Correlating ambient seismic noise allows us to image the subsoil in various contexts and at different scales. Applying this technique to anthropogenic seismic noise can be challenging when the spatial distribution of the sources is not uniform. We have addressed the feasibility of exploiting this kind of noise in addition to microseismic noise to extend the reconstruction of Rayleigh-wave dispersion at periods between 0.2 and 1 s. We used data acquired with two small aperture arrays ( stations with a 200 m helical distribution) deployed near the deep geothermal site of Rittershoffen (Alsace, France). In this region, the sparse human activity causes strong seismic noise, whose nonuniform spatial distribution limits our ability to determine the surface wave velocity between stations using the classical noise correlation technique at periods of less than 1 s. We have used double beamforming to isolate the noise sources that contribute constructively to the empirical Green’s function between the two arrays and recovered the Rayleigh-wave dispersion curve at periods less than 1 s. Using a probabilistic inversion, we found that such data, combined with surface wave measurements at periods greater than 1 s, are helpful to improve the reliability of and profiles at depths down to the deep-geothermal reservoir (2.5 km). Such profiles are helpful in a geothermal context because they improve the location of induced seismic events, necessary for reservoir monitoring and risk assessment.