**9. Conclusions**

**7.5. Signal aliasing**

108 Tsunami

**8. Dealing with false alarms**

responds to them over a several month period.

straight coastline.

We discuss two approaches for the reduction of false alarms:

**8.1. Search for correlations between** *q***-factor alerts from adjacent HF radars**

A real tsunami would be seen at two coastal locations say 30 km apart, for example, within a definable time window (say 15 min). Tsunami waves refract as they move into the ever‐ shallower water of the continental shelf near the coast. This means they tend to arrive perpen‐ dicular to the shore. This forces similar arrival times at neighboring locations along a nearly

Hence, if a high *q*‐factor peak at Radar M arrives at a given range within 15 min of a high *q*‐ factor peak at Radar N, the presence of both raises the probability that a real tsunami is being seen by orders of magnitude. Likewise, if a high *q*‐factor at Radar M has no counterpart at

Strong echoes from close‐in ranges can alias into distant range cells in systems that transmit while receiving. This "ringing" phenomenon produces spurious echoes in distant range cells that can be erroneously interpreted to be tsunami echoes. These far‐out spurious echoes are easily distinguished from real tsunami echoes, as they appear at the same time as the real tsunami echoes from close‐in ranges. If they were caused by a real tsunami, they would appear earlier at distant ranges. Also current amplitudes extracted from the spurious echoes are usually independent of depth. Fundamental, linear shallow‐water wave physics [1, 6] shows that the current amplitudes always decrease with depth. Observations from Chile of the Japan

The first tsunami alert is indicated when the *q*‐factor exceeds the preset limit. Due to the varying background current and noise/interference effects described in the last section, the detection might be a false alarm. The trade‐off in the selection of the *q*‐factor threshold is as follows: if the *q*‐factor limit is set too low, the peak will certainly produce a *q*‐factor alert, but many non‐ tsunami false‐alarm detections may be generated, degrading performance and operational acceptability. If it is set too high, false alarms will be eliminated, but then the first actual tsunami peak may be missed. This is the classic "probability‐of‐detection vs. false‐alarm rate" trade‐off encountered by warning sensor systems. An acceptably low false‐alarm rate for a given tsunami intensity and detection distance from shore are the parameters to be optimized at each site. Judging by the strength of the *q*‐factor signals seen in the Japan tsunami, it would appear that a tsunami having a run‐up height of 1 m should easily be detectable with very low false‐ alarm rates in uncontaminated radar spectra, using the detection methods described earlier. The value of the *q*‐factor limit defining a tsunami detection is site specific and needs to be studied for the site under consideration. The most effective way to handle this for a given site is to study the background currents/extraneous effects and how the *q*‐factor algorithm

tsunami [4], discussed in Section 3.4.3, appear to show indications of "ringing."

The use of HF coastal radars to detect and warn of approaching tsunamis was proposed nearly 40 years ago [1]. Radars measure the tsunami wave's orbital velocity, unlike all other tsunami sensors that measure its height. However, it was not until the 2004 tragic Banda Aceh event that killed a quarter of a million people that attention was seriously focused on developing a local warning system. After the significant 2011 Tohoku, Japan tsunami, enough radars were in place worldwide to gather a radar‐signal database that allowed the development of detection and warning methodology.

Several papers ensued demonstrating a detection algorithm and developments followed toward a methodology to capitalize on this near‐field monitoring capability [2–3, 5–6]. The present work presents a summary of these prior works and also covers recent work on tsunami simulation and site evaluation that has not been reported elsewhere.

In the last 5 years, tsunami events were detected by 21 SeaSonde HF coastal radars. Times between detection and arrival at shore (as confirmed by tide gages) ranged between 1 and 43 min. This alert time for a tsunami of given intensity depends principally on the offshore bathymetry. A shallow shelf edge affords a much longer alert time than a steep drop‐off. We have examined and compared approximate and exact methods for estimating the arrival. The latter include the solution of exact linear partial differential equations, especially suited to near‐field HF observations; in the future, this will also allow the radar‐measured orbital velocity to be related to the height of the approaching wave.

We have also described steps toward simulation of tsunami height and velocity. This allows the assessment of the warning possible at a given site location based on the offshore bathy‐ metry, background currents, and noise. False‐alarm rate and probability of detection are the metrics against which performance is evaluated. Methods of increasing the latter while decreasing the former are examined. These include correlations with similar alerts from other nearby radars and with reports on seismic events that may constitute tsunamigenic sources.

After radars on the US East Coast detected a meteotsunami in 2013 that provided an alert that was missed by conventional methods, NOAA recognized the potential that the US Integrated Ocean Observing System (IOOS) could offer for tsunami warning. IOOS consists of 130 SeaSonde radars around the US coast. The NOAA Tsunami Warning Program office and IOOS entered into a partnership with Codar Ocean Sensors to optimize this capability based on steps outlined in this report and organized its transition to the two operational tsunami warning centers in Hawaii and Alaska.

It is clear that early local detection of incoming tsunamis by deployed radar systems is now within the capabilities of existing technology.
