Prediction and assessment of water quality index for surface water using 1D CNN and SVR models.

Water quality monitoring is essential for managing and protecting surface water resources. Traditionally, assessing water quality relied on time-consuming laboratory analyses, which were prone to errors and often limited in accuracy, efficiency, and scalability. Conventional methods for calculating the water quality index (WQI) aggregate various water quality parameters into a single value to represent overall water quality. However, these traditional approaches often fail to capture the complex, nonlinear relationships between water quality parameters and their temporal and spatial variability, especially under fluctuating environmental conditions. Therefore, these limitations can lead to poor decision-making regarding overall water quality. Adopting reliable artificial intelligence (AI) methods for predicting the WQI is essential for achieving accurate predictions by leveraging computational processing to effectively approximate the WQI from complex, combined input variables. This approach enables the use of real-time monitoring and forecasting, which is crucial for implementing sophisticated, adaptive methodologies capable of handling large datasets and delivering more robust models for water quality assessment. This study investigates the efficacy of a one-dimensional convolutional neural network (1D-CNN) and support vector regression (SVR) in predicting the WQI of surface water. The WQI was first calculated analytically using the weighted arithmetic index method (WA-WQI) and used to assess the surface water quality at Tilesdit Dam in Bouira, Algeria. The performance of the prediction models was evaluated using three statistical metrics: the coefficient of determination (R2), root mean square error (RMSE), and mean absolute percentage error (MAPE), based on a 9-year dataset (2009-2018) of six key parameters from the study area. The 1D-CNN model demonstrated significantly higher performance metrics during both the training phase (R2 = 0.9989, RMSE = 0.60, MAPE = 0.51) and the testing phase (R2 = 0.9962, RMSE = 0.57, MAPE = 1.06), outperforming the SVR model, which showed lower performance in both phases: training (R2 = 0.9597, RMSE = 3.61, MAPE = 0.57) and testing (R2 = 0.976, RMSE = 1.68, MAPE = 3.01). Thus, the proposed approach to surface water quality assessment offers effective, adaptive, real-time solutions for advanced control strategies, resulting in more efficient water resource management in the study area. This method represents a significant advancement over conventional analytical techniques and supports proactive water resource management.