In drylands, rainfall is typically delivered during short-lived and localised convective storms, the characteristics of which determine how water is partitioned into different terrestrial stores. However, rainfall datasets used in hydrological modelling and assessments of water resources are typically derived from climate models that are too coarse to represent convective processes occurring at scales smaller than the model. In this paper, we quantify the impact of climate model representation of convection on the simulated water balance at four locations in the Horn of Africa: a humid site in the Ethiopian Highlands, a semi-arid site in southern Kenya, an arid site in eastern Ethiopia, and a hyper-arid site in northern Somalia. We benchmark the novel pan-Africa convection-permitting climate model (CP4A) and its parameterised counterpart (P25) against high-resolution satellite-derived gridded datasets of rainfall (IMERG) and PET (hPET). The comparison shows that explicitly resolving convection improves the characterisation of rainfall frequency, intensity, and the relative contribution of low vs high-intensity rainfall to annual totals. We also demonstrate that the representation of convection can impact model PET. However, differences in PET between CP4A and P25 are more muted relative to rainfall, and both can capture seasonal and diurnal PET dynamics. To establish how climate model representation of convection can impact hydrology, we then ran a series of one-dimensional hydrological model experiments along an aridity gradient across the Horn of Africa using Hydrus 1-D, where at each of our four sites, Hydrus was driven by rainfall and PET from CP4A and P25 (and hPET). The ‘drizzle’ bias in P25 means that when rainfall is propagated through Hydrus, wetting fronts are more restricted to upper soil layers, resulting in higher evaporative losses, lower soil moisture, and bottom drainage in drylands. While at our humid site in the Ethiopian Highlands, there are minimal differences in hydrological outcomes; in drylands, the more intense and intermittent rainfall in CP4A means surface runoff is up to ten times higher and bottom drainage up to 25 times higher. We conclude that dryland hydrology is highly sensitive to climate model representation of convection and that forcing hydrological model projections with convectional climate models that parameterise the average effects of convection risks underestimating future crop health, groundwater availability, or flood risk.