The seismic activity in a 1000 sq. km region along the Middle Durance river in the South of France has increased significantly the last two years - in this period there have been more than 30 felt earthquakes. In addition, there have been numerous felt and damaging historical events in this region with intensities up to VII-VIII. As an aid in assesing the earthquake hazard in this tectonically active area, it is important to have both accurate earthquake locations and a good understanding of the relative and absolute uncertainties in these locations. In addition, because of the structural complexity and large velocity contrasts in this region, it may be important to use earthquake location procedures valid for 3D, heterogeneous media.

Using recordings from the 13 seismic station IPSN network in the Durance region, we obtain probabilistic earthquake locations using an efficient grid-search algorithm for three-dimensional media. For this algorithm, we calculate once and store the P and S traveltimes between each station of the network and all nodes of a regular spatial grid, using an Eikonal finite-difference scheme (Podvin and Lecomte, 1991). Next, for each event we apply a suite of successively finer grid-searches over hypocenter locations to obtain an estimate of the spatial probability density function (PDF) for the hypocenter following the approach of Tarantola and Valette (1982).

For a layered model, we present and compare the hypocenter locations and uncertainty estimates obtained with the non-linear, grid-search approach to those for an iterative, linear algorithm (Hypo71). We next compare the grid-search PDFs for the layered model with those for a preliminary 3D velocity structure obtained from seismic reflection profiles. We discuss the uncertainty in hypocenter and depth estimates in relation to the network geometry and the choice of 1D or 3D structure. We also present some hypocenter PDFs which have multiple minima and strongly non-elliptical confidence regions.