Spatial model development for resource management decision making and strategy formulation : application of neural network (Mounds State Park, Anderson, Indiana)

No Thumbnail Available
Authors
Guisse, Amadou W.
Advisor
Gimblett, H. Randy
Issue Date
1993
Keyword
Degree
Thesis (M.L.A.)
Department
Department of Landscape Architecture
Other Identifiers
Abstract

An important requirement of a rational policy for provision of outdoor recreation opportunities is some understanding of natural processes and public concern and /or preferences. Computerized land use suitability mapping is a technique which can help find the best location for a variety of developmental actions given a set of goals and other criteria. Over the past two decades, the methods and techniques of land use planning have been engaged in a revolution on at least two fronts as to shift the basic theories and attitudes of which land use decisions are based. The first of these fronts is the inclusion of environmental concerns, and the second is the application of more systematic methods or models. While these automated capabilities have shed new light on environmental issues, they, unfortunately, have failed to develop sufficient intelligence and adaptation to accurately model the dynamics of ecosystems.The work reported proceeds on the belief that neural network models can be used to assess and develop resource management strategies for Mounds State Park, Anderson, Indiana. The study combines a photographic survey technique with a geographic information system (GIS) and artificial neural networks (NN) to investigate the perceived impact of park management activities on recreation opportunities and experiences. It is unique in that it incorporates both survey data with spatial data and an optimizing technique to develop a model for predicting perceived management values for short and long term recreation management.According to Jeannette Stanley and Evan Bak (1988) a neural network is a massively parallel, dynamic systems of highly interconnected interacting parts based on neurobiological models. The behavior of the network depends heavily on the connection details. The state of the network evolves continually with time. Networks are considered clever and intuitive because they learn by example rather than following simple programming rules. They are defined by a set of rules or patterns based on expertise or perception for better decision making. With experience networks become sensitive to subtle relationships in the environment which are not obvious to humans.The model was developed as a counter-propagation network with a four layer learning network consisting of an input layer, a normalized layer, a kohonen layer, and an output layer. The counter-propagation network is a feed-forward network which combines Kohonen and Widrow-Hoff learning rules for a new type of mapping neural network. The network was trained with patterns derived by mapping five variables (slope, aspect, vegetation, soil, site features) and survey responses from three groups. The responses included, for each viewshed, the preference and management values, and three recreational activities each group associated with a given landscape. Overall the model behaves properly in learning the different rules and generalizing in cases where inputs had not been shown to the network apriori. Maps are provided to illustrate the different responses obtained from each group and simulated by the model. The study is not conclusive as to the capabilities of the combination of GIS techniques and neural networks, but it gives a good flavor of what can be achieved when accurate mapping information is used by an intelligent system for decision making.

Collections