A neural network approach to selecting indicators for a sustainable ecosystem

The determination of indicators for monitoring the natural environment to ensure the ecological sustainability of areas under stress is problematic and prone to bias. Further, because of limitations in the resources available for data collection, such indicators should be selected carefully and their number confined to the minimum required to effectively monitor the system under study. Statistical techniques have traditionally been used to help select indicators, but often data lack the requirements of valid statistical analyses: minimal noise, variables linearly separable, variables able to be assigned numerical values, and high dimensionality. In this study, an alternate robust technique, artificial neural networks, is used to examine ecosystem data in multidimensional space and to select the minimum number of measured indicators that have the greatest weight in maintaining a sustainable ecosystem. The case study involves selecting indicators for ensuring the completion of ecologically sustainable army training conducted in a mixed grass prairie ecosystem. A neural network model was created that reduced the number of required measured indicators from 62 to 12, while minimizing researcher bias. Key words: ecological sustainability indicators, neural networks, military training.