Reports and Papers

Fedra, K. (1995)
Decision support for natural resources management:
Models, GIS and expert systems.
AI Applications, 9/3 (1995) pp 3-19.

Application Examples
EIA for Water Resources Development Projects

The development of large-scale water resources projects such as dams and reservoirs, as well as irrigation schemes, or flood control is increasingly faced by opposition on environmental or socio-economic grounds. Environmental impact assessment (EIA) is a required component of almost all such projects, and it is a typical example of a complex problem involving difficult assessment and trade-offs among a diverse group of actors.

To support a screening level assessment at an early stage of project planning offers the possibility to introduce environmental and social concerns early on before opposite views become entrenched and all the more difficult to reconcile, but it also implies that very little data may be available initially.

The MEXSES system described below (Fedra et al., 1991) has been implemented as a rule-based expert system, using hierarchical checklists to perform screening level environmental impact assessment. The current prototype system, developed for the Lower Mekong, is geared toward the assessment of water resources development projects such as dams and reservoirs, hydropower and irrigation schemes, flood control, navigation, aquaculture, etc.

The structure of the assessment process is based on the Asian Development Bank's Environmental Guideline Series (ADB, 1988). The indicators used to assess a given project are based on checklists of items specific to a project type, covering environmental as well as selected socio-economic topics, each indicator being rated on a qualitative scale, from not significant to major.

In the current prototype a system of hierarchical checklists is used with a rule-based deduction process including a recursive explain function and a knowledge base browser, both connected to a hypertext system. The browser and explanation function display the near natural language rules; hypertext links them to a handbook style definition and explanation of the terms and concepts used in the system as well as general background information on environmental impact assessment, including numerous examples, to provide a tutorial framework for the assessment.

Selecting a project from a list (or from the map of the integrated GIS) retrieves data already available on a specific project or project alternative. The Environmental Checklists are organized by project types, and grouped into problem classes. They include problems due to location, planning and design problems, problems during the construction phase, problems during project operation, and finally, environmental enhancement measures, which looks at possible enhancement or mitigation strategies. Project types include reservoirs and dams, hydropower projects including transmission lines, irrigation projects, fisheries and aquaculture, and could also include infrastructure projects (roads and highways, sewerage, water supply, etc.), navigation, erosion control, etc.

The checklists are designed to guide the analyst through a reasonably complete set of expected environmental impacts for a given project type. Subproblems or basic indicators covered in the checklists include, for example for a reservoir project, environmental impacts from, or in terms of:

resettlement; watershed degradation; encroachment upon precious ecosystems; encroachment on historical/cultural values; watershed erosion; reservoir siltation; impairment of navigation; changes in groundwater hydrology, water logging; seepage and evaporation losses; migration of valuable fish species; inundation of mineral resources/forests; other inundation losses and adverse effects; earthquake hazards, and local climatic change.

To provide an assessment for each item in the checklists, analysts can choose/set a value and then ask the system to check their hypothesis. This triggers a backward chaining inference system that will attempt to establish all the necessary preconditions for the result (the hypothesis) specified. If some of the required facts are unknown, the inference procedure will ask the user the necessary questions. As a final result, the user's assessment will either be confirmed or rejected.

Alternatively, the analyst can start a forward-chaining inference procedure, where the system will reason from the available data to arrive at a classification of impacts. Again, missing information will have to be supplied by the analyst in a question--answer dialogue.

The answers the user provides to the various questions posed are taken from a menu of possible answers offered by the system from its knowledge base. Most descriptors or variables used can be symbolic as well as numeric, and the user can choose the appropriate format depending on the information at hand; defaults associated with the various symbolic labels are offered, and an additional layer of context-sensitive help, explaining the various terms and concepts, as well as the background for each question, the range of possible answers, and illustrative examples are provided in the graphical interface through hypertext.

By using information that is likely to be available at an early project stage, the system will attempt to determine the expected severity of a given subproblem such as, eg., watershed erosion, by using rules that, for example, consider climatic and topographic data, soil and slope conditions, vegetation cover and land use, management practices, etc.

MEXSES uses a straight-forward knowledge representation, combining an object oriented design for the descriptors, the basic elements in the inference procedure, with near natural language rules (Fedra et al, 1991; Fedra 1992). As the basic object of the system, descriptors are the concepts or terms used in the knowledge representation; they are linked through the rules, that allow to derive values for a given descriptor from other descriptor values through the standard logical operations of first order logic.

Descriptors are defined as part of the knowledge base of the system. The definition includes basic characteristics such as name, type, and units of measurement, and a list or range of legal values the descriptor can take. Depending on the descriptor, these values can be symbolic, numeric, or both. Descriptor objects also know about methods they can use to establish their values in a given context. These methods include questions to ask of the user (Figure 4), data base or GIS references that trigger the appropriate retrieval or estimation function and rules.

For example, data from meteorological records or flow data can be retrieved from the nearest appropriate station, or interpolated where necessary; values for elevation, land use, soil type, slopes, or population density can be retrieved from the respective topical maps. In addition to rules and simple algebraic expressions and formulas that can be expressed within the rule syntax, methods can also be references to complex numerical functions and entire simulation models that can be used to obtain an appropriate value. In the interactive dialogue, the user can choose between different methods; priorities of methods, ie., which one should be tried first, are also defined in the descriptor definitions and can be dynamically modified through rules.

Finally, descriptors can have alternative sets of (partial) definitions to be used depending on the context and under rule control. Rules can result in the absolute assignment of descriptor values, their relative, incremental modification, or they can be used to control the inference strategy depending on context.

© Copyright 1995-2016 by:   ESS   Environmental Software and Services GmbH AUSTRIA | print page