Automatic calibration for wobble errors in shallow water multibeam bathymetries*

Tianyu YUN , Xianhai BU ,**, Zhe XING , Zhendong LUAN , Miao FAN , Fanlin YANG ,4

1 College of Geodesy and Geomatics, Shandong University of Science and Technology, Qingdao 266590, China

2 National Marine Data and Information Service, Tianjin 300012, China

3 Key Laboratory of Marine Geology and Environment & Center of Deep Sea Research, Institute of Oceanology, Chinese Academy of Sciences, Qingdao 266071, China

4 Key Laboratory of Ocean Geomatics, Ministry of Natural Resources, Qingdao 266590, China

Abstract The wobble errors caused by the imperfect integration of motion sensors and transducers in multibeam echo-sounder systems (MBES) manifest as high-frequency wobbles in swaths and hinder the accurate expression of high-resolution seabed micro-topography under a dynamic marine environment.There are many types of wobble errors with certain coupling among them. However, those current calibration methods ignore the coupling and are mainly manual adjustments. Therefore, we proposed an automatic calibration method with the coupling. First, given the independence of the transmitter and the receiver,the traditional georeferenced model is modif ied to improve the accuracy of footprint reduction. Secondly,based on the improved georeferenced model, the calibration model associated with motion scale, time delay,yaw misalignment, lever arm errors, and soundings is constructed. Finally, the genetic algorithm (GA) is used to search dynamically for the optimal estimation of the corresponding error parameters to realize the automatic calibration of wobble errors. The simulated data show that the accuracy of the calibrated data can be controlled within 0.2% of the water depth. The measured data show that after calibration, the maximum standard deviation of the depth is reduced by about 5.9%, and the mean standard deviation of the depth is reduced by about 11.2%. The proposed method has signif icance in the precise calibration of dynamic errors in shallow water multibeam bathymetries.

Keyword: multibeam echo-sounder systems; footprint reduction; wobble errors; genetic algorithm;automatic calibration

1 INTRODUCTION

Multibeam echo sounding technology is one of the main approaches for seabed topography and geomorphology survey, with the characteristics of high precision, full coverage, and high effi ciency(Hughes Clarke, 2000; Li et al., 2008). Due to the dynamic working environment of multibeam echosounder systems (MBES) and the characteristics of the multisensor combination of the system itself. The sounding results could be aff ected by various errors,including outliers caused by the system noise or ocean environmental noise, refraction errors caused by the incomplete surface sound velocity or sound velocity prof ile, tidal correction errors caused by imperfect tidal measurements or the inappropriate tidal calibration model, static draft errors of the transducer caused by the mass change of the carrier itself,dynamic draft errors caused by the uneven motion of the vessel, transducer installation errors and its calibration errors, nonconcentric array errors between sectors, and integration errors (Hare, 1995; Hare et al.,1995; Hughes Clarke, 2003a, b; Church, 2020). Except for the outliers, other errors can be divided into staticerrors and dynamic errors according to the nature of the measurement. The static errors are usually in larger magnitude and have received extensive attention.There are many mature methods to calibrate the static errors (Kammerer, 2000; Yang et al., 2013; Zhao et al.,2014; Hamilton et al., 2014; Guan et al., 2019; Bu et al., 2020). Compared with the static errors, the magnitude of the dynamic errors related to the motion sensor and the vessel motion is small, usually within the special order of standard in the International Hydrographic Organization (IHO) S-44 (IHO, 2008).The essence of wobble errors is the dynamic errors caused by the imperfect integration of motion sensors and transducers in MBES. When the magnitude of the dynamic errors is greater than 0.2% of the water depth(Maingot et al., 2019), the wobble errors may manifest as high frequency ribs or wobbles perpendicular to the heading direction in the high-precision shaded relief maps. The existence of wobble errors will aff ect the presentation of microtopography, thereby misleading the accurate analysis of seabed topography.

The problem of wobble errors in the MBES was f irstly proposed by Hughes Clarke, Ocean Mapping Group (OMG) of New Brunswick University in Canada, and the error sources and manifestations of these errors are discussed in detail (Hughes Clarke,2003b). Hughes Clarke realized the discrimination of these error sources based on the statistical method by analyzing correlations between motions and acrosstrack prof ile-averaged residuals in the f lat area.Inspired by Hughes Clarke’s method, Yang et al.(2009, 2010) used wavelet analysis or Fourier transform to separate the high frequency wobble errors and the terrain trend. Then, a similar statistical method was used to locate errors. However, the above studies only focus on analyzing and discriminating single or dominant errors, ignoring the correlation between the errors, and do not give an eff ective method to eliminate these wobbles directly. To eliminate the errors existed in the motion sensor,Demkowicz et al. (2003) analyzed the inf luence of the roll error based on the simulated data and proposed to use the Kalman f ilter to correct wobbles in the bathymetries. Zhao et al. (2014) proposed a method of calibrating these errors based on the spectrum characters of seabed-topography trend and multibeam bathymetric data. Although the methods of Demkowicz and Zhao could iron these wobbles, the two methods are of strong subjective. When calibrating the installation errors of the transducers in the bathymetric data, Bjorke (2005) developed a method based on the soundings’ georeferenced model.The installation parameters were solved by the least square algorithm, with the correlation between the parameters considered. Based on Bjorke’s georeferenced model, Seube and Keyetieu (2017) and Keyetieu et al. (2018) realized the calibration of the multibeam installation errors and the motion time delay. However, the georeferenced model adopted by Bjorke ignored the independence of the transmitter and the receiver, which can lead to inherent errors in the bathymetric data and hinder the calibration results of these small wobble errors. Maingot’s method considered the independence of the transmitter and the receiver, whereas the calibration process was complicated and the specif ic calibration model was not given (Maingot et al., 2019).

Consequently, limitations are as follows in the existed methods: 1) only useful for wobbles caused by single or dominant error; 2) lack of objective method considering the correlations between the errors; 3) lack of an accurate and effi cient georeferenced model to eliminate these wobbles. To solve the problems above, this paper proposed a method calibrating the wobbles caused by single or multiple errors automatically. When constructing the georeferenced model, the independence between the transmitter and the receiver is considered, and the more simplif ied model is deduced specif ically. Then,the wobble errors are considered as parameters in the improved georeferenced model to construct the calibration model. Finally, the genetic algorithm (GA)is used to search for the optimal error parameters with soundings in the f lat area.

2 ANALYSIS OF ERROR SOURCES OF THE WOBBLES

Fig.2 Three types of wobble manifestation (Hughes Clarke, 2003b)

As indicated in Maingot’s study (Maingot et al.,2019), when the magnitude of wobble errors is greater than about 0.2% of the water depth, the high-frequency wobbles are easy to appear, as shown in Fig.1.According to Hughes Clarke’s study, the error sources can be divided into three types: 1) motion scale, time delay, and yaw misalignment; 2)XandYlever arm errors; and 3) inaccurate surface sound speed correlated with vertical motion or rolling. Although the three kinds of errors all lead to wobbles, their manifestations are diff erent in shallow water. The f irst type of error results in a linear tilting of the acrosstrack prof iles (Fig.2a); the second type of error causes a linear lifting and falling of the across-track prof iles(Fig.2b); and the third type of error leads to a complex nonlinear tilting of the across-track prof iles (Fig.2c).Since the third type of error is related to the external factor out of the MBES, thus, this paper only focuses on addressing the f irst and second types of errors in the MBES.

2.1 Time delay Δ t

Due to serial port transmission or buff er setting, a certain time delay may exist between the motion sensor and the MBES. According to the research(Hughes Clarke, 2003b), it is found that the roll error caused by time delay Δthas the greatest inf luence on depth (Fig.3). When time delay Δtexists, the depth error is def ined by Eq.1.

For instance, a typical wave period of 10 s with a roll amplitude of 3° may provide a maximum rate up to 0.019°/ms (Hughes Clarke, 2003b). Taking the edge beam with beam steering angleθbeing 60° as an example, whenris 2°, a time delay as small as 3 ms may cause depth errors up to 0.2% of the water depth,thereby leading to wobbles.

Fig.3 Schematic diagram of depth distortion after introducing time delay or motion scale

2.2 Motion scale Δ ρ

The output value of the motion sensor is a factor associated with the motion time series, that is, the error is a scale with the real value (Hughes Clarke,2003b). The larger the real value is, the greater the error is, and vice versa. The roll error caused by the motion scale Δρalso has the greatest inf luence on the depth (Fig.3). When there is a motion scale Δρ, the depth error can be obtained by Eq.2.

Fig.4 Inf luence of yaw misalignment on the measured roll and pitch when the pitch of vessel exists (Yang et al.,2010)

For a general roll and pitch with the magnitude of 4° and 2°, respectively, yaw misalignment as small as 2° will cause depth errors beyond 0.2% of the water depth and lead to wobbles.

2.3 Yaw misalignment Δ κ

Ideally, the longitudinal axis (X-axis) of the motion sensor should be parallel to the longitudinal axis of the vessel to ensure that the measurements of the motion sensor can accurately ref lect the motion of the vessel. However, due to imperfect manual installation,it is inevitable to introduce a certain angle bias Δκ(Fig.4). According to the eff ect of yaw misalignment(Yang et al., 2010), the relationship between the actual vessel rollr, pitchp, and the sensor output rollr′,pitchp′ can be calculated as Eq.3.

Fig.5 Main coordinate systems in the MBES georeferenced model

2.4 Lever arm errors [Δ x, Δ y]

Lever arm errors are the off set errors of the motion sensor relative to the multibeam transducer. Since there is a certain off set between them, under the inf luence of the attitude, the transducer will produce an induced heaveHinduce. When errors exist in the lever arm, uncertainty will be introduced into the induced heave, which leads to the overall f luctuation of the swath. According to Hughes Clarke (2003b),lever arm errors inXandYdirections have a great inf luence, and the induced heave error caused by Δxand Δylever arm errors can be calculated as Eq.4.

For a general roll and pitch with the magnitude of 4° and 2°, respectively, both Δxand Δyas small as 1 m will cause the induced heave error up to 0.035 m.And these errors also will lead to wobbles when the water depth is approximately less than 17 m.

3 AUTOMATIC CALIBRATION METHOD OF WOBBLES

Since Bjorke’s georeferenced model does not consider the independence of the transmitter and the receiver, the accuracy of the model is low, which is unfavorable to the high-precision calibration of these small wobble errors. Therefore, this traditional model is improved f irstly, and the calibration model coupled with these wobble error parameters is constructed based on the improved georeferenced model. Finally,the GA is used to search for the optimal error parameters.

3.1 Improvement of traditional georeferenced model

The coordinate systems involved in the georeferenced model include the transducer coordinate systemO-XtYtZt, the attitude coordinate systemO-XmYmZm, the vessel coordinate systemOXvYvZv, and the local lever coordinate systemO-NED(Yang et al., 2017). By using the sounding information of multibeam and measurements from each auxiliary sensor, the depth information under the multibeam transducer can be transformed to the geographical coordinate system based on a series of transformations,as shown in Fig.5.The IMU is generally located in the center of the vessel and the transducer is generally located on the side of the vessel or at the bottom of the vessel.

The coordinates of the soundings calculated based on the georeferenced model in the geographical coordinate system (Bjorke, 2005; Barnard, 2012) can be expressed as:

where [0,Ssin(θr+θR),Scos(θr+θR)]Tis the coordinates of the soundings in the transducer coordinate system;Sis the oblique distance from the acoustic center of the transducer to the sounding, which can be obtained by ray tracing;θris the received beam steering angle;θRis the sound ray refraction angle; [Xoff,Yoff,Zoff]Tis the off set between the acoustic center of the transducer and the center of the global navigation satellite system (GNSS)antenna in the vessel coordinate system; [XG,YG,ZG]Tis the geographical coordinates of the GNSS antenna center in Earth-Centered Earth-Fixed (ECEF) coordinate system; Rwis the rotation matrix consisted of the longitude and latitude (Wikipedia, 2021); Rattand Ralignare the rotation matrices composed of the attitudes and installation angles, respectively; the rotation matrices of the angles in three directions are expressed as in Eq.6.

Fig.6 Inf luence of roll and pitch on the transmitted beam pattern and received beam pattern, respectively

where R(r), R(p), and R(κ) are the rotation matrices of roll (r), pitch (p), and yaw (κ), respectively.

In general, when multibeam transducers are installed, the transmitter is often parallel to the direction of the bow- and stern- line, while the receiver is perpendicular to the bow- and stern-line. According to the virtual concentric array (VCCA) method and beamforming theory (Beaudoin et al., 2004; Teng, 2012), the transmitted beam is more sensitive to the attitude of pitch rotation, while the received beam is more sensitive to the attitude of the roll rotation, as shown in Fig.6.

To derive the beam incidence vector (θ,Φ) considering the independence of the transmitter and the receiver,with the inf luence of the txsteer angleθt(Hamilton et al., 2014), the coordinate of the unit vectorU=[0 0 1]Talong theZ-axis in the transducer coordinate system is def ined by Eq.7, whereas with the inf luence of the rxsteer angleθr(Hamilton et al., 2014) it is def ined by Eq.8.

Then, the beam incidence vector in the transducer coordinate system can be calculated by Eq.10.

R′attand R′alignare the rotation matrices composed of the attitudes and installation angles considering the independence of the transmitter and receiver, respectively;ris the roll at the receiving time;pis the pitch at the transmitted time;κis the yaw at the transmitted time; ΔRis the roll installation angle of the receiver; ΔPis the pitch installation angle of the transmitter; ΔHis the yaw installation angle of the transmitter;Hdraft,Htide,Hheave,andHinduceare the draft of the transducers, the tide, the heave, and the induced heave, respectively.

3.2 Calibration model coupled with wobble errors

According to Eq.3, when time delay Δt, motion scale Δρ, and yaw misalignment Δκexist simultaneously, the output rollr′ and pitchp′ are given in Eq.12.

H′induceis the induced heave with the inf luence of the above four errors.

3.3 Solving the error parameters

This paper proposesg(Δt, Δρ, Δκ, Δx, Δy) which represents the depth diff erence between the plane f itted by the least square method and the real-time estimated seabed. When the estimation ofg(Δt, Δρ,Δκ, Δx, Δy) approaches zero, the corresponding parameter solution is the optimal solution. Typical seaf loors, however, are not always planar. Fast changes in seaf loor complexity, induced by the sudden appearance of such high frequency features, including sand waves, are likely unsuitable for analysis(Maingot et al., 2019). Therefore, to facilitate the elimination of wobble errors, f lat areas are selected to avoid the inf luence of complex terrain changes on the performance of the proposed method. Since the wobble errors always manifest themselves as acrosstrack ribbing, symmetrical soundings on both sides of the swath along the track are selected for the calibration.

Fig.7 Flow chart of the calibration method based on the GA

Sinceg(Δt, Δρ, Δκ, Δx, Δy) is diffi cult to linearize and solve by the least square method, the GA (Aytug et al., 2003; Goyal and Agrawal, 2013; FuRui et al.,2019) is used in this paper. The GA is proposed by Holland (1992), which can quickly search for the global optimal value. According to the principle of the GA, the f itness function is def ined by Eq.14:

whereiis the serial number of soundings in the selected area;Nis the number of soundings in the whole region.

The process of using the GA to automatically invert the best estimation of the corresponding error parameters can be divided into the following steps.

(1) Improving the georeferenced model;

(2) Introducing the corresponding error parameters into the improved georeferenced model;

(3) Calculating the coordinates of the soundings and f itting the plane;

(4) Encoding the feasible solutions of the corresponding error parameters by the real number;

(5) Using the random undirected initialization method to initialize the population in the corresponding error parameter range, according to the number of population set;

(6) Making each individual of the population evolve to generate new individuals, according to the set crossover probability and mutation probability(After testing, the crossover probability and mutation probability set in this paper are 0.5 which can not only ensure the optimal results but also ensure the computational effi ciency.);

(7) Using the wheel selection method to calculate the probability of each individual being left, according to the f itness value of the new individual, and making the retained individuals form a new population;

(8) Repeating steps (6)-(7), if the set number of iterations is not reached or the optimal f itness value is less than the set threshold, or continuing step (9);

(9) Selecting the optimal individual which is the optimal solution of the corresponding error parameters, according to the f itness value.

The specif ic process is shown in Fig.7.

4 EXPERIMENT

Both simulated data and measured data are used to verify the proposed method. These datasets are preprocessed with noise removal, ray tracing, patch testing (Hughes Clarke, 2003b), and tide correction.In the simulated experiment, wobble errors are artif icially introduced. In the experiment of measured data, the measured data with wobble errors are selected. Then, symmetrical soundings in the f lat area on both sides of the swath along the swath are selected to construct the f itted plane and invert the wobble error parameters.

4.1 Simulated experiment

The data used for the simulated experiment are selected from the public dataset collected in Wellington Port, New Zealand in 2012 (Reson Company, 2012), which are acquired by Kongsberg EM2040. The average depth of the survey region is about 20.3 m. As described in Section 2, the common wobble problems are mainly caused by the motion scale, time delay, yaw misalignment, andXandYlever arm errors. Therefore, the above wobble errors are taken as the representative for analysis. To verify the proposed method, these above wobble errors are artif icially introduced and set to common values to get visible wobbles (Hughes Clarke, 2003b). The swath in the blue rectangle at the bottom of the survey region is used to analyze the results before and after calibration, as shown in Fig.8.

Fig.8 Survey region of simulated data

4.1.1 Single error

In the original data, a 1.02 motion scale, 15-ms time delay, 2° yaw misalignment, and 1-mXandYlever arm errors are introduced respectively. Then, the artif icial errors are calibrated with the proposed method.

As shown in Fig.9, under the inf luence of motion scale, time delay, and yaw misalignment, the edge of the swath has obvious wobbles (Fig.9a, c, e). While the whole swath undulating phenomenon is introduced due to theXandYlever arm errors (Fig.9f). After calibration with the proposed method, the introduced wobble errors are eliminated. With our method, the f inal wobble error parameters of motion scale, time delay, yaw misalignment, and lever arm errors are approximately 1.018, 14.094 ms, 1.782°, 1.074 m,and 1.028 m, respectively, which are close to the originally introduced values.

Fig.9 Illumination topographic maps before and after calibration of simulated data

Fig.10 Histograms of depth diff erences between original data and data before and after calibration under four situations

To verify the accuracy of our method, the histograms of depth diff erences between the original data and the calibrated data are shown in Fig.10. After calibration, the distributions of depth diff erences are more concentrated near 0 under four situations. And indexes of the depth diff erences between the original data and data before and after calibration, such as the maximum absolute depth diff erence (maximum), the standard deviation (σμ), the mean depth diff erence (μ),and the percentage of soundings with depth error exceeding 0.2% of the water depth (γ) are calculated to evaluate the calibrated data. The results are shown in Table 1.

As shown in Table 1, when a single error exists, the maximum absolute depth diff erences caused by 1.02motion scale, 15-ms time delay, 2° yaw misalignment,and 1-mXandYlever arm errors are 0.073, 0.110,0.090, and 0.097 m, respectively, resulting in the depth error exceeding 0.2% of the water depth. After calibration, the maximum absolute depth diff erences are reduced to 0.010, 0.015, 0.011, and 0.008 m,respectively. Theσμandμof the depth diff erence are reduced from centimeter level to millimeter level.Therefore, the accuracy is controlled to under 0.2% of the water depth.

Table 1 Statistic of depth diff erences between original data and data before and after calibration under four situations

Fig.11 Illumination topographic maps before and after calibration with four errors introduced

Table 2 Statistic of the four error parameters after calibration

Table 3 Statistic of depth diff erences between original data and data before and after calibration in the case of four errors introduced

4.1.2 Multiple errors

In this situation, a 1.02 motion scale, 10-ms time delay, 2° yaw misalignment, and 0.5-mXandYlever arm errors are introduced to the original data at the same time. Then, the artif icial errors are calibrated simultaneously.

Due to the coupling of these errors and the limited amount of data in the selected area, the calculated parameters have a certain uncertainty. To verify the stability of our method, the mean value (μ) and the standard deviation (σμ) of each error parameter after 10 times of automatic calculations are counted as shown in Table 2. Theσμof each error parameter is small. Therefore, our method has good stability and takes into account the coupling between error parameters. As shown in Fig.11, under the inf luence of motion scale, time delay, yaw misalignment, andXandYlever arm errors, the swath has obvious wobbles(Fig.11a). After calibration with the proposed method,the introduced wobble errors are eliminated (Fig.11b).As shown in Table 2, the mean values of motion scale,time delay, yaw misalignment, and lever arm error parameters are approximately 1.018, 10.528 ms,1.815°, 0.520 m, and 0.477 m, respectively, which are close to the originally introduced values.

Fig.12 Histograms of depth diff erences between original data and data before and after calibration in the case of four errors introduced

Then, the histograms of depth diff erences between original data and data before and after calibration in the case of four errors introduced are shown in Fig.12.After calibration, the distributions of depth diff erences are more concentrated near 0. The indices of Maximum,σμ,μ, andγare shown in Table 3.

As shown in Table 3, when these four errors exist simultaneously, the maximum absolute depth diff erence caused by 1.02 motion scale, 10-ms time delay, 2° yaw misalignment, and 0.5-mXandYlever arm errors is 0.160 m, resulting in the depth error exceeding 0.2% of the water depth. After calibration,the maximum absolute depth diff erence is reduced to 0.015 m, and theσμandμof the depth diff erence are reduced from 0.025 m and 0.027 m to 0.002 m,respectively. The accuracy of the calibrated data is within 0.2% of the water depth.

Fig.13 Illumination topographic maps before and after calibration of measured data

Table 4 Statistic of error parameters of measured data

Fig.14 Histograms of standard deviations of the depth before and after calibration of measured data

4.2 Experiment of measured data

The measured data are provided by the Maps and Geospatial Products open platform of the National Oceanic and Atmospheric Administration (NOAA)and selected from the dataset collected in the Gulf of Mexico off shore, which are acquired by Simrad EM1002. The average depth of the survey region is about 179.8 m. These three swaths with wobble errors are selected to analyze the results before and after calibration. In the process of calibration, the motion scale, time delay, yaw misalignment, and lever arm errors are calculated at the same time because it is unknown whether each error exists.

The mean value (μ) and the standard deviation (σμ)of each error parameter after 10 times automatic calculations are counted as shown in Table 4. As shown in Fig.13, the swaths have obvious wobbles(Fig.13a). The wobble errors are eliminated after calibration with the proposed method (Fig.13b). As shown in Table 4, the mean error parameters of motion scale, time delay, yaw misalignment, and lever arm errors by our method are approximately 1.003,21.900 ms, 0.527°, 2.232 m, and 1.005 m, respectively.Therefore, it can be concluded that the wobbles in the swaths are mainly caused by theXandYlever arm errors and time delay. Theσμof each error parameter is small, which shows that our method has good stability.

Since the wobble errors always manifest themselves as across-track ribbing, the data before and after calibration are gridded to evaluate the calibrated data.TheX-axis of the grid is perpendicular to the track,and theY-axis is parallel to the track. The grid resolution is 28 m along the track and 7 m across the track, so the average number of soundings in each grid is about 8. The standard deviations of the depth in each grid are calculated before and after calibration,respectively. And the histograms of standard deviations of the depth are shown in Fig.14.

After calibration, the maximum standard deviation of the depth decreases from 0.729 m to 0.686 m,which is reduced by about 5.9%. The mean standard deviation of the depth decreases from 0.215 m to 0.191 m, which is reduced by about 11.2%. This further illustrates the eff ectiveness of our method.

5 DISCUSSION

5.1 Selection criterion for the f lat terrain

When choosing the f lat terrain area to calibrate the wobbles, the absolute curvature of the seabed larger than 0.2°/m corresponds to gradual slopes(Mohammadloo et al., 2020). Thus, for a surface patch with curvature smaller than this threshold, a linear function can be used, otherwise, a bi-quadratic f it should be employed. However, this selection criterion for the f lat terrain is complicated and timeconsuming. In practice, the f lat terrain is selected by visual inspection to construct the relatively f itted plane. The feasibility of the visual inspection method is also indicated by the signif icant calibrated results for both the simulated and the measured data in this paper.

5.2 Eff ect of the natural relief and roughness

The data with high roughness, such as sand waves,and bedrock, seriously decrease the accuracy of the inverted results (Maingot et al., 2019). To verify the eff ect of the natural relief and roughness, the data in the red rectangle (Fig.8) are selected to invert the corresponding parameters. Under this situation, the calibrated wobble error parameters of motion scale,time delay, yaw misalignment, and lever arm errors in the four cases are approximately 1.015, 15.168 ms,4.968°, 1.472 m, and 1.342 m, respectively. The calibrated results are far from the originally introduced values. Thus, when the proposed method is used to calibrate wobble errors, data with protruding objects and high roughness should be avoided.

5.3 Limitations of the proposed method

Although our method can eliminate wobbles caused by the f irst and the second types of errors, for errors introduced by the third type of error, i.e.,inaccurate surface sound speed correlated with vertical motion or rolling, the proposed method is not suitable. In addition, using a plane as the f itted surface also limits the general performance of our method.When facing a more complex seabed, the accuracy of our method will be aff ected. To improve the universality of our method the eff ectively f itted quadric surfaces (Hughes Clarke, 2003b; Maingot et al., 2019) can be used as constraints to improve our method.

6 CONCLUSION

This paper focuses on the subtle wobble errors caused by the imperfect integration of motion sensors and transducers in shallow water multibeam bathymetries. To address the wobble errors in multibeam bathymetries, a precise calibration model associated with motion scale, time delay, yaw misalignment, and lever arm errors is constructed.And the GA is used to obtain the optimal values of these error parameters with soundings in the f lat area.Experimental results showed that our method can eff ectively eliminate the wobbles caused by single or multiple wobble errors in shallow water areas, and the accuracy of the calibrated data is controlled within 0.2% of the water depth. The proposed method has signif icance in the precise calibration of dynamic errors in shallow water multibeam bathymetries.

7 DATA AVAILABILITY STATEMENT

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.