Spatial prediction of groundwater salinity in multiple aquifers of the Mekong Delta region using explainable machine learning models
Published in Water Research, 2024
Recommended citation: Jeong, H., Abbas, A., Kim, H. G., Van Hoan, H., Van Tuan, P., Long, P. T., Lee, E., & Cho, K. H. (2024). Spatial prediction of groundwater salinity in multiple aquifers of the Mekong Delta region using explainable machine learning models. Water Research, 266, 122404. https://doi.org/10.1016/j.watres.2024.122404
Salt is spreading through the groundwater of the Mekong Delta, where millions of people depend on wells for drinking water and irrigation. Rather than reduce the problem to a single number, this work predicted three separate chemical signatures of salt across the delta’s stacked layers of groundwater, comparing nine models built on field measurements. Chloride, the clearest signature, was predicted well enough to explain 94% of the variation, while the two subtler indicators proved harder, at 78% and 67%. Opening the models up to see what drove each prediction revealed that the dominant cause of the salt changes with both the indicator and the depth, and that the volume of water people pump out came through strongly for several of them. That is what makes the results usable for deciding where to intervene, rather than only for forecasting where salt will show up next.