The Faidherbia tree (Faidherbia albida) is frequently used as an intercrop in Sahelian agroforestry parklands due to its multi-purpose advantages and reverse phenology. However, its effect upon the water balance remains unclear, due to the challenges in directly measuring water fluxes in the underlying vadose zone. Mechanistic hydrological models can be inversely calibrated on transient observations and used to partition different hydrological components, but the computational burden of the analysis can become impractical if the model itself is computationally expensive. To overcome this limitation, and to provide novel insights into the hydrological role of Faidherbia, we combine a low-fidelity, one-dimensional hydrological model (HYDRUS-1D) with a kriging-based correction function to emulate the response of a high-fidelity, two-dimensional axisymmetric description of the system (HYDRUS-2D). Multiannual measurements of soil moisture and sap flow in a Senegal agroforestry parkland are used in conjunction with Bayesian inference to calibrate the resulting validated multifidelity surrogate, and to inversely estimate soil hydraulic and root water uptake parameters. Results show that the model can reproduce observations with good accuracy and limited uncertainty for both the calibration and the validation phases, and also confirm the phreatophytic behaviour of Faidherbia by indicating the existence of a moderately compensated root water uptake. Moreover, a local sensitivity analysis suggests that a fully compensated uptake could potentially reduce groundwater recharge by 13%. Interestingly, estimated soil hydraulic parameters hint at the possibility of root-induced changes in soil hydraulic properties that mimic preferential and/or macropore flow, resulting in sustained recharge fluxes (≈ 26% of the annual precipitation). The analysis indicates that overall, Faidherbia could have a net positive effect upon the water balance in arid areas.