Assessing Climate Factors and Cyanobacterial Abundance on Microcystins Prediction Using Artificial Neural Network: A Case Study in Malaysia’s Drinking Water Reservoir

Abstract

Toxic cyanobacterial blooms often lead to contamination with cyanotoxins, particularly microcystins. This study aims to examine microcystins persistence in a selected public water supply system and predict their concentration at various points based on climate factors and cyanobacterial abundance. Using the Enzyme-Linked Immunosorbent Assay (ELISA) method, microcystins concentrations were quantified at various points of the water supply system, including the raw water intake, reservoir, water treatment plant outlet, and distribution system. The highest microcystins concentration was detected at the reservoir with a mean concentration of 2.63 μg/L. An artificial neural network (ANN) model was developed to predict microcystins concentration. Rainfall, temperature, chlorophyll-a, phycocyanin (BGA-PC), and mcyE gene copy numbers were used as inputs, while microcystins concentrations at various water sampling points served as outputs of the multilayer perceptron ANN. Using the Statistical Package for the Social Sciences (SPSS, ver. 29), three networks with scaled conjugate gradient, sigmoid functions, and one hidden layer with 4 to 13 neurons were trained and validated to determine the best configuration that fits the observed data. The result shows a satisfactory prediction at the reservoir (Point 2) with low values of error (root mean square error = 0.065) and high coefficient values (R² = 0.894) between experimental and predicted values, which are below the maximum value of the actual concentrations. Phycocyanin (BGA-PC) and chlorophyll-a had the most positive effects in predicting microcystins concentrations. These results indicate that ANN modelling can be a reliable tool for predicting microcystins contamination in drinking water reservoir.