Estimation of water quality parameters based on time series hydrometeorological data in Miaowan Island

Abstract

Water quality parameters (WQPs), such as dissolved oxygen (DO), chemical oxygen demand (COD) and chlorophyll (Chl), are important indicators of ecosystem system. The easy availability of hydro-meteorological parameters (HMPs) provides an important tool for estimating WQPs. In this study, using three empirical machine learning (ML) algorithms, namely Multi-Layer Perceptron (MLP), Random Forest (RF), and M5 Model Tree (M5T), and based on a large amount of time series in situ monitoring of HMPs and WQPs data over a six-month period in Miaowan Island, a new ML model was developed to estimate DO, COD, and Chl in a simple and cost-effective manner. Through feature selection, the input HMPs for ML the ML models include temperature, salinity, depth, air pressure and relative humidity. The results of the accuracy evaluation showed that the RF-based model was the optimal model for estimating DO, COD, and Chl with R2 values of 0.987, 0.992, and 0.965 on the testing set, respectively. With the RF-based model, the WQPs at two sites of Miaowan Island were estimated over a temporal sequence, and the estimated results are highly consistent with the measurements obtained from IEEIoTS. Furthermore, we extended the application of the RF-based model to estimate DO in Zhanjiang Bay throughout August 2023. This extension was based on in situ monitoring of HMPs obtained from WQMS, and comparison with the measured DO. They have corresponding temporal trends but with variations in values, potentially attributable to the inherent normality of the model. The results suggest that the RF-based model based on HMPs information provides a practical approach for estimating WQPs.

Keywords

• Water quality parameter
• Hydrometeorological
• Island
• Machine learning