Application of Artificial Neural Networks (ANN) to Primary Production Time-series Data

Abstract

An artificial neural network (ANN) model was applied for predicting primary productivity (PP) from a 12 year time series (1985-1996) of monthly observations on a set of environmental and climatic variables from the Gullmar Ford (south-western Sweden). Results indicate a good fit between observed and predicted PP values. ANN can be regarded as a novel tool for primary production modelling and more generally when the numbers of environmental and climatic co-variates are large. ANN models fitted the data with a lower root mean square error of prediction (RMSEP) than more conventional and classic methods, such as multiple regression. Predictions of future changes in primary production from the same set of input variables using network set-ups with PP leading the input variables by 1, 2 and 3 month lags indicated that RMSEP was about the same as for the case with no lag. These results show the possibility of generating patterns of future fluctuations in primary productivity using ANN.