Water Quality Prediction using LSTM: A Deep Learning Approach at Wat Makham Station, Chao Phraya River, Thailand

¹ Department of Mechatronics, Faculty of Technical Education, Rajamangala University of Technology Thanyaburi, Pathum Thani, Thailand
² Department of Electromedical Technology, Faculty of Vocational Study, Sanata Dharma University, Indonesia
³ Department of Mechatronics, Faculty of Vocational Study, Sanata Dharma University, Indonesia

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

This study develops a Long Short-Term Memory (LSTM) neural network for forecasting water quality parameters at the Wat Makham Station based on data collected from the Chao Phraya River, Thailand, for nine months. The study used IoT sensors to collect real-time values for ten water quality indicators: Turbidity (TURB_NTU), Optical Dissolved Oxygen (HDO), Dissolved Oxygen Saturation (HDO_SAT), Spatial Conductivity (SPCOND), Acidity/Basicity (pH), Total Dissolved Solids (TDS), Salinity (SALINITY), Temperature (TEMP), Chlorophyll (CHL), and Depth (DEPTH). The study identified water quality indicators through the implementation of an LSTM model following application of data cleansing techniques, using mainly the Interquartile Range (IQR) method for outlier detection. The results confirm that prediction accuracy varied across parameters. For stable indicators, very high prediction accuracy was achieved: for pH, MSE = 0.0064, MAPE = 0.89%, RMSE = 0.0800, and RMSPE = 1.12%; for salinity, MSE = 0.0006, MAPE = 10.55%, RMSE = 0.0246, and RMSPE = 41.14%. Temperatures were predicted with high confidence also: MAPE = 2.59% and RMSPE = 3.24%. In contrast, highly volatile parameters were difficult to predict; Turbidity MAPE = 32.87% and RMSPE = 109.22%; Chlorophyll MAPE = 38.64% and RMSPE = 190.15%.