Forecasting the growth of Ambystoma mexicanum via artificial neural networks based on water quality parameters in intensive culture systems

Abstract

Mexico is home to 17 species of the genus Ambystoma. Among them, 15 have some risk category, and the Ambystoma mexicanum is endangered. Different projects are focused on breeding them in aquaculture, where the growth rates of organisms are highly important. In this research, five parameters were selected as input variables: water temperature, pH, ammonia, nitrite, and nitrate nitrogen. These parameters were selected because they affect the growth of the species through their negative effects on the metabolism of the species. A computer model for growth forecasting is subsequently proposed. A backpropagation artificial neural network (ANN) was used to estimate weights via water quality parameters. The model has a hyperbolic-sigmoid transfer function for the hidden layer and a linear transfer function for the output layer. The data from the samples were generated over 90 days, divided into groups, and subjected to different diets. A trial-and-error approach optimized the number of hidden layer nodes and identified seven optimal neuron nodes. Then, autoregressive models were used to forecast individual parameters as the ANN input. The model obtained was able to predict weight within seven days with an error of less than three grams. The simulation results reveal that the ANN model efficiently forecasts the weight in Ambystoma intensive culture.