Prediction and elucidation of algal dynamic variation in Gonghu Bay by using artificial neural networks and canonical correlation analysis

Abstract

This paper describes the training, validation and application of recurrent neural network (RNN) models to computing the algal dynamic variation at three sites in Gonghu Bay of Lake Taihu in summer. The input variables of Elman's RNN were selected by means of the canonical correspondence analysis (CCA) and Chl_a concentration as output variable. Sequentially, the conceptual models for Elman's RNN were established and the Elman models were trained and validated on daily data set. The values of Chl_a concentration computed by the models were closely related to their respective values measured at the three sites. The correlation coefficient (R 2 ) between the predicted Chl_a concentrations by the model and the observed value were 0.86-0.92. The results show that the CCA can efficiently ascertain appropriate input variables for Elman's RNN and the Elman's RNN can precisely forecast the Chl_a concentration at three different sites in Gonghu Bay of Lake Taihu in summer.