**3.3. Detection of the tsunami signal in HF radar data**

**Figure 4** shows band velocities observed at Kinaoshi, Hokkaido (A088), for three 2‐km bands

**Figure 4.** Band velocities from Radar A088. Distance from radar: 6–8 km (blue), 8–10 km (red), and 10–12 km (black). Time span: March 11, 2011 (a) surrounding the tsunami arrival, from 12:46 to 19:46 JST and (b) close to the tsunami

The arrival of the tsunami is indicated by the commencement of distinctive oscillations in

Further examples of band velocities from radars on both sides of the Pacific close to the tsunami

**Figure 5.** Time series of band velocities showing the typical appearance around the tsunami arrival time for three adja‐ cent 2‐km area bands. (a) A087 (Hokkaido, Japan), (b) YHS2 (Oregon, USA), (c) BML1 (California, USA), and (d) PREY

velocity with a period of about 40 min, which are strongly correlated over range.

ranging from 6 to 12 km from the shore.

84 Tsunami

arrival, from 14:16 to 15:16 JST.

(California, USA).

arrival time are shown in **Figure 5**.

Two effects distinguish tsunami velocities from the background in **Figures 4** and **5**: (a) velocities in neighboring bands are strongly correlated after the arrival of the tsunami and (b) the velocity oscillations are clearly visible above the background. These effects appear to be characteristic of tsunami band velocities, as they occur in all the radar data from Japan and the US West Coast that we have analyzed. They form the basis of a simple pattern detection procedure. At a given time, a factor (which we call the *q*‐factor) is defined which signals the tsunami arrival when it exceeds a preset threshold. The steps in the detection algorithm are as follows:


Details of the detection algorithm are given [6].

Positive *q*‐factor values indicate the tsunami velocity at the wave peak is moving toward the radar, negative values indicate that it is moving away.

To set the operational threshold signaling a tsunami detection, an extended data set obtained under normal conditions is analyzed to produce *q*‐factors. A threshold value is then selected. There is a trade‐off in the threshold selection: if the *q*‐factor limit is set too low, the peak will certainly indicate a detection, but there may be many false‐alarm detections. If the threshold is set too high, there will be few false alarms, but then the tsunami arrival may not be detected.

The tsunami signal can be evident for some time after arrival; however, our detection method is optimized to apply to the first arrival.
