The location procedures of seismic events are influenced by two major classes of errors, the error in picking individual seismic phases and modeling error due to the departure of the real Earth from the reference model used in the location. Both classes of error influence the estimate of location and it is difficult to separate them.
The role of picking errors can be assessed by a nonlinear analysis using a Monte Carlo procedure. Arrivals times are perturbed with random numbers drawn from a normal distribution, and the event is relocated using these perturbed arrival times. By repeating the procedure many times, a cluster of locations is obtained, which can be used to investigate the effects of picking errors on the hypocenter. This analysis is insensitive to velocity-model errors as these are fixed for a given combination of stations and phases. Some care must be exercised when analysing multidimensional distributions in two-dimensional slices because of a projection effect.
The modeling error due to the influence of lateral heterogeneity in the Earth is examined by comparing the locations of the same event using different combinations of phases and network geometries, which reinforces the need to use arrivals other than P for accurate depth resolution. The sensitivity of P arrivals to changes in depth are swamped by model errors, and inclusion of depth-sensitive phases such as pP is highly recommended. The effect of picking errors on location is found to be much smaller than the mislocation caused by neglecting lateral heterogeneity when only P arrivals are used. Consequently, the Monte Carlo analysis, which is primarily aimed at picking errors only, is most appropriate when multiple phases have been used to more accurately constrain the hypocenter, especially for the depth component.
Altering the type of phase data used in the location plays a similar role in changing the network geometry, in that both are mechanisms that influence the nature of the constraint on the hypocenter. By relocating events with network geometries corresponding to the different magnitudes, it is found that the location of the event can be affected significantly by the magnitude, and when using robust statistics to describe earthquake residuals, the mislocation can occur in a systematic manner. The effect is marked in regions with significant lateral variations in seismic velocities. For example, low-magnitude events in the Flores Sea are found to be dragged toward Australia as a result of the fast paths to Australian stations relative to the iasp91 reference velocity model.