Assessing water quality of an ecologically critical urban canal incorporating machine learning approaches

This study assessed water quality (WQ) in Tongi Canal, an ecologically critical and economically important urban canal in Bangladesh. The researchers employed the Root Mean Square Water Quality Index (RMS-WQI) model, utilizing seven WQ indicators, including temperature, dissolve oxygen, electrical conductivity, lead, cadmium, and iron to calculate the water quality index (WQI) score. The results showed that most of the water sampling locations showed poor WQ, with many indicators violating Bangladesh's environmental conservation regulations. This study employed eight machine learning algorithms, where the Gaussian process regression (GPR) model demonstrated superior performance (training RMSE = 1.77, testing RMSE = 0.0006) in predicting WQI scores. To validate the GPR model's performance, several performance measures, including the coefficient of determination (R2), the Nash-Sutcliffe efficiency (NSE), the model efficiency factor (MEF), Z statistics, and Taylor diagram analysis, were employed. The GPR model exhibited higher sensitivity (R2 = 1.0) and efficiency (NSE = 1.0, MEF = 0.0) in predicting WQ. The analysis of model uncertainty (standard uncertainty = 7.08 +/- 0.9025; expanded uncertainty = 7.08 +/- 1.846) indicates that the RMS-WQI model holds potential for assessing the WQ of inland waterbodies. These findings indicate that the RMS-WQI model could be an effective approach for assessing inland waters across Bangladesh. The study's results showed that most of the WQ indicators did not meet the recommended guidelines, indicating that the water in the Tongi Canal is unsafe and unsuitable for various purposes. The study's implications extend beyond the Tongi Canal and could contribute to WQ management initiatives across Bangladesh.