Salinity Prediction in Coastal Aquifers of the Vietnamese Mekong River Delta Using Innovative Machine Learning Algorithms

Abstract

Groundwater salinization is a severe issue, causing various problems to human health, agriculture, ecosystems, and infrastructure in many coastal regions across the world. However, this phenomenon is difficult to predict with high accuracy. In this study, we propose and verify a new artificial intelligence approach for predicting groundwater salinity and identifying the main factors of salinization. The coastal aquifers of the Mekong River Delta (Vietnam) were selected to test the new approach. In the proposed approach, Extreme Gradient Boosting (XGB) was used to build a groundwater salinity model, and Genetic Optimization (GO) was employed to optimize the model parameters. Gaussian Processes (GP) and Random Forests (RF) were also used as a benchmark for the model comparison. For this regard, a groundwater salinity database with 215 groundwater samples and 20 driven factors related to hydrology, geology, geography, and anthropogenic activities was prepared. Performance of the models was assessed using Correlation Coefficient (r), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and Mean Absolute Error (MAE). The results show that the proposed GO-XGB model yields high performance both on the training dataset (r = 0.999, RMSE = 18.450, MAPE = 2.070, and MAE = 4.864) and the validation dataset (r = 0.787, RMSE = 141.042, MAPE = 87.250, and MAE = 74.993). The proposed GO-XGB model performed better predictive result compared to the benchmark, GP, and RF. Among the 20 factors, groundwater level, vertical hydraulic conductivity, lithology, extraction capacity, horizontal hydraulic conductivity, distance to saline sources, and well density are the most important factors to groundwater salinization prediction.