Neural network based modelling of environmental variables: A systematic approach

Abstract

Feedforward artificial neural networks (ANNs) that are trained with the back-propagation algorithm are a useful tool for modelling environmental systems. They have already been successfully used to model salinity, nutrient concentrations, air pollution, and algal growth. These successes, coupled with their suitability for modelling complex systems, have resulted in an increase in their popularity and their application in an ever increasing number of areas. They are generally treated as black box models that are able to capture underlying relationships when presented with input and output data. In many instances, little consideration is given to potential input data and the internal workings of ANNs. This can result in inferior model performance and an inability to accurately compare the performance of different ANN models. Back-propagation networks employ a modelling philosophy that is similar to that of statistical methods in the sense that unknown model parameters (i.e., connection weights) are adjusted in order to obtain the best match between a historical set of model inputs and corresponding outputs. Consequently, the principles that are considered good practice in the development of statistical models should be considered. In this paper, a systematic approach to the development of ANN based forecasting models is presented, which is intended to act as a guide for potential and current users of feedforward ANNs that are trained with the back-propagation algorithm. Issues that need to be considered in the model development phase are discussed and ways of addressing them presented. The major areas covered include data transformation, the determination of appropriate model inputs, the determination of an appropriate network geometry, the optimisation of connection weights, and validation of model performance.