Integration of GIS and machine learning for saltwater intrusion forecasting in Ho Chi Minh city, Vietnam under climate change

This study applied machine learning/ deep learning, and geographic information systems to predict saltwater intrusion in Ho Chi Minh City, Vietnam using monthly salinity, temperature, rainfall, and tide datasets between 2008 and 2024. The results indicated that the Long Short-Term Memory outperformed the Random Forest model with higher accuracy across almost stations – five out of six stations, particularly at An Ha station (NSE = 0.92 and PBIAS = -2.12%). In the future, the forecasting results show that serious salinity intrusion in coastal areas such as Can Gio district, and there is a risk of spreading deep into the area around Nha Be, and Ong Thin stations. Specifically, salinity peaks in 2029 (spread fast and deep with peak value of 13.84‰) and 2033 (spread slowly with peak value of 14.16‰), following by a slight decline in 2035. Moreover, using geographic information systems to spatial analysis highlights high-risk zones, contributing to support the management of water resources, sustainable urban planning orientation and promoting adaptive solutions to protect the ecosystem and livelihoods of the coastal residential community affected.