Multivariate statistical analysis of hydrogeochemical and stable isotopic data for characterising groundwater dynamics of an anthropized alluvial aquifer

In France, alluvial aquifers provide 45% of the freshwater used for drinking, agriculture, and industry (Maréchal and Rouillard, 2020). These aquifers are often hydraulically connected to rivers and are particularly vulnerable due to their proximity to the surface. This proximity makes them sensitive to anthropogenic pressures both quantitatively and qualitatively. The current study aims to better understand groundwater-surface water interactions in an anthropized alluvial aquifer in the lower Rhône Valley. This goal will be achieved using multivariate statistical methods on hydrogeochemical and stable water isotope data. Two datasets were analyzed. One dataset contained inert tracers (Cl⁻, Br⁻, δ²H and δ¹⁸O) and included 667 water samples (40 rainwater, 110 surface water and 517 groundwater). The other dataset contained major and minor ions (Ca²⁺, Mg²⁺, K⁺, Na⁺, Cl⁻, SO₄^2-, alkalinity, NO₃⁻, Br⁻ and U(VI)) and stable water isotopes (δ²H and δ¹⁸O) data., and included 374 water samples (37 surface water and 337 groundwater). First, a hierarchical cluster analysis (HCA) was applied to both datasets to better understand groundwater recharge and the geochemical processes that control groundwater chemistry in the study area. We used the recently developed t-distributed stochastic neighbor embedding (t-SNE) (Van Der Maaten and Hinton, 2008) method and principal component analysis (PCA) to assist with the cluster analysis and visualization. When HCA, PCA, and t-SNE were applied to inert tracers dataset, the results first revealed the distribution of groundwater on the study site between two recharge sources: the Rhône River and rainfall. Water samples collected along the Rhône River boundary are characterized by highly depleted δ²H and δ¹⁸O signatures, indicating the significant influence (up to 80%) of the Rhône on the recharge of the alluvial aquifer in this area. In contrast, groundwater samples collected in the northern and northwestern parts of the site showed highly enriched δ²H and δ¹⁸O signatures, similar to those of rainfall indicating dominant recharge from local precipitation. Others water samples are characterized by intermediate δ²H and δ¹⁸O signatures, falling between the signatures of rainfall and the Rhône River, indicating mixed recharge. These water samples are predominantly distributed in the southern part of the site. Second, the results revealed the response of the alluvial groundwater to exceptional climatic events. In 2022, particular isotopic signatures with higher deuterium excess and high chloride concentrations were observed in rainwater and the Rhône River, which are reflected in groundwater. This suggests an influence of continental air masses originating from the Sahara Desert (Xu-Yang et al., 2025). When HCA, PCA, and t-SNE were applied to the second dataset containing hydrogeochemical and stable water isotope data, the results identified water samples with high uranium and chloride concentrations, likely due to historical pollution. This study shows that t-SNE is a promising tool for assessing groundwater-surface water interactions in an alluvial aquifer when used to assist in cluster analysis. Compared with PCA, t-SNE can better identify hidden information and perform much better with complex, nonlinear hydrogeochemical, and stable water isotope data.

 

References

Maréchal and Rouillard, 2020.https://doi.org/10.1007/978-3-030-32766-8_2

Van Der Maaten and Hinton, 2008.Res.9,2579–2625.

Xu-Yang, et al., 2025.https://doi.org/10.1126/sciadv.adr9192