Dmitry Yu. Mikhin1, Oleg A. Godin1, Sergey V. Burenkov1, Yury A. Chepurin,2
Valerii V. Goncharov2, Vladimir M. Kurtepov2 and Viktor G. Selivanov1

1 P.P.Shirshov Oceanography Institute of the Russian Academy of Sciences (OI RAS)
23 Krasikova St., Moscow 117851, Russia

2 Moscow Institute of Physics and Technology (MIPT)
9 Institutskii Per., Dolgoprudny, Moscow region 141700, Russia

The Ocean Acoustic Tomography (OAT) is based on the idea that the variations of the sound speed field within some ocean region may be inferred from the changes in travel times of acoustic signals propagating between a number of moored sources and receivers located on the periphery of the area under study. In this scheme the vertical resolution of the sound speed field is mainly due to the acoustic multi-paths in each vertical slice, while the resolution in horizontal plane is acquired combining the data obtained from intersecting propagation paths between different moorings, similar to the medical computer-aided tomography (CAT). This approach will be further referred to as the traditional OAT. Though the number of crossing propagation paths grows as the number of moorings squared, it may be insufficient for adequate mapping of ocean mezoscale structure. Analytic estimates and numerical simulations demonstrate that the quality of horizontal mapping may be greatly improved by augmenting the existing array of acoustical moorings with a movable ship-based source or receiver [1, 2]. In this paper we describe a field experiment devoted to such dynamic or moving ship tomography, present some preliminary results of the data analysis and interpretation.

A tomographic network of seven moored transceivers was deployed in the Western Mediterranean basin in early 1994 by IfM (Kiel, Germany), IFREMER (Brest, France) and WHOI (Woods Hole, U.S.A.) in the framework of the European project THETIS-2 [3]. This 10-month long sea trial was based on traditional tomography technique. The locations of THETIS-2 transceivers are shown in Fig.1. They are labeled according to the notation used by THETIS-2 participants. The devices were moored near the sound channel axis approximately at 150 m depth. The transceiver H had a central frequency of 250 Hz, all the others operated at a central frequency of 400 Hz. The signals emitted by the transceivers were also utilized in the Moving Ship Tomography experiment, MOST [4]. It was fulfilled by research groups from OI RAS and MIPT during the expedition on-board R/V 'Akademik Sergey Vavilov' of the Russian Academy of Sciences. Simultaneous operation of the traditional tomographic network and the specialized acoustic R/V in the same area created a unique opportunity to assemble the experimental data for adequate comparison of traditional and dynamic tomography under well-controlled environmental conditions, and to reveal possibility of assimilation the data obtained through different tomographic methods.

Fig.1. General layout of THETIS-2 and MOST experiments. The markers stand for: THETIS-2 moorings (bell-like symbols - ), points of measurement during the first (squares) and second (triangles) parts of the MOST experiment, locations of XBT casts (asterisks). The solid lines show the propagation paths connecting THETIS-2 transceivers.

The MOST experiment covered a quadrangle zone with apices at locations of W2, H, W3 and W5 moorings. On the first stage of the experiment lasted from June 10 to June 26, 1994, the R/V has steaming along the W2-H-W3-W5 path (Fig.1), making stations for recording the tomographic signals each 8 hours (according to the transmission schedule). Each acoustic reception was accompanied by CTD measurements to a depth of at least 700 m. To study the possibility to infer the time variability of the ocean from dynamic tomography data, the measurements along W2-S leg were repeated on the second stage of the experiment in mid-July, 1994 (Fig.1). For this period the hydrographic conditions in the region have altered from those typical to spring to the summer conditions. In addition to our CTD measurements a series of XBT casts along the H-W3 path (asterisks in Fig.1) was fulfilled in the framework of THETIS-2 project.

The single receiving hydrophone was deployed from the R/V which drifted with engines turned off to decrease the R/V-generated noise. The receiver depth usually was about 270÷300 m. It was measured either with a high-frequency acoustic system or with a pressure gauge. Hydrophone position 150÷200 m off the sound channel axis was chosen to maximize the number of resolvable ray arrivals. The motion of the R/V was tracked using the Global Positioning System. The recorded signal was complex demodulated and correlated against a replica of the transmitted one. A special computer algorithm was developed to measure the Doppler frequency shift due to the R/V drift. The code is based on iterative processing of acoustic data and uses satellite navigation for the initial approximation. The relative velocity of the receiver with respect to the source was estimated with an accuracy of 1 cm/s that allowed to eliminate the Doppler distortion of the time scale and to average coherently all sequence periods of the emitted signal, yielding a total processing gain of 36÷42 dB (for the number of transmitted periods varying from 10 to 40). The applied technique allowed to extract the signals of transceivers S, W2, W4 and W5 up to a range of 350 km, and the signals of the most powerful low-frequency transceiver H over all area of the experiment. As a result, at each station the signals of 4÷6 transceivers were recorded. The collected data carry information about the sound speed field in several hundreds vertical slices in the area under study.

The present paper is concerned with interpretation and preliminary inversion of the acoustic receptions from the source H accomplished along H-W3 leg of the MOST experiment. The CTD data showed the sound speed field in the H-W3 vertical slice was almost range-independent below 500 m depth. The sound channel axis was located at 90-100 m depth near the H mooring and gradually dipped to 160 m horizon in the vicinity of W3. Simultaneously the sound speed at the surface grew from 1516 m/s near H to 1530 m/s near W3. The propagation conditions were characterized by very high vertical gradients of the sound speed above the waveguide axis which amounted to 0.5 s-1.

The sound speed field calculated from CTD measurements was used to verify the predictability of acoustic propagation. Sound velocity between the stations was found by linear interpolation. No CTD data were available in the vicinity of the H source. The nearest CTD cast at station #774 was located approximately 41 km apart. The medium between the transceiver and the nearest station was assumed range independent. The predicted and measured pulse responses (the latter is the correlation function) at a station #788 located 560 km off the source H (see Fig.1) are compared in Fig.2. An adiabatic normal mode model was used for this simulation. Taking account of the uncertainty in the absolute propagation range, on this and the subsequent plots the time axis for the measured signal was shifted to match the first peaks of the two curves. The overall correspondence between the measured and simulated pulse responses demonstrates that the quality of acoustic measurements is acceptable and the collected hydrographic information is complete enough. However, the measured and predicted patterns are not identical. There is a good agreement between the early arrivals corresponding to resolved steep eigenrays. For the later, flatter rays the pulse response is more complex and erratic. Its maxima are due to interference of multiple eigenrays not resolvable for the signal bandwidth used. The agreement between simulated and measured arrival patterns in this time interval is poor. The analytic estimates show that at distances over 300-350 km the final part of the signal should consist of discrete peaks corresponding to separate normal modes, which can be incorporated into the data set for tomographic inversion. This effect is seen in the

Fig.2. The measured (solid lines) and simulated (dashed lines) arrival patterns at the station #788. The final part of the time interval underlined with heavy line is shown in more detail on the lower plot. The arrivals corresponding to separate low-order normal modes are marked with corresponding numbers. The arrows show their presumable location on the experimental curve.

Fig.2. The theoretically predicted arrivals of adiabatic modes number 2, 3 and 4 are pointed out (the receiver is located deeper than the lower turning point of the first mode, hence, its amplitude is small). The lower plot in Fig.2 shows the similar maxima are observed in the measured arrival pattern. They arrive somewhat earlier then predicted by theory. This effect is partially explained by the heating of the near-surface water layers that does not affect the near-axial late arrivals and speeds up the early ones. Comparison of CTD and XBT data shows the sound speed increase reached 10 m/s while the R/V passed from H to W3. Another possible reason for the observed discrepancy is that the sound speed profile (SSP) at the source location could differ from that at the station #774.

The first step toward the complete reconstruction of the sound speed field in the H-W3 vertical slice is the tomographic inversion of the SSP at the source H from the data collected at the nearby stations. The results of such tomography may be verified against the independent data of direct XBT measurements. However, for propagation conditions in the Mediterranean sea the traditional linear inversion is not trivial even at short ranges. The main problem consists in ray identification. This fact is illustrated by the predicted and recorded arrival patterns at station #775 presented in Fig.3. At 62.5 km from H only

Fig.3. Reconstruction of the SSP at the location of the H mooring.
Bottom: the arrival pattern at the station #775. The solid line shows the experimental correlation function, the dotted and dashed lines - the pulse responses predicted for the initial and reconstructed media, respectively. The vertical lines stand for the eigenrays in the initial medium (dotted lines correspond to bottom-reflected eigenrays).
Top: SSP's at the source (left) and their deviations from the profile at the station #774. The dotted line shows the initial approximation - the same SSP as at the station #774, the dashed line - the profile found via tomographic inversion, the solid line - the SSP calculated from XBT data of the cast #2 (see Fig.1).

four early arrivals corresponding to steep rays may be identified. These rays carry quite limited information about the sound speed field in the upper layers. The subsequent maxima of the arrival pattern appeared due to interference of several eigenrays. Moreover, the trajectories of these flat near-axis eigenrays experience strong non-linear distortions under the influence of relatively small variations in the sound speed profile. At this ranges the modes have not separated in time yet and the only identifiable feature in the late part of the pulse responce is the final cutoff. To summarize, in the framework of ray- and mode-travel-time inversion there would be only 5 equations for the SSP variations, 4 of which are strongly dependent. Obviously, the arrival pattern presented in Fig.3 contains more information.

To take full advantage of available data and to avoid problems with identification and linearization it is natural to match the whole pulse responce without extracting separate rays or modes. The matched field processor in time domain minimizes the RMS discrepancy between the simulated and measured dependencies of signal amplitude vs. time. The predicted function depends on the sound speed field and the experiment geometry used in the trial model. The method requires repeated calculations of the arrival pattern on each step of iterations, which were performed in the framework of ray theory. This technique was applied to reconstruct the SSP at the location of the H mooring. The results of inversion are presented in Fig.3. The sought-for variations of the SSP were described with 4 empirical orthogonal functions built from historical data corresponding to the season and place of the MOST experiment. The reconstructed profile is in good agreement with available XBT measurement. Of special note is much better coincidence between the overall duration of the signal as compared to the initial approximation. More careful analysis showed that extending of the theoretically predicted pulse responce is due to the joint action of two effects:

- reduction of the sound speed near the waveguide axis more affect the late flat eigenrays than the early steep ones;

- the final cutoff corresponds to the flattest eigenray available. Varying the sound speed field one may choose this eigenray to have the zero grazing angle at the source or at the receiver (depending on where the sound speed is higher). Such eigenray will posses the longest propagation time possible for the given experimental geometry.

Together these reasons brought into correspondence the latest maxima of the measured and simulated arrival patterns. After that the experimental curve decays slower then the predicted function. This discrepancy may be partially accounted for wave phenomena which were ignored in our ray-theory model.


The 1994 expedition of R/V 'Akademik Sergey Vavilov' could not be organized without thorough assistance of Acad. Prof. L.M. Brekhovskikh. His support is greatly appreciated. The authors are grateful to Dr. Uwe Send (IfM, Kiel), Dr. Yve Desaubies (IFREMER, Brest) and Prof. Michael Taroudakis (IACM/FORTH, Heraklion, Crete) for their efforts in coordinating the MOST experiment with THETIS-2 project, and to Dr. Claude Millot (CNRS/COM, Toulon) for providing the XBT data. Work supported in part by INTAS, project 93-0557, RFBR, project 95-05-14616, and ISF, project M3P000.


[1] Munk, W. and C. Wunsch, "Observing the ocean in 1990s" // Phyl. Trans. Roy. Soc. London, A307 (1982), 439-464.

[2] Cornuelle, B., W. Munk and P. Worcester, Ocean Acoustic Tomography from ships // J. Geophys. Res., 94 (1989), 6232-6250.

[3] Send, U., Basin-scale tomography in the Western Mediterranean // J. Acost. Soc. Amer., 97 (1995), 3248-3249 (A).

[4] Godin, O.A., S.V. Burenkov, D.Yu. Mikhin, S.Ya. Molchanov, V.G. Selivanov, Yu.A. Chepurin and D.L. Aleynik, An experiment on dynamic acoustic tomography in the Western Mediterranean // To appear in Doklady Akademii Nauk, 1996.