Excessive groundwater pumping exacerbates aquifer depletion and poses a major threat in regions all around the world that already suffer from overuse or climate change. In this situation, accurate and reliable predictions of long-term aquifer water balances are key prerequisites to manage groundwater sustainably. Compared to spatially explicit numerical models, lumped hydrological models are computationally fast, lean on data requirement and more accessable for quantifying uncertainty. However, lumped hydrological models are mainly designed to simulate river discharge only, not aquifer storage. Consequently, calibration only includes stream flow data. In this study, we hypothesize that we can extend a lumped hydrological models (here HBV) towards a lumped geohydrological model (LGhM) by additional, designated terms for water budget and groundwater storage. The model building is inspired by the geometry and hydrogeological large-scale properties of the catchment's aquifers. Underground flow routing resembles major groundwater flow paths. The model is calibrated and evaluated on both groundwater storage data and surface discharge data. We apply our LGhM to a MODFLOW-based virtual reality describing the unconfined Wairau Plain Aquifer, New Zealand. We consider and discuss specifically river-groundwater exchange processes, long-term forecast of aquifer storage dynamics, and groundwater depletion in a hypothetical, persistent drought. Our model evaluation shows very plausible predictive capabilities in 40-year forecasts with synthetic weather time series and several years of groundwater depletion in the extreme drought case.