Evaluation of real-time tsunami earthquake discriminants

Tsunami earthquakes generate a disproportionally large tsunami for their seismic moment. For a tsunami warning center, they are especially difficult to detect in real-time since magnitude alone is insufficient to issue an alert. Recently, several methods have been developed to identify tsunami earthquakes, including: various energy magnitude estimates (e.g., MED, Lomax et al, 2007), the theta discriminant (Newman & Okal, 1998), RTerg (Newman & Convers, 2010), TACER (Convers & Newman, 2013), mHFER (Hara, 2007), and rupture duration, TR (Lomax & Michelini, 2009 & 2010).

Each method makes particular assumptions about the rupture process and subsequent tsunami generation that lead to the use of different algorithms to estimate the radiated seismic energy and/or rupture duration. However, the various methods are essentially comparing these estimates to the long period seismic moment, in order to identify the unusually long durations, slow ruptures, and small stress drops that characterize tsunami earthquakes.

We test the tsunami earthquake discriminants on a dataset of subduction zone earthquakes containing several tsunami earthquakes with the goal of determining which methods (or combination of methods) are well-suited to real-time implementation at the U.S. tsunami warning centers in Hawaii and Alaska. Of particular interest is the ability of each method to correctly identify known tsunami earthquakes with a minimum of false positives and with a minimum of apriori assumptions about any individual event that might bias a real-time detection system.