Non-perennial streams play a crucial role in ecological communities. However, the key parameters and processes involved in stream intermittence remain poorly understood. While climate conditions, geology and land use are well identified, assessing and modeling the groundwater controls on streamflow intermittence remains a challenge. In this study, we explore new opportunities to calibrate process-based 3D groundwater flow models designed to simulate stream intermittence in groundwater-fed headwaters. Streamflow measurements and stream network maps are jointly considered to constrain aquifer’s effective hydraulic properties in hydrogeological models. The simulations were then validated using visual observations presence/absence of water, provided by a national monitoring network in France (ONDE). We tested the methodology on two pilot catchments with unconfined shallow crystalline aquifer, the Canut and Nançon (Brittany, France). We found that streamflow and expansion/contraction dynamics of the stream network are both necessary to calibrate simultaneously hydraulic conductivity K and porosity θ with low uncertainties. Conversely, calibration resulted in accurate prediction of stream intermittence - in terms of flow and spatial extent. For the two catchments studied, the Canut and Nançon, hydraulic conductivity is close reaching 1.5 x 10 ^-5 m/s and 4.5 x 10 ^-5 m/s respectively. However, they differ more by their storage capacity, with porosity estimated at 0.1 % and 2.2 % respectively. Lower storage capacities lead to higher fluctuations in the water table, increasing the proportion of intermittent streams and reducing perennial flow. This new modeling framework allowing to predict streamflow intermittence in headwaters can be deployed to improve our understanding of groundwater controls in different geomorphological, geological, and climatic contexts. It will benefit from advances in remote sensing and crowdsourcing approaches that generate new observed data products with high spatial and temporal resolution.