PREDICTION OF TOTAL NITROGEN AND PHOSPHORUS IN GUCHENG LAKE USING ARTIFICIAL NEURAL NETWORKS COMBINED WITH FACTOR CORRELATION ANALYSIS

Gucheng Lake has been suffering from eutrophication due to increased pollution and nutrient loads discharged into the watershed. Based on artificial neural networks (ANNs) and a 4-year record of water quality data (from 2006 to 2009), this study proposes an early-warning model for eutrophication aiming to predict the concentration of total nitrogen (TN) and total phosphorus (TP) of Gucheng Lake with a lead time of one week. To develop such data-driven models efficiently, a comprehensive sampling strategy is adopted to ensure that most relevant predictors for TN and TP are retained. Factor correlation analysis is then employed to further eliminate noisy predictors. The preferable selecting ranges of correlation coefficient values are proven to be [-1, -0.5] and [0.5, 1]. As a result, 6 and 18 input variables are filtered from 75 potential input variables to develop the TN and TP prediction models, respectively. The prediction models can achieve high performance. The validation results of TN (TP) showed that the correlation coefficient of 0.9915 (0.9945) and the RMSE of 0.0684 (0.0015), which have demonstrated the potential of ANN models to predict TN and TP conditions at Gucheng Lake.