It is essential to be able to model the surface–subsurface flow processes that occur within and at the vicinity of channels and ditches because of their major role in groundwater recharge in arid regions. The main aim of this study was to evaluate a three-dimensional Richards’ equation-based modelling approach for simulating groundwater recharge. Another aim was to evaluate various strategies for estimating the hydraulic properties of the soil–aquifer system near the ditch. The case study consists of two infiltration experiments on a ditch in a catchment located in South-France. One experiment was used for calibrating the soil hydraulic properties by inverse modelling and the other for validation. The simulations were performed with the SWMS_3D simulation code, which computes three-dimensional saturated–unsaturated water flow and solute transport. For calibration and validation, two criteria were considered: piezometric heads at different distances from the ditch and hydraulic head profiles. Four scenarios of hydraulic parameter distribution were tested: an ‘‘a priori’’ scenario with observed values obtained by classical measurement techniques and three calibration scenarios. These are ‘‘homogeneous isotropic’’, ‘‘heterogeneous isotropic’’ and ‘‘heterogeneous anisotropic’’ scenarios. The results obtained show that a threedimensional Richards’ equation-based approach could satisfactorily simulate groundwater recharge by seepage losses from ditches in case of induced surface ponding experiment. Although a multi-criteria calibration was performed, three sets of soil hydraulic properties of various complexity that could reproduce satisfactorily the flow experiments were found. Finally, the results obtained also showed that representing the variability of soil properties in greater detail did not necessarily improve water flow simulation. The Akaike information criteria was used in order to discriminate between the well-calibrated scenarios. The ‘‘homogeneous isotropic’’ scenario represents the best compromise between goodness of fit and parsimony. The study also indicates that hydraulic heads are not necessarily sufficient for the thorough evaluation of groundwater flow models