<p>Precise knowledge of the sources of seismic noise is fundamental to our understanding of the ambient seismic field and its generation mechanisms. Two approaches to locating such sources exist currently. One is based on minimizing the misfit between estimated Green's functions from cross-correlation of seismic noise and synthetically computed correlation functions. This approach is computationally expensive and not yet widely adopted. The other, more common approach is Beamforming, where a beam is computed by shifting waveforms in time corresponding to the slowness of a potentially arriving wave front. Beamforming allows fast computations, but is limited to the plane-wave assumption and sources outside of the array.</p><p>Matched Field Processing (MFP) is Beamforming in the spatial domain. By probing potential source locations directly, it allows for arbitrary wave propagation in the medium as well as sources inside of arrays. MFP has been successfully applied at local scale using a constant velocity for travel-time estimation, sufficient at that scale. At regional scale, travel times can be estimated from phase velocity maps, which are not yet available globally at microseism frequencies.</p><p>To expand MFP&#8217;s applicability to new regions and larger scales, we replace the replica vectors that contain only travel-time information with full synthetic Green's functions. This allows to capture the full complexity of wave propagation by including relative amplitude information between receivers and multiple phases. We apply the method to continuous recordings of stations surrounding the North Atlantic and locate seismic sources in the primary and secondary microseism band, using pre-computed databases of Green's functions for computational efficiency. The framework we introduce here can easily be adapted to a laterally homogeneous Earth once such Green&#8217;s function databases become available, hopefully in the near future.</p>