Elena Tolkova

Tsunami Support Scientist

I graduated from the Radiophysics department of Nizhny Novgorod State University in Russia in 1985 and stayed with the University for 15 years after the graduation. My Kandidat Nauk (Ph.D equivalent in Russia) dissertation, "Interferometric methods of observation of small angular size incoherent objects through the Earth's atmosphere," described three new interferometers, each with an original data collecting/processing algorithm, designed to resolve fine details of celestial bodies if ever directed into the night sky... but only one of the three made it from a project into a hard copy and neither made it out of a lab, to the field.

I joined JISAO in 2005 to work with the NOAA Center for Tsunami Research. In particular, I work on Empirical Orthogonal Function (EOF) analysis, for its exceptional ability to discover hidden patterns and to separate sufficient from unessential.

My first experience with EOFs was to use them for tidal predictions. Tidal signal continuously transforms. Within a week, two high tides per day can be replaced by just one of completely different amplitude. Scholastic approach to tidal prediction is to analyze an existing record (optimally a year-long) for sines and cosines at specific frequencies, and then use those harmonic components to synthesize the tidal signal any time in the future. But to de-tide an ongoing tsunami record, we only need to know a tidal component for the next few hours. Is there a simpler way to predict the tide a short time ahead?

An image showing day-long consecutive tidal fragments arranged in rows displays so smooth pattern that continuing it just one day further seemed almost certain to me. EOF analysis came as the right tool. It analyzes a set of patterns (tidal fragments of specific length, in this example) for its statistical modes. It turned out, any tidal fragment under three day in length anywhere in a deep ocean can be composed of the same several modes, or shapes, combined with different coefficients. The same-anywhere set of the three-day-long shapes and a two-day-long tidal record at a location happened to be all that needed to predict the next day tide at this location, with no less accuracy than that of the traditional methods.

Next time, my colleague and I were looking for very particular statistical modes, that would coincide with resonance (normal) modes of a semi-enclosed water basin. Resonance modes of bays and harbors determine how they respond to tsunamis. Like a rod of length L clamped at one end, a bay oscillates at frequency c/4L, where c is the wave velocity, and L is the length of the bay. Except, how exactly long is a bay open to an ocean? Existing methods for determining the normal modes of bays and harbors start with assumptions about where the bay ends and the ocean begins. Consequently, their results are only as good as was the initial guess about the boundary between the basin and the ocean. However, tsunami models can reproduce tsunami wave evolution in a bay and thus capture the normal oscillations without any such assumptions. What remains is to extract those oscillations from the tsunami data. It turned out, the EOF analysis of a data set composed of multiple tsunami wave-fields can do that, and can be very accurate.

I'm an all-weather biker. With the rest of my family (my husband, son and daughter), I like hiking steep forest trails in the mountains, biking flat paved roads in the valleys, traveling around, and taking pictures of new places.