SCOPE 5 - Environmental Impact Assessment

5.1 INTRODUCTION 

This chapter is written for the assessor, the individual directly responsible for designing an EIA. The intended reader may be a project administrator in government or industry, or he may be the head of a committee charged with developing an independent assessment. In any case, he has a specific and well-defined role in the decision process and must interact with the other role-players. He is directly involved in the strategic evaluation and he is presumed to have a technical staff that participates in the technical evaluation of a proposed 'action'. 

The chapter is written with the conviction that a conceptual framework needs to be formulated before the EIA methods described in Chapter 4 are applied. If the assessor simply follows existing, pre-packaged methods, the results will fall short of their potential. 

An outline for such a framework is presented in this chapter and is described in more detail in a related book, Adaptive Environmental Assessment and Management (Holling, 1978). Both presentations prescribe two major activities. The first is the orchestration of resources and people -scientists, managers, and decision. makers-into a process that integrates their talents and concerns. The second activity utilizes computer-based simulations/scenarios to synthesize and focus the work in an explicit manner. Five illustrative case studies are described in the above-mentioned book; a sixth example is discussed in this chapter and in Appendix 5. 

The chapter begins by defining a simulation model, describing its essential characteristics, and identifying the criteria which will establish the need for such a model in an EIA. Then, assuming that the use of a simulation is appropriate and worthwhile, the chapter gives advice on how to start, on what the decision-maker will need to do, and on what he will need to ask his staff to do. After a brief description of a simple policy analysis which will determine whether or not it is worth continuing with the development of the simulation, the processes of modelling and validation are outlined so that our administrator will know what the technical experts concerned with these stages are doing. The use of simulations in complex policy analysis and possible forms of presentations of the results of the analysis are then described. Finally, a very brief description is given of possible developments in simulation techniques relevant to EIAs.

Other relevant SCOPE publications on simulation models include HRI (1976) and SCOPE 9 (1978). 

5.2 WHAT IS A MODEL? 

The models used in EIAs are simplified representations (sometimes 'caricatures') of reality. We can never produce a perfect copy, but if we can mimic the most important features of reality , then the model will be recognizable and useful. 

Models can be sub-divided into three main classes: 

1. a scaled-down copy of a physical object such as a ship or tall building; 
2. a mathematical representation of a physical or biological process, e.g., of the spread of pollution from a chimney, or the movement of a weather disturbance across a region; 
3. an exploratory representation of complex relationships amongst physical, biological, and socioeconomic factors or indicators, some quantitative, others qualitative. 

This chapter is about the third class of model, often called a simulation or a scenario. In its simplest form, this kind of representation is extremely useful in the first stages of an EIA, helping to synthesize the widely diverse information reaching the assessor through many specialists. As the simulation model becomes more and more complex, it becomes less and less relevant to the EIA process. In fact, the tendency towards complexity , leading to the construction of mathematical extravaganzas, has given the modeller a poor public image in some cases.* 

Throughout the rest of this chapter, the word 'model' will be used in the sense of a simulation, i.e., of a procedure for exploring relationships amongst several variables, some of which may be in qualitative form. The simulation may be incapable of validation, relating to management alternatives that have not been and will never be implemented. For example, when three or four alternative sites for a nuclear power station are considered, validation data will become available only for the alternative that is selected, and then only several years or decades later. The predictions for the other alternatives cannot be verified.

5.3 SHOULD I USE A SIMULATION MODEL? 

The essential feature of an EIA is the provision of choice between a range of alternatives. Any choice will affect several heterogeneous 'elements': physical, ecological, and sociological. Further, these elements are usually interrelated in complicated ways and there is a mass of information. The mass may be small; it may have obvious and not so obvious gaps in it, and it is likely to appear , initially, in a thoroughly indigestible form. The previous chapters have described methods for ordering this information, for displaying the links between elements, and for evaluating the alternatives. When will it be valuable, or even essential, to go beyond these techniques and set up a simulation? In the following sub-sections, we propose a series of criteria. 

(*Some modellers believe that simplicity is a requirement for only the output of a model, not for the model itself. This may be true in some cases.)

5.3.1 Volume of Data 

The best known feature of computers is their ability to use large amounts of data. Typically, such data belong to a few simple categories -e.g., hours worked and pay scales- and the calculations performed are, individually, straight-forward- e.g., wages paid. It is the volume and speed of the calculations which are impressive. 

Criterion 1. If your assessment will require the handling of large numbers of simple calculations, a computer-based mathematical model will generally be desirable. 

5.3.2 Complexity of Environmental Relations 

The interconnected nature of the elements in the environment poses special problems for impact assessment because the linkages between these elements are often far from simple. If we have two related elements, representing an action and an impact, the simplest assumption to make is that when we alter one element slightly, the other element will change slightly and proportionately (Figure 5.1 (1)). The technical term for such relationships is 'linear'. Very often in natural or social systems, however, the assumption of linear relationships is false. An action may produce an impact, but increasing the action may not significantly increase the impact (Figure 5.1 (2)). Alternatively, a gradually increasing action may produce negligible change until a point is reached at which dramatic alterations in impact occur (Figure 5.1 (3)). Both of these relationships are technically described as 'non-linear'. In the former, we may over-estimate the probable impact of increased action by assuming linearity; in the latter, we may not foresee a potential catastrophe. Further, these responses may be displaced by the system and appear as impacts at points structurally or geographically distant from the action. 

In the descriptive methods outlined in Chapter 4, we usually assume either explicitly or implicitly that the links between action and impact are of the simple linear type, although we may wonder what would happen if they were not. Also, there is a limit to the number of links which we can comprehend at any one time, particularly when some of these links are direct and some are indirect. Exploration of the possible influences of non-linearities and of indirect links is frequently possible by the use of mathematical models, and where there are many such links non-linearities can best be explored by the use of computers. 

Criterion 2. If there are many complex links between the elements of your environmental impact assessment, the use of a computer-based model will be of great value.

5.3.3 Time-Dependent Relations 

The natural world is not static. Flows of energy and matter, and changes in these flows, are not only usual but also sometimes necessary for the maintenance of viable ecosystems. Conditions that appear to be static may be slowly changing or may represent only a temporary equilibrium condition amongst several processes acting in opposite ways. Because man's actions alter these relations, analysis of the time-dependent processes may be necessary to predict the future. Of particular importance is the need to search for possible feedback mechanisms amongst the various environmental, sociological, and economic processes. 

Sometimes not only the scale of the changes to be imposed by a development project but also the rate at which these changes will be introduced affects the final equilibrium state of the system. In some cases, the impact might be less if the rate of development slowed down.

In other cases, changes may be set in motion leading to impacts that are perceptible only a long time after the project has been completed. If, for each of the links, the relationships which affect the changes can be defined, including delays and time-lags, then the overall changes can be estimated. In mathematical terms, the analysis would then be dynamic (time-dependent) and not static. 

Criterion 3. If the affected processes are time-dependent, a computer-based simulation will usually be helpful. 

5.3.4 Explicit Relations 

One apparent disadvantage of a model is that every element and every link must be defined explicitly. It is not enough to say 'wildlife will decrease through lack of food'. It is necessary to define the species of deer, to estimate the size of the present population, the food, the rate of migration, the present death rate, and so on. In fact, this apparent disadvantage is actually a major advantage. The nature of the modelling process forces into the open hidden assumptions which may have too little basis. It reveals areas where information seems inadequate, and, especially, it makes the participants in the assessment, who may have very different backgrounds, aware of each other's problems. 

Criterion 4. If increased definition of assumptions and elements will be valuable in drawing together the many disciplines involved in the assessment, a simulation model may be helpful. 

5.3.5 Uncertainty 

When the elements and links in a model have been defined, it is likely that very few will have the exactness of simple elements like 'width of right-of-way' or 'height of dam'. Many will have wide limits to their probable values, either through a lack of knowledge or because they do really vary in space and time. If the average value of each element is used as a basis for the simulation, then the model will produce only a single, apparently exact, result of the consequences of an environ. mental change. 

A watershed analysis that uses average river flows will fail to account for impacts that result from once-a-year or once-in-a-hundred-year extremes. Even when several input values are included, the exactitude of the predictions will still be an illusion, although information on the probable ranges of input values may provide indications of the probable range of effects. Wherever feasible, the best approach is to begin with frequency distributions for all of the relevant input variables. This may result in a seemingly unmanageable data set, but with a computer-based model it is a relatively easy task to explore the implications of various combinations of parameter values. If, for example, we are examining the effect of a dam on a deer population, we may find that for all values in our assumed range, we predict only slight decreases in the size of the population. In this case, no further information needs to be sought. Similarly, if the effect is shown to be consistently large, then clearly it should be carried further into the impact evaluation. Where, however, the effect is shown to be highly variable and dependent upon the input assumptions chosen, considerably more information may be required about the link between dam construction and deer populations before the assessment is complete. This is called a sensitivity analysis.

Criterion 5. If some or all of the relationships of the assessment can only be defined in terms of statistical probabilities, or in terms of a range of values for input assumptions that must be explored, then a model will be useful. 

5.3.6 Major Knowledge Gaps

The above example illustrates the circular nature of the modelling process and the value of undertaking a simulation early in the EIA. But there is another question that needs to be asked. How important is the deer population? This value judgement is a critical part of the assessment. It defines what must go into the simulation and in how much detail it needs to be described. 

It is essential that inadequacies in the data or in the assumptions are not conveniently lost within the computer. Facts and values must not become confused. It is the role of the assessor to relate them. Because answers are usually required quickly, it is no help to start a long-term research programme (although a first attempt at simulation can reveal critical gaps in existing data). Finally, in contrast to scientific research, experimental tests of the model are not normally possible in environmental impact studies. 

The EIA assessor must take into account many different types of regional, political, and social factors. Can these be fitted within the constraints imposed by the nature of simulation modelling? If the assessor considers that some factor is so important that it must be included, but is so ill-defined that even the main features of its interactions with other aspects cannot be expressed, or should not be expressed, quantitatively, then modelling may not be an appropriate technique. However, a simulation model need not necessarily cover all of the impacts of an action. Segments of the environment can sometimes be modelled and the conclusions can be embedded in the remaining, non-quantifiable parts. 

Criterion 6. If there is no possibility of defining some of the elements and their relalionships in the assessment, there is no point in attempting to include them in a model. 

5.3.7 Relevance to Policy Questions

In the best of all possible worlds, it would seem that computers and models are admirably suited to cope with the complexity of impacted environmental systems. A perfect tool, surely, for impact assessment? Unfortunately, many of the early products fell far short of their promise. It is not that the potential was not there. It is rather that the understandably enthusiastic excitement of at last being able to grapple with complex problems led to grandiose dreams. These dreams generated simulations that often were 'caricatures of reality', but better caricatures are found in fairy tales and science fiction. Nevertheless, that swell of enthusiastic effort was necessary and useful. A very small sub-set of models and modelling approaches has now emerged that has directly contributed to assessing impacts of large-scale developments, by identifying unanticipated benefits and dangers that have subsequently become fact, by forcing modification of the assessment procedure, and by illuminating critical areas of weakness in data and policy. 

Generally, the useful and usable techniques developed to date have emerged from very small groups of scientists, who have had a well-defined and rather narrow focus and have made great efforts to link the modelling effort, from the start, with policy questions and with vigorous data validation. If we forget about the early 'over-selling' of the technique, some useful progress has been made . 

Criterion 7. In general, be cautious of applying any integrated set of modelling techniques and procedures unless they have been designed, from the start, with policy questions as the first consideration. 

5.3.8 Costs 

Finally: how much will it cost? With costs varying from country to country, the best we can do is describe the minimum requirements in terms of personnel and facilities. A barely adequate assessment is feasible with three scientists (having, between them, experience in several resource areas and in modelling), one computer programmer, and a secretary. In addition, there must be access to a small computer, to data, and to resource specialists. 

In Appendix 5, an example is discussed of a simple but useful preliminary assessment of part of a major hydroelectric development in Canada. This assessment took three weeks with a staff similar in size to that described above and cost approximately US $8,000 in staff and computer time. An essential ingredient, however, was a five-day workshop involving 20 policy and resource specialists from the contracting agency. The costs of their travel and salaries are not included. 

Criterion 8. Construction of a simulation model for an EIA will require a minimum of three professional staff, two support staff, and access to a small computer and to the necessary data and specialists. 

5.4 HOW DO I START? 

The basic criteria described above suggest that, whatever the apparent complexity of the problem, we can at least attempt to use a model in an EIA. There is no magic formula for finding the entrance to the problem, but we should set up a clear strategy at the beginning. From the outset, we have to use all our resources and expertise to impose our goals on the solution, instead of allowing the problem to impose on us a solution which is unsatisfactory. How do we attain this desirable position? The strategy to be employed will depend upon an evaluation of the problem, an evaluation which can only be performed after we have adequately delimited the problem itself in terms of the questions to be posed and the associated scale and complexity. 

The major practical limitation is usually time, because the assessment will be required by a certain date. A second limitation is usually one of facilities. Do you have the necessary computing facilities, and people with the necessary expertise to prepare the programmes? These purely administrative constraints may determine not merely whether or not it is practical to use a mathematical model but also the limits of the scale and complexity of the model. Leaving the latter aside for the moment, we will focus our attention on the technical problems. 

5.5 WHAT DO I DO?

5.5.1 Delimitation of the Problem 

From the previous chapters, it has become clear that the problems of EIA are interdisciplinary , impinging on almost every facet of human interest. However, our strategy will start by imposing some specific limits to the real universe surrounding the problem. Answers to the following four questions will help us reduce the problem to a manageable size: 

1. What Classes of Output will be Needed to Make Decisions? From the whole host of variables involved in the problem, only a fraction of them (or a certain combination of them) will be relevant to the final decision. We see here a first step in weighting. Only the group of people involved in the decision-making process will be in a position to select the variables relevant to the decision; the consequences of this selection have wide repercussions on the operational aspects of the EIA. 

2. What Are the Geographical Limits to the problems?

Although human technology has proved to be capable of producing effects at the global scale, with few exceptions we will be able to place geographical limits on the size of the problem. Again, this is an arbitrary limitation which usually reflects the interests involved, and helps to indicate the desired strategy. In consideration of the effect of environmental changes on the fishing industry , for example, global effects such as international price increases of cheap protein due to fish stock depletion could be included in the analysis if desired, or these international aspects ignored and the problem reduced to a regional or national geographical scale. 

By restricting the problem to too small an area, important factors may be ignored; by trying to take in everything, the problem may become unmanageable. The preliminary analysis may indicate, however, that certain aspects can be omitted. The failure of the anchovy fishery in Peru (Idyll, 1973) was caused by a combination of abnormal water movements and very heavy fishing. The abnormal environmental conditions off Peru appear to be part of large-scale anomalies in the circulation of water in the Pacific, which themselves depend on fluctuations in the atmospheric general circulation. Is it necessary to include all these factors in an EIA model? The answer is 'no'. The model requires information only on how often this adverse condition is likely to occur and what effect it has on survival of young fish spawned by the adults in the stock. 

It is important to distinguish between an EIA simulation and the model that would be required for a general research programme. In the long run, an understanding of the environmental changes off Peru not only can improve our knowledge of the basic factors controlling the fish population, but also can indicate how certain types of environmental fluctuations would affect other stocks, as yet unexploited, in quite different parts of the world. Excluding these types of study from an impact assessment model is not denying the long-term need for research. 

3. What Are the Time Horizons of the Impact? 

The assessment of a given environmental impact has to be performed in relation to a given period of time. Once more, there is no simple way to define this dimension, and the decision will depend on the many specific factors surrounding a given problem. Nevertheless, at least a word of caution is desirable. Frequently, the events involved in environmental impacts are characterized by non-linear processes, or by lags between cause and effect, so that consequences which are negligible during one period of time may become important if that period is extended. 

4. What Are the Sub-Systems Affected by the Model? 

The previous sections have described some of the problems of setting boundaries of time and space to the model. The result, in technical terms, will be a listing of elements and of the links between them, either as a table or a flowchart. The number of elements may be relatively small or very large. The links may also be large in number although each link is of a relatively simple kind, or there may be complex interactions at many points. The next stage in the delimitation of the problem is to see if this mass of elements and links needs to be, or can be, considered as a group of sub-systems. 

How can sub-systems be identified? Are there, in the table or flowchart representation of the problem, smaller areas which, although highly interconnected internally, have relatively few links with other parts of the system? If so, we may regard such areas as valid sub-systems. They may be geographical or structural, representing groups of people, organisms, or activities. This decomposition into sub-systems is useful, not only for the strategic analysis of the problem, but also for the management of the assessment, as it will partially divide the relevant elements into groups, each of which can initially be examined separately by those with specialized knowledge of a particular sub-system. Hence, the economist, the wildlife biologist, the limnologist, have a clear focus for their expertise.

5.5.2 Strategic Evaluation of the Problem 

For any major development, there is always a set of possible alternatives. The initial generation of these alternatives is a crucial step, because it provides the reference frame which will largely determine the kind of information you will need, as well as the type and usefulness of the model to be constructed, and the universe of more detailed alternative options which will need to be assessed. The development may not even be feasible as originally proposed, but may be feasible or more useful when considered in some alternative form. 

The initial generation of alternatives may be greatly helped by some rules for providing a systematic reference frame. While it would be impossible, and often of no value, to present a complete list of alternatives for many projects, a few guidelines may be of assistance. 

Usually, the most obvious proposal for a development in a particular region is the one which is expected to produce the maximum benefit (either economic, social, etc.) assuming nothing goes wrong. However, it is important to look for alternatives which will imply a minimal cost if things do go wrong. In addition, one may look for alternatives with a high probability of being successful (low probability of failure ), even if the potential benefits are not very high. 

We may sometimes suppose that all of these considerations have been taken into account in the original proposal, but it is advisable to separate them explicitly. For instance, in a development plan for a particular region, the proposal may be to build a large oil-fired power station, taking advantage of the efficiency of centralization and of the benefits of economy of scale. This is the 'maximum-benefit' approach. However, the consequences of a failure of the assumptions in the EIA might lead to unacceptable air quality episodes. We might, therefore, as an alternative, consider the construction of a series of small power stations, so that the emissions from one of them would not affect the whole region; some of the power stations could even be modified if unsuspected environmental consequences were to become evident. This alternative may well be less efficient, from a traditional point of view , but safer than building a big power station, and might be regarded as the 'minimal cost of failure' approach. Finally, we might propose the construction of medium-sized stations in a planned sequence, modifying the project whenever unexpected effects are detected. This would be the 'maximum probability of success' approach, and while it may be less efficient than the first alternative, it will entail less risk. 

The above three viewpoints (i.e., the maximum:-benefit, the minimum cost of failure, and the maximum probability of success) can be distinguished at each major step of development planning. 

5.6 WHAT DO I ASK MY STAFF TO DO?

Now that you have constructed a strategic bounding and evaluation of the problem, the first and obvious essential is to gather together all the available information and to identify the people who can contribute to the model -usually specialists of various kinds, including system analysts and computer programmers. Some of these people will be consultants, brought in to help your own staff, and some of them must have a broad policy of view of the problem. What do you ask them to do? There are several specific guidelines which may be useful. 

(a) Initial variable identification and organization. Having carefully identified the problem within the strategic framework developed above, and listed the essential variables, the following steps will be necessary: 

1. Organize the variables into separate classes identified according to some common properties. 
2. Specify hypotheses concerning the interactions between classes of variables and illustrate these graphically. Some thought should be given to the form of the independent and dependent variables in order to facilitate interfacing with the rest of the model.
3. Identify , for each interaction, all reasonable alternative hypotheses and make rough estimates of maxima, minima, and thresholds. Retain these subsequent tests of the sensitivity of the simulation model to various alternatives and extremes. 

(b) Assigning degrees of precision. When a problem can be divided into sub-systems, it is important to have approximately the same degree of precision in each sub-system. The best way to do this is to make an initial estimate of the required or possible precision for each sub-system, identifying inputs, model detail, and outputs. The choice of the appropriate level of precision should be a joint effort by you and your staff and should be based on the kind of questions you want answered, the time available for the study, and the quality of the data. 

(c) Construction of a flow diagram. A wide choice of conventions is available for drawing flow diagrams, based on control system theory, cybernetics, and information theory. The best conventions seem to be the simplest, in which one symbol designates an input or output, another an intervention, and a third a process. The same symbols are used throughout both the model and its constituent sub-models. 

(d) Interaction table. If separate sub-systems are analysed independently by sub-groups of your staff, one of the most difficult tasks is to ensure effective interfacing between sub-models. The device that seems to work best is an interaction matrix, which identifies the inputs each sub-system expects to receive from others. 

The ease and success of the whole simulation is enhanced when each sub-model group is asked to provide outputs to the other sub-models rather than informing the other sub-groups what they intend to provide (Holling, 1978). This technique avoids the common temptation, by discipline-oriented sub-groups, to generate excessively detailed models of little relevance to the other sub-groups. This technique of 'looking outward' requires that the subgroups ask 'What is needed of this sub-model?' rather than 'What would we like to include?'. The participants are then asked if the output expected of them as input to another sub-system is the kind that they intend to produce. Very often it is not, and compromise proceeds in a series of steps until a consistent table of interactions between sub-systems is achieved.

By this time, all the necessary steps have been taken to permit the start of the actual modelling. Your staff has largely been responsible for generating an interaction table which shows exactly which variables are affected by possible changes in each of the other variables. Moreover, it should be possible to suggest the direction of the influence. For example, if nutrients in a reservoir increase, they will have a stimulating effect on some variables and possibly an inhibiting effect on others. A further level of qualitative information might indicate different ranges over which the effect might be zero, plus, or minus. Finally, enough insight might have been gained by this to weight, subjectively, the intensity of the cross-variable effects. 

The above sequence of steps can be developed very quickly; for example, once the data for the James Bay study described in Appendix 5 were collated, the group of 20 workshop participants completed all the above steps in two days. Not only can the interaction table be prepared rather quickly but also it provides, in a concentrated form, an immense amount of qualitative information which itself could form the basis for a preliminary assessment. Alternatively, if the resources, time, and information are not available for an extensive assessment and evaluation, the same table could be the basis for a formal evaluation. If this is the case, the process should no longer be controlled by your staff, but by you, with staff assistance. At this point, you need a simple procedure for policy analysis, aimed at producing a rough qualitative picture of the impact of each of the alternative development proposals. 

5.7 HOW DO I MAKE A SIMPLE POLICY ANALYSIS?

Three sets of information are necessary for the first step in the analysis. 

1. You must identify the important system elements, or variables, and their interactions. This information has been provided by your staff in the interaction table. 
2. You must identify and organize the key impact indicators in relation to the overall goals of the project. 
3. You must specify the policy or management actions which together define any one plan. 

5.7.1 Developing Impact Indicators 

The strategic evaluation should have identified the major impact categories in relation to the original goal of the project. A broad list of categories was given in Table 3.3, Chapter 3, together with indicative examples which represent some possible social, economic, physical, and ecological consequences. 

Because these classes are broad and general, they must be disaggregated into variables which are measurable and relevant. For example, social indicators must be made specific in terms of jobs, leisure time, or similar variables. Having developed a list of indicator variables, it is then often necessary to express them in the most relevant forms. For example, it might be more useful to express 'jobs' not in absolute terms, but as a rate of change, or as 'jobs per capita', or as 'diversity of job opportunity' .In this way, a list is prepared of significant impact indicators, i.e., of measurable variables with direct policy relevance. 

But you must keep in mind that your list will never be complete, or completly adequate. Relationships will always be omitted whose existence you know but whose form you do not. There will, as well, be missing relationships whose identity you do not even suspect. And what is true of these relationships is equally true of the overall objectives of the development. The societal objectives that seem so clear at the moment can dramatically shift, leaving society with a policy and a system which cannot itself shift to meet these new needs. The growing demand for EIA procedures is one symptom of such a shift of objectives. An assessment based solely on the presumption of sufficient knowledge can therefore lead to approval of a plan that could not easily be modified later to absorb the unexpected. 

Few systems-ecological, economic, and social-are in a state of delicate balance, poised precariously in some optimum state. If they were, they would not last, because all systems experience traumas and shocks from time to time. The ones that survive are, in fact, those that are able to absorb changes, i.e., those that have a considerable degree of internal resilience. Resilience, in this sense, is indicated by the maximum external stress (magnitude and rate of change, positive as well as negative) that a system can accept before collapsing or shifting to a fundamentally different behaviour. A review of the concept can be found in Holling (1973). 

In addition to the traditional indicators used in an EIA, it would be useful to have a category that gave some sense of the resilience of the system-of its capacity to absorb the unexpected. The key requirement of resilience indicators is that they measure the degree to which alternative options are foreclosed. 

But how can these resilience indicators be developed? There are three different classes: 

1. Resilience Indicators of Benefits 

The net economic and social benefits of development or policy proposals are often emphasized in environmental assessments. But there are resilience counterparts to these impact indicators. If the development were to fail unexpectedly, or if social objectives were to shift to such an extent as to require removal of the project or policy, there would be an associated cost. A computer model provides an explicit way of measuring this cost of failure, by merely programming such a hypothetical event during the course of the computer simulation. Example: regional insect pest control projects can have several forms. One might be intensive and extensive insecticide spraying. Another might combine cultural practices with limited and controlled application of insecticide at critical times or in critical places. Both policies might achieve generally favourable and not too dissimilar results. Suppose however, that the insecticides were removed suddenly as a result of rising costs or new government regulation. In the first case, the removal could produce intensive outbreaks covering large areas, with disastrous effects on benefits. In the second case , the loss of benefits could be minor .

The impact of policy failure can be evaluated with the aid of computer model in many such instances, providing a measure not of the relative fail-safe features of the proposed programme or policy, but of its degree of safe-failure. 

2. Resilience with Respect to Systems Boundaries 

Social-ecological systems are dynamic in the sense that their structures and functional interrelations themselves establish the outer limits of resilience. Example: phosphates added to an aquatic ecosystem are incorporated into existing biogeochemical cycles. But there is a limit to the amount that can be added without upsetting the integrity of the cycle. Therefore, an indicator that expresses the total amount of phosphate added to the system should be matched with one that expresses the relative amount in relation to the system boundary for phosphate. 

Sometimes the simulation model itself can be used to identify these thresholds; in other cases with less knowledge, the boundary is merely an educated guess. Again, the task is first to identify the significant social, physical, and ecological variables; and second to add a resilience dimension that measures the magnitude of each variable in relation to the system boundary or standard. 

3. Resilience Indicators of Social and Environmental Capital 

Reserve funds and resources are available for development projects and policies. Whenever the unexpected occurs, however, the decision-maker may be forced to draw from these reserves to modify existing programme. This course of action forecloses future options to a certain extent, the degree of foreclosure being dependent upon the relative sizes of the withdrawals and of the reserves. Thus a resilience dimension should be attached to the various indicators of social and environmental capital. Example: the development of a recreational area would produce certain impacts, which can be evaluated in terms of traditional recreational indicators. But the development is drawn from a land-bank of fixed size, with certain intrinsic qualities for absorbing recreation. A resilience indicator should therefore be paired with each recreational indicator . 

5.7.2 Developing Policy and Management Actions 

In any one development, there are several internal options for action during the construction and post-construction phases. Some relate directly to the project itself while others are indirect actions. In the James Bay example of Appendix 5, there were three sets of actions: those directly concerning power production ( e.g., schedule of dam construction, character of water diversion and water flow controls), those concerning environmental quality (e.g., silt control, tree clearance in reservoirs, sewage treatment, etc.), and those affecting demand (e.g., recreational controls, job allocation between insiders and outsiders, road access, etc.). The essential feature is that classes of policy and management actions must be decomposed into specific, definable actions.

This process is identical to the effort your staff made to decompose the environmental system into the system variables, and it is identical to the effort you performed to decompose project goals into impact indicators.

5.7.3 Putting the Pieces Together 

You now have the three elements necessary to develop the first rough assessment: the system variable interaction table, the list of impact indicators, and the list of policy actions. Your goal is to develop a table of actions vs. impacts. This table is Box IV in Figure 5.2. It is the interaction table of the system variables acting on one another (in Box I in Figure 5.2) which allows you to do this. In the complete analysis you will use the model that you are creating, but in the meantime the interaction table in Figure 5.2 will provide a preliminary policy assessment and an indication of adaptations needed in the assessment activity. 

Briefly, two intermediate tables are developed. The first is designed to show how each action is likely to affect each system variable (Table II). The second shows how each system variable is related to each impact variable (Table III). The action vs. impact table (Table IV) is formed by linking Tables II and III through Table I as indicated in Figure 5.2 .Such a table is similar to those described in Chapter 4, and the process of weighting the impacts (in order to express degrees of importance) is the same as that described earlier. With tables of this kind for each of the alternative plans, it should then be feasible to reject the most extreme proposals, leaving a smaller set for later discussion and decision. 

5.8 WHAT HAPPENS IN THE MODELLING PROCESS? 

Now that the problem has been defined in terms of its boundaries, its sub-systems, its possible variables and their couplings, we are ready to begin the modelling process itself, assuming that we have decided to proceed beyond the stage of the simple policy analysis. It is at this point that the expertise of the applied mathematician, already used as a consultant in the strategic and tactical definition of the problem, becomes paramount, and some understanding of his role is necessary if you are to retain the necessary control of the impact assessment process. 

The mathematician's expertise will first be exercised in the choice of the kind of model to be used. In this, he will be guided by the size of the problem, the nature of the various classes of variables, and by the degrees of uncertainty present in the relationships between them. His choice will be influenced by his knowledge of the various 'families' of models which already exist and which have previously been applied to similar problems, in much the same way that a field naturalist is guided by his knowledge of the natural families of plants in his identification of an unfamiliar flower.

Broadly, his choice will lie between the following classes of models: 

(a) Deterministic vs. Probabilistic Models 

In the former, all the relationships are constructed as if they were governed by fixed natural laws-the uncertainties and random fluctuations are not built into the model. In the latter, some or all of the relationships which are defined by statistical probabilities are included explicitly in the model, whose output then directly represents the consequences of those probabilities. This is sometimes called the Monte Carlo approach. 

(b) Linear vs. Non-Linear Models 

Although it may be convenient to assume that relationships between variables are linear, most practical problems require the more complex assumption of non-linearity . 

(c) Steady-State vs. Time-Dependent Models 

Steady-state models compute a fixed final condition based on a fixed pre-action condition, whereas time-dependent models incorporate the way actions affect processes that may eventually produce impacts. 

(d) Predictive vs. Decision-Making Models 

Predictive models enable the consequences of particular decisions to be explored, while decision-making models indicate which of the decisions is 'best' in some defined way. 

When a computer is used in conjunction with a mathematical model, the computer programme must be unambiguous. The resulting 'algorithm' must define the model in sufficient detail for its essential features to be communicated to other experts. Once the mathematician has tested the algorithm to ensure that all the component parts operate correctly, he seeks to validate it with respect to the real world system he is studying, searching for possible inconsistencies or unrealistic results. It is easier and more direct to discover when a model is wrong than to determine when it is right. By modifying the model at this point and subjecting the resulting version to further analysis, he continues the process of improving the model within the limitations of the time and resource constraints of the impact assessment process.

In this connection, the mathematician uses two essential techniques. First, he employs a sensitivity analysis, already mentioned in Section 5.3.5. Second, the mathematician searches for the maximum simplification of the model which is consistent with its value in a predictive or decision-making process. Frequently, it is possible to show that parts of the model which have been developed to satisfy theoretically important considerations have relatively little effect on the final outcome of the modelling process. In such cases, simplification of the model is both desirable and readily achievable. 

5.9 HOW IS THE SIMULATION VALIDATED? 

Repetitions of analysis and refinement can, in theory, continue indefinitely, but in an EIA they will usually be brought to a halt by the need to provide results quickly. Indeed, there may be too little time to develop the model to the degree that would be desirable in a research investigation. At some earlier stage, an attempt at validation will usually therefore be necessary . 

We should admit, however, that validation (the matching of our model with reality) in EIAs is not easy. Sometimes, the only apparent validation which can be achieved is the matching of future performance of the environmental system with the expectation from the model -a test which hardly meets the criteria of good science. Nor does it contribute to the decision-making process that seeks the assessment. Nevertheless, some confirmation of the appropriateness of our model can be obtained. 

First, the analysis which was necessary in the refinement of the model will have given us some confidence that the behaviour of the modelled system is consistent with our expectations. Where it has been possible to divide the total system into sub-systems, the behaviour of these sub-systems, singly and in aggregate, will have reinforced our knowledge of the dynamics of the system. If, as is likely, the behaviour of an aggregated system runs counter to our intuitive expectations, we wi1l have been forced to reconsider the basis of our 'common sense' expectation. In this way, confidence in the value of our model, as at least a working approximation, will have been increased.

Second, experimentation with model systems may indicate critical experiments which would enable a valid test of the model to be carried out as a direct appeal to nature, consistent with the logic of the scientific method. Such a test may seem relatively unlikely in EIAs, where the time-scale for the assessment is limited. But the model may indicate a specific, focused experiment that can contribute significantly to the validation; alternatively, existing experimental evidence that had not yet been considered may be suggested for testing the predictions of the model system. 

Third, where it has been possible to undertake surveys to obtain the necessary data for the construction and parameterization of mathematical models, it may be desirable to hold back a certain proportion of the data so that they may be used in an independent test of the hypothetical model derived from the main data set. In this way, the inconsistency of formulating and testing an hypothesis on the same set of data can be avoided. 

Whatever method is used in an attempt to validate the model system, one of the paramount advantages of mathematical models dominates the argument at this point. In contrast to all other forms of reasoning, the mathematical model is explicit in its statement of the relationships between the variables and of the assumptions underlying the model. If anyone disagrees with these assumptions or relationships, he has only to replace them by some equally explicit set to verify that the changes make corresponding changes in the expectations of the EIA. 

5.10 HOW DO I USE THE OUTPUT FROM THE SIMULATION IN A COMPLEX POLICY ANALYSIS? 

Once a model has been satisfactorily validated, the next step is to select from amongst the set of possible alternative policies or actions that have been generated. The problem is: 'Which is the best choice?' 

Suppose you are confronted with a set (Say, A, B, C, D, E, F) of alternative policies or actions, generated by some kind of model. For each of the alternatives, it is feasible to estimate the probability of being right or wrong* on some objective basis. That is, according to the uncertainties involved in the construction of the model, and according to the likelihood of a critical hypothesis being wrong, you may allocate (or be given) the degree of confidence to be placed on the success or failure of the policy. For each of the alternatives, you may also have an appreciation (economic, social, political, etc.) of the benefit of succeeding or the cost of failing, and this appreciation can be given a numerical weight, or at least a ranked order. 

Given this information, there are different ways of choosing, which can be best il1ustrated by a hypothetical example. Suppose you have six alternative policies or actions, their associated probabilities, and the relative weights to be applied to the consequences of being right or wrong: 

*For simplicity, we assume that a prediction is right or wrong. In some cases, success or failure must be measured on a continuous rather than a binary scale, but this has no effect on our line of reasoning.

From these two sets of values, it is possible to estimate in relative terms for each alternative: 

1. The probable loss (the probability of failing multiplied by cost of failing); 
2. The probable benefit (the probability of succeeding multiplied by the benefit of succeeding); 
3. The most likely net benefit (the probable benefit minus the probable loss).

The total information might be presented as follows: 

This table may be used to make the 'best' choice from amongst the six alternatives, using several different criteria for defining the word 'best'. 

The first criterion is trivial, and consists of choosing the alternative which has the greatest probability of success (lowest of failure) without considering the size of benefits or costs associated with success or failure. Using this criterion, either alternatives D, E, or F would be chosen. 

A second criterion consists of choosing the alternative which provides the highest gain if successful (alternative B, with a possible benefit taken as 100 in the example). This criterion has been widely used, either explicitly or implicitly, sometimes with disastrous consequences. No account is taken of the consequences of the action being wrong, or of the probability of the action being right. 

A third criterion is to choose the alternative which produces the lowest cost in case of failure, which is in a sense the safest choice. Using this criterion, alternative C (with a loss of 15 if the alternative is wrong) would be selected. 

A fourth criterion is to use the alternative which provides the highest probable gain, i.e., to select the alternative which takes into account both the magnitude of the possible benefit and the probability of succeeding. In this case, alternative F (probable gain of 72) is chosen. A fifth criterion is to pick alternative E, which has the lowest probable loss (-2). Fina1ly, the sixth criterion is to select the alternative with the highest value of the most likely net benefit, which takes into account both the probable benefit and the probable loss; in the case under consideration, this is alternative D (+36). Alternative A is not chosen using any of the above criteria. 

The above simple example is intended to make the following points: 

1. There are many different criteria for choosing alternatives, i.e ., there are many ways of deciding what the words 'best' or 'worst' mean in a given context.
2. Some evaluation of the likelihood of failure or success and of the respective losses and benefits is necessary for the alternatives to be evaluated.
3. The six different selection criteria defined above can be grouped into two classes, according to whether the aim is to maximize the gain or to minimize the loss, i.e., ambitious vs. cautious strategies. Our ignorance about the behaviour of complex systems, particularly environmental systems, is so vast that it is often foolish to adopt anything but a cautious view of the outcome. Failure to do so, i.e., attempting to achieve the highest possible benefit without proper consideration of the risk involved, has already transformed golden dreams into black nightmares in many parts of the world. 

Rabinovich (1977a) has suggested additionally that the choice of a decision from amongst several alternatives should depend in part on the problems likely to be encountered in implementing each alternative. The optimal decision in a stable and technologically advanced country may be sub-optimal in a country that is technologically backward or is politically unstable. Rabinovich identifies the following kinds of implementation difficulties that could arise: 

1. once a decision is made, there are long delays before it is carried out; 
2. once a decision is made, some actions relating to the project are carried out, others are not; 
3. once a decision is made, all actions are implemented with no lags but implementation is technically deficient; 
4. once a decision is made, and correctly and timely implemented, there is either: (i) no monitoring carried out to verify the adequacy of the decision, or (ii) no control measures to prevent violations to regulations resulting from the decision. 

Rabinovich has illustrated his ideas with a simulation model (GURI) for a large hydroelectric power scheme in Venezuela (Rabinovich, 1977b). 

5.11 HOW CAN THE RESULTS BE PRESENTED? 

Perhaps the greatest potential Achilles' heel for those assessors who include model-generated information in EIAs lies in communicating the results. Two problems are of importance here. First, the assessor receives an unsettling if not alarming mass of information that he must synthesize before communicating with the decision-maker. Because of the need to test many alternative hypotheses, the computer print-out may sometimes greatly exceed the data input. Second, the credibility of the model-generated information may be viewed by the decision-maker with unbridled optimism at best and untempered hostility at worst. The potential seriousness of both these problems may, oddly enough, increase in proportion to the model's capacity to help analyse complex problems. 

To overcome these difficulties, the assessor should, in the first place, produce information that fits the interpretative capabilities of whatever person(s) he must communicate with. Pragmatically, the final information is inappropriate if it exists in one form only (such as tables). Secondly, the assessor should be able to explain the algorithms, i.e., to state clearly the ways in which raw data have been converted to finished information within the computer. The information's credibility is in jeopardy if the decision-maker considers that the information has passed through one or more 'black boxes' where it has been transformed in mysterious ways. 

Regardless of whether the model has been adopted for use on a computer, it is at this stage of assessment that computer-aided communication forms (actually communication models) can be of immeasurable value. With a common set of data, a computer system can simultaneously produce a wide variety of specialized displays, including flowcharts, tables, matrices, graphs, response surfaces, maps, and reports in traditional prose form. With such a graduated series of displays which trade off depth of explanation (credibility) for simplification (ease of interpretation), almost any decision-maker can locate a display form which suits his interpretative abilities and through which he can build an understanding and belief in more or less complex forms of assessment (see Figure 5.3). In this manner, two or more decision-makers with differing interpretative abilities have a common communication package from which they may achieve an understanding of each other's viewpoint.

5.12 ARE THESE IDEAS AND METHODS OF ANY VALUE IN A DEVELOPING COUNTRY? 

The assessor in a developing country with few data and few resources may think this chapter is not for him. But it is precisely in developing countries that the simulation technique is of greatest value, providing guidance in the choice of environmental parameters to include and exclude in the analysis.

Certainly a million-dollar model should not be used, and perhaps such a model is not cost-effective even in a developed country. However, a simple simulation costing only a few thousand dollars at most may provide an initial focus, saving time and money over the long run and, more importantly, providing a more useful environmental assessment. 

5.13 WHAT DEVELOPMENT OF THESE IDEAS CAN I EXPECT IN THE NEAR FUTURE? 

The description and evaluation of modelling concepts relevant to EIAs given above is based on present-day incomplete knowledge of the behaviour and mathematics of complex systems. This knowledge is rapidly growing, and while it is impossible to predict major advances and achievements, some generally promising directions can be identified. These are described in the following paragraphs, sometimes employing the technical language of the specialist.

It is likely that the present trend towards more precise and complex models for specific situations will continue. Some advances to be expected in this direction are the development of better procedures for the analysis of the role of spatial characteristics of systems, the development of a theory of self-organizing systems, improvement of the techniques of search for optimum sets of conditions, and the establishment of techniques for conflict reconciliation in complex systems. Of particular relevance to environmental problems would be the development of procedures of efficient search for optimality in hierarchical systems, where the optimization at each sub-system level must be reconciled with optimality of the higher levels of the system. 

In addition, a trend towards generalization of environmental model structures is beginning to emerge, and we can expect the development of flexible, easy-to-use, computer languages for environmental simulation, allowing the user to concentrate on the conceptual problems, without having to worry about how to communicate with the computer. Closely related is the possibility of developing functional Packages, or modules, for essential and invariant ecological processes, which could be combined in different ways for each particular problem in much the same way as physical laws are introduced currently in the construction of meteorological and hydrological models.

Further a trend towards model simplification and analysis may be expected. Such steps will allow critical and focused examination of key components the model- thus point to the key components of the impacted system. These analyses will include considerations of qualitative model behaviour system stability and measures of resilience and robustness. 

Finally, a trend simpler initial models may be expected better adapted to cope with uncertainty and the important qualitative aspects of environmental problems. Such models by using semi-quantitative or even non-numerical mathematics should be cheaper in terms of data and resource requirements than most current types of models and still should provide rigorous answers to the basic questions posed. This approach may be expected to help developing countries in resolving the current dilemma between, on the one hand, the urgent need to understand their ecosystems in order to provide adequate management and, on the other hand, the severe restrictions on available money and on the number of specialized scientists who can be deployed. 

However, no consideration of future possibilities should be allowed to conceal the more important fact that techniques already exist to take the EIA far beyond the subjective and speculative stage at which it is frequently practised at present.