2003 Seattle Annual Meeting (November 2–5, 2003)

Paper No. 6
Presentation Time: 3:10 PM

DECISION TREES WITHIN EXPERT SYSTEMS FOR MODELLING ECOLOGICAL KNOWLEDGE RELATED TO WATER LEVEL MANAGEMENT FOR PALUSTRINE WETLANDS


SOJDA, Richard S., Northern Rocky Mountain Science Center, USDI Geological Survey, 212 AJM Johnson Hall, Montana State University, Bozeman, MT 59717-3492, sojda@usgs.gov

An expert system was developed to suggest seasonal water levels for palustrine emergent wetlands in the Northern Rocky Mountains.  Recommendations were based on knowledge of vegetation, hydrology, water quality, soils, weather, wetland successional dynamics, and migratory bird management objectives. The knowledge base was developed using standard knowledge engineering techniques from the field of artificial intelligence. Pertinent knowledge was gleaned from a small team of experts and the published literature and then encoded using a commercial expert system development shell. This process resulted in 1882 production rules mostly organized into 43 decision trees. Such rules are of the form:

 IF "the wetland exhibits a classic pattern of vegetation zonation..."
 AND "...within each zone, vegetation is relatively monotypic"
 AND "...vegetation does not exhibit a clumped pattern"
 THEN "stable water levels appear to be occurring"

Individual rules were linked graphically into decision trees, representing broader components of ecological knowledge.  Correct interpretation of questions posed by the system is dependent on judicious wording of those questions, providing specific definitions to the user where necessary, and allowing user input only from lists of specific responses.

Here, I focus on two examples of decision trees.  One guides an assessment about stable water levels in a wetland based on simple questions posed to the user related to vegetative species presence and interspersion patterns. Another similarly assists the overall system provide management recommendations for promoting Sago pondweed (Potamogeton pectinatus) using information from the user on water quality and the recent history of surface water hydrology.

One of the values of expert system technology is that it allows knowledge engineers to make existing expertise available to others. Often such an approach is most valuable in situations where it is not currently feasible to procedurally model causal relationships because 1) empirical data from scientific experiments is lacking, or 2) the complexity of the system of interest prevents mathematical descriptions of those causal relationships.

It is my opinion that both of these conditions exist for many aspects of montane wetlands in the Northern Rocky Mountains.