Aerospace and Electronic Systems Magazine March 2018 - 55

Dzvonkovskaya
utes of measurements as a trend estimation. Further, the detection scheme uses the sign of the residuals; therefore, every binary
value is considered to have a Bernoulli distribution. Following
Laplace's law of succession applied to conditionally independent
random variables having a Bernoulli distribution, the probability
of a detected tsunami is, therefore, calculated in each grid point
of radar coverage, where the residuals are estimated. The final
alert decision is based on the ordered statistics estimation for the
probability map. This approach is already implemented in the
tsunami alert software and delivers spatial information regarding
an alert every 33 seconds. An example of a tsunami detection
and corresponding alert maps is shown in Figure 5 for the case
of the 2011 tsunami measured in Chile. It is clearly visible in this
sequence of images that the first tsunami wave needed less than
10 minutes to cross the shelf, which is only between 10 and 20
km wide.
The final decision on a tsunami alert is done automatically
by software using a three-level tsunami alert, i.e., no tsunami, a
possible tsunami, and a tsunami threat (see example in Figure 6).
The vertical bar chart shows tsunami threat values calculated from
real measurements in Chile. The first alert message arises at 04:27
UTC. In the future implementation of this software, the values in
the red alert zone would be immediately transmitted to the nearest
TEWS server.
Another approach is used by the direction-finding SeaSonde
ocean radar. Because the radar is limited in angle resolution for
currents with a high update rate, in [23], it is proposed to use
measured radial velocities at three adjacent times and three 2-km
range bands to calculate a newly determined q factor that signals
the tsunami arrival when it exceeds a preset threshold. The tsunami detection q factor at time t is defined to be the product of
the correlation function, the velocity increment function, and the
velocity deviation function correlations. The method is empirical
and does not contain a probabilistic or statistical treatment for
deriving the threshold for this q factor. The comparison between
the peak q factor derived from the measured velocities during the
2011 tsunami and the initial water level obtained from the neighboring tide gauge exhibited that the water level was too small
to allow the algorithm to pick up the tsunami's initial approach
(see [23] for further details). It is expected that the analysis can
be improved by incorporating additionally known tsunami characteristics to produce earlier detection and cut down on the risk
of false alarms.
Another example in [18], shown for the NJRC radar system,
proposed the possibility for tsunami detection by using a cross
correlation between distant observation points on a coverage grid
(as compared with using the range bands as in SeaSonde case)
in which 60-minute data streams of filtered current velocities in
time and space were cross correlated. When the cross correlation
increased significantly together with the passage of the first tsunami wave, it was judged that the ocean radar had detected the
tsunami at the grid points where the correlation was calculated.
In this routine, the tsunami effect starts when current velocity exceeds a radar velocity resolution of 4.78 cm/s. Additionally, the
influence of a ship echo in the radar spectra was also discussed in
[18]. When a measured ship Doppler frequency appears close to
MARCH 2018

Figure 6.

Three levels of tsunami alert (green: no tsunami; yellow: possible tsunami; and red: tsunami threat). The first appearance of the 2011 Tohoku
tsunami in Chile corresponds to a vertical bar in the red zone at 04:27
UTC, March 12, 2011.

the tsunami-shifted Bragg peak, it leads to an incorrect estimation
of the surface current, in general, and to false detection by using
the correlation technique.

THE DETECTION OF TSUNAMIS ON THE BASIS OF TIME
SERIES OF RADAR SIGNALS
Generally, most attention is directed toward coastal tsunami hazards governed by near-field tsunami sources, in which cases tsunami propagation times may be too small for a detection by deep or
shallow water buoys and cannot be precisely predicted by models
similarly as for far-field sources. To detect a tsunami in deeper
water, beyond the continental shelf, the authors in [26] proposed a
new detection algorithm that does not require "inverting" currents
from radar range-Doppler frequency spectra. Instead, their method
is based on computing spatial correlations of the raw radar signal
at pairs of radar cells located along the precalculated tsunami wave
rays, shifted in time by the tsunami propagation time along the ray.
A change in pattern of these correlations indicates an approaching
tsunami. By contrast to previously mentioned algorithms, applying
this algorithm for idealized tsunami wave trains and bathymetry, it
was concluded in [26] that the arrival of tsunami currents as low as
background values of 5-10 cm/s could be inferred; thus, tsunami
detection could take place in deeper water beyond the continental
shelf.

HF RADAR SYSTEM INTEGRATION INTO EARLY WARNING
SYSTEMS
Traditional TEWS relies mainly on seismic observations for early
detection of potentially tsunamigenic sources. Moreover, sea-level observations from deep-sea bottom pressure recorders (BPR)
together with surface-moored buoys and coastal tide gauges are

IEEE A&E SYSTEMS MAGAZINE

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine March 2018