Self-localization of buoyless fishing gear and other objects on the sea floor

There are many approaches to remotely locating assets in the ocean. For example, fishermen mark the location of fixed fishing gear (e.g., pots or traps) set on the sea floor using surface buoys attached to the gear with rope. There is also a myriad of methods to remotely locate assets using active acoustics [reviewed by Tan et al. (2011)]. The vast majority of applications to date involve the localization of a high value asset, such as a diver, moored instrument, or autonomous underwater vehicle (AUV), using one or more reference stations (e.g., buoy, ship) whose position is accurately known. For example, moored scientific and industrial equipment is often located using an acoustic transponder attached to the equipment and a similar transponder deployed from a ship; range can be estimated from the two-way acoustic travel time measured by the transponders, and several ranges measured at different bearings relative to the equipment on the sea floor can be used to estimate its location. Transponders can also be used with directional receivers [e.g., ultra-short baseline (USBL) receiver arrays] to simultaneously measure range and bearing to estimate position. Most of the time, each of these methods relies on a localization process that occurs somewhere other than at the asset that needs to be localized, such as aboard a ship (exceptions include automated navigation of an AUV). As such, these methods are unhelpful when an object moves of its own accord to a new location that is outside of the detection range of the localization process (e.g., fixed fishing gear or scientific equipment moved in a storm or dragged to a new location by trawled nets). If the object could self-localize, then it could report its position to passing ships so that it can be relocated by its owner.

We were motivated to create an acoustic self-localization capability for a specific problem: the development of buoyless fixed fishing gear to substantially reduce lethal entanglements of large whales, particularly the critically endangered North Atlantic right whale (Eubalaena glacialis). North Atlantic right whales number less than 400 animals total (Pettis et al., 2021), and the species has been declining over the last decade due to unsustainable entanglements in fixed fishing gear and ship strikes (Corkeron et al., 2018; Sharp et al., 2019). Buoyless fishing is defined as trap or pot fishing without the use of end lines and buoys, and it holds the promise to significantly reduce entanglements of large marine vertebrates, including whales, sharks, and turtles. Implementation of buoyless fishing faces many economic, cultural, and regulatory difficulties, but the principal technical challenge is marking the location of the gear on the sea floor to avoid gear conflict, such as fishermen setting long lines of traps (called trawls) on top of one another or mobile fishermen (e.g., trawlers, draggers) dragging nets through fixed fishing gear (Baumgartner et al., 2019). We propose that acoustic modems on fixed fishing gear could take advantage of a network of acoustic modems attached to fishing vessels to both self-localize and transmit vital information about the location, ownership, and status of the fishing gear.

Self-localization is ubiquitous in consumer and industrial products thanks to global navigation satellite systems (GNSSs) like the Global Positioning System (GPS). These systems use a constellation of satellites whose positions are accurately known to broadcast microwave radio transmissions toward earth. Using both the timing of the transmissions sent from several satellites and the data encoded therein, a receiver can calculate its own position on earth. Unfortunately, these microwave radio transmissions are strongly absorbed by seawater, so self-localization via GNSS alone is impossible below the surface of the ocean. However, a similar concept can be developed for the ocean by using a “constellation” of fishing vessels whose positions are accurately known (via GNSS) that broadcast periodic acoustic transmissions. By passively receiving these transmissions, acoustic devices on fishing gear can use both the timing of the transmissions and the data encoded therein to calculate their own position on the sea floor. These location data, as well as encrypted ownership information, can be transmitted to approaching fishing vessels for display on the vessels' chart plotters to avoid gear conflict and can be stored by the passing vessel and transmitted to a cloud database when the vessel returns to shore for decrypted access by gear owners, regulators, and fishery enforcement agencies. Unlike GNSS satellites, fishing vessels need not be dedicated solely to the task of providing information for fishing gear to self-localize (i.e., they can go wherever their fishing operations require), nor does the self-localization process require that transmissions from passing vessels be received simultaneously by the acoustic device attached to the gear.

In this paper, we describe a method we call successive acoustic receive time (SART) self-localization, which takes inspiration from the silent positioning scheme of Cheng et al. (2008) and computational methods from Baumgartner et al. (2008). Cheng et al. (2008) used time-difference-of-arrival calculations for a passively listening system and several serially broadcasting nodes; these nodes must be present at the same time to localize, whereas SART allows nodes to be present asynchronously. Baumgartner et al. (2008) solved time-difference-of-arrival equations to track tagged whales in real time using similar computational methods to those described in the supplementary material.¹ Unlike these traditional time-difference-of-arrival approaches, however, SART uses differences in travel times from the same moving node (a passing ship) instead of differences in arrival times at multiple fixed nodes to localize. For our method, we assume that (1) both mobile and fixed fishing vessels are equipped with a GPS receiver and an acoustic modem to send and receive encoded data acoustically (we call this modem the ship modem), (2) fishing gear (or other equipment) on the sea floor is equipped with a similar acoustic modem (we call this modem the trap modem), and (3) power usage for the modems on the sea floor is minimized to maximize battery life. Note that we make no assumptions about the trap and ship modems having synchronized time. The SART method was implemented in a research acoustic modem as a proof of concept, and we present preliminary data on its localization accuracy.

The acoustic modem carried by a fishing vessel (hereafter called the ship modem) transmits a message at a regular interval (e.g., every 60 s) that contains the position of the vessel's modem transducer (latitude, longitude, and depth), the exact time of transmission, and the nominal interval at which this message is transmitted. Any modem attached to gear on the sea floor (hereafter called the trap modem) receives these data and stores them in a table along with the exact time of reception. If the trap modem has already self-localized, it will transmit to the ship its estimated position along with encrypted ownership information, and those location data will be displayed on the ship's chart plotter to minimize gear conflict. If the trap modem has not yet collected enough messages to self-localize, it can transmit to the ship a GPS-derived location that was stored when the trap modem was last at the surface (i.e., at the gear deployment location). This GPS position could be supplied by an attached GPS or telemetered to the trap modem via a radio modem (e.g., Bluetooth) on the ship. When the vessel returns to shore, all messages received by the ship modem during its time at sea can be uploaded to a cloud database for selective and authorized access by gear owners, regulators, and fishery enforcement agencies. With these data, a gear owner on shore could learn the location of his or her fishing gear that has moved from its original deployment location, thus allowing successful recovery of moved gear.

SART self-localization relies on measuring successive differences in transmission times and successive differences in reception times to estimate the location of the trap modem. These differences are stored in a message table that grows over time as more vessels pass the gear; with the addition of more messages, the accuracy of the location estimate improves. For localization methods that rely on traditional ranging or ranging with a directional receiver, each ship that passes gear must independently determine its position, but the SART self-localization method accumulates information from multiple ship passages to determine and improve its position estimate.

The ship modem transmits a message from its transducer at position (X_i, Y_i, Z_i) at time t_i (i is a counter such that i = 1 for the first message, i = 2 for the second message, and so on), and the message is received by the trap modem's transducer at position (x, y, z) at time $tir$ (Fig. 1 and Table S1¹). The message contains X_i, Y_i, Z_i, and t_i, and $tir$ is measured by the trap modem. The slant range (S_i) between the ship modem's transducer and the trap modem's transducer at time t_i is

We assume that z (the depth of the trap modem) is known, either from an integrated pressure sensor or telemetered via radio modem (e.g., Bluetooth) from the ship's fathometer immediately prior to deployment (although this latter method is problematic if the trap modem moves from the original deployment location and the water depth varies considerably in the deployment area).

The ship modem is programmed to transmit its message at a regular interval $Δt$⁠, so at the next transmission, $ti+1=ti+Δt$⁠, the slant range is now

Note that we assume the trap modem is stationary on the sea floor between t_i and t_i + 1. In fact, we will assume that the trap modem is always stationary during a ship's passage past the trap modem, and we will use significant changes in diagnostics from the calculations below to detect if the trap modem has moved between ship passages.

The slant range can also be estimated as

where c is the speed of sound in water and $tir−ti$ is the travel time for message i. The difference between slant ranges for two successive ship modem messages is

Note that since $ti+1r−tir$ and $ti+1−ti$ are measured independently, the ship and trap modem clocks do not need to be synchronized to estimate the difference in the slant ranges. Moreover, if acoustic propagation conditions cause refraction and lengthen travel times relative to a straight-line path between the ship and trap modems, the estimation of $Si+1−Si$ is unaffected as long as the propagation conditions do not change temporally over $Δt$ and spatially over the distance the ship travels in $Δt$⁠.

By repeatedly measuring $ti+1r−tir$ and $ti+1−ti$ as well as X_i, Y_i, Z_i, X_i+1, Y_i+1, Z_i+1, and z, values of x, y, and c can be estimated using the equations presented in the supplementary material.¹ Initial estimates of x, y, and c are required for these calculations. For x and y, the last known location of the trap modem is used. Prior to successful self-localization, the last known position is a GPS-derived location measured immediately before deployment; after self-localization, the last known position is the last self-localized location. The speed of sound (c) is initialized to 1500 m s⁻¹. A proposed hardware design for systems to implement the trap and ship modem functions described here is presented in the supplementary material.¹

The success of underwater localization algorithms depends a great deal on the orientation or geometry of reference nodes relative to the object being localized (in this case, the positions of the ship modems relative to the trap modem). To assess this geometry, we calculate what we call a geometry score and only attempt self-localization if the geometry is adequate. Upon reception of a new message from a ship modem, an origin for a Cartesian coordinate system is established as either (1) the last good self-localized position, (2) a GPS-derived location measured immediately before deployment, or (3) the average of all ship positions in the message table. All earth coordinates (latitude and longitude) for the ship messages and the last known position of the trap modem are converted to eastings and northings relative to this origin, and all further calculations are carried out in these Cartesian coordinates. Prior to attempting to self-localize with the latest message using the equations above, two checks are completed to determine if it is appropriate to conduct the self-localization. First, there must be at least four travel time difference estimates (m_i) to estimate x, y, and c. Second, the geometry of ship modem transducer locations contained in the message table must be suitable for localization, which we assess with the geometry score.

The difference in successive travel times is a measure of whether the ship modem is moving toward or away from the trap modem. If $ti+1r−tir<ti+1−ti$⁠, then the ship is moving toward the trap modem, and if $ti+1r−tir>ti+1−ti$⁠, then the ship is moving away from the trap modem (see Table S1¹ and Fig. 2). We use this to assess the geometry by creating a 10 × 10 grid of locations over a 4 × 4 km area centered at the origin of the Cartesian coordinate system (Fig. 2). For each of these 100 grid locations, we determine the percentage of travel time differences derived from the message table that are consistent with the true motion of the ship(s) assuming the trap modem is at that grid location. A travel time difference is considered consistent with the motion of a ship for a particular grid location if either (1) the ship is moving away from that location and $ti+1r−tir>ti+1−ti$ or (2) the ship is moving toward that location and $ti+1r−tir<ti+1−ti$⁠. For a good geometry, very few grid locations will be consistent with all of the measured travel time differences. The geometry score is computed as 100 minus the percentage of grid locations for which all travel time differences are consistent with the true ship motions. We only attempt a SART self-localization if the geometry score is 95 or higher.

A program incorporating the calculations above was developed in the C programming language on a MacBook Pro laptop and tested with simulated data. The program was ported to a Woods Hole Oceanographic Institution MicroModem 2 acoustic modem (Gallimore et al., 2010) for use in real time as the trap modem. A second MicroModem interfaced with a personal computer was used as the ship modem; the computer ran an application designed to send messages to the trap modem at a fixed time interval and to visualize both the ship and trap modems' locations. The ship modem transmitted to the trap modem the following information: a unique ship modem identifier, the ship modem transducer location (X, Y, Z), the date and time of the message transmission (t), and the interval between message transmissions (⁠$Δt$⁠). The trap modem received this transmission (at time t^r), added the message to the message table, computed a geometry score, and, if the geometry score was 95 or higher, attempted a SART self-localization. After receiving three transmissions from the same ship, the trap modem transmitted a message to the ship containing a unique trap modem identifier, its last known position, the source of that position (i.e., GPS-derived or SART self-localized), and other diagnostic information. The trap modem was programmed to only send this trap-to-ship message once; the trap modem kept track of ship modem identifiers to whom it had sent this trap-to-ship message and did not communicate again with those ship modems until some nominal interval had passed (e.g., 1 h). In addition to self-localizing, the trap modem was also capable of responding to ranging messages sent by the ship modem that allowed the distance between the ship and trap modem to be estimated using two-way travel time. The source code for SART calculations and the MicroModem 2 implementation of the trap and ship modem described here are freely available from the authors upon request.

On November 11, 2019, we moored the trap modem 1.5 m off the sea floor in a water depth of 14 m in Buzzard's Bay, Massachusetts from the R/V Tioga. From the sea floor to the sea surface, the mooring consisted of a 300-lb Dor-Mor anchor, a 0.5-m length of chain, the 1.5-m long trap modem cage, a 28.5-m length of synthetic rope, and an A-3 Polyform float. Two 36-cm Panther floats were suspended roughly 1 m above the trap modem cage, and lead was inserted into the rope 12 m below the A-3 Polyform float to create a catenary above the trap modem. The 2:1 scope was necessary to account for tidal currents in this area. An internally recording hand-held GPS was affixed to the A-3 Polyform float to record the location of the float during the demonstration. After deployment, the R/V Tioga used a profiling conductivity-temperature-depth (CTD) instrument to measure sound speed for comparison to the SART estimates (this CTD-derived measurement was not used to initialize the SART calculations) and then steamed in a circle around the deployment location to collect two-way travel time measurements and estimate the location of the trap modem using ranging [Fig. 3(a)]. The R/V Tioga then conducted three transects away and past the trap modem while sending SART messages every 45 s. The ship modem identifier was changed as the ship turned on a new transect to make it seem to the trap modem that it was being passed by three different ships [Fig. 3(a)]. The trap modem self-localized using the SART messages, and the accuracy of the self-localizations was estimated relative to the position estimate determined by ranging. These accuracy estimates were compared to the accuracy of the buoy in representing the true location of the trap modem on the sea floor.

Nine two-way travel time measurements were collected at roughly 300 m range [average = 294 m, standard deviation (SD) =17.5 m] and 40° angular intervals relative to the moored trap modem. From these measurements, the location of the trap modem was estimated using ranging [Fig. 3(a)]. Although this method of localization has errors of its own, we will treat this ranging estimate as the true location of the trap modem and will estimate the accuracy of the SART localization relative to this range position estimate.

After completing the ranging survey, the ship then traveled away from the trap modem to the north while sending SART messages [mimicking a fishing vessel that had just deployed the trap modem on fixed fishing gear; “Ship 1” in Fig. 3(a)]. After the completion of the northerly transect, the ship modem identifier was changed, and the ship turned and steamed to the southeast while sending SART messages [“Ship 2” in Fig. 3(a)]. After completion of this southeasterly transect, the ship modem identifier was changed one last time, and the ship turned and traveled to the west-southwest while sending SART messages [“Ship 3” in Fig. 3(a)]. The trap modem calculated its first position halfway through the ship 2 transect and calculated updated positions throughout the rest of the ship 2 and ship 3 transects.

The mooring float was, on average, 20.1 m (SD = 1.8 m, median = 20.7 m, interquartile range = 18.3–21.4 m) from the trap modem, while the trap modem SART self-localization estimate was, on average, 6.8 m (SD = 2.6 m, median = 5.5 m, interquartile range = 5.0–9.2 m) from the trap modem [Figs. 3(b)–3(d)]. The SART location estimates varied over time, but they settled to 5.2 m from the trap modem toward the end of the ship 3 transect [Fig. 3(d)]. Sound speed estimated from the CTD measurements changed monotonically from 1491.9 m s⁻¹ at the surface to 1493.9 m s⁻¹ near the sea floor for an average water column sound speed of 1492.9 m s⁻¹ (SD = 0.646 m s⁻¹). SART sound speed estimates were 1.3 m s⁻¹ lower than the CTD estimate, averaging 1491.6 m s⁻¹ (SD = 1.088 m s⁻¹).

SART self-localization offers several advantages over traditional ranging. Ranging requires that each ship that wishes to know the location of gear on the sea floor conduct a survey around each piece of gear to obtain its position, as we did at the start of our field demonstration [diamonds in Fig. 3(a)]. In most cases, fishermen do not have time to conduct such a survey for every piece of gear on the sea floor. SART self-localization allows information from passing ships with ship modems to be used to determine and refine the estimated position of the gear. Traditional ranging requires regular acoustic transmissions between the ship and a transponder on the gear; every time the ship sends a ranging transmission, the transponder must respond to allow a two-way travel time and slant range to be calculated. SART localization uses passive listening to collect messages from passing ships and responds only once to a passing ship to share its location and other information (e.g., registration number, owner). This minimizes the acoustic transmissions of the trap modem and reduces both power consumption and the amount of noise introduced into the environment. Finally, the use of acoustic modems for SART offers the capability to share information (encrypted or not encrypted), such as location, registration number, owner, deployment location, deployment time, and gear type, which is information that is necessary for both fishermen and enforcement agencies. Simple transponders do not allow the communication of such data.

Traditional ranging with USBL arrays is another method of localizing gear that provides both distance (slant range) and bearing to transponders on the sea floor from which position can be estimated. While surveys are not required to locate gear as with traditional ranging, transponders affixed to gear would be required to respond to every ship transmission, thus consuming power and introducing more acoustic noise than SART self-localization. If the transducers could act as modems, then transmissions could be limited in the same way as described above for SART self-localization (i.e., the transponder affixed to the gear keeps track of which ship modems it has communicated with in the recent past). However, localization with this method is sensitive to bearing accuracy. For example, a bearing error of ±1° results in an error of ±17.5 m at 1000 m, and a bearing error of ±2° results in an error of ±34.9 m at 1000 m. These errors improve with decreasing range (±1° results in an error of ±1.75 m at 100 m, ±2° results in an error of ±3.49 m at 100 m), which implies that traditional ranging with a USBL receiver array would almost certainly require multiple responses from the transducer affixed to the gear to acquire sufficient accuracy to avoid gear conflict.

One of the potential drawbacks of SART self-localization is that it cannot produce a position estimate until an adequate geometry of observations is obtained. In our simulations, this does not occur until the first ship passes by the trap modem [Fig. 2(c); assuming the vessel that deployed the gear sends SART messages as it exits the area as shown in red filled circles in Fig. 2]. In our implementation of the trap modem, the first ship to approach the gear will receive from the trap modem a GPS-derived surface deployment location, not a SART self-localization. After the first ship passes abeam of the gear, the geometry becomes sufficient for the trap modem to begin SART self-localization. Thereafter, the trap modem will communicate the SART self-localized position to passing ships. A system that uses transponders and directional transducers would allow for localization immediately after deployment (i.e., would not need to rely on a GPS-derived surface deployment location) but would likely require more power and would produce more noise (i.e., more transmissions) than a SART self-localization system. However, a combination of the two systems (i.e., a SART self-localization system with directional transducers attached to the ship modem) would solve this problem completely. If the trap modem reported anything other than a SART-localized position (e.g., a GPS-derived surface deployment location), the ship modem could use the bearing and two-way travel time to estimate a position. Once the geometry allowed SART self-localization, the bearing to the trap modem would no longer be used to localize the trap modem.

SART self-localization takes advantage of the fact that many vessel operators have an interest in knowing what is on the sea floor and where it is located. This same network of vessels has the capacity to collect information from bottom-mounted gear and deliver it to a shore-side cloud database for dissemination to appropriate parties, including gear owners, regulators, and enforcement. Gear location marking is essential for the successful implementation of buoyless fishing, but it also offers an opportunity for other seafloor users, such as ocean researchers, to locate and communicate remotely with their equipment and to avoid costly interactions with other users. Moreover, SART self-localization offers the prospect of significantly reducing lost gear and equipment, thereby saving on replacement costs and reducing marine debris.

We are grateful for the expert assistance of the following Woods Hole Oceanographic Institution engineers and technicians: Keenan Ball, Dennis Giaya, John Kemp, and Meghan Donahue, as well as the captain and mate of the R/V Tioga, Pete Collins and Leah Saunders. We also thank two anonymous reviewers for helpful suggestions that improved this article. This work was funded by the Woods Hole Oceanographic Institution's Innovative Technology Program.