Examination of Multiple Linear Regression (MLR) and Neural Network (NN) Models to Predict Eutrophication Levels in Lake Champlain

Abstract

Eutrophication is one of the main causes of the degradation of lake ecosystems. In this paper, multiple linear regression (MLR) and neural network (NN) methods were developed as empirical models to predict chlorophyll-a (Chl-a) concentrations in Lake Champlain. The models were developed using a large dataset collected from Lake Champlain over a 24-year period from 1992 to 2016. The dataset consisted of monitoring depth (Depth), total phosphorus (TP), total nitrogen (TN), alkalinity (RegAlk), temperature (TempC), chloride (Cl) and secchi depth (Secchi). Statistical analyses showed that TP, Secchi, TN and Depth demonstrated strong relationships with Chl-a concentrations. The simulation results revealed that both the MLR and NN models performed well in predicting Chl-a concentrations, especially for low to moderate concentrations of Chl-a (<7.5 μg/L). The NN model showed better accuracy and generalization performance in comparison with the MLR model for both the training and verification processes. In addition, both the developed MLR and NN models produce good results when used to predict Chl-a concentrations from 2017 to 2021. However, neither the MLR nor NN models can accurately predict high Chl-a concentrations (> 7.5 μg/L). These models can be useful for improving lake management and providing early warnings regarding the problem of eutrophication.

Keywords: neural network (NN), multiple linear regression (MLR), eutrophication, chlorophyll-a (Chl-a) prediction, Lake Champlain, empirical models