope, purpose, and validity. EXPERT SYSTEMS / 22 The scope dimension ranges from general, common statements to specific, focused statements. The purpose dimension ranges from descriptive (factual) to prescriptive (procedural) statements. The validity dimension ranges from 100% certain to 100% uncertain. Presently, the knowledge bases of most expert systems contain mainly specific, descriptive knowledge. It is difficult to incorporate general and prescriptive knowledge in an expert system [Wolfgram et al., 1987]. Examples of knowledge that are difficult to incorporate into an expert system include general problem-solving knowledge and metaknowledge (knowledge about knowledge). This limits the types of problems that expert systems technology can be applied to. It is important to understand that there are many different components of._ knowledge in order to be able represent knowledge effectively. Parsaye and Chignell (1988) propose five components of knowledge. These components are: naming, describing, organizing, relating and constraining. Naming involves selecting a unique name for an object so there is no confusion concerning which object is being referred to. Describing involves noting the important properties about an object. Organizing involves putting objects into conceptual categories such as classes or hierarchies. Once we have described and organized objects, we need to describe the relationship between EXPERT SYSTEMS / 23 them. Part of the skill in describing a relationship is in choosing the right level of analysis and deciding whether to include particular entities in a relationship. Constraints that govern the properties of objects should also be defined. 2.7 Knowledge Representation Many approaches to representing knowledge have been suggested [Newell and Simon, 1972; Minsky, 1975; Quillian, 1968; Stefik and Bobrow, 1986]. Some of these knowledge representation techniques have evolved from studies of the way humans appear to store information. Since an expert system is attempting to model the reasoning of the human expert, a knowledge representation method must allow for knowledge structures that are similar to those of the human mind. A comprehensive list of knowledge representation techniques would include: 1. Semantic nets: Some of the earliest proposed types of knowledge structures are referred to as semantic nets. These nets specify how objects are related and also allow for the inheritance of properties. Nodes and links are the fundamental units used to represent knowledge [Parsaye and Chignell, 1988]. 2. Production rules: These are examples of simple types of EXPERT SYSTEMS / 24 knowledge representation. This knowledge structure allows knowledge to be stored in simple IF - T H E N rule formats. The structure makes it easy to encode heuristics (rules of thumb) into an expert system. 3. Logic: Knowledge can also be expressed in the form of logic. However, this representation method is cumbersome and is not commonly used. However, there are many languages that are suited for this type of representation such as PROLOG and LISP. 4. Frames, scripts: More recently proposed forms of knowledge representation structures are frames and scripts. These structures are basically mechanisms to package knowledge. They allow for properties to be inherited. The concept of inheritance makes it possible to specify properties that apply to a whole class of entities. Research has also shown that people store information in their brain in some form of package structure. Scripts permit reasoning based on expectations about what should happen next in stereotyped situations [Parsaye and Chignell, 1988]. 5. Objects: The object-oriented methods of knowledge representation share a number of features with frames and EXPERT SYSTEMS / 25 semantic networks. Knowledge is viewed as a set of objects, each of which is capable of exhibiting certain behaviours. Actions can be taken by invoking a method. A method defines how an object is allowed to behave in response to a message from another object. 6. Blackboard: Blackboard-based representation is a less common form of knowledge representation. There are three major components to the blackboard representation method [Walters and Nielsen, 1988] : the knowledge sources (expertise), the blackboard (knowledge storage and representation), and the control (problem-solving strategy). The advantage of this method is that knowledge can be stored in a modular way. Different knowledge representation methods can also be employed. 7. Model: The model-based representation is best used when it is desirable to represent a system by more than a simple list of facts and rules. The model provides a complete specification of the system being examined [Walters and Nielsen, 1988]. Reasoning is done within a complete model context by matching the current situation to the model. In essence, the reasoning is done from first principles [Turban, 1988]. The potential exists for \"transportability\" of the EXPERT SYSTEMS / 26 knowledge base. For example, a system for diagnosing electronics problems could potentially be used for a wider class of electronic devices rather than just one specific device. Parsaye and Chignell (1988) give a checklist of what to look for in a knowledge representation method. The quality of the basic knowledge structures, storage mechanisms, retrieval mechanisms, and representation environment are all important factors to be considered. 2.8 Knowledge Acquisition Knowledge acquisition (KA) can be defined as the \"process of obtaining the public and private knowledge used by an expert skillful in solving problems in a constrained and restricted domain.\" [Walters and Nielsen, 1988, page 5]. It is this process that is usually termed the \"bottleneck\" in the expert system development process. It is also the critical part of the development process because the success or failure of the system will be determined by the results of the K A stage. Because of the importance of the knowledge acquisition process, it has been discussed extensively in the literature in the past several years [Kidd, 1987; Hart, 1986; McGraw and Harbison-Briggs, 1989]. The knowledge acquisition phase is complicated by several factors. First of all, it may be difficult to find a suitable expert in the domain who has enough time to devote to the development of an expert system. In most cases, the main EXPERT SYSTEMS / 27 reason for building the system is to increase the availability of expertise in the domain. However, the slow and lengthy K A phase will actually serve to decrease the availability of the expert in the short term. A second complicating factor concerning the knowledge acquisition phase is what was referred to earlier in this chapter as the \"paradox of expertise\". This phenomenon occurs when an expert is so proficient in reasoning through a problem that he is not able to explain the intermediate steps of reasoning in such a level of detail that is required to implement it on a machine. There are three general categories of approaches to knowledge acquisition. These approaches are interviewing experts, learning by being told, and learning by observation [Parsaye, 1988]. The first category, termed \"interviewing experts\" is the classic knowledge acquisition technique. The knowledge engineer interviews the experts and proceeds to elicit concepts and knowledge about the domain. This interview can be very structured or unstructured. An unstructured interview might be used to elicit terms and concepts in an early stage of knowledge acquisition whereas a structured interview could be used in the later stages to elicit facts and problem-solving strategies. Hoffman (1987) describes different types of structured interviews (i.e. method of \"familiar\" tasks, constrained processing tasks, limited information tasks, etc.) and discusses how they can increase the effectiveness of the K A process. EXPERT SYSTEMS / 28 The second approach, learning by being told [McGraw and Harbison-Briggs, 1989], essentially involves the expert being able to communicate his knowledge through some sort of K A tool. This could involve using an automated KA tool or participating in tasks such as card-sorting or scale development in order to define the knowledge and expertise in a domain. In this approach, the expert is responsible for expressing and refining most of his/her own knowledge. The repertory grid is a much researched technique which has proved to be very useful for classification-type problems. Essentially, this technique involves rating examples according to a series of different constructs. Algorithms are then available to translate these ratings into rules that can be incorporated into an expert system. The third approach, learning by observation [Parsaye, 1988] can also involve automated or non-automated KA methods. Induction, otherwise known as machine learning, has been studied quite extensively. This process involves the expert presenting detailed examples of problems as well as their solutions. Algorithms are then available which can determine a set of rules that could be used to obtain the same solutions. The more examples that are provided, the more comprehensive the set of rules that is produced. This technique has proved to be useful in situations where the paradox of expertise does not allow the expert to communicate his/her reasoning effectively. Protocol analysis is another technique that fits into this category. This technique involves observing the expert while he/she is doing the task. Usually, the expert is asked to EXPERT SYSTEMS / 29 provide a verbal transcript of his/her thought processes as he/she solves the problem. Various authors have identified different stages of knowledge acquisition. McGraw and Harbison-Briggs describe five stages (see Figure 3): identification, conceptualization, formalization, implementation, and testing. Commonly, there is an iterative aspect to these stages such that many of them will probably be repeated several times during the development of the system. Sometimes, this development process is called rapid prototyping. There have also been some attempts to define some more formal methodologies. KADS [Breuker and Wielinga, 1987] is an attempt to define a more formal methodology. They define three stages in their methodology: orientation, problem identification, and problem analysis. For each of these stages they define the purpose, the type of knowledge, and the K A techniques that should be used. Various models for the K A process have been proposed [Dhaliwal and Benbasat, 1988]. An extensive set of criteria for selecting a knowledge engineering technique is illustrated in Figure 4. There are many different factors which will affect the results of the K A process. These factors include the knowledge acquisition tool, the characteristics of the expert, the characteristics of the knowledge engineer, the stage of EXPERT SYSTEMS / 30 * Identify Problei Characteristics Identification * Identify Concepts i Retina Requicenents Conceptualization i Refine Concepts * Organise Knowledge Forualizatlon Refine Design * roruuiate Rules Inpleientation * Validate Rales i Refine Representations Testing FIGURE 3 Stages of Knowledge A c q u i s i t i o n (adapted from McGraw and Harbison-Briggs, 1989) EXPERT SYSTEMS / 31 Purpose of Technique - e l l c l t a t i o n -analysis -representation -validation Domain C h a r a c t e r i s t i c s -deep knowledge vs. Bballov knowledge TasX type - c l a s s i f i c a t i o n , diagnosis, debugging, etc. Kno¥ledge Type -concepts -structure of knowledge -problen-solvmg strategies Stage of Development - I n i t i a l stage vs. late stage C h a r a c t e r i s t i c s of E x p e r t -articulate vs. in a r t i c u l a t e -personality type -novice vs. seasoned expect C h a r a c t e r i s t i c s of XE -doialn knowledge -Interpersonal s k i l l s - a b i l i t y to evaluate knowledge C o s t of Technique -tiae and aoney costs FIGURE 4 C r i t e r i a f o r Selecting a Knowledge Engineering Technique EXPERT SYSTEMS / 32 development of the system, the type of knowledge, the domain characteristics, the task type, and the cost of the technique. 2.9 Knowledge Validation There are some special concerns regarding knowledge-based systems when compared to traditional information systems. Specifically, it is the problem of validation. Validation is usually handled informally - not in a systematic fashion. However, almost as many validation methods as acquisition methods have been proposed. Validity is defined as \"the degree of homomorphism between a representation system, i.e., expert system and the system that it is supposed to represent, i.e., expertise source\" [Vandierendonck, 1975]. An associated concept is the one of verification. Verification can be defined as \"the demonstration of consistency, completeness, and correctness of the software at each stage and between each stage of the software development life cycle\" [Adrion et al., 1982]. The distinction between validation and verification is important. The concept of validation includes elements of how well the conceptual model represents the real model and how well the implemented model represents the source and conceptual model. Validity can also be defined at many different levels such as how closely each minute chunk of knowledge in the system correlates with chunks of knowledge in the expert's memory. EXPERT SYSTEMS / 33 Validation is becoming a more important issue as system developers in the expert system domain build larger and larger systems that are having an increasingly important impact. The issue of maintaining these large systems is also becoming a critical issue. Therefore, many authors are proposing a range of validation techniques corresponding to the range of elicitation techniques. These techniques include knowledge-base walkthroughs with source and non-source experts and the use of different K A tools such as protocol analysis to assess validity. Benbasat and Dhaliwal (1988) propose a validation-based life cycle model (see Figure 5). From this model they propose a range of techniques to be used at different stages of the development of the system. For example, a different validation technique would be used in the modelling phase and the system construction phase. It appears that increasingly formal approaches to validation will become more popular just as knowledge acquisition techniques have become more formalized. 2.10 Knowledge Engineering with Multiple Experts Expert systems are difficult enough to properly develop using just one expert. When multiple experts are consulted, a host of additional issues become important. More demands are placed on the knowledge engineer because he/she must be able to reconcile and integrate the knowledge from different sources. Several group K A techniques have also been proposed for use with multiple experts [McGraw and Searle, 1988]. EXPERT SYSTEMS / 34 IDEALIZED' KNOWLEDGEBASE .:. b reality); :. OTHER 'ALTERNATIVE KNOWLEDGE-BASES . IN REALITY ; i$ •SOURCE* - * KNOWLEDGE-BASE IN REALITY . IMPLEMENTED KNOWLEDGE-BASED SYSTEM CONCEPTUAL KNOWLEDGE-BASE Notes: 1 Conceptual and alienation validation oonttltute Knuwtodtft Afq<*iWnn Valdalon. 'implementation validation can b* termed VWIficabon. *Bolh functional and representational validation constitute Knowtodga-Syitam Valdaflon. FIGURE 5 A Validation-Based L i f e Cycle Model of the Knowledge-Based System Development Process (reprinted with permission of the authors) EXPERT SYSTEMS / 35 The first issue to consider is whether the knowledge domains of the different experts are overlapping. If so, then the knowledge engineer will have to consider different problem solving strategies and viewpoints. The knowledge engineer can try to integrate these viewpoints by himself or he can do it in a group session. Group K E sessions can take many forms. McGraw and Searle (1988) discuss three different techniques : brainstorming, consensus decision-making and the nominal group technique. Brainstorming involves a group session where participants are encouraged to express their ideas in rapid succession. Each idea is evaluated afterwards by the group. The focus in this technique is on the quantity of ideas. In consensus decision-making, the emphasis is placed on finding the best solution to the problem. Alternative solutions to a problem are proposed and then each one is voted on. The nominal group technique involves soliciting ideas anonymously from each group member. Each idea is then ranked anonymously by each group member. This technique is useful when there is the possibility of group members not being able to express their thoughts freely. EXPERT SYSTEMS / 36 Other issues associated with multiple experts are: how to validate the knowledge base, and how skilled the knowledge engineer should be in group dynamics and in integrating knowledge. 2.11 Hypertext The concept of hypertext is becoming more relevant to the expert systems field. Many expert system software packages now include hypertext-like features [Stoddard, 1988]. Hypertext may be defined as \"a combination of natural language text with the computer's capacity for interactive branching, or dynamic display ... of a nonlinear text ... which cannot be conveniently printed on a conventional page\" [Nelson, 1967]. It is essentially a more complex method of handling information and has many advantages over the \"linear\" fashion of managing information. Hypertext may be envisioned as a series of nodes and links. Each node corresponds to \"chunks\" of information. Each node may be connected to many other nodes via links [Conklin, 1987]. The user is able to \"jump\" to other nodes in a hypertext system by activating pointers. A popular hypertext system is the HYPERCARD system that is available on the Macintosh computer [Williams, 1987]. This system also includes the use of graphic images so it could be termed a hypermedia system. Many possibilities for managing information exist with a hypertext system. EXPERT SYSTEMS / 37 The user has the option of choosing his own path through the information base. There is much potential in the forestry field for this technology. This is because there is a large base of information that the forest manager has to use in order to help make his decisions. An expert system coupled with a hypertext system would make it possible for a forest manager to use one computer system for all his/her information needs thus reducing the need to maintain large sets of reference manuals. 2.12 Integration of Expert Systems into the Environment Although many expert system prototypes have been developed in the past ten years, very few have actually been successfully implemented [Pedersen, 1988]. Many of the \"technical success - operational failures\" could have been avoided if the developers had properly considered how the expert system should have been integrated into the environment. This entails the examination of many different issues. The user interface of the expert system is very important. The system should provide advice at a suitable level for the user. The focus should be on -having the user .understand the causal structure underlying the problem [Gordon et al., 1987]. Therefore, the system dialogue should be at a level the user can understand. Some expert systems vary the technical level of their dialogue in response to the perceived technical ability of the user. EXPERT SYSTEMS / 38 The users should also be educated as to the limits of expert system technology. This is necessary because research has shown that many expert systems are only used for the most difficult problems [Gordon et al., 1988]. Unfortunately, it is at this end of the problem spectrum that expert systems are most \"brittle\" in that the quality of their advice rapidly degrades. Therefore, a user who uses the system mainly for these sort of problems will have a distorted view of the reliability of the system. Another issue to consider is how the expert system will integrate with existing systems. Cholawsky (1988) states that data problems are a major cause of implementation failures. These types of data problems range from nonautomated data to incomplete data to inappropriate data. The exact data needs of the expert system must be analyzed carefully so that the user has to manually type in as little data as possible. Data checks should be placed to avoid erroneous data. There are also business issues that need to be considered for a prototype system to be implemented successfully. The system must be cost-justifiable, for if it is not, then the commitment of senior management will be jeopardized. Also, maintenance of the system is very important. One of the most successful expert systems in use is DEC's X C O N system which is experiencing problems with high maintenance costs [Newquist, 1988]. It does not make good sense to replace dependence on an expert with dependence on expert system programmers. EXPERT SYSTEMS / 39 2.13 Expert Systems i n the Forestry Domain Expert system prototypes have been built for several problems in the forestry area. These problems include the planning of forestry roads [Thieme et al., 1987], the dispatch of fire control resources [Kourtz, 1988], and the treatment of tree fungi [Rust, 1988]. A demonstration of expert system technology for fire effects problems has also been built [Starfield and Bleloch, 1983]. A comprehensive system for developing fire prescriptions is now being built at the University of Missoula [Reinhardt, 1987]. Another system for prescribed burn planning is also being developed by the Canadian Forest Service [Hawkes and Lawson, 1986]. Expert systems technology has considerable potential in the forestry domain. Many of the problems and decisions can't be treated quantitatively and are best solved using the experience and skill of an expert. Many of these problems also require the integration of multiple sources of knowledge and expert systems technology can help in this regard. The fire effects domain is quite suitable for the application of expert systems technology. There are only a few experts in this field and there are many people who can make use of their expertise. Much of the expertise needed for problem solving consists of narrow domain knowledge. There are many different sources of knowledge that are relevant to the decisions and the EXPERT SYSTEMS / 40 decisions can't easily be treated algorithmically. The ideal structure and form of a fire effects system is unclear. It is not immediately apparent as to how global the reasoning framework should be for a fire effects expert system. There are many diverse ecosystems in BC and prescribed fire interacts quite differently with each of these ecosystems. However, it is not operationally or economically feasible to develop an expert system specifically for each type of ecosystem. Dyer (1989) discusses some of the problems of adapting an expert system to diverse geographical areas. She proposes a layered structure for the expertise in an expert system whereby the local heuristics can be incorporated into a geographically- standardized knowledge base. An expert system for predicting fire effects could make use of this approach. CHAPTER 3. PRESCRIBED FIRE AND FIRE EFFECTS 3.1 Introduction Muraro (1971) defined fire effects as the combined result of the immediately evident effect of fire on the ecosystem in terms of biophysical alterations or population reduction as well as postfire influences. This definition indicates that there are both short term and long term components to fire effects. In order to study these two components, fire effects researchers have had to study old burns as well as monitor recent burns. The first section of this chapter discusses fire effects research in BC. Subsequent sections discuss the various ways in which fire interacts with site properties, the terminology of fire behaviour, the procedures involved in conducting a prescribed burn, the BC site classification system, the limiting factor concept and the status of prescribed fire decision aids in BC. 3.2 Prescribed Fire and Fire Effects Research i n B.C. Many different groups in B.C. are involved in research concerning prescribed fire and fire effects. The BC Ministry of Forests (MOF) and the Canadian Forest Service (CFS) are two of the more influential groups involved. Figure 6 shows the links between the researchers, the various coordinating groups and the Ministry of Forests. The Ministry of Forests has the critical responsibility of identifying research needs and transferring technology from the researchers 41 PRESCRIBED FIRE / 42 Research Reeds Croi tbe n i n i a t r y of fo r e s t s mit (tg Oittrict) t nb-Vx t O p i H l l o n i l t Land Banageatnt Acumias A3 3B33ieilt OJLA Coordinating Oroups (A •» C)« B.C. Forest flesaarch cooacll (A • c. BasearcB Beads] (tor Beeional a. Provincial Besaarcn Ad>laor, Caul t t tn (A • C) crs/Bor BecaarcB wresignt n u g a n i earning (BesaarcB item, A » o Provincial-Industrial coordinating Counties {BoseercB Hods and A « c. e.g. silviculture coanlttoss] aetignaL Besearci council (Raitaxcb loading] rtBIC (RosearcD isnain BID u-BouseJ FOBIinX and tha Palp aad Paper Besearcn Instiling ol Canada (Baaearch lending aad la-houie) CPS Science SibTenlioa Prograas (A I C] • . j . paor contracts to nniversity He3eo.1cb.er3 ~ 1 Land Hanageant Orgaaisatioi, (•a Biuatrj ol Foiamlt lestercB Braicn) organisation (eg. CFS. CIS] MlnivercltLes Private [Rdmtrr consultants •Assessaeat aid Coordinating troops aar Be Brpagsad ana faaaarcb neede ny flev directly to tat raaaarchars FIGURE 6 Flow of Research Needs from the MOF to Forestry Researchers (adapted from Hawkes et a l . , 1984) 42 to the operational staff. PRESCRIBED FIRE / 43 In order to make the best use of limited research funds and research personnel, a Prescribed Fire Research Advisory Committee (PFRAC) has been established. This group includes representatives from the Ministry of Forests, the Canadian Forest Service, industry and academia. The primary functions of this group are as follows [Hawkes et al, 1984]: 1. to further develop and revise the strategic plan for prescribed fire and fire effects research. 2. to prioritize, advise and help facilitate research projects to meet the requirements of the strategic plan. 3. to be given the mandate to develop cooperative, multi-agency research projects and studies. 4. to develop implementation plans for research results, decision — aids and training needs. 5. to recommend policy, planning, and operational procedure changes required in various agencies in BC to improve the use of prescribed fire. The basic goal of this group is to promote closer links between fire research PRESCRIBED FIRE / 44 and the forest resource management process. In order to accomplish this goal, the strategic plan defines a prescribed fire management system comprised of seven phases. These seven phases are: 1. Operational plan. 2. Pre-harvest assessment. 3. Treatment alternatives evaluation and selection. 4. Prescribed fire prescription and plan development. 5. Prescribed fire application. 6. Prescribed fire monitoring. 7. Prescribed fire evaluation. For each of these phases the strategic plan attempts to define operational problems and technology transfer needs. Standardized methodologies are also proposed to document and evaluate prescribed burns. Figures 7a, 7b, and 7c illustrates these seven phases. The Ministry of Forests conducts its prescribed fire research through several different internal groups. In Victoria, the Protection Branch, the Research Branch, and the Silviculture Branch do research into prescribed fire and fire effects. The Ministry of Forests also conducts much of its prescribed fire and fire effects research through its six regional offices located in: Smithers, Prince George, Nelson, Kamloops, Williams Lake, and Vancouver. Most of these offices have a research silviculturalist, a research pedologist, a research ecologist, a wildlife ecologist, a pathologist and a hydrologist. PRESCRIBED FIRE / 45 OPERATIONAL PLAH Pbyscial and b i o l o g i c a l s i t e paraaeters* Inventory t o t a l resource I I s i l v i c u l t u r e end protection l i r e ueasgeaeat expertise policy end practice I engineering end (Ire aaaageaent eipertlse I what coabinstions of barvest systeas. tree regeneration systeas. and s i t e treataente v l l l occur l a t i e harvesting ores? 0 f i l l prescribed l i r e De part ot tbe s i t s treataent alternatives'' 0 Hov should harvesting schedules, segue&ces. systeas. and block layout be developed to use prescribed (Ire cost-offactively? _____ general s u i t a b i l i t y o( prescribed (Ire on s p e c i f i c s i t e s -road developaent plan -block layout standards sensitive to prescribed burning reguiresents -logging sequence to ainlouae p r e « j C T l n « 1 t\\ra r r m t r n l T l i k PHE-HARVEST ASSESSMENT (Cut DlOCk l e v e l ) advanced regeneration status species s i l v l c s seed and seedling a v a i l a b i l i t y acceptable regeneration delay stocking sta\"i=»rii« _ ^ engineering and cost constraints vatersbed predicted post-harvest residue and organic setter conditions plantable spots insect and disease concerns ecological constraints Vast possible regeneration aetbods? Vbet possible harvest systeas? Vkst possible s i t e preparation technlgues to peralt regeneration al the area? . Treatment A l t e r n a t i v e s TREATMENT ALTERNATIVE EYALUATIO^AND^SELECTION cost/benefit analysis vkicb coul consider: -treataent costs -positive or negative long tare productivity changes -positive or negative protection e i l e c t s ( l i z e . insect, disease) X -regeneration e s t a b l l s - i e n t . s u r v i v a l , and growth standards -positive or negative vatarshed changes -positive or negative v l l d l i f a . range and l i s b habitat etfects Vbat i s tbe optinua a n at post-harvest s i t e conditions on tbe cut block to achieve the s p e c i t l c land aanagenent ohjectives? O p L i i l i i i o treatment a c t i v i t i e s (e.g. nigD leaa yara, DioadcasL Dura and plant) aa Undoing y PRESCRIBED FIRE PRESCRIPTION AND PLAN DEVELOPMENT FIGURE 7a Proposed Prescribed F i r e Management System (adapted from Hawkes et a l . , 1984) PRESCRIBED FIRE / 46 PRESCRIBED FIRE PRESCRIPTION AJTD PLAN BEVEL OP KENT post-bervest s i t e treetaent (Ire lapact objectives translated into ( l r c predictor lapact c r i t e r i a post-barvest slosh and s o i l organic natter condition ecological constraints (translate* Into (Ire lapact c r i t e r i a 1 predicted f i r e saoke aenageaent bebavlour i n cot Block and adjacent stand v a t e i . acce and topograpny H a r v e s t i n g layout logging aetnod t l a i n g Vbat ar I e g II I l t l B t a r \"\"^\"\"aa**\" i b a t bu F i r e Vbat are tae rrurn objectives' e g alasb f u e l and dult reduction, and i l n e r a l s o i l exposure e tae rrurn conditions needed' re veatber and l u s l noistare are tbe f i r e control requirements'' I b rn techniques. Ignition systeae. and patterns w i l l be used? Vbat are tbe saoke aanageaent xequlxeaeata' Vaat are tbe aop-up requirements' Regeneration species Iteas per hectare planting aetbod Prescribed F i r e Prescription. Plan and Burning Peralt PRESCRIBED FIRE APPLICATION a v a i l a b i l i t y o( i g n i t i o n systeis Burn day acceptable'' - v i t n i n presciption' -burn conditions -control - i g n i t i o n f e a s i b i l i t y -BOp-Up -saoie dispersion BtsnlardiMd Mttedolnqr o Tbe Burn PRESCRIBED FIRE MOKITORIXG FIGURE 7b Proposed Prescribed F i r e Management System PRESCRIBED FIRE / 47 PRESCRIBED FIRE tiOUITORIHG aiesb trio soil organic M t t « l reduction conditions tor the burn eey rtra control •eiion Burn tecknieues. Ignition nrstena. and patterns used I Hop-vp net L O S S . 8«ok« dispersal pattern I H a r v e s t i n g Harvest to n l n l a i z e residue P i r e Vint vere tbe M r s mpacts' Vbat vere tbe barn conditions'' Vbat control actions vere taken'' vnat Banting tecnnigues. l a m c i o n systems, and patterns vere used'' Vbat vere tbe nop-op reqolrenents' Vbat was tbe saoke d i s p e r s a l ' Vbat vara tbe presribed burning c o s t s ' R e g e n e r a t i o n Advanced regeneration'' Plantation establishment, s u r v i v a l , and grovth'' I vaste assessment Burn Docunentation (including escaped l i r e analysis) I Post-barvBst assessment Survival p l o t s PRESCRIBED FIRE EVALUATION ( i r e lapact translated into s i t e treatment objectives I burn conditions, p r e s c r i p t i o n t i r e actions f i r e control plan burn tecnnigues and bum plan techniques I saoke dispersal smoke nanageaent pien I Vere tbe s i t e treatment objectives net' Vere prescribed burn conditions net' vere tzeatnent costs acceptable Any t i r e control problems' 01b burn techniques including equipeaent. I g n i t i o n systeas. and pattern meet tbe f i r e control plan and s i t e treatnent objectives'' Vere there any saoke management problens' Prescribed f i r e analysis standardised netnodolagy Feedback FIGURE 7c Proposed Prescribed F i r e Management System PRESCRIBED FIRE / 48 The goal of much of this region-based research is to develop ecosystem-specific guidelines for using prescribed fire [Mackinnon, 1988]. Through the use of biogeoclimatic maps, these guidelines can be extrapolated to similar sites throughout the province. This is seen as the best method to translate research results to operational procedures. 3.3 Effects of Fire on Soil Fire can have a very significant impact on soil properties. The effect of fire on soil properties will vary with the fire intensity and the amount of organic matter that is consumed. Fire can also be beneficial or detrimental to the short and long term productivity of the site. These changes can be temporary or permanent. Although it is convenient to classify these changes into different categories, it is important to note that most of these factors tend to interact with each other resulting in a complex series of changes to the site [Feller, 1982]. Kimmins (1987) classifies the effects of fire on soil into three types of changes: physical, chemical and biological. 3.3.1 Physical Changes Physical property changes can be further classified into four categories: PRESCRIBED FIRE / 49 organic matter, structure and porosity, moisture, and temperature. Organic matter: Fire consumes organic matter by transforming dead foliage and fine roots into its constituent minerals. Although the mineralization of dead organic matter occurs naturally through the action of microbes, fire can accomplish this transformation much faster. As well, although much of the nutrients may be lost from the site in the form of fly-ash, the amount of readily available nutrients on the site almost always increases immediately after a fire. Structure and Porosity: Many instances have been documented concerning fire-induced changes in structure and porosity. Fire has been known to break down the structure of the soil and result in a hydrophobic (water-repellent) layer. Since fire consumes organic matter, the porosity of the soil will almost always be affected by fire. Increased surface water flow can occur after a fire resulting in accelerated erosion of the topsoil. The amount of erosion that occurs will be influenced by the slope, the depth of the organic and/or mineral soil layers, the texture, the intrinsic erodibility of the soil, the fire depth of burn, the climate, and the revegetation rate [Feller, 1982]. Moisture: Moisture is the third type of physical change. Most of the changes in soil moisture content are due to the loss of foliage. The soil becomes wetter as less moisture is lost through interception and transpiration. The soil can also become drier if the soil is quite coarse textured [Kimmins, 1987]. This PRESCRIBED FIRE / 50 results from the decreased moisture retaining capacity of sandy textured soils as compared to humus rich soils or clay type soils. The moisture status of the soil is affected by the organic matter content of the soil, the soil moisture content during the burn, the fire intensity and depth of burn, and the climate. Soil Temperature: Soil temperature is affected by the aspect, the forest floor thickness, the fire intensity and depth of burn, and the climate. The effect of fire is an increase in soil temperature because fire blackens the soil surface resulting in greater heat absorption. In B.C., the soil temperatures during the growing season are cold compared to the temperatures required for optimal seedling growth. Therefore, fire is almost always beneficial for the soil temperature regime of the site. 3.3.2 Chemical Changes Chemical changes to the soil can be classified into two types [Kimmins, 1987]: changes to pH, and changes to site nutrient capital and nutrient availability. pH: Changes in soil pH primarily result from the basic (alkaline) nature of ash. Therefore, fire will almost always raise the pH of the soil. The pH value of the soil relates to the solubilities of key nutrients and correlates positively with the rate of decomposition of organic matter. PRESCRIBED FIRE / 51 Site Nutrient Capital and Nutrient Availability: The second type of chemical changes are changes to site nutrient capital and nutrient availability. Fire almost always results in a reduction in total site nutrient capital due to volatization (rapid evaporation during fire). This occurs because much of the minerals are lost through smoke and fly-ash. However, there is usually an increase in the amount of available nutrients because fire converts undecomposed organic matter into a soluble form that is more readily available to plants. 3.3.3 Biological Changes The final type of soil property changes are biological changes. Fire induces changes in the biological properties of the soil primarily due to soilheating although fire-induced pH changes can also affect the biological properties of the soil. Fauna: Most of the meso fauna and micro fauna are killed during a fire. These fauna can be important to energy flows between different parts of the ecosystem because of the contribution of microorganisms to the process of decomposition. A site can usually be recolonized with meso and microfauna in just a few years. However, this new set of fauna can be quite different from the old set of fauna. Vegetative Cover: Fire will also cause changes to the vegetative cover on the PRESCRIBED FIRE / 52 site. Although fire will kill most of the existing vegetation cover on the site, the vegetation that regrows may be lusher than the vegetation that existed before the fire. The increased nutrient availability is usually regarded to be the cause of the lusher vegetation regrowth. However, the vegetation complex may be quite different than that which existed before the burn. Many different factors will influence the degree to which fire will affect each of these soil properties [Kimmins, 1987; Feller, 1982]. Figure 8 summarizes the relationship between these factors and the soil properties. 3.4 Fire and Ecosystem Processes The previous section discussed how fire induced changes in soil properties. The real effect of these changes is a change in energy flows between the different parts of the ecosystem and the biogeochemistry of the site. This in turn affects the productivity of the site. Fire can have a major impact on ecosystem processes. It should also be noted that fire-induced changes to energy flows are not usually static. For example, changes in pH can be temporary or permanent. The major processes that affect energy flow between the different parts of the ecosystem are the processes of leaching, decomposition and erosion. Fire can affect each of these ecosystem processes. PRESCRIBED FIRE / 53 SOIL CHARACTERISTIC SOIL FACTORS OTHER FACTORS Physical Properties • Structure & Por o s i t y and Organic Hatter Content 1. slope 2. depth of organic and/or a i n e i a l s o i l layers 3. texture 4. c r e d i b i l i t y -aggregate s t a b i l i t y -parent n a t e x i a l 1. f i r e depth of burn 2. c11nate 3 revegetation rate - tlolsture Status 1. organic Batter content 2. a o i s t a r e content of s o i l during tbe burn 1. l ire i n t e n s i t y and depth of burn 2. c l l B o t a - Teaperature 1. aspect 2. (orest f l o o r thickness 1. f i r e i n t e n s i t y and depth of burn 2. c l l D B t e cnealcal Properties - S i t e l u t n e n t C a p i t a l and Mutrient A v a i l a b i l i t y 1. c a t i o n exchange capacity 2. f o r e s t f l o o r depth 3. f o r e s t ( l o o r n u t r i e n t content *. mineral s o i l n u t r i e n t content 1. f i r e i n t e n s i t y and depth of burn 2. c l l n a t e 3. s l a s h consumption - pa- 1. organic natter content 2. clay a l n e r a l content 1. f i r e i n t e n s i t y and depth of bum 2. s l a s h consumption B i o l o g i c a l Properties - fauna 1 organic natter content 2. noisture content of s o i l during the burn 3. presence or absence 1. fire I n t e n s i t y and depth of burn 2. type of plant species regenerated a f t e r the burn - Vegetative Cover 1. presence or absence 2 tolerance to burning 1. l i r e i n t e n s i t y and depth of burn FIGURE 8 Factors Determining the Impact of Prescribed F i r e on S o i l s (adapted from F e l l e r , 1982) PRESCRIBED FIRE / 54 For example, fire can initially induce a change in the pH of the soil. This will affect the concentration of microorganisms in the soil which will affect the rate of decomposition. The solubility of the various mineral ions will also be affected by the pH. This relates to the amount of leaching that will occur. 1 The pH will also affect the vegetation that is on the site which will in turn affect the amount of erosion. 3.5 Fire Severity and Fire Behaviour In order to define the effect of fire on the ecology on a site, it is necessary to clarify some of the fire behaviour terminology. There presently exists much confusion in the literature because researchers have been using terms such as \"cool\" fire and \"hot\" fire. These fire classifications are very subjective and do not easily allow for direct comparisons between fires on different sites [Alexander, 1982]. The terms \"fire intensity\" and \"fire impact\" are used quite commonly in the literature. A suitable definition of fire intensity can be found in Alexander (1982). The fire intensity, or energy output, is defined as the product of PRESCRIBED FIRE / 55 available fuel energy and the fire's rate of advance. However, fire intensity can be a misleading number because a fire can be intense but last only a short time. Conversely, a fire could be not very intense but could last a long time. Both these fires could have similar effects on a site. Therefore, fire effects researchers usually use other terms besides fire intensity. \"Fire impact level\" is a term first proposed by Muraro (1977). His prescribed fire predictor/planner uses a single impact level to characterize a fire in terms of its effects on the duff thickness, amount of mineral soil exposure, and slash loading. The duff thickness is the thickness of the forest floor (see Section 3.10 for a more complete definition), the mineral soil exposure is the amount of soil that is exposed on the surface (i.e. where there is no forest floor), and the slash loading refers to the weight of slash (logging detritus) still present on the site. Although the term impact level is widely used, it is not always the most appropriate term to use because it portrays the impact of fire on a site in terms of one single number. It would be better to think of fire impact levels in terms of a number of different site properties. Therefore, the term fire severity could be a more appropriate term (M. Feller4, pers. comm.). The severity of a fire would be defined in terms of the amount of duff consumed or the amount of mineral soil exposed. For example, a fire could be classified as being severe in terms of duff consumption and 4 Professor of Forestry, U n i v e r s i t y of B r i t i s h Columbia. PRESCRIBED FIRE / 56 moderate in terms of mineral soil exposure. Most of the research concerning fire severity has been directed toward predicting slash consumption, duff consumption and mineral soil exposure by relating these parameters to the preburn moisture codes (i.e. Drought codes and Duff Moisture codes) and the slash loading on the site. The moisture codes can be determined quite easily for most sites and are already used quite extensively by the Forest Protection Branch to determine fire danger ratings for sites throughout the province. These moisture codes are supposed to reflect the moisture at various depths within the duff. The codes define a moisture gradient which has an important effect on the impact of a fire on a site. For example, a burn conducted on a site where the surface is dry and the underlying soil layers are wet will probably result in a \"low\" severity fire. For any given site, it is also possible to manipulate the impact of a fire on a site by burning under particular weather conditions and at certain times of the year. A spring burn will generally be less severe than a fall burn because the underlying duff layers are still quite moist due to the snow melt [Anonymous, 1985]5. 5 This document i s a manual j o i n t l y published by Macmillan-Bloedel and the BC M i n i s t r y of Forests. PRESCRIBED FIRE / 57 3.6 Vegetation and Prescribed Fire Fire can have a dramatic effect on the vegetation complex of a site. In particular, fire usually converts the vegetation on a site to an earlier successional state such as a grassland. Fire is used quite extensively throughout the world to convert areas for use as grazing pastures. Without the intervention of fire, a site will reach its climax successional state, that is, the most mature state that the vegetation complex will reach. Fire usually changes the vegetation complex on a site because of the different tolerances of the species to fire. A very severe burn can actually kill all the vegetation on a site. Any subsequent regrowth on the site will only result from inseeding or through some sort of human intervention such as seeding or planting. Vegetation control can be the primary objective for a prescribed burn. Burning is done on the site to establish a time window for the tree seedlings to grow and take hold. However under certain situations, burning can dramatically increase the vegetation cover on the site. For example, a site with a moderate concentration of aspen on it before the burn will almost always become a dense aspen thicket after a burn (E. Hamilton6, pers. comm.). This is because fire induces suckers in the stems of the aspen plant. 6 Research Branch, BC Forest Service. PRESCRIBED FIRE / 58 Kimmins (1987) has defined the following five categories of mechanisms by which vegetation adopts to fire: 1. adaptations to fire in the vegetative stage. 2. adaptations to fire in the reproductive stage. 3. effects of fire on the germination phase. 4. evolution of increased inflammability. 5. other adaptations. Adaptations to fire in the vegetative stage include species that have developed fire-resistant bark. Other adaptations in this category include species that have nonflammable tissues, as well as species with rhizomes (horizontal, underground stems). Adaptations to fire in the reproductive phase include stimulations of flowering in some species. For some species, seed dispersal can also be affected by fire. Seed germination of some species can be influenced by fire. Some species are termed seedbankers in that they commonly have dormant seeds below the surface of the soil. The heat from a fire induces germination of these seeds. The mineral soil exposure induced by burning can also influence the germination success of some species. This phenomenon may be partially attributable to the increased moisture content of some mineral soils (E. Hamilton, pers. comm.). PRESCRIBED FIRE / 59 Evolution toward inflammability is believed to be another adaption to fire. That is, the species may evolve towards having more flammable tissues. Although this species will be consumed by fire, most of its competitors will be consumed as well. The adapted species can usually recblonize the site quite easily. Other adaptations to fire include fire-resistant bark, fire resistant needles and rapid elevation of the terminal bud and foliage. Fire primarily influences vegetation through duff consumption and soil heating. For any given burn, it is possible to define a depth of lethal temperatures: any species which sprouts from a point above this depth of lethal temperature will usually be killed by a fire. However, since fire does not usually affect a site uniformly, a certain percentage of the species will normally survive. The tolerances of different species to fire is only generally known. Much of the information regarding fire tolerance is summarized in Haeussler and Coates (1986). PRESCRIBED FIRE / 60 3.7 Prescribed Burning Procedures i n BC If prescribed burning is to be done on a site then this information should be considered at an early stage in the forest resource planning process. The cutblock must be laid out so that a treatment can be applied uniformly and economically over the whole site. The method for handling slash on the site will also be affected by the site preparation method. On sites where whole tree harvesting is done, an effort must be made to leave some slash on the site if burning is to be done afterwards. The site preparation treatment to be used is first noted on the Pre-Harvest Silvicultural Prescription form (FS 711A). Appendix 2 contains a copy of this form. A more thorough analysis of site preparation alternatives is usually done after harvesting of the site. The Site Preparation Guide (Appendix 3) is designed to help lead the land manager through the evaluations of the various alternative treatments. If a prescribed burn is to be done, then a burning plan (Appendix 4) is usually required by most forest regions. The resource manager records his/her burning prescription on this form. When the burn is done, a Prescribed Burn Analysis form (Appendix 5) is required. This form is used to document the site and weather conditions at the time of burn, the costs associated with this burn and a summary of how the PRESCRIBED FIRE / 61 fire affected some readily observable site characteristics. Chapter 4 discusses prescribed burning procedures in more detail. The prescribed burning decisions are also modeled using a system analysis technique known as a Data Flow Diagram (DFD). 3.8 Prescribed Burning Decision Models Many prescribed burning decision models have been developed to aid forest resource managers in decisions involving prescribed burning. The primary decision involves considering whether fire is the best site preparation alternative for any given site both economically and ecologically. Secondary decisions involve determining the combination of moisture indices and weather conditions to burn at. Fullerton and Martell (1984) have used a decision analysis framework to define management alternatives for a jack pine sand flat cutover. This framework is primarily concerned with detennining whether fire is the most suitable site preparation treatment. They defined a total of seven decision variables which the land manager can manipulate to achieve his/her objectives. The site preparation method, the regeneration stock type, the seeding intensity, the thinning policy and the thinning year are included in these decision variables. A total of forty-eight state variables defined the environment. Both these set of variables were incorporated into a FORTRAN program which could PRESCRIBED FIRE / 62 show how each management strategy would affect the economic value of a site. The output of this model is the net present worth of the site as a result of a particular management strategy and inclusive of the costs of following that strategy. Radloff and Yancik (1983) have also attempted to model the prescribed burning decision framework. Their model considers the important variables such as fire behaviour, operational costs, risks to resources and property as well as the uncertainties of these variables. Probabilities are used to assess the potential for fire escapes given different weather scenarios. The output of their model is an expected value of the cost of a burn which reflects the various probabilities of events occurring. Raybould and Roberts (1983) have defined a prescription matrix. This matrix aims to show how different elements of a fire prescription can interact with each other. These factors can be offsetting and therefore it may still be possible to burn to obtain the desired effects even when the weather and site conditions are out of range for the prescription. 3.9 The Ecosystem Classification System i n B.C. The biogeoclimatic ecosystem classification system (BEC) forms the basis for much of the forestry interpretation work in the province. Land is mapped on the basis of ecosystem type in order to prescribe guidelines for forest PRESCRIBED FIRE / 63 management on an ecosystem-specific basis. The BEC uses vegetation, climate and soil data to map an area, thus, it is an integrative classification system [Pojar et al., 1987]. The first step involved in classifying an area is to identify zonal (climatic climax) ecosystems. These are ecosystems that are intermediate in properties (i.e. soil moisture, soil nutrients, light and heat) to the other ecosystems in the area. These sites are generally found on gently sloping areas. The vegetation and soil on these sites are primarily a product of the climate in the area. The plant association is the basic unit of vegetation classification and is defined as a floristically uniform vegetation unit. Each plant association can be transformed into a site association because the vegetation on a site is a function of the biotic potential of the site. Species with relatively narrow ecological amplitudes form the basis for much of the mapping. However, if the vegetation on a site has been disturbed by events such as fire or logging, then other factors must be given more weight in order to classify the site. These factors include soil properties such as texture and humus form. A plot of soil moisture regime versus soil nutrient regime for a subzone defines an edatrophic grid. Site associations can be represented as unique areas on this grid. Forestry interpretations and management recommendations are done on a PRESCRIBED FIRE / 64 site association basis. Each site association has a recommended tree species and a recommended treatment. However, even though two sites may have an equivalent vegetation complex, they can have quite different site properties. Some site factors may compensate for each other. Fire can therefore affect these two sites differently and this is why the recommended site treatments for an ecosystem should not always be adhered to. 3.10 Soil Development and the Forest Floor It is important to discuss how the various soil layers are formed in order to understand how fire affects soils. The development of soil horizons is related to energy flows within the ecosystem and of course has a direct impact on the potential productivity of the site. Organic matter is deposited continually on the site in the form of litterfall. Depending on the chemical and biological conditions, this organic matter will gradually be decomposed into its mineral constituents. Leaching (percolation of waters through the soil) may remove much of the minerals from the site and this process will be the major determinant of soil fertility. In fact, most soils are classified by the extent to which they have been leached. Because additions of litterfall to the site usually occur faster than decomposition, a forest floor will normally develop. This forest floor is a permanent buildup of organic matter. In mature forest floors, three distinct PRESCRIBED FIRE / 65 layers can generally be identified. These are the L , F, and H layers. The L layer is the litter layer. The F layer is the fermentation layer and is an intermediate layer composed of fine roots and partially decomposed Utter. The H layer is the humus layer and is generally darker in colour and moister than the overlying F layer. Forest floors can be classified on the basis of their humus type. Three major categories are usually used to classify humus forms. These categories are mors, moders and mulls. Mors are formed in cool moist climates where the decomposition process is slow [Kimmins, 1987]. There is usually very little intermixing of the forest floor with the underlying mineral soil. The forest floor is very important for these types of sites. The opposite extreme is a mull humus form. This humus is quite rich and there is usually a lot of intermixing of the forest floor with the underlying mineral soil. Frequently, an Ah (organically rich mineral soil) layer is formed in the underlying mineral \"soil as a result of this intermixing. The forest floor is not as important to this site's productivity. Moders are humus forms that are intermediate in properties between mors and mulls. The term duff is also sometimes used when forest floors are discussed. The term duff refers to the intermediate layer of more or less decomposed organic material underlying the litter layer [Klinka et al., 1981]. However, the term duff does not refer to any part of the underlying mineral soil even in the case of mulls where there may be a lot of organic material in the mineral soil. A PRESCRIBED FIRE / 66 common measure of the severity of a fire is the amount of duff that is consumed. 3.11 Limiting Factor Concept The limiting factor concept is used quite commonly in the ecological modelling field. This concept states that the growth rate of a plant species will be determined by the level of the factor that is the most limiting (least optimum). Figure 9 illustrates the limiting factor concept as applied to an agricultural example [Brady, 1974]. In this example, potassium is the most limiting factor in the figure on the left hand side. As long as the amount of potassium was not increased, it would not be possible to increase the crop production by increasing the amounts of the other elements. On the right hand side of the figure, the potassium level has been increased and it is now nitrogen that is the most limiting factor. The hmiting factor concept can also be applied to tree production in forestry. The Site Preparation Guide uses the limiting factor concept when it asks the user to indicate whether the moisture or nutrients is limiting on a site. Although the limiting factor concept is an oversimplification of reality (some factors can in fact compensate for others), it is a very easy concept to understand and implement in any sort of manual or computerized system. PRESCRIBED FIRE / 67 FIGURE 9 Example Demonstrating Li m i t i n g Factor Concept (adapted from Brady, 1974) 3.12 Prescribed Fire Decision Aids PRESCRIBED FIRE / 68 A major goal of the prescribed fire research advisory committee (PFRAC) is to transfer technology from the research groups to the field staff. Educational workshops, field manuals and decision aids are some of the different ways to enable this technology transfer. One of the earliest decision aids developed was the Prescribed Fire Predictor/Planner (PFP). This decision aid is composed of a series of tables that relate weather, fuels and topography to fire behaviour parameters (ease of ignition, rate of spread, and difficulty of control) and measures of fire severity (duff consumption, slash consumption, and mineral soil exposure) [Muraro, 1977]. The PFP was an attempt to ease the task of developing prescriptions to meet burn objectives. However, this decision aid was never intended to produce definitive predictions of the impact of a fire given a set of site conditions. The user of the decision aid is supposed to incorporate much of his own knowledge into the development of the prescription [Anonymous, 1985]7. A need was also recognized in the early 1980's to determine the sensitivity of a site to fire. It was known that fire could not be used on all sites and an attempt was made to classify a site's sensitivity to fire on the basis of several readily observable site characteristics. Klinka et al. (1984) developed a one page 7 This document i s a manual j o i n t l y published by Macmillan-Bloedel and the BC M i n i s t r y of Forests. PRESCRIBED FIRE / 69 key to classify sites in the Vancouver Forest Region (see Figure 10). The emergence of expert systems technology has created some renewed interest concerning new possibilities for decision aids. These decision aids can now be made more complex and more comprehensive than a set of tables or a one page key could allow. The PB Planner expert system was conceptualized in 1987 in order to address the need for decision aids incorporating this form of technology. This system is being developed by Bernie Todd of the CFS at the Petawawa National Forest Research Centre. The system is envisioned to cover all aspects of the prescribed burning decision process from the development of prescriptions to the scheduling of multiple burns to the modelling of ignition patterns and fire behaviour. The Fire Effects Expert System Project is attempting to build one module from this overall system as well as examining the issue of knowledge engineering in this area. PRESCRIBED FIRE / 70 START I Soil <25cm arid/or coarse fragment content >80% Jtio Sbpe >80% 4 NO Sbpe 50-80% NO Slope 33-50% Seepage water, water table, gkrying. or periodic flooding NO Humus form Mull or Moder with Ah horizon NO Humus term >20cm trick kNO Soil dark coloured fligh C content) NO Well developed Ac OR toil parbcle sire coarse OR soil <50cm NO \" 1 YES YES YES NO YES YES YES YES 4 NO M 0Humus form >20cm thick • NO Soil tenure silly • NO o Soil panicle sin coarse 0 Wall developed Aa horizon 4 NO o I YES Humus form <20cm Buck 6 well developed Ae OR sail particle sin coarse OR soil 30 degree slope, then this would change the position of soil temperature on the numeric index scale. The effect of fire on the site would be modelled by examining how fire affected each site factor. In this example fire would only affect humus form. However, the expert did not believe that this would be a completely viable approach. It would not be necessary to define a numeric index corresponding to each growth factor. Instead, he believed that it would be possible to directly evaluate the degree of limitation of all growth factors, given a set of site conditions. The most limiting factor could then be determined by comparing the limiting level of all these factors. The degree of limitation of each of the \"1 H O G H O TJ O C/> CD 0-3 Ct-UI rt H-O 3 t\"1 H-3 H-rt H-3 iq o rt o H O o 3 O CD •O rt ECOBTSTEB ASSOCIATIOI CDTj/Ql DIRECT OR iroiBECT GBOVTH FACTOR SOIL TEMPERATURE V0RNAL LIMITING LEVEL HUMESIC INDEX • 1.0 »0.9 .0 8 »0. 7 •0.6 •0. S •0 « •0.3 •a. 2 •0. 1 •0.0 -a I -0.2 -0.3 -a. 4 — o s -0.6 -0.7 -0. 8 -0.9, -1.0 NUMERIC IITIO RELATED TO LIMITATION SLIGHT 00 DERATE SITE FACTOR 1 ASPECT NORTH EAST OR VEST SITE FACTOR 2 > 30 DEGREES SITE FACTOR 3 lUHtrs FORM HOR > 15 CM. noR < i a ca. HOR < S CM. COMMEJTS: Fire affects only raie of tae s i t e factors (I.e. aliens fore). Tberefore. t i r e v u i nave a negative efect on tnis ecosystea. A illicit auans layer acts to Insulate tbe s o i l . p a s GO 173 oo BUILDING T H E EXPERT SYSTEM / 82 growth factors would also be determined after burning. The seventh and eight sessions were conducted at UBC. They were done in the MIS lab on the second floor of the Henry Angus building. Part of the sessions were devoted to demonstrating system prototypes that the researcher had developed at various stages of the project. None of the systems had implemented the limiting factor concept though. For most of the time in these two sessions at UBC, the limiting factor concept was discussed for each growth factor. The expert took most of the notes for these sessions. Because several growth factors weren't covered in these sessions, it was decided that the expert would work on these back in Nelson and send the results to the knowledge engineer. Between the eighth and ninth sessions, the researcher implemented the limiting factor concept for most of the growth factors. Much of the system dialogue had to be written at this point and the researcher had difficulty composing some of this dialogue. This occurred because the concepts were complex but had to be explained in simple terms. The ninth and tenth sessions were done back in Nelson in December. These sessions were primarily conducted around the research branch computer in the basement of the Ministry of Forest building. The dialogue was reviewed in these sessions and much time was spent showing the expert how the reasoning BUILDING T H E EXPERT SYSTEM / 83 strategy of the system worked. The rule format of VP-EXPERT made this process quite easy. Work was also started on how the limiting factor concept would incorporate vegetation. It appeared that this would be a complex part of the program to implement. The eleventh and twelfth sessions were also conducted in Nelson. These sessions were devoted to refining the system. Particular attention was paid to the system dialogue. At one point in the eleventh session, another expert was brought in to clarify the reasoning in one of the modules. The knowledge engineer spent several hours discussing the root rot module with the regional pathologist, Don Norris. These last two sessions were quite productive. 4.1.3 Knowledge Engineering Sessions with B. Hawkes Brad Hawkes is a fire researcher with the Canadian Forest Service ~at the Pacific Forestry Research Centre in Victoria. Most of his research relates to models predicting burn impacts as well as models to describe fire behaviour. He is currently working with Bernie Todd of the CFS in the development of the PB Planner expert system. Two sessions were conducted with this expert. Both were held in Victoria at the CFS bunding in the conference room. The first' session was devoted to developing a conceptual model of how fire impacted site factors. Models for predicting slash consumption, duff consumption, etc. were discussed as well. An BUILDING T H E EXPERT SYSTEM / 84 attempt was also made to clarify the terminology of the domain. The expert constructed the models using the blackboard in the conference room. The second session with the expert was mainly devoted to reviewing the current status of the project. Computer produced block diagrams of the previous sessions were also reviewed (see Appendix 8). It became apparent that there was a lack of research results in terms of models for predicting fire impact. More would be known when the previous summers research would be analyzed. It was decided that this part of the expert system would be constructed primarily using the PFP to predict impacts. The expert would provide advice as to where these predictions could be improved. Slash loading was noted as being missing from the PFP model. 4.1.4 Knowledge Engineering Sessions with other Experts Professor Mike Feller of U B C and Bill Beese of Macmillan-Bloedel were two other experts that were consulted. Mike Feller is known as an expert in the ecological effects of slashburning. He authored a 1982 publication on the ecological effects of Slashburning in BC [Feller, 1982]. He appeared to be a very articulate expert and appeared quite able to explain concepts in the domain. Both sessions were conducted in his office at UBC. There were many distractions as students from the university came to BUILDING T H E EXPERT SYSTEM / 85 his office for advice. He appeared to be aware of all the slashburning research that was going on in BC. However, he was not of the viewpoint that it would be easy to predict the ecological effects of slashburning on any particular site. From his experience, he has observed a substantial amount of conflicting research results. Professor Feller stated in the second session that he would probably not have enough time to devote to the project. One session was done with Bill Beese of Macmillan Bloedel in Nanaimo. He is regarded as an expert in silviculture. He also has a lot of general expertise in prescribed burning. Most of this session was devoted to going over the conceptual model that had been developed with the other experts. The assumptions that had been made were reviewed. Much time was also spent discussing the typical decisions that have to be made when a site is slashburned. 4.2 Prescribed Burning Procedures In order to develop a system to model the ecological effects of slashburning, it is first necessary to study the procedures and decisions involved. This depends on the objectives of the decision maker. Hence in this section, the motivations of the different players, the information needs and types of BUILDING T H E EXPERT SYSTEM / 86 decisions are analyzed. Figure 13 shows the decision process for the prescribed burning problem. This figure shows the information needed and the sequence in which the decisions are made. The decision making process starts when the land is initially evaluated as to its value for commercial timber harvesting. Other possible uses include wildlife, recreation and tourism. If a parcel is deemed suitable for commercial harvesting then depending on visual concerns or environmental concerns, it is either selected for clearcutting or selective cutting. The site is mapped, classified ecologically and a cutblock is then laid out for the company to harvest. This cutblock should be laid out so that a site treatment can be applied uniformly over the site once it has been harvested. The size of the cutblock should also fit the needs of the company that wants to harvest it. In the BC interior, many of the cutblocks are quite small in order to accommodate the needs of the small-sized logging companies. A pre-harvest silvicultural prescription form must be submitted to the government at this point. The site needs to be classified according to the Biogeoclimatic Ecosystem Classification (BEC) system in order to use this form. Stocking standards for replanting the site must also be noted. These standards must be met in order to conform to Section 88 of the Silviculture Act. The BUILDING T H E EXPERT SYSTEM / 87 COMPANY FORESTER RESOURCE ASSISTANT \\ t l E l o b a r r e g u l r e a e n t s r e g i o n a l p r i o r i t i e s I \" \\ Doteruine bianagenent bjectlves f o r area f —\\ Assess side p o t e n t i a l hi r v e s t l n g s l t e l s p b a r a c t e r l a and c l a s s i c / s i t e l e a s i t e b b a r a c t e r i s t l c b Lay oat cutDlocx SILVICULTDHAt FICEH Evaluate S i t e lPreparatlon| A l t e r n a t i v e p r e s c r i p t i o n aXa s i t e data i t e and i n i t i a l f e a s i b i l i t y data Determine b a r v e s t i n g l e t b o o c a t b l o c k s p a c i t i c a t i o n s p r e s c r i p t i o n Pre-Barvest S i l v l c u l t u r a l P r e s c r i p t i o n S i t e P r e p a r a t i o n Guide PISE PI OTECTION r e g i o n a l knavledge 1 Determne Z e e s l b l l i t r ot t i r e h i s t o r i c a l veatber data F i r e Veatber Database BUSKER burn p l a t d a i l y veatber l n f o r a a t i o n E i e c n t e burn plan FIGURE 13 Prescribed Burning Decision Process Model BUILDING T H E EXPERT SYSTEM / 88 standards are set for each ecosystem association by the forest region. A site treatment method is tentatively identified in order to start anticipating how the site should be prepared for planting. The way the slash is treated is critical to how the site is to be prepared. If the site is to be burned then slash has to be left on the site, preferably scattered uniformly rather than in piles so as to achieve a uniform burn coverage on the site. A set of silvicultural objectives should also be drafted at this stage. These could include the objective of sanitizing the site from insects, improving the soil temperature, achieving a minimum number of planting spots or enabling easy planting access. An initial feasibility assessment is done to make sure that it is possible to meet the silvicultural objectives with the preferred treatment. In the case of burning, an initial estimation is done to determine what the burn objectives would have to be in order to meet the silvicultural objectives. Then an assessment must be made of how easy it would be to meet these burn objectives. Once the site is harvested, the site treatment options must be reevaluated. This is done using the Site Preparation Guide (form FS 117). The user of this form is led through the decisions necessary to evaluate the different site preparation alternatives. Risks of burning are evaluated too. There is also an economic assessment of the cost of the different treatments and an assessment BUILDING T H E EXPERT SYSTEM / 89 of the site sensitivity to fire. If the site is to be burned then a burn prescription must be defined as well. These burn prescriptions are usually developed in the winter before the site is to be burned. From the burn objectives, a set of moisture indices (Duff Moisture Codes, Drought Codes and Fine Fuel Moisture Codes) for burning needs to be determined because the moisture content and moisture gradient of the soil correlate with the impact that fire will have on a site. These moisture indices are usually derived by using the Prescribed Fire predictor (PFP). Alternatively, the user could look at previous burns in the area. The site sensitivity to fire must always be kept in mind when a prescription is developed. For several areas of the province, site sensitivity to fire keys have been developed. The user of these keys is able to obtain as output a qualitative rating of how sensitive the site is to fire. One burning manual has related these sensitivity ratings to PFP impact levels [Anonymous, 1985]. The forest manager can use other sources of information to complement the site sensitivity to fire keys. The publication, \"Autecological Characteristics of Selected Species that Compete with Conifers in British Columbia: A Literature Review\" [Hauessler and Coates, 1986] provides information on the response to burning of selected vegetation species. The regional ecologist and silviculturalist might be consulted as well. BUILDING T H E EXPERT SYSTEM / 90 Most prescriptions are also run through a weather database program in order to predict the chances of obtaining a particular prescription in a given week. The manager might need to reassess the situation if the probability of achieving a prescription is quite small. Once the decision is made to burn, a burning permit must usually be obtained. The district office would make sure that the chance of escape is acceptably small. The district staff might also provide some input into the burning prescription. Of course, there may be many given sites scheduled for burning in a particular area. The stock that is coming in and the potential productivity of the site will affect the decision of which sites to burn first and whether to burn at all. If the site is to be burned, then a decision must be whether to extensively document this burn as part of the Prescribed Fire Assessment Research. When a site is documented for this database, then a large number of site measurements must be made. These measurements involve sampling the soil and measuring the slash within the different size classes at various points within the site [Trowbridge et al., 1987]. The results of all these measurements are filed with the Protection Branch. At present, this data is partially analyzed but the results of this analysis is not given to the people who conducted the burn or the people who made the slash measurements. Even if the site is not documented as a research burn, a Prescribed Fire BUILDING T H E EXPERT SYSTEM / 91 Analysis form must be completed. Site measurements pre and post burn are noted on this form but the level of detail is much lower than the Prescribed Burn Assessment form. Currently, there does not appear to be any procedures in place to evaluate how well a site treatment worked. If the planting was a total failure then the forest company could be asked to replant. However, besides the stocking standards guidelines, there are no firm policies in place to ensure that the quality of the restocking will be good. 4.3 Conceptual Model of Fire Effects To understand the fire effects problem, it is necessary to have a conceptual model of how fire affects site properties which in turn affect the overall productivity of the site. Figure 14 illustrates the conceptual model developed in this project. In the expert systems field, it is actually the knowledge engineer's conceptual model that is implemented in the system, not the expert's [Addis, 1987]. However, it is the goal of the knowledge engineer to produce a conceptual model that is as close as possible to the expert's conceptual model. In this model, fire is a process which acts upon the different properties of the site. Fire affects the pre-burn vegetation complex and pre-burn soil characteristics such as duff thickness and mineral soil exposure. The impact of fire on these two sets of factors is determined by the burning weather FRE-BITB1\" VEGETATION COMPLEX IEE-BURJ son. CHARACTERISTICS -fall tnlclness -aineral soil BIJOSttlB 7 IRE ST-BURH VEGETATIOH COMPLEX POST-RURI SOIL dARACTEXISTICS -doit tbiciness -nlneral s o i l eiposure SITE FACTORS AFFECTITO FIRE SEVERITY -s lash loading -slope -dull loistare code -drDuglt code 1THER IIDIRECT GROlTH FACTORS -root rot BURIIIG COIDITIOHS -Bind -leather flaring burn IRE-BURN DIRECT iROVTB FACTORS -light -aaisttue -nutrients -teaperature PHT3ICAL. CHEMICAL AID BIOLOGIC SOIL PROCESSIS POST-BDRI DIRECT GIOTTE FACTORS - l i g a t -aoistaie -nutrients -teiperatare BUILDING T H E EXPERT SYSTEM / 93 conditions and other site factors which affect fire severity. The post-burn vegetation complex and the post-burn soil characteristics affect the direct and indirect growth factors for the site because of their effect on the soil processes of leaching, decomposition and erosion. These growth factors in turn affect the growth of the tree seedling. 4.4 Motivations of Groups Involved i n Prescribed Burning There are several different groups involved in the prescribed burning decision framework. There are the research personnel, the regional and district management within the Forest Service, the logging companies, and the crews that do the burning. The motivations of all these groups must be considered because they all could affect the success of an expert system in the prescribed burning domain. The objectives of the Prescribed Fire Research Advisory Committee (PFRAC) have already been mentioned in a previous section of this thesis. This group's main goal is to see that prescribed fire is used properly in BC. Some of the public is looking for an opportunity to condemn prescribed fire even though the alternatives such as chemical or mechanical treatments are probably more detrimental to the environment. A new focus in prescribed burning research has been on smoke management. It is now a priority in many areas to burn when the smoke ventilation will be best which is not always the best conditions BUILDING T H E EXPERT SYSTEM / 94 for the optimal preparation of the site. It should be mentioned that prescribed fire has almost been legislated out of existence in Oregon because of smoke management problems. Many of the goals of the PFRAC are also similar to those of the Forest Service's regional and district management. However, in the Forest Service the emphasis is on implementing policies uniformly and staying within budget. In some forest regions, the Forest Service does most of the burning. This occurs when smaller companies are involved. In these cases, the government also assumes all the costs of reforestation. However, the regional and district management is also committed to staying within budget. For example, they might not be able to pay overtime to burning crews when all of the cutblocks' optimal burning prescription windows coincide. The companies involved in logging have differing views concerning the whole reforestation process. The smaller companies appear to take a shorter-term view and are mostly interested in just meeting the standards set by the Silviculture Act. The larger companies appear to have a much stronger commitment to the quality of the reforestation process. These companies usually harvest under tree farm licenses. Their logging licenses are usually for longer time periods and are usually renewable. This means that the same company will be harvesting an area rotation after rotation. Therefore, the logging companies involved are motivated to take a longer-term view of the reforestation process. In fact, wood volume forecasting models are sometimes BUILDING T H E EXPERT SYSTEM / 95 generated so that the needs of the sawmills and the wood production from the site (which include wood from spacing and thinning) are in balance. Another issue involved in the reforestation process is that the length of time necessary to generate optimal seedling stock. Since the tree nurseries in the province have been privatized, it has been possible to obtain seedling stock that is more mature and is of higher quality. Companies now order their stock several years in advance to ensure the stock has reached optimal maturity. Therefore, the harvesting and replanting schedules of the various sites must be coordinated to ensure the availability of suitable seedling stock. 4.5 Structure of the Expert System The prototype of the Fire Effects expert system was implemented in VP-EXPERT. This tool is generally not considered to be a good shell to use on large projects partly because of its inherent limitations in terms of rule base size and partly because of its slow speed [Press, 1988]. However, it does allow some types of graphics and it allows hypertext. The Fire Effects expert system knowledge base is segmented into modules. For each separate growth factor, at least one module is used. Figure 15 shows the structure of the Fire Effects expert system. Besides the growth factor modules, there are some additional modules that serve to obtain parameters that are needed for several different modules. It was advisable to obtain most BUILDING T H E EXPERT SYSTEM / 96 FIREFFEC D e t s r m n e g e n e r a l s i t e p a r a n e t e r s and a c c e s s s i t e i n t o n a t i o n I r o n database i FUELCHAR D e t e m i n e f a c t o r s a f f e c t i n g f i r e s e v e r i t y and p r e d i c t d u f f consumption, m n e r a l s o i l exposure and s l a s h consumption • VEG 1 - 7 D e t e m i n e i n i t i a l v e g e t a t i o n complex i n c l u d i n g a e l g t i t and d e n s i t y ROOTROT Assess degree o l i m i t a t i o n of f u n g n • SOILNUTR Assess degree of I m i t a t i o n o l s o i l n u t r i e n t s SOILTEHP Assess degree of i m i t a t i o n of s o i l t e n p e r a t u r e i AIRTEMP Assess degree of i m i t a t i o n of a i r t e n p e r a t u r e V HYGROTOP Assess degree of i m i t a t i o n of s o i l n o i s t u r e \" \" ' \" ~ ~ • \"\"\"\" ' \" EVALLItl D e t e m i n e v h i c n i s nost l i n i t l n g f a c t o r i n presence or absence of f i r e • FIGURE 15 Structure of the F i r e E f f e c t s Expert System BUILDING T H E EXPERT SYSTEM / 97 of these parameters all at the same time rather than asking for them only when the system needs them. This was done in order to improve the flow of the system dialogue and to allow the user to concentrate on the outputs of the system. Typical parameters that are prompted for by the system include information on the ecological classification of the site, the slash loading, the duff depth, and the presence or absence of an Ah layer (organic layer in mineral soil). The output of the system is a determination of whether fire is beneficial or detrimental for the site. An explanation as to how fire will influence each growth factor is also given. The overall structure of the system is to have it function as a management gaming and learning tool. The users are encouraged to experiment with several different burning prescriptions to see how each are predicted to affect the site. The output of the system is not viewed as the most important part of the system, but rather the learning process involved in seeing how fire will affect a site is the most important part. 4.6 Typical Interaction with the System The typical user of the system is a forest manager who has received an education in forestry but is only vaguely aware of how fire influences site factors. This user would be the person who is developing the fire prescription BUILDING T H E EXPERT SYSTEM / 98 to be used to burn the site. The system could also be used as an auditing tool to check parameters on certain prescriptions that have been executed. The first series of screens lets the user specify all the background information needed for the site (see Appendix 9). This information includes data concerning the ecosystem classification for the site. The F U E L C H A R module then prompts the user for moisture codes and slash loading parameters in order to obtain a prediction of the duff consumption and mineral soil exposure following the burn. The user is then lead through the vegetation module, the root rot module, etc. The dialogue structure for each module is very similar in format. For each growth factor, the system describes the principal site characteristics that affect it. The limiting values for that factor, before and after fire, are then determined. Questions that are not applicable for an area are not asked. For example, only certain types of root rot are problems in some parts of the province. The system will only ask for information if it is pertinent to the particular site. As much dialogue as possible is incorporated into the system to make sure that the user understands the output and the limitations of the system's predictions. A graph of the limiting values of all the growth factors is presented in the E V A L L I M module. The user is then prompted to see if he/she wants to vary the moisture codes to see how different severity fires will affect the ecology of BUILDING T H E EXPERT SYSTEM / 99 the site. If so, then the system redisplays all the modules, but the user is not permitted to change data other than the moisture codes. 4.7 Comparison of Different Expert System Software There are many expert system development tools on the market today. These products range from user-friendly expert system shells such as VP-EXPERT to logic-based programming languages such as PROLOG. Recently, there has been a trend toward very sophisticated expert system shells. In fact, the term \"knowledge programming environment\" better characterizes these types of systems. Thus, it might be that no software will be the right tool for every project. In fact, the features desired of a tool will change as the project progresses in the expert system lifecycle [Citrenbaum et al., 1988]. For example, features that are important in the feasibility stage will differ from the features that will be important in the knowledge acquisition phase and the implementation phase. For the feasibility phase of the project, the tool should have an easy to understand rule format. It should also be easy to develop the user interface to simulate how the final system might look. The knowledge acquisition and prototype development phase requires an extensive range of programming support features. These features could include BUILDING T H E EXPERT SYSTEM / 100 system checks for consistency and completeness of the knowledge base, a debugging facility, a. bridge to external databases and languages, knowledge dictionary support and a knowledge base browsing facility. The implementation phase of the project requires a different set of features. Portability is important as any system would probably have to implemented in several different hardware environments. The flexibility of the user interface is important too. Finally, the performance of the implemented system is very important because this will affect user acceptance of the system. For this project, VP-EXPERT was used as a development tool, and three other development tools were evaluated. VP-EXPERT was used for development work primarily because of its ease of use and flexibility. It is important to be able to quickly alter a prototype of the expert system between knowledge engineering sessions. The rule format employed in VP-EXPERT seems quite easy for the expert to understand. However, VP-EXPERT is a comparatively slow tool when used in a production environment. It also lacks the ability to incorporate forms of knowledge representation other than rules (i.e. frames or objects). Therefore, other tools such as TURBO-PROLOG might be better suited for a production environment. For this project TURBO-PROLOG and the expert system shells G U R U and NEXPERT were assessed through the use of demonstration packages and industry software reviews. Figure 16 compares the different products on a range of criteria. There are o o I 0) H H-01 O 3 O t-h \"3 O C H W TJ (D h r+ Co •< CO f t (0 3 o (D < (D (-• o 3 r+ O O i VP-EXPERT TURBO PROLOG GDRU NEXPERT Type of Tool ; Elaple rale-based shell rest aacrocaaputer version al PROLOG sophisticated rule-based sh e l l Sophisticated object-oriented shell Xnovlaflge B i p n t n t i t i o n s supported Bales Rales a rrases Rales Roles a Objects (fraaes) cost lor Basic systen -Jiso con. •11SO cdn -16000 cdn. -17000 cdn Bun-Tine aodule cost -1150 cdn. free -1300 cdn. -I1SQ0 cdn Usage i n the narlstplace very popular very popular soaevhat popular quite popular Peitoraance Slav very cast aodcrate aoaerace i i Graphics Capability goad very good good very goad Zase ol Learning . i vary easy hard aoderataly easy noderately easy o s BUILDING T H E EXPERT SYSTEM / 102 large price differences between the four systems. VP-EXPERT is one of the least expensive packages but it is also a very slow package [Press, 1988; Stoddard, 1988]. NEXPERT is quite sophisticated but is much more expensive than the other tools. Although the development environment of NEXPERT is easy-to-use, the production environment must be programmed almost completely in the C language. This makes it time-consuming to prototype a system interface. TURBO-PROLOG is very fast and inexpensive but it is cumbersome to program rules directly into the system. CHAPTER 5. DISCUSSION AND CONCLUSIONS 5.1 Discussion of Knowledge Engineering The knowledge engineering phase is the critical phase of any expert systems development project. As was mentioned in the previous chapter, the development of the Fire Effects expert system presented many difficulties of which the main ones are listed below: 1. Decision-making experience: The experts in the domain aren't accustomed to making the types of decisions that this system is attempting to model. The experts that were consulted for this project are primarily researchers who are normally accustomed to designing experiments to test their conceptual ideas and theories. The experts were very seldomly approached with a problem that involved predicting the ecological effects of a given fire. 2. Area-specific knowledge: Many of the experts have had experience in only a few different geographical areas. This appeared to bias the approach they used to reason about fire effects. 3. Commitment: It was noted in the previous chapter that the 103 SUMMARY AND CONCLUSIONS / 104 experts were all quite busy people. This is usually the case with experts in every field. However, in this project all of the experts did not report to the sponsor of the project. Therefore, it was difficult to develop the commitment that was needed for the project. It appeared that some sort of upper-level approval of the project would be beneficial so that the experts would feel that they were partly responsible for the success of the project. 4. Conflicting viewpoints: Each expert had diverging views on how fire affected site properties. Because of this, the knowledge engineer had to ensure that he was able to reconcile these conflicting views. Each expert also had an issue that he/she was particularly concerned with. For example, one of the experts was concerned that whether the site was winter-logged or summer-logged had more ecological impact a site than whether it was slashburned. However, the other experts did not judge this issue to be as significant. 5. Knowledge engineering time interval: The time gap between K E sessions caused some difficulties. Because there were several experts involved in the project, it was not possible to see each expert as frequently as would be desirable. Much time was also spent at the start of each session reviewing SUMMARY AND CONCLUSIONS / 105 the results of the previous session. 6. Lack of Familiarity with Technology: Most of the experts in the field were not very familiar with expert systems technology. This had some negative effects on the productivity of the sessions. This lack of familiarity was a hindrance when it was necessary to demonstrate the system and explain the flow of reasoning to the experts. 7. Uncertainty of Knowledge: The uncertainty of the knowledge in the domain caused problems in the knowledge engineering phase. Many experts did not know why certain phenomena occurred and were not sure about their own ideas regarding how different factors interacted. 8. Familiarity of KE with Fire Effects Domain: The knowledge engineer was not very familiar with the problem area. This was viewed as a serious problem because of the complex theories involved in fire effects models. Most of these factors acted to hinder the development of the expert system, however, there were many factors about this project which have a positive effect on the viability of the system. These are listed below: SUMMARY AND CONCLUSIONS / 106 1. Experts: They were researchers in the domain and they were also articulate people. All of them had advanced degrees and were used to presenting, defending and explaining their ideas. They did not react adversely to having their ideas challenged. However, it was sometimes difficult to obtain their intuitive ideas because they were not used to presenting ideas that they could not factually defend. 2. Practice in model development: The experts all had much practice developing conceptual models. Some of them preferred to take the lead in the construction of the model. 3. Expert's motivation: The experts were all committed to transferring technology to the field staff. Therefore, they were interested in any new approach to accomplish this transfer. Also, they were used to acting in a consultative role. • . ... Many different knowledge engineering techniques proved to be successful in this environment. However, the main point to note is that the technique had to be adjusted to each expert. Some experts required a more structured interview format than others. 1. Block diagrams; prototyping: It was important to have some SUMMARY AND CONCLUSIONS / 107 sort of product after each session. If the elicited knowledge could not be implemented in a prototype, then it was useful to produce some sort of a block diagram of the conceptual model of the expertise. This block diagram served two functions. First of all, it forced the knowledge engineer to review the elicited knowledge in his mind. It also served to refresh the expert's memory at the start of the next session. 2. Discussion of background issues: At several points during the series of sessions, the knowledge engineer shifted the discussion away from the conceptual modelling and discussed various issues concerning expert systems technology. These diversions served to allow the expert a break from intense thinking. 3. Alternative problem approaches: One thing that was apparent in the knowledge engineering sessions was the vast amount of work that was accomplished when the knowledge engineer presented a different way of approaching the problem. This forced the expert to do some deeper thinking. The experts seemed to enjoy participating in extensive brainstorming discussions about how to solve the problem. summary, there were positive and negative factors that affected the SUMMARY AND CONCLUSIONS / 108 success of the knowledge engineering phase of this project. Much of the work was accomplished in the sessions when the knowledge engineer was able to motivate the expert to do some intense thinking. For this to be possible, the knowledge engineer has to be very familiar with concepts in the domain. It was also necessary for the knowledge engineer to adapt the format of the session to suit each expert. 5.2 Discussion of Development and Production Environments The previous chapter summarized the characteristic features of four different software environments. Different features are required of an expert system environment at different phases of the project. At the feasibility and knowledge acquisition phases of the project, it is desirable that the environment is flexible and the coding format is quite readable so that the experts can understand it easily. It also has to be easily modifiable in order to demonstrate to the expert what effects changes in his/her reasoning will produce. VP-EXPERT was found quite suitable for most of the development work that occurred. Its hypertext and screen mapping features were an improvement over the traditional text scrolling display that most expert systems use. A missing feature was the ability to import graphics such as scanned images directly into the system. Many concepts in forest ecology are most easily explained using diagrams and it would be beneficial to have the ability to display these diagrams in an expert system. SUMMARY AND CONCLUSIONS / 109 Also, VP-EXPERT was lacking in sophisticated forms of knowledge representation. Because most of the entities in the forestry domain can be put into some sort of hierarchical structure, much time could have been saved if the development environment had incorporated frame, or object-like properties. Attribute values could have been specified at the class level. If necessary, these values could have been overridden at the object level. For example, much of the knowledge in the vegetation module could have been implemented in some sort of hierarchical structure. Separate classes could be specified for trees, shrubs, and herbs. Default tolerances to burn severity could then be specified at the class level instead of the entity level. A production or implementation environment needs its own set of features. It needs to be able to access or integrate easily with current and planned databases and systems. Some expert system tools have special links to different types of databases. For example, NEXPERT can interface directly with ORACLE databases. Other expert systems such as TURBO-PROLOG can be called up as executable modules from within other programs. This is important for the system to be successful as it has to be incorporated into other systems within the Ministry of Forests such as the system to document prescribed burns. Cholawsky (1988) indicates that many prototype implementation problems result from data problems. These are: voluminous data, nonautomated data, SUMMARY AND CONCLUSIONS / 110 incomplete data, erroneous data, and inappropriate data. Therefore, much thought should be given as to the data requirements of the expert system. Much of the information that the Fire Effects Expert system uses is currently recorded on various forms such as the Pre-Harvest Silvicultural Prescription form and the Site Preparation Guide. However, much of the information that is recorded on these forms is not recorded in a standardized format. For example, information concerning vegetation is recorded differently on both forms. The programming code should be locked out from the users; that is, they should not be able to access the source code once the system is implemented. Usually, the best way to do this is to have a compiler for the system. This has the dual effect of making the system run faster as well as hiding the actual program code from the users. Much effort has to be devoted to the user interface of the system. Many companies have found that 70% of the development effort goes into the interface of the system and only 30% goes into the reasoning part of the system [Berry and Broadbent, 1987]. Since most of the potential users of Fire Effects system have limited familiarity with computers, it is important that this system has a good user interface. The system also has to be portable across many different environments. A desirable feature would be the ability to distribute runtime versions inexpensively. This would make it easier for the forestry companies to adopt SUMMARY AND CONCLUSIONS / 111 the system. Another consideration is version control. It has to be easy to distribute new versions of the program. Finally, the microcomputer environment is preferable because with microcomputer-based systems there is no need for special communication facilities and expenses. 5.3 Discussion of the Prototype System The prototype of the fire effects expert system was developed in VP-EXPERT. This expert system shell is quite flexible and it incorporates a number of advanced features such as hypertext and graphic images. However, the graphics were cumbersome to the programmer and it was difficult to produce graphic objects. A simple bar graph requires a substantial amount of code to produce and is not generated very quickly on the computer when the code is executed. The system also runs slowly on anything less powerful than an IBM PC-AT. Typical response times for some operations such as generating a bar graph were in the order of thirty seconds to one minute on an AT computer. The limiting factor approach was easy to implement on the system. At least one module was devoted to the reasoning for each growth factor. A pre-burn limiting level is determined initially for each growth factor. The limiting level after the fire is then determined. Since these limiting levels are all based on a common numeric scale from 1 to 8, it is a very easy process to determine which is the most limiting growth factor. SUMMARY AND CONCLUSIONS / 112 Currently, the system only interacts on one level with the user, that is, each time the system is consulted, the user obtains the full dialogue. A desirable feature of the system would be the ability to provide different levels of user interaction. For example, a user with a strong academic background in forestry would probably desire a strong technical emphasis to the dialogue, whereas a forester with limited formal training would probably want the dialogue to explain each concept in simpler terms. In addition, the user should be able to adjust the level of detail of the system dialogue. The system is envisioned to perform many different functions. Its uses include: site management, education, a basis for group discussions, prediction/validation purposes, and standardization purposes. 1. Site preparation: A primary function of the system would be as a site preparation tool to assist foresters in developing fire prescriptions. The user would input all the information for a particular fire. The system then makes predictions as to the ecological effect of a particular fire. Much of the usefulness of the system derives from the ability of the user to easily vary several parameters within a fire prescription to test the effects of these changes on the system's predictions. This can be viewed as a sensitivity analysis on a particular fire. SUMMARY AND CONCLUSIONS / 113 2. Education: The education of forest managers is another important use for the system. Users can learn from the explanation of the reasoning that the system provides for its predictions. 3. Group discussion: This system could also serve as the basis for group discussions. The beliefs of the different experts could be discussed quite easily because contradictory views about how fire interacts with site factors would become quite apparent. This is because the knowledge of the different experts would be documented in the system and in the conceptual models that are developed. 4. Prediction I validation: The system can also be used to give a prediction based on the actual parameters of a fire that occurred. This is so the system prediction can be kept on file and compared to what actually occurs. This is seen as a good way to test the validity of the experts' predictions and improve the data/knowledge base. 5. Standardized reasoning: The system is also set up so that the same reasoning framework is standardized across the province. However, the system accesses a number of DBASE SUMMARY AND CONCLUSIONS / 114 files that contain specific information about each ecosystem association. These DBASE files could be modified by Research Branch personnel in the forest regions. Finally, it should be mentioned that most of the system's predictions are purposefully formulated using imprecise wording. This is to ensure that the user understands that the system is only producing crude estimates of fire effects and that the system's predictions should be treated with caution. 5.4 Discussion of Expert Systems i n Prescribed Burning There is considerable potential for expert systems technology in the prescribed fire domain. The previous section mentions the many different functions an expert system could perform. Essentially, the main benefit derived from the system would be an increased transfer of technology between the researchers and the field staff. Expert system technology is suitable for the prescribed burning domain for many different reasons: 1. Qualitative nature of decisions: Most of the decisions in the prescribed burning field are qualitative in nature. Many of the experts in the field use heuristics to aid in making their decisions. SUMMARY AND CONCLUSIONS / 115 Large Solution Space: Most decisions in the prescribed burning field involve a large number of possible solutions. Expert systems technology is particularly well suited for handling these types of problems. Diversity of Knowledge Sources: Much of the knowledge needed to make a decision is located in many different sources. These sources include: manuals and guides produced by the Forest Service, experts within the Forest Service and within academia, and databases on various computers. An expert system would essentially allow centralized access to most of the knowledge that is relevant to a decision. Scarcity of experts: There are very few experts in the prescribed burning domain. Most of these experts have extensive research commitments and have very little time to act in a consultative role for the field staff. Narrow breadth of domain: Most of the knowledge in the prescribed burning domain consist of very narrow, domain-specific knowledge. These factors make this domain ideal for expert systems technology. However, an expert system must be adaptable to a wide geographical area so that the system's benefits will outweigh its costs. SUMMARY AND CONCLUSIONS / 116 There are potentially many difficulties in applying expert systems technology to this field. These difficulties include: 1. Uncertainty of knowledge and complexity of domain: Most of the experts in the prescribed burning field are very unsure about how prescribed fire interacts with site factors due to the complex nature of the domain. Additionally, many of these experts have conflicting views. 2. Uncertainty of payoff: The benefits derived from the use of expert systems technology are very intangible. Substantive economic benefits won't be realized in the short term. 3. Lack of commitment of senior management: Experts systems technology requires the commitment of senior management. The development of an expert system requires many years and requires much time from the experts. In the prescribed burning domain, many of these experts are in different agencies. This means that the senior managers of these agencies must cooperate in order to properly develop an expert system. 4. User acceptance of technology: Potential problems could occur SUMMARY AND CONCLUSIONS / 117 regarding user acceptance of an expert system within this domain. Many of the potential users of the system are not very familiar with computer technology. Many different decision aids have been developed for prescribed burning domain. Each type of decision aid has its own set of advantages and disadvantages. The author of this thesis proposes that decision aids be categorized into six types: the \"cookbook\", the \"workbook\", the simple expert system, the detailed expert system, the mathematical modelling, and the expert system simulation approaches (see Figure 17). For illustration purposes, these six types have been plotted on an axis labelled \"Complexity of Technique\". Any approach can be made as complex or as simple as the designer wants. However, it is proposed that each approach is best suited for dealing with a problem at a particular complexity level. It should be noted that it is difficult to establish clear definitions of the six different types of approaches. The \"cookbook\" approach is currently used extensively by the Forest Service. This approach may be defined as the use of a limited number of factors in order to produce a small decision hierarchy. Usually, these decision hierarchies take the form of one page keys. Site Sensitivity to Fire keys are examples of this type of approach. Although these keys are easy to use, they do not appear to result in an appreciable increase in user understanding of the relationship between prescribed fire and ecology. SUMMARY AND CONCLUSIONS / 118 The \"workbook\" approach has not been used in the prescribed fire domain but it represents a viable approach to the ecological effects of fire problem. An example of the \"workbook\" approach is the \"Decision Making Profile for Vegetation Management Options\" [MacDonald, 1987]. This approach may be defined as a technique in which the user is asked a number of questions that are relevant to the decision. Usually, the reasons for asking each question are given. The user then combines the results of the questions using a procedure specified in the workbook. The result is a recommendation concerning a particular decision. This technique enables various factors to be weighted to arrive at a conclusion with the constraining factor being the amount of computation that it requires the user to perform. The author of this thesis suggests that the expert systems approach for this area can be categorized into two levels: simple and complex expert systems. A simple expert system is differentiated from a complex expert system on the basis of how it models physical processes. A complex expert system can explicitly model the external physical processes. It might also be able to model the time varying nature of the effects of fire on ecology. A simple expert system would just contain knowledge about how some site factors affect some other site factors. The mathematical modelling approach is best exemplified by the FORCYTE program. This program was developed by Dr. J . P. Kimmins of UBC [Kimmins, 1987] in order to model the flows of nutrients within an ecosystem. Although SUMMARY AND CONCLUSIONS / 119 not specifically designed for slashburning, it nevertheless could be used to model the effects of slashburning on ecology. This approach may be defined as the use of quantitative models to directly model physical processes. The final type of approach can be termed an expert system simulation. This technique would involve elements of mathematical modelling and expert system technology. It could provide interpretations of the modelling results and answer queries about the actual model itself. As Figure 17 illustrates, expert systems technology has the potential to address problems at a more complex level. However, we feel that it is best to progress gradually along the \"complexity of technique\" axis. Therefore, at this time, no attempt should be made to develop complex expert systems that model directly physical processes. Instead, simple expert systems should be developed to enable researchers and users to become more familiar with the technology. 5.5 Strategy for Further Expert Systems Research — Expert system technology is expensive to implement because of the knowledge engineering effort. Expert systems usually take a long time to develop. Not only does the knowledge engineer have to be spend time but so do the experts. The knowledge engineering effort could be a substantial drain on the expert's time while the benefits can be quite intangible. Expert systems might also require a considerable amount of maintenance once they are Low Complexity of tecnnlque COOKBOOK -aieapllflM By l i t e • e i i t l l i * l t y to r i r i ley -approach It m r - H - t i i •no alapla bat does net iced to a n t uaintanaiM Mderlylag physical factort t proeesata High &EBFU EXPERT BT8TE8 -generellly mart eaaplti lata sorkbool lyyroaeh -assert ayataa dees a l l the calealatiaaa -only cenidsrs factors tad loaa aat try ta • i l l i c i t l y aadal tsysical processes otnui rarsicu IOKLUIO -eieapllfled by FORCYTft syete -coaaUtri factor! and processes -eaa't give erplanatlan vomaor -ao carreat eieaple la prescribed (ire doaala -spproacn allots leigatlag factors to be asad -caa erplala purpose of each euestloa DETAILED EXPERT 8T&TEM -eeaelders processes as veil as factors -eaa s i mi dyaaalc ehaage la factor aita tlae 1 e states of astrleats soils processes aacb as leaching aad decdeposition are atfectlag It m i r r s r s m i i n i A f i o v -caa give aiplanatlen -caa aaeeer general eaestlon* about aadal i SUMMARY AND CONCLUSIONS / 121 implemented. This occurs because knowledge is seldom stable within a domain. Therefore, the following strategy is proposed for further expert systems research within the domain: 1. Future planning and coordination of expert systems development should be done at a high level. This allows for experts from different departments to be assigned to a project. It will also allow for the expert system to be integrated into current and future databases and systems. 2. It is possible to draw on academic work in the field of expert systems. Labour might be less expensive in this environment. Current research in the academic environment concerning automated K A tools has the potential to decrease the development time and expense of an expert system. 3. The emphasis should be on developing a number of small systems rather than one large system. In this way, it is possible to review the development strategy frequently in order to direct efforts towards the systems which will produce the most benefits. Also, the introduction of each new system will serve to gradually familiarize the users with expert systems technology. SUMMARY AND CONCLUSIONS / 122 4. Specific standards should be developed for all the information regarding site factors. These standard should be developed in conjunction with the expert system developers. The standards would help alleviate \"data problems\" that might occur (i.e. inappropriate data, erroneous data, missing data). 5. Establishing some contact with American researchers doing work in the forestry expert systems field [Reinhardt, 1987; Martin, 1987] would be useful to examine how they are approaching the same types of problems. 6. Research into prescribed burning should be coordinated with the development of expert systems. This is necessary because there are many \"gaps in knowledge\" in the domain. For example, the response of vegetation to fire is only known for a few ecosystems in the province. / 123 BIBLIOGRAPHY Addis, T. R., \"The Role of 'Explanation' in Knowledge Elicitation\", International Journal of Systems Research and Information Science, 1987, Vol. 2, pp. 101-110. Adrion, W.R., Branstad, M.A., and Cherniavsky, J.C. , 'Validation, Verification, and Testing of Computer Software\", A C M Computing Surveys. Vol. 14, No. 2, June 1982, pp. 159-192. Alexander, Martin E . , \"Calculating and Interpreting Forest Fire Intensities\", Canadian Journal of Botany. Vol. 60, No. 4, 1982, pp. 349-357. Anonymous, A Guide to Prescribed Broadcast Burning in the Vancouver Forest Region. BC Min. For., 1985. Beerel, Annabel C , Expert Systems: Strategic Implications and Applications. Ellis Horwood, 1987. Benbasat, Izak and Jasbir S. Dhaliwal, \"A Framework for the Validation of Knowledge Acquisition\", 3 rd Knowledge Acquisition for Knowledge-Based Systems Workshop, 1988, pp. 1-1 - 1-18. Berry, Dianne C. and Donald E . Broadbent, \"Expert Systems and the Man-Machine Interface. Part Two: The User Interface\".Expert Systems. February 1987, Vol. 4, No. 1, pp. 18-28. Boose, John H. , \"A Survey of Knowledge Acquisition Techniques and Tools\", 3rd AAAi-Sponsored Knowledge Acquisition for Knowledge-Based Systems Workshop. November 1988, pp. 3-1 - 3-23. Brady, Nyle C , The Nature and Properties of Soils. 9 th edition, Macmillan, 1984. Breuker, Joost and Bob Wielinga, \"Use of Models in the Interpretation of Verbal Data\", in Knowledge Acquisition for Expert Systems: A Practical Handbook, edited by Alison L. Kidd, Plenum Press, 1987, pp. 17-42. Chandler, Craig, Cheney, Phillip, Thomas, Philip, Trabaud, Louis, and Dave Williams, Fire in Forestry: Volume II Forest Fire Management and Organization. John Wiley & Sons, 1983. Chandrasekaran, B., \"Expert Systems: Matching Techniques to Tasks\", in Artificial Intelligence Applications for Business, edited by W. Reitman, Ablex Publishing, 1984, pp. 41-64. / 124 Cholawsky, Elizabeth M. , \"Beating the Prototype Blues\", A l Expert. December 1988, pp. 42-49. Citrenbaum, Ronald, Geissmann, James R. and Roger Schultz, \"Selecting a Shell\", A l Applications in Natural Resource Management. Vol. 2, No. 1, 1988, pp. 3-12. Conklin, Jeff, \"Hypertext: An Introduction and Survey\", Computer. September 1987, pp. 17-41. Curran, Mike, \"Short and Long-Term Effects of Slashburning on Soil Properties, Tree Growth and Nutrition on Some Coastal B.C. Sites\", Northwest Forest Fire Council Annual Meeting Proceedings. 1986. Davies, Miranda and Simon Hakiel, \"Knowledge Harvesting: A Practical Guide to Interviewing\", Expert Systems. Vol. 5, No. 1, February 1988, pp. 42-50. Dhaliwal, Jasbir S. and Izak Benbasat, \"A Model for the Empirical Investigation of Knowledge Acquisition Techniques\", Proceeedings of the Second International IFIP/IFAC/IFORS Workshop on Artificial Intelligence in Economics and Management. January, 1989. Dyer, Rosemary M . , \"Adapting Expert Systems to Multiple Locations\", A l Applications in Natural Resource Management. Vol. 3, No. 1, 1989. Feller, M . C , The Ecological Effects of Slashburning with Particular Reference to British Columbia: A Literature Review. BC Min. For., 1982. Fullerton, J .M. and D. L . Martell, \"Decision Analysis for Prescribed Burn Planning\", in Forest Fire Management Symposium, edited by B.J. Stocks, R.G. Elliott and J.D. Walker, Canadian Forestry Service, 1984, pp. 88-91. Gordon, Sallie E . , Gill, Richard T. and Thomas A. Dingus, \"Designing for the User: The Role of Human Factors in Expert System Development\", A l Applications in Natural Resource Management. Vol. 1, No. 1, 1987, pp. 35-45. Haeussler, S. and D. Coates, Autecological Characteristics of Selected Species that Compete with Conifers in British Columbia: A Literature Review. BC Min. For., 1986. Hamilton, Evelyn H . and H . Karen Yearsley, Vegetation Development after Clearcutting and Site Preparation in the SBS Zone. BC Min. For., 1988. / 125 Harmon, Paul and David King, Expert Systems: Artificial Intelligence in Business. John Wiley & Sons, 1985. Hart, Anna, Knowledge Acquisition for Expert Systems. Kogan Page, 1986. Hawkes, Brad and Bruce Lawson, \"Documentation of Prescribed Fire Behaviour and Effects on Forest Fuels\", in Prescribed Fire-Forest Soils Symposium Proceedings, edited by R.L. Trowbridge and A. Macadam, BC Min. For., 1982, pp. 93-114. Hawkes, B.C. and B.D. Lawson, \"Prescribed Fire Decision-Aids in B.C. : Current Status and Future Developments\", Northwest Forest Fire Council Annual Meeting Proceedings. 1986. Hawkes, Brad, Lawson, Bruce and John Parminter, \"Prescribed Fire and Fire Effects Research in Forest Resource Management in B.C.\", Unpublished BC Forest Service document, 1984. Hayes-Roth, F. , \"Knowledge-Based Expert Systems\", IEEE Computer. October 1984, pp. 263-273. Herrod, R. and M . Smith, \"The Campbell Soup Story: An Application of A l Technology in the Food Industry\", Texas Instruments Engineering Journal. Vol. 3, No. 1, 1986, pp. 16-19. Hoffman, Robert R., \"The Problem of Extracting the Knowledge of Experts from the Perspective of Experimental Psychology\", A l Applications in Natural Resource Management. Vol. 1, No. 2, 1987, pp. 35-48. Kidd, Alison L . , \"Knowledge Acquisition-An Introductory Framework\", in Knowledge Acquisition for Expert Systems: A Practical Handbook, edited by Alison L. Kidd, Plenum Press, 1987, pp. 1-15. .-Kimmins, J.P., Forest Ecology. Macmillan Publishing, 1987. Klinka, K. Guide for the Tree Species Selection and Prescribed Burning in the Vancouver Forest District. Second Approximation, BC. Min. For., 1977. Klinka, K Green, R. N. , Courtin, P. J . , and F. C. Nuszdorfer, Site Diagnosis. Tree Species Selection, and Slashburning Guidelines for the Vancouver Forest Region. BC Ministry of Forests, 1984. Klinka, K., Green, R.N., Trowbridge, R.L. and L . E . Lowe, Taxonomic Classification of Humus Forms in Ecosystems of British Columbia. First Approximation, BC Min. For., 1981. / 126 Kourtz, Peter, \"Expert System Dispatch of Forest Fire Control Resources\", A l Applications in Natural Resource Management. Vol. 1, No. 1, 1987, pp. 1-8. Latham, Don, \"VP-Expert: A Review\", A l Applications in Natural Resource Management. Vol. 2, No. 1, 1988, pp. 60-61. Macdonald, Vanessa, Decision Making Profile for Vegetation Management Options. BC Min. For., 1987. Mackinnon, Andy, \"BCFS Research Branch In-House and FRDA Funded Prescribed Fire Effects Research\", unpublished Forest Service document, 1988. Martin, Robert E . , \"Fire Effects, Education and Expert Systems\", U.S. Department of Agriculture Gen. Tech. Rep. PSW-101. 1987, pp. 140-144. Mathieson, Kieran, \"Prototyping Expert Systems\", A l Applications in Natural Resource Management. Vol. 2, No. 2-3, 1988. McDermott, J . , \"Rl: A Rule-Based Configurer of Computer Systems.\", Artificial Intelligence. September 1982, pp. 39-88. McGraw, Karen and Karan Harbison-Briggs, Knowledge Acquisition: Principles and Guidelines. Prentice-Hall, 1989. McGraw, K. L. , and M . R. Searle, \"Knowledge Elicitation with Multiple Experts: Considerations and Techniques\", Artificial Intelligence Review. No. 2, 1988, pp. 31-44. Minsky, M. , \"A Framework for Representing Knowledge\", The Psychology of Computer Vision, edited by P. Winston, McGraw-Hill, 1975. Muraro, S.J., \"Prescribed-Fire Impact in Cedar-Hemlock Logging Slash\", Can. For. Serv. Publ. No. 1297. 1971. Muraro, S. J . , \"The Prescribed Fire Predictor\", Fire Control Notes. January, 1977, pp. 26-29. Nelson, T. H . , \"Getting It Out of Our System\", Information Retrieval, edited by G. Schechter, Thompson Books, 1967. Newell, A. and H . Simon, Human Problem Solving. Prentice-Hall, 1972. Newquist, Harvey P. Ill, \"Struggling to Maintain\", A l Expert. August 1988, / 127 pp. 69-71. Parsaye, Kamran, \"Acquiring and Verifying Knowledge Automatically\", A l Expert. May 1988, pp. 48-63. Parsaye, Kamran and Mark Chignell, Expert Systems for Experts. John Wiley & Sons, 1988. Pedersen, Ken, \"Connecting Expert Systems and Conventional Environments\", A l Expert. May 1988, pp. 26-35. Pojar, J . , Klinka, K., and D.V. Meidinger, \"Biogeoclimatic Ecosystem Classification in British Columbia\", Forest Ecology and Management. No. 22, 1987, pp. 119-154. Press, Larry, \"Eight-Product Wrap-Up: PC Shells\", A l Expert. September 1988, pp. 61-65. Quillian, M.R., \"Semantic Memory\", Semantic Information Processing, edited by M . Minsky, MIT Press, 1968, pp. 227-270. Radloff, David L . and Richard F. Yancik, \"Decision Analysis of Prescribed Burning\", Fire Management Notes. Vol. 44, No. 3, pp. 22-29. Raybould, Steven and Tom Roberts, \"A Matrix Approach to Fire Prescription Writing\", Fire Management Notes. Vol. 44, No. 4, 1983, pp. 7-10. Reinhardt, Elizabeth, \"An Expert System for Designing Fire Prescriptions\", U.S. Department of Agriculture Gen. Tech. Rep. PSW-101. 1987, pp. 223-225. Rust, Marc, \"White Pine Blister Rust Hazard Rating: An Expert Systems Approach\", A l Applications in Natural Resource Management. Vol. 2, No. 2-3, 1988, pp. 47-50. Smith, Donald L . , \"Implementing Real World Expert Systems\", A l Expert. December 1988, pp. 51-57. Starfield, A . M . and A.L. Bleloch, \"Expert Systems: An Approach to Problems in Ecological Management that are Difficult to Quantify\", Journal of Environmental Management. No. 16, 1983, pp. 261-268. Stefik, M. and D. G. Bobrow, \"Object-Oriented Programming: Themes and Variations\", A l Magazine. Vol. 6., No. 4, 1986, pp. 40-64. Stoddard, Joan, \"VP-Expert v.2.0\" A l Expert. October 1988, pp. 73-77. / 128 Thieme, Ronald H. , Jones, Don D., Gibson, Harry G., Flicker, Jon D. and Reisinger, Thomas W., \"Knowledge-Based Forest Road Planning\", A l Applications in Natural Resource Management. Vol. 1, No. 1, 1987. Trowbridge, R., Hawkes, B., Macadam, A., and J . Parminter, Field Handbook for Prescribed Fire Assessments in British Columbia: Logging Slash Fuels. BC Min. For., 1987. Turban, Efraim, Decision Support and Expert Systems. Macmillan Publishing Company, 1988. Vandierendonck, A. \"Inferential Simulation: Hypothesis-Testing by Computer Simulation\", Nederlands Tijdschrift voor de Psycholgie. Vol. 30, 1975, pp. 677-700. Walters, John and Norman R. Nielsen, Crafting Knowledge-Based Systems. John Wiley & Sons, 1988. Williams, Gregg, \"HyperCard\", Bvte Magazine. December 1987, pp. 109-117. Wolfgram, Deborah, Dear, Teresa and Craig S. Galbraith, Expert Systems for the Technical Professional. John Wiley & Sons, 1987. / 129 APPENDIX 1 - PROJECT PROPOSAL /130 FACULTY OF COMMERCE AND BUSINESS ADMINISTRATION UNIVERSITY OF BRITISH COLUMBIA PRESCRIBED FIRE EXPERT SYSTEMS PROJECT PROPOSAL Introduction The purpose of this document is to provide a brief outline of a research project that will investigate the feasibility of applying expert systems technology to the prescribed burning (PB) field. The goals of this project, the manner in which the study will be done, and the required resources are also described in this document. Nature of the Problem Area Prescribed burning refers to the use of fire as a site preparation tool to achieve specific land management objectives. In British Columbia, it is used primarily to prepare logging sites for replanting. Although mechanical and chemical methods can be used to prepare a site, prescribed burning is usually the least expensive site preparation tool. However, prescribed burning is not suitable for all sites because it can have very deleterious effects on some ecosystems and because it can be unsafe to conduct a prescribed burn under certain environmental conditions. Much research has been conducted concerning the complex decision-making process of how severe a burn should be and whether the current environmental conditions are acceptable for a burn. At this time, various expert system projects are currently underway in the U.S. and Canada to investigate the suitability of expert systems to this problem domain. Project Goals The primary goal of this project is to investigate the suitability of expert systems technology for tbe prescribed burning decision-making process. The potential benefits of expert systems in this area will be examined as well as problems associated with knowledge engineering. From the results of this project, it should be possible to recommend the direction for future work in this area. /131 Background In early 1988, Bernie Todd of the Canadian Forestry Service commenced work on an expert system called \"The PB Planner\". The general outline of this system is described in Appendix 1. This expert system was envisioned to include just about all the decisions that relate to prescribed burning on clear-cut areas. It is expected that the system will be implemented in three stages. Stage 1 involves the decisions that relate to a specific site whereas Stage 2 and Stage 3 relate to the regional scheduling of burns on different sites and simulating fire behaviour, respectively. Presently, Bernie Todd is working on the Silvicultural Objectives and Site Sensitivity to Fire Modules of the PB Planner system. Within the next several months, he expects to complete a prototype system using experts from the Prince George region. From this prototype, he then expects to extend the system to the rest of B.C. and Canada. The Management Information Systems division in the Faculty of Commerce and Business Administration at UBC became interested in the use of expert systems in this area through discussions with people in the FEPA (Forestry Economics and Policy Analysis) project and the Protection Branch of the BC Ministry of Forests. This project interested us because expert systems are an important development in information systems and, therefore, some of our current research is in this field. Systems Approach to this Problem An essential component of an expert systems project is an analysis of the \"systems\" issues that pertain to a particular system. It is important that the objectives of the system and how they relate to the users are examined. A system can be a technical success but an operational failure. With expert systems there are several additional issues to be considered. These issues include what will be the sources of expertise and how will the knowledge base be maintained once the system is operational. Another concern is how this system will fit in with existing procedures and systems. Issues Concerning Expert Systems Technology Expert systems technology is only suitable for certain types of problems. These problem domains primarily involve situations where there are only a few experts in the /132 area and there are many people who could benefit from their expertise. It is important to note that an expert system relates to a well defined and narrow domain that does not require general common knowledge. These problem domains are also characterized by the existence of many different factors that must be considered in making a decision. In these problem domains the experts use heuristics or \"rules of thumb\" to combine these factors in their reasoning process. Usually, the experts will apply different weights to these factors depending on the situation they are analyzing. Although manuals embody lots of useful knowledge and, in particular, list the factors that should be considered in a particular problem, their use can be problematic. First, people may have difficulties in integrating knowledge from varied sources and second, manuals do not accurately describe how to combine the various factors. However, it is generally possible to simulate this reasoning process in an expert system. An essential component of the development of an expert system is the use of knowledge engineering techniques to elicit expertise from the experts. Although many automated tools have been developed to facilitate this process, they are only suitable for relatively simple types of problems. For most types of problems, a knowledge engineer must interview the expert to determine the reasoning process the expert uses. In certain situations, it is useful to provide the expert with test problems and then examine the experts reasoning process in each of these test problems. The knowledge engineering process is generally an iterative process whereby the knowledge engineer models the reasoning process of the expert by constructing a prototype of the system. The knowledge engineer shows this prototype to the expert after each knowledge engineering session and the expert is queried as to how closely the system models his reasoning process. The knowledge engineer refines the prototype according to the results of each session. This prototyping process is necessary because most experts have problems describing their own reasoning processes. Only after the knowledge engineer and expert are satisfied with the prototype, should the detailed programming of the system be done. It is at this stage that a decision is made as to exactly how the system will be programmed. Integration with existing systems, maintainability of the knowledge base and efficiency of the system will be concerns at this time. It is important that the programming environment decision is made after most of the knowledge engineering has occurred. Otherwise, the selection of the programming environment will influence the way the reasoning process is modeled. The final stage of an expert systems development project is the operational testing /133 of the system. This testing usually involves the comparison of the performance of experts (including ones not interviewed) and the expert system on test cases. From the results of this comparison, it should be possible to determine how closely the system replicates the knowledge of the expert and how similar the expert reasons relative to other experts. It should also be remembered that the expert system performance as assessed in this way might not reflect the performance in reality. This is because, in practice, the user does not need an' expert system to handle simple problems and hence, tbe expert system will probably only be used for problems that are considerably more complex than thee test scenarios. The user should also be queried at this stage to see that he is familiar with the terms that the system uses and whether the human interface of the system is adequate. Proposed Problem We propose to develop an expert system that will encompass a well defined part of the overall PB Planner system. This expert system relates to the Fire Effects module. This system would focus on the short and long term effects of fire on the productivity of a particular site. As we view it, the inputs to this system will comprise of the following: -all the fuel parameters that are necessary to assess the burn severity (fuel loading and continuity, fuel moisture, fuel species type and size class distribution) -the species type, coverage, and characteristics of the existing site vegetation -the species type and characteristics of the species to be planted -various soil and site characteristics The system would then calculate the response of various ecological parameters to the severity of the burn. The erodibility of the soil, the long and short term nutrient availability, and the long and short term effects on competing vegetation would be some of the parameters that would be calculated. Using these intermediate parameters, the system would then be able to derive an estimate of the effect of a particular fire on long and short term site productivity. Appendix 2 describes a preliminary conceptual model for this expert system. An important concern in this project will be tbe integration of this work with the work presently being done by Bernie Todd. It may eventually be necessary to reconsider the structure of the PB Planner system. It will also be beneficial to /134 It would be advisable to bring Bernie Todd out to Victoria to discuss the overall system and how this project will integrate with the overall project. At this time, it might also be beneficial to reevaluate the structure of the PB Planner system. Finally, some funds may be required to purchase software (expert system 'shells') and possibly to upgrade a microcomputer's memory to be able to run these shells. It is assumed that in meetings with experts there will be access to microcomputers on which the knowledge base prototype will be demonstrated and examined for feedback. Immediate Work Plan The first step in the work plan would be to meet in Victoria with a few experts as well as the project sponsors. At this meeting, the objectives of the system would be clarified. A conceptual model of the system should be discussed at this time and users and experts be identified. A reassessment of the viability of expert system technology for this problem domain could also be done at this time. The knowledge engineering process would occur over the next several months. This will include knowledge acquisition and prototype implementation. It is planned to complete this process during the summer. At the end of the summer, a meeting with Bernie Todd could be scheduled. The prototype of the expert system will also be demonstrated to the project sponsors at this time. Mike Johnston Yair Wand June 16,1988 /135 G e n e r a l FLQhCHAST o f t n e P i P l a n n e r APPENDIX 1 I S I L V I L C U L T U R A L | i O B J E C T I V E S I r V.. S I T E .* I I S E N S I T I V I T Y - . , I I ' P F P I M P A C T J I L E V E L S I I I v I O N - S I T E F I R E I I FACTORS / R I S K S I — > OFF S I T E R I S K S / V A L U E S I I—> I - F U E L HAZARD | I FACTOR | — > I I G N I T I O N TECHNIQUES I I ANO PATTERNS |-C05T I CONSIDERATIONS |-MOP-U? FACTO* I — > A V A I L A B L E • RESOURCES I BURN O B J E C T I V E IMPACT L E V E L . ->l I I I 1 PB PLAN I F I R E E F F E C T S < — I V A L U E I A V A I L A 6 L E < — I WEATMERCMXaX) I P F P RANKS / <--I L E V E L S 1 j 1 1 1 1 E F F J C T / R I S K | 1 K — 1 Or NO BURN | 1 1 1 1 — 1 1 1 1 1 1 1 PRE-BURN | 1 1 K ~ l WEATHER I V V I SPOT WX < — | FORECAST I SMOKE < — | MANAGEMENT APPENDIX 2 PRELIMINARY C O N C E P T U A L MODEL OF FIRE E F F E C T S S Y S T E M FIRE SEVER IIY CHARACTERISES Fu«l Loading Fu«l Continuity Sb« Class DM. Fusl Spaclaa Ago of Fat I HK4C DC • TREE SPECIES CHARACTER 6TICS IN THIS ECOSYSTEM Shod* Totoranc* Sta<]« R«qulr»m«nts Nutria nt R«qulnsm«nts Molstura Raqulrennnts SOIL AND SITE CHARACTERISTICS Nutriant Supply Awllabk Nutrients Molstura Supply Erodlbinty FIRE SEVERITY! LEVEL VEGETATION CHARACTERISTICS \\'j«t