Constructed wetlands (CWs) have been proved to successfully reduce the ecological impact of urban runoff on receiving water bodies. Simplified models predicting CW performances would ease the work of designers and stakeholders, but for proper parameterization, such models require large datasets that are difficult to observe due to the randomness of storm events. An alternative to observation lies in the use of robust mechanistic models. Here, we explore this approach using the HYDRUS-1D environment as mechanistic model and with particular emphasis on hydrodynamics. Based on experiments carried out on a 20 m2 pilot-unit, we performed both calibration and validation of the model. The singularity of the approach hinges on modeling the throttled outflow with a virtual porous layer having a particularly low saturated hydraulic conductivity Ks. This allows simulation of the saturated bottom reserve implemented in French vertical-flow CW to treat sewer overflow. After van Genuchten parameters were estimated by laboratory and field experiments, sensitivity tests highlighted that a, n and Ks of the filter layer and virtual Ks predominantly influence water flow during desaturated and saturated drainage, respectively. These parameters were then optimized by inverse modeling, resulting in a van Genuchten parameter set that provides a good fit between measured and simulated data. However, validation showed that the prediction failed when simulated events significantly differed from the event used for calibration, due to an uncorrect value of the Ks of the virtual layer. A workaround was suggested, with the implementation of a new boundary condition reproducing singular head losses.