Local characterization of groundwater systems is critical for managing and protecting vulnerable resources. Geophysical methods can provide dense imaging of subsurface parameters to delineate lithological boundaries and water tables for hydrogeological investigation. Though, using a single geophysical method for determining lithologies can yield erroneous interpretations as different lithologies can have similar properties. By using several geophysical methods, it is possible to reduce this risk and better assign likely lithologies to subsurface units. We present two case studies where transient electromagnetic (TEM) and surface nuclear magnetic resonance (SNMR) are used in combination to delineate hydrogeological structures. Novel spatially constrained inversion in SNMR was used to provide horizontal consistency between soundings. Three coincident parameters, resistivity from the TEM measurements and water content and relaxation time from the SNMR measurements were used in a K-means clustering scheme to resolve subsurface structures. The K-means clustering was evaluated with a silhouette index to pick the number of clusters. After clustering, each cluster was assigned a hydrogeological description based on the distinct features in the three parameters, e.g. a low resistivity, high water content, and high T2* is assigned as saltwater saturated sand. In the first case study, the clusters enabled improved resolution of a regional water table in an unconfined aquifer setting by the multi-geophysical approach. The water table estimates were positively evaluated against multiple boreholes within 500 m of coincident geophysical models. The second case study illustrates how clustering, of SNMR and TEM models, can delineate saltwater intrusion in an island coastal aquifer, which would not be possible with any of these methods individually. Additionally, the clustering resolved the main shallow aquifer on the island. Our work illustrates how the combination of geophysical data can be used to improve resolution of hydrogeological layers and reduce interpretational bias.