The discovery and publication by Barrick in 1979 that HF radars could detect a tsunami wave from its orbital velocity¹ lay dormant for 25 years until the devastating 2004 Banda Aceh event provided a "wake-up call." Although no HF radars were in place to observe that event, work began on hypothesizing the idealized tsunami current patterns one might expect to see from a single radar. It was recognized that the nearshore bathymetry would be a dominant factor that defined its surface signature. The problem lay in finding this tsunami pattern among the complex background flow constituents that these radars normally observe, and doing this rapidly with an automatic algorithm. This hypothetical pattern work gave way to reality after the March 2011 cataclysmic tsunami that wreaked havoc on Japan and whose strong waves were observed in the U.S. and South America. This time, over 20 SeaSondes were in place and able to record a trove of data that proved invaluable in creating the first HF radar algorithm ever that could detect the approaching tsunami signal offshore -- pulling it out of the background flows -- and provide an alert.

The algorithm that proved successful was based on the temporal characteristics of the tsunami wave rather than its spatial pattern; the latter will be added later. Damaging tsunamis have periods between 20 - 40 minutes. This separates them from wind-wave and swell motions, but is shorter than the 6-hour half-period of tides. However, normal spectral analysis methods require several periods to identify the desired peak, and this is too long to wait when lives are at stake. Therefore, to find the onset of this unique signature quickly among the other background flows, the algorithm we developed and demonstrated is based on the following:

Locator map for April 2012 tsunami event in the Indian Ocean.

• Running averages and standard deviations of radial velocity flows from the radar back in time are being continually computed in bands parallel to the bathymetry (depth) contours. Then we:
• Calculate a "velocity-deviation function" from this background as a product over several adjacent bands.
• Calculate a "velocity-increment function" as the magnitude change over three successive time steps of 2 - 4 minutes each.
• Calculate a "velocity-correlation function" over the three or four depth-oriented bands. This must then increase or decrease in at least three consecutive time steps.
• Finally, the above three "functions" are multiplied together to create our "q-factor", which -- when it jumps from quiescent by a significant amount -- becomes the "tsunami alert" that is transmitted to a national coordination center.

The above algorithm was tested offline but in a automatic manner on data from 14 SeaSondes for the intense March 2011 Japanese tsunami: in both Japan and the U.S.; and at different frequency bands². In all cases, it detected the tsunami before it was seen at the coast by tide gages. And in all of these locations, the shallow shelf did not extend far offshore -- a particularly challenging scenario.

How does it work for weaker tsunamis? We had a chance to test this with an April 2012 magnitude 8.6 event in the Indian Ocean, followed by a weak 8.2 aftershock. [See locator map].