Human activity, and especially intensive agriculture, has more than doubled reactive nitrogen delivery to Earth’s ecosystems, creating eutrophic conditions in aquatic and coastal environments around the world. In the past several decades, efforts have been made to reduce agricultural inputs of nitrogen, but the decrease of the nitrate concentration in rivers also depends on natural factors. The aquifer exerts a dual control on nitrate pollution. On one hand, the circulation of groundwater induces a temporary storage of nitrate, responsible for the nitrate legacy. On the other hand, the denitrification occurring in the aquifer lowers the nitrate concentration. Characterizing the groundwater residence time and the denitrification in the aquifer is thus necessary to predict the nitrate trajectories in agricultural catchments. This PhD thesis proposes to extract information about the aquifer from in-stream measurements. The method is tested on a 35 km2 agricultural catchment located in Brittany (France), in a fractured bedrock aquifer. First, a sensitivity analysis evidences that the nitrate concentration can be predicted with the knowledge of a limited number of parameters. Then, the possibility of getting these parameters in headwater streams during low flow is evaluated. Investigation of gas exchanges between the stream and the atmosphere shows that the use of gas tracers at the catchment scale is challenging. However, in situ nutrients injections reveal a very low effective denitrification in the stream and the hyporheic zone, which does not significantly modify the nitrate concentration. Finally, spatially distributed measurements of silica and nitrate in headwater streams provide a representative view on the processes of transport and denitrification in the aquifer. It highlights that the scale of homogenization is larger than the studied catchment and allows to hypothesise global control factors on the processes.