Innovative technologies in field experimentation have advanced the conceptual understanding of groundwater-surface water interactions, in particular at patch to reach scales. Up-scaling of this knowledge often requires generalisations in numerical models. Recent studies of the importance of small-scale processes and conditions at larger reach and catchment scales have sparked discussions to what degree common simplifications made in groundwater-surface water modelling may influence the ability to simulate interface processes realistically and contribute to informed decision making for river basin and aquifer management. In this paper, we address one specific question: Does increasing complexity improve the performance of groundwater-surface water models across the groundwater – surface water interface? A 3D model of the Tern River, UK, was developed to investigate groundwater flow paths, residence time distributions and groundwater-surface water exchange. The model was set up to test two comparative parameterisations: (1) homogeneous representation of the shallow subsurface and (2) heterogeneous subsurface geology utilising extensive core data and Ground Penetrating Radar (GPR) surveys of the area and in particular the streambed interface. Both the models were compared for residence time distributions and development of preferential flow paths. The models were validated against continuous hydraulic head readings at piezometers and Distributed Temperature Sensors (DTS)-based information of groundwater-surface water exchange. The heterogeneous model predicted increased lateral flow and altered preferential flow paths around low conductivity structures and differences in residence times within the site that was controlled by the subsurface structure. The differences between the homogeneous and heterogeneous subsurface models indicate that increased model complexity produced more accurate representation of the site conditions. In comparison, the homogenous model with simplified subsurface conditions failed to adequately represent interface exchange flow patterns and residence time distributions in the subsurface. Importantly, quantified residence time distributions in the model including heterogeneity helped to identify biogeochemical hotspots. Results on a limited number of case studies confirm that the atmospheric tracers of the last 50 years give accurate estimates of cumulative residence times and of the transport fate of some contaminants like nitrates. It already appears that the quality of the prediction does not always come from the capacity of the transit-time model to fit the actual transit-time distribution but also from the nature of the prediction or of the hydrological regime. Access to a wide range of well-informed and calibrated models taken as synthetic references should be developed to confirm and refine these early conclusions.