Gas migration in rocks results from natural and artificial processes. Understanding gas migration matters for the Comprehensive Nuclear-Test-Ban Treaty (CTBT), to improve the detection of underground nuclear explosions by their radioactive gases. This work concerns many other fields in Earth Sciences, for fundamental as well as applied science. Issues in improving the detection and the understanding of gas migration in geological media are the following. What are the driving forces of gas migration from depth to the surface? How much of the gases produced at depth do arrive at the surface? Does this migration lead to temporal delays and dilution between production and breakthrough at the surface? To answer these questions, this thesis is dedicated to the identification of gas transport mechanisms in fractured rocks, from both field experiments and numerical simulations. The Roselend Natural Laboratory (French Alps) is a unique facility for studying gas transport in the unsaturated zone at the field scale, representative of natural processes. Parameters and external forcings have been determined. A tunnel and an isolated chamber, at 55 m depth, as well as boreholes at depth and at the surface, allow to monitor gases that are present in the rocks. Pneumatic properties of the rocks, permeability and porosity, were determined at scales ranging from 1 to 55 m, from both pneumatic injection tests and pressure monitoring as well as from computational studies of fluid flow and transport in porous media. Inverse modeling was used to quantify the associated uncertainties. The results underline the strong spatial heterogeneity of fractured media. The natural dynamics of three gases, CO2, SF6 and 222Rn, was monitored continuously for more than one year. The results, interpreted with numerical simulations, determined that the processes controlling the natural dynamics, or baseline, of gases are atmospheric pressure fluctuations and water movements. Such water movements also explain the gas anomalies observed in the tunnel. Two tracing experiments were also conducted between the isolated chamber at the end of the tunnel, and the surface. SF6 was injected first with a 167 mbar overpressure, followed by $^3He$, one year later, with only 20 mbar overpressure. Monitoring of tracer gases at the surface was conducted for more than one year after the injections. Rapid tracers breakthrough (10 to 50 h) was observed at the surface, with a strong dilution of 10^4 up to 10^6. This is explained by gas advection in a limited number of fractures, in response to the gas overpressure occurring at the injection point, even if it is small. In the months following the injection, tracers migrate more slowly in the whole fracture network, in response to barometric pumping. These transport mechanisms, identified and quantified, are now to be included in a comprehensive model taking into account the interactions of gases with the geosphere in order to improve the detection of underground nuclear explosions by their radioactive gases.
