Automated systems often declare seismic events with fixed depths. In SeisComP, for example, scautoloc fixes the depth at 10 km when it determines that the depth resolution is poor.
Well, automated processing is all about the P’s: P distribution, P proximity and P density. Robust picking on other phases (e.g., the S) continues to elude the automated processing community. The result of this P-centric world is that when there are lots of well distributed P-waves close to the source and spaced with good azimuthal coverage, automated processing systems can determine the depth. When one or more of these elements is missing, then depth resolution is often poor, a fixed depth is assigned and a spurious or misleading depth avoided.
Here are a few real-life examples (click for larger images):
Example 1: The importance of epicenter location with respect to network
Example 1a: Event epicenter outside of network; Depth 10 km
Example 1b: Event epicenter inside of network; Depth 45 km
Example 2: The importance of stations local to the epicenter
Example 2a: No stations local to epicenter; Depth 10 km
Example 2b: Stations local to epicenter; Depth 422 km
Example 3: The importance of good azimuthal coverage
Example 3a: Poor azimuthal coverage; Depth 10 km
Example 3b: Great azimuthal coverage; Depth 48 km