An integrated hydrogeological modelling approach applicable to hard-rock aquifers in semi-arid data-scarce Africa was developed using remote sensing, rainfall-runoff modelling, and a three-dimensional (3D) dynamic model. The integrated modelling approach was applied to the Hout catchment, Limpopo Province, South Africa, an important agricultural region where groundwater abstraction for irrigation doubled during 1968-1986. Since the 1960s, groundwater levels in irrigated areas have displayed extended periods of decline with partial or full recovery in response to major decadal rainfall events or periods. The integrated dynamic 3D hydrogeological flow model, based on the One-Water Hydrologic Flow Model (MODFLOW-OWHM), helped to understand recharge and flow processes and inform water use and management. Irrigation abstraction was estimated based on irrigated crop area delineated using the Landsat Normalized Difference Vegetation Index (NDVI) and crop water requirements. Using groundwater level data, the model was calibrated (2008-2012) and validated (2013-2015). Estimated mean diffuse recharge (3.3 ± 2.5% of annual rainfall) compared well with estimates from the Precipitation Runoff Modelling System model. Recharge and groundwater storage showed significant inter-annual variability. The ephemeral river was found to be losing, with mean net flux to the aquifer (focused recharge) of~1.1% of annual rainfall. The results indicate a delicate human-natural system reliant on the small but highly variable recharge, propagating through variable pumping to an even more variable storage, making the combined system vulnerable to climate and anthropogenic changes. The integrated modelling is fundamental for understanding spatio-temporal variability in key parameters required for managing the groundwater resource sustainably.