Using Case-Based Reasoning to Learn About Ecological Engineering.

摘要

Ecological engineering, the practice of designing, creating or manipulating, and monitoring ecosystems, is applied for a variety of purposes benefiting both human society and the natural environment, often integratively. While there are basic principles that help practitioners in the development and implementation process, at this time there is no comprehensive theory that guides the design of ecosystems. In order for such theory to be developed, extensive knowledge about the interactions between ecosystem constitution and comportment, and ways to analyze and integrate this knowledge, are needed. Consequently, the ability to qualitatively and quantitatively evaluate large datasets in a multivariate fashion is required. Thus, the objective of this project was to investigate the use of case-based reasoning as a method of gathering and analyzing large sets of ecological data not only for prediction but for engineering purposes, a previously untested application.;To maximize the number of cases to be analyzed without limiting the inputs to only known systems described in the literature, a virtual ecosystem and simulation platform was created. Simulation outputs and values for applied measures were compiled into a case base for use with a case-based reasoner to attempt to predict the results of several additional randomly created virtual ecosystems. Actual results were compared to the predicted results. The accuracy of the predictions made by the case-based reasoner varied, but they were more than 75% accurate 83.3% of the time. An initial attempt was made to apply this approach to "engineering" ecosystems for specified performance levels within the virtual ecosystem framework. While the targeted values of persistence were not obtained, the "engineered" virtual ecosystems were more persistent overall than the randomly created systems, with an average ratio of 0.40527 surviving species to initial species versus an average persistence of 0.20750 for the random systems. This is indicative of the potential of this novel approach for data analysis in ecological engineering.