Predicting fish yield of African lakes using neural networks

Artificial neural network (ANN) approaches to modelling and prediction of fish yield as related to the environmental characteristics were developed from the combination of six variables: catchment area over maximum area, fishing effort, conductivity, depth, altitude and latitude. For a total of 59 lakes studied, the correlation coefficients obtained between the estimated and observed values of abundance were significantly high with the neural network procedure (r adjusted=0.95, P<0.01). The predictive power of the ANN models was determined by the leave one out cross-validation procedures. This is an appropriate testing method when the data set is quite small and/or when each sample is likely to have ‘unique information’ that is relevant to the model. Fish yields estimated with this method were significantly related to the observed fish yields with the correlation coefficient reaching 0.83 (P<0.01). Our study shows the advantages of the backpropagation procedure of the neural network in stochastic approaches to fisheries ecology. Using the specific algorithm, we can identify the factor influencing the fish yield and the mode of action of each factor. The limitations of the neural network approaches as well as statistical and ecological perspectives are discussed.