The integration of ecological, neural and spatial modelling for monitoring and prediction for semi-arid landscapes

Land degradation in the southern central part of the South Island, New Zealand is a pressing environmental issue. There is a requirement for a method encompassing integrated monitoring and modelling. This paper describes an approach that combines ecological modelling using regression methods and neural networks with spatial information including remotely sensed information. Regression techniques are used initially for modelling with result of r(2) = 0.68. Prompted by some residual concerns about disaggregation, a series of neural networks were designed. The result of these latter tests produced r(2) = 0.86. A study was undertaken to analyse these results and to use the models for extrapolation. The models were useful for the prediction and simulation of ground based land monitoring data, and successfully predicted the condition of a large area of land when applied to a spatial dataset. Copyright (C) 1996 Elsevier Science Ltd.