The use of a large number of sensors is becoming more common in seismology at both the global scale for deep Earth studies, and at the exploration geophysics scale for monitoring and subsurface imaging. Seismic arrays require array processing from which new type of observables contribute to a better understanding of the wave propagation complexity. This thesis deals with a subset of these techniques. It first focuses on a way to select and identify different phases between two source-receiver arrays based on the double beamforming (DBF) method. At the exploration geophysics scale, the goal is to identify and separate low-amplitude body waves from high-amplitude dispersive surface waves. At the continental scale, as the source arrays are uncommon, the cross-correlation (CC) method of broadband ambient seismic noise can be used to evaluate the Green's function between two receiver arrays. The combination of DBF and CC is applied on Transportable Array (USArray) data to construct high-resolution phase velocity maps of Rayleigh and Love waves. Finally, at the global scale, by using a large number of sensors, it is shown that body waves can emerge form CC of continuous records in the 5-100s period band. We also analyze the contribution of strong earthquakes and particularly their long lasting reverberated coda. We compare it to the contribution to correlations of the continuous background sources associated with the ocean-crust interaction. The reconstructed body waves constitute a valuable supplement to traditional earthquake data to image and to monitor the structure of the Earth from its surface to the inner core.
