Machine Learning–Based Early-Warning Systems for Salinity Intrusion in the Mekong River Delta, Vietnam

Abstract

The Mekong Delta is renowned as one of the world’s most productive regions for rice cultivation. However, it faces significant challenges due to salinity intrusion, where seawater from the South China Sea flows upstream into the delta area. Early warning systems that can assess the severity of salinity intrusion events are crucial in mitigating its negative impacts. In this study, various machine learning strategies are presented to forecast salinity intrusion in the Mekong Delta. The available data are fully utilized using the principal component analysis technique in conjunction with 13 advanced machine learning algorithms. The results demonstrate that logistic regression, support vector classification, and quadratic discriminant analysis models consistently achieve accuracies higher than 86% across most data sets. Additionally, random forest, extra trees, gradient boosting, and bagging classifier models demonstrate accuracies of 95% and 100% for specific data sets. These findings highlight the effectiveness of machine learning models in forecasting salinity intrusion and present a range of algorithms and data sets that can be employed for accurate predictions in the Mekong Delta region.