The estimation of earthquake source parameters using seismic arrays has become a powerful research tool over the last decade. More recent studies use back-projection of combined seismic arrays with accurate time corrections, first arrival alignment, and filtering. Although numerous publications show good results using this technique as a post processing tool, real-time reliable source area estimation remains a challenging task. The main shortcoming in real-time processing is spurious results due to seismic noise and off-path arrivals causing random source coordinates, not related to the true source, to be chosen by search algorithms. As a consequence, wide adoption of this approach by Early Warning Centres has not occurred. However, properly chosen constraints on the source parameters can restrict the search field and significantly reduce spurious results, making the final estimate useful for warning systems. To achieve this goal for operational Tsunami Early Warning we introduce the following concepts in source back-projection: only nodes previously used to pre-compute tsunami scenarios are used to form seismic beams; and the possible source search area is limited by a circle that expands with a velocity less than 3.5km/sec from the epicentre. Our final results show that with this approach robust solutions can be obtained in real-time. The most significant information for tsunami modelling, such as the location of the rupture area relative to the epicentre, is retained in the solution. Using a search grid comprised of nodes previously used for tsunami modelling gives an important advantage over an arbitrary dense global grid because the algorithm finds the best matching nodes using observed data, thereby removing human ad-hoc decision making. Tests of the algorithm on a number of recent large earthquakes show that it is a practical approach for Tsunami Early Warning Centres.
