Tsunami detection by High Frequency Radar using a Time-Correlation Algorithm: performance analysis based on data from a HF radar in British Columbia

A High-Frequency (HF) radar was installed by Ocean Networks Canada in Tofino, BC, to detect tsunamis from farand near-field sources on the Pacific Ocean side of Vancouver Island; in particular, from seismic sources in the Cascadia Subduction Zone. Based on a classical analysis of the Doppler spectrum, this HF radar can measure ocean surface currents up to a 85-110 km range depending on sea state. However, an inherent limitation of detection of small and short-lived tsunami currents is the conflicting requirement for short integration time and sufficient accuracy (resolution) of the Doppler spectra. This limits a direct tsunami detection typically to shallow water areas over the continental shelf where tsunami currents have become sufficiently strong due to wave shoaling. To overcome this limitation, the authors have recently proposed a new detection method, referred to as “Time-Correlation Algorithm (TCA)”, that does not require inverting Doppler spectra for the tsunami currents and can thus potentially detect an approaching tsunami in deeper water, beyond the continental shelf. This algorithm is based on computing space-time correlation of the raw radar signal in different radar cells aligned along precomputed tsunami wave rays, and time-shifted by the precomputed tsunami propagation time between cells. A change in pattern of such correlations indicates the presence of a tsunami. They validated the TCA with numerical simulations for both idealized (Grilli et al., 2016a) and realistic (Grilli et al., 2016b, 2017) tsunami wave trains and seafloor bathymetry, using data simulated with a radar simulator. Here, the TCA is for the first time applied to actual radar data measured with the ONC WERA HF radar and numerically modified by a synthetic tsunami current. Using a state-of-the-art long wave model we perform tsunami simulations with realistic source and bathymetry, and combine the resulting currents with the background currents and radar backscattered signal measured by the HF radar system. This combination makes it possible to evaluate the performance of the proposed TCA detection algorithm, based on an experimental rather than numerically simulated, data set of radar signal. Our findings confirm that an actual detection can be achieved beyond the continental shelf, where tsunami currents are small (as low as 5 cm/s), in deeper water than when using an algorithm based on a direct inversion of currents from the measured radar Doppler spectra. INTRODUCTION The use of shore-based High Frequency (HF) radars to detect incoming tsunami waves was proposed almost 40 years ago by Barrick (1979) and, more recently, was supported by numerical simulations (see, e.g., (Lipa et al., 2006), (Heron et al., 2008), (Dzvonkovskaya et al., 2009), (Gurgel et al., 2011) and (Grilli et al., 2016a)), and by HF radar measurements of the Tohoku 2011 tsunami made in Japan (Hinata et al., 2011; Lipa et al., 2011, 2012), in Chile ((Dzvonkovskaya, 2012)), and in Hawaii ((Benjamin et al., 2016)). HF radars are routinely used to estimate surface currents based on the analysis of the Doppler spectrum over a grid of radar cells, covering a sweep area of tens to hundreds of kms, and new radars have been installed or existing radar used as a means of detecting tsunami currents as well. To do so, tsunami detection algorithms were proposed (see some of the above studies), based on identifying the oscillatory nature of tsunami currents, in space and/or time in the radar measurements. There are, however, some intrinsic limitations to this detection method. In order to be detectable, the magnitude of the tsunami current should be larger than the threshold of accuracy of the Dopplerbased estimation. This accuracy depends mainly on the radar frequency and integration time used to compute the Doppler spectrum. Estimating small currents requires large integration times, but this is in contradiction with the oscillatory nature of tsunami currents, as a large integration time will cause averaging and significant decrease of these currents, which will become less measurable. Hence, the integration time must remain short enough (up to a few minutes) to avoid averaging out the tsunami induced current, which will reduce the resolution and detection threshold of the Doppler-based methods. In practice, this limits a direct detection of tsunami currents by way of Doppler shifts to the continental shelf where currents have become sufficiently large due to shoaling; hence, this also means small warning times, unless there is a very wide shelf. Recently, (Grilli et al., 2016a,b, 2017) proposed a new detection method, which we refer to as the “Time Correlation Algorithm” (TCA), which does not require inverting Doppler spectra for the currents, but is based on the space-time correlation properties of the radar signal along pre-computed tsunami wave rays. By performing numerical radar and tsunami simulations for both idealized (Grilli et al., 2016a) and realistic (Grilli et al., 2016b, 2017) seafloor bathymetry, we found that the TCA detection algorithm could be applied beyond the continental shelf, in deeper water, where tsunami currents are small (a few cm/s). Here, for the first time, we report on the implementation of the TCA using actual measured radar data measured off of the Pacific Ocean side of Vancouver Island (radar deployed at Tofino, BC), together with simulated tsunami currents using a state-of-the-art tsunami propagation model. As it has been customary in other related works, we superimpose the effect of tsunami currents on the measured radar signal by the introduction, in each radar cell, of a phase shift depending on a current memory term (see details below). We show results for a Mw 9.1 far-field seismic source in the Semidi Subduction Zone (SSZ; Fig. 1). The numerical generation of a tsunami for this source was presented elsewhere and we refer to (Grilli et al., 2016b, 2017) for details. Fig. 1: Zoom-in on area of the Pacific Ocean covered by the 2 arc-min grid G0, with initial surface elevation (color scale in meter) of the Mw 9.1 SAFRR seismic source in the Semidi Subduction Zone (SSZ); boundaries of nested model grids off of Vancouver Island, BC, are shown as black boxes: G1 (0.6 arc-min), G2 (270 m), and G3 (90 m) WERA HF RADAR SYSTEM USED IN THIS WORK To mitigate the elevated tsunami hazard along the shores of Vancouver Island in British Columbia, Canada, Ocean Networks Canada (ONC) has been developing a Tsunami Early Warning System combining instrument deployment on the seafloor as part of their Neptune Observatory, and a shore-based High-Frequency (HF) radar installed near Tofino (BC) (Fig. 2), which has been operational since April 2016. This WERA HF radar, with carrier frequency fEM = 13.5 MHz, can detect and estimate ocean radial currents up to a 85-110 km range depending on sea state, as the propagation losses increases with sea surface roughness. An example of a typical current map is shown in Fig. 2. The radar sweep area is outlined in Fig. 4 and is covered by radar cells, within which the received radar signal is averaged. The cells all have a radial length ∆R = 1.5 km and an angular opening ∆φr = 1 degree in the azimuthal direction; the detection sector of the sweep area is 120 degree, implying that cells are 1.48 km wide at a 85 km range and narrower closer to the radar (cell area: ∆S= R∆R∆φr increases with range). The orientation of the radar array of 12 antennas (275 deg. from N, clockwise; centered at 49◦ 4’ 24.82” N, 125◦ 46’ 11.55” W; Fig. 2) is such that one side of the sweep area boundary is nearly parallel to the coastline southeast of Tofino, and the array length (110 m) allows for approximately a 12 degree azimuthal resolution. In the direction finding algorithm, as the radar signal is processed for overlapping angular windows, surface currents are estimated in a larger number of radar cells with a 1 degree azimuthal resolution. Fig. 2: The ONC WERA HF radar site in Tofino, BC. Fig. 3: Typical map of radial surface currents measured by the WERA HF radar installed in Tofino, BC, by inversion of the Doppler spectra. THE TIME-CORRELATION ALGORITHM (TCA) According to first-order Bragg theory, the complex backscattered signal received at time t for a given radar cell p is of the form: Vq(t) = α e−2iπ fBt +α− e+2iπ fBt (1) where α± are constant complex coefficients function of sea state, range, and radar calibration, and fB = √ g/(πλEM) is the Bragg frequency (λEM is the electromagnetic wavelength). In the presence of a constant current U , the complex radar signal experiences a Doppler frequency shift fU = −2U/λEM and is thus multiplied by a complex exponential e2iπ fU t . For a variable current in time, U(t), the Doppler frequency shift is obtained through the integration of the instantaneous Doppler frequency fU (t) =−2U(t)/λEM , and the radar time series is multiplied by the complex exponential eiM(t), where