Many types of ecological or environmental problems would benefit from models based on people's knowledge. To create ecological models with both expert and local people's knowledge, a multi-step fuzzy cognitive mapping approach is proposed. A cognitive map can be made of almost any system or problem. Cognitive maps are qualitative models of a system, consisting of variables and the causal relationships between those variables. We describe how our cognitive mapping research has been used in real environmental management applications. This research includes examining the perceptions of different stakeholders in an environmental conflict, obtaining the perceptions of different stakeholders to facilitate the development of participatory environmental management plans, and determining the wants and desires for resettlement of people displaced by a large scale dam project. Based on our research, which involved six separate studies, we have found that interviewees complete their cognitive maps in 40-90 min on average. These maps contain an average of 23 ± 2 S.D. variables with 37 ± 3 S.D. connections. People generally put more forcing functions into their maps than utility variables. Fuzzy cognitive mapping offers many advantages for ecological modeling including the ability to include abstract and aggregate variables in models, the ability to model relationships which are not known with certainty, the ability to model complex relationships which are full of feedback loops, and the ease and speed of obtaining and combining different knowledge sources and of running different policy options.

No abstract

12We review the use of artificial neural networks, particularly the feedforward multilayer 13 perceptron with back-propagation for training (MLP), in ecological modelling. In MLP 14 modeling, there are no assumptions about the underlying form of the data that must be met as 15 in standard statistical techniques. Instead the researchers should make clear the process of 16 modelling, because this is what is most critical to how the model performs and how the results 17 can be interpreted. Overtraining on data or giving vague references to how it was avoided is 18 the major problem. Various methods can be used to determine when to stop training in 19 artificial neural networks: 1) early stopping based on cross-validation, 2) stopping after a 20 analyst defined error is reached or after the error levels off, 3) use of a test data set. We do not 21 recommend the third method as the test data set is then not independent of model 22 development. Many studies used the testing data to optimize the model and training. 23Although this method may give the best model for that set of data it does not give 24 generalizability or improve understanding of the study system. The importance of an 25 independent data set cannot be overemphasized as we found dramatic differences in model 26 accuracy assessed with prediction accuracy on the training data set, as estimated with 27 bootstrapping, and from use of an independent data set. The comparison of the artificial 28 neural network with a general linear model (GLM) as a standard procedure is recommended 29 because a GLM may perform as well or better than the MLP. If the MLP model does not 30 predict better than a GLM, then there are no interactions or nonlinear terms that need to be 31 modelled and it will save time to use the GLM. MLP models should not be treated as black 32 box models but instead techniques such as sensitivity analyses, input variable relevances, 33 neural interpretation diagrams, randomization tests, and partial derivatives should be used to 34 make the model more transparent, and further our ecological understanding which is an 35 important goal of the modelling process. Based on our experience we discuss how to build an 36 MLP model and how to optimize the parameters and architecture. The process should be 37 explained explicitly to make the MLP models more readily accepted by the ecological 38 research community at large, as well as to make it possible to replicate the research. 39 40

Fuzzy cognitive mapping was used to develop a participatory ecosystem management plan for Uluabat Lake, Turkey. Interviews were conducted with stakeholders belonging to six different groups. Lake pollution was the most central and most mentioned variable for stakeholders. Stakeholder groups agree that lake pollution is negatively affecting ecosystem health and thereby local economies. Thus, reducing lake pollution was chosen as the overall goal for the management plan. Possible ways to reduce lake pollution and increase ecosystem health were seen differently by the different groups. Hunters, factory managers, NGO personnel, and local people thought industry was the main cause of lake pollution, while officials from the government and local municipalities thought roads and urban development contributed the most to lake pollution. Generally the stakeholder groups did not perceive their own actions as affecting the lake as strongly as other groups thought. For example, factory managers viewed factory pollution as negatively affecting the lake but less strongly than the other groups did. According to policy option simulations, reducing lake pollution had positive effects on all variables, especially fish, birds, animal husbandry, irrigation, agriculture, and the ecological balance of the lake. Results of this analysis were used to facilitate meetings among stakeholder groups and to develop a participatory ecosystem management plan. The analysis was useful for pointing out the similarities as well as the differences among the groups. It also helped the facilitators understand the focus of each stakeholder group and enabled them to suggest activities in which each group would want to participate.