Marine compressed air source array primary acoustic field characterization from at-sea measurements

: The primary acoustic ﬁeld of a standard seismic survey source array is described based on a calibrated dataset collected in the Gulf of Mexico. Three vertical array moorings were deployed to measure the full dynamic range and bandwidth of the acoustic ﬁeld emitted by the compressed air source array. The designated source vessel followed a speciﬁed set of survey lines to provide a dataset with broad coverage of ranges and departure angles from the array. Acoustic metrics relevant to criteria associated with potential impacts on marine life are calculated from the recorded data. Sound pressure levels from direct arrivals exhibit large variability for a ﬁxed distance between source and receiver; this indicates that the distance cannot be reliably used as a single parameter to derive meaningful exposure levels for a moving source array. The far-ﬁeld acoustic metrics’ variations with distance along the true acoustic path for a narrow angular bin are accurately predicted using a simpliﬁed model of the surface-affected source waveform, which is a function of the direction. The presented acoustic metrics can be used for benchmarking existing source/propagation models for predicting acoustic ﬁelds of seismic source arrays and developing simpliﬁed data-supported models for environmental impact assessments. V C 2022 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license (http://creativecommons.org/ licenses/by/4.0/). https://doi.org/10.1121/10.0011678


I. INTRODUCTION
Seismic surveys are used to remotely explore the geological structures at and below the seafloor by collecting reflected acoustic signals and analyzing their properties.In a conventional survey, acoustic signals are generated by a source array towed by a vessel along a designated path.A conventional source array is a spatially extended broadband sound source, consisting of multiple single or cluster compressed air sources (frequently referred to as airguns) as array elements.(In this paper, such an array is referred to as a source array.)The generated sound energy propagates in complex and variable ocean environments (oceanic waveguide), thus, the analytical analysis and prediction of the acoustic field and source direction pattern in the far-field are challenging (Eliseyevnin, 1991;Tashmukhambetov et al., 2008).Typically, 20 or more source elements are arranged spatially into a rectangular array.Timed releases of the compressed air from individual sources lead to bubbles forming within the water.Individual bubbles expand first, and then pulsate, interfere, and decay.As the result, a short intense sound pulse, mostly carrying low frequency (<500 Hz) acoustic energy down toward the seafloor, that is effectively coupled and transmitted deep into the bottom layers is produced (Caldwell and Dragoset, 2000).The number of compressed air sources in an array, their locations and volumes, and firing pressures and times (synchronization) determine the pulse shape, duration, frequency content, and source array directivity pattern.Some acoustic energy propagates laterally into the water column and may cause environmental disturbances; therefore, the seismic surveys are subject to regulatory compliances in many countries.
In 2018, the U.S. National Marine Fisheries Service (NMFS) published updated guidelines identifying the acoustic thresholds at which individual marine mammals experience changes in their hearing sensitivity (either temporary or permanent) from exposure to underwater anthropogenic sound sources (National Marine Fisheries Service, 2018).Similar guidelines exist for fishes and sea turtles (Popper et al., 2014).The models describing an airgun source array (Laws et al., 1990;Ziolkowski, 1970;Ziolkowski et al., 1982) and predicting the sound pressure levels (SPLs; Abadi and Freneau, 2019;Ainslie, et al., 2016;Keen, et al., 2018;MacGillivray and Chapman, 2012;Tashmukhambetov et al., 2008;Prior et al., 2021) have been developed, particularly to assess potential environmental impacts of planned surveys.However, the model predictions do not always agree well with each other or the limited data available from the field measurements (Ainslie et al., 2019).It is theoretically and practically attractive to predict the temporal source waveform, based on the array design, and then use it to model broadband acoustic field in the far-field of the array at distances greater than D 2 =k, where D is the largest array dimension and k is the wavelength of interest.The source waveform, as defined by ISO (2017), is independent of the distance as one tries to predict the far-field of the array and a) Electronic mail: nas@louisiana.eduonly the directional dependence should be taken into account (Ziolkowski et al., 1982;Gisiner, 2016).Historically, the source models focused on simulating the low frequency bandwidth relevant to collected geophysical data because the low frequency signal components penetrate deep into the bottom and are measured on their return to the water column to provide the information about sub-bottom structures.One of the key drivers for the presented study was a widely recognized need for the far-field calibrated measurements of the broadband acoustic field and radiation pattern of a source array in deep water (to prevent contamination from bottom reflections) to benchmark the existing and newly developed models to simulate acoustic radiation of a source array into water for the broad range of directions and distances from the source.The acquisition of the far-field broadband controlled dataset for model benchmarking, validation, and accuracy quantification was the shared priority among researchers, industrial partners, and regulatory agencies.
Over the years, prior to 2007, industry conducted a number of source characterization studies.However, these datasets primarily characterized the vertically propagating (down-going) sound energy over the bandwidth of seismic interest (0-500 Hz).A broadband (up to 25 kHz) source characterization study was planned to provide information over a broad range of emission angles, azimuths, and ranges to help answer many of the questions and concerns related to the three-dimensional (3D) emission of high frequency sound energy from industrial source arrays, particularly into the water column.A limited source characterization study had been conducted in 2003 as part of the Minerals Management Service (MMS) Sperm Whale Seismic Study (SWSS) project.The study was executed by the Littoral Acoustic Demonstration Center (LADC) consortium members and funded by the Industry Research Funders Coalition (IRFC).The LADC Environmental Acoustic Recording System (EARS) buoys in a bottom-moored, vertical deployment configuration were used in the experiment.The study successfully characterized the close to vertical, broadband (5 Hz-25 kHz) output of a source array.The field work had also demonstrated that the existing EARS recording equipment provided an effective means of collecting such data and consistent and robust data processing results were generated (Tashmukhambetov et al., 2008).In 2007, the LADC was contracted by the International Association of Oil and Gas Producers (IOGP) Sound and Marine Life Joint Industry Programme to collect acoustic data from a standard source array, commonly used during either offshore oil and gas exploration surveys or academic research studies, in the Northern Gulf of Mexico (GOM) to characterize the watercolumn acoustic field (Newcomb et al., 2009).A broadband 3D source characterization study (across a wide range of propagation directions and distances) using the dedicated industry source array was conducted September 3-19, 2007.The far-field acoustic field metrics and source waveforms, derived from the field dataset, are presented in this paper.The terminology, formula symbols, and calculation algorithms are consistent with the international standard, ISO (2017).
The rest of the paper is organized as follows.Section II describes the study design, acoustic measuring system, and details of the system calibration and data acquisition.Section III presents the methods of calculating acoustic metrics from the dataset.In Sec.IV, the field metrics are discussed and modeled.Sections V and VI summarize the results, draw the conclusions, and discuss future research directions.

II. EXPERIMENT
The study was conducted in the northwestern GOM. Figure 1 shows the geographic location of the study site.The local water depth is approximately 1500 m.
Acoustic data were recorded by the four-channel EARS G2 (generation 2) buoys developed by the Naval Oceanographic Office for the LADC academic research (Sidorovskaia et al., 2005;Newcomb et al., 2009).The nominal bandwidth was 25 kHz per channel.Using four 120 GB hard drives for data storage in each buoy, a maximum recording period of 14 days was achieved.The system gains, filter characteristics, and pre-amplifier settings were modified prior to the field work in the laboratory to optimize the EARS performance for the given set of experimental requirements.The 16-bit analog-to-digital converter (ADC) yielded a 96-dB dynamic range.Four hydrophones were connected to each EARS buoy.Two hydrophones (desensitized and sensitive) were collocated at each desired depth to increase the effective dynamic range of the measurements and prevent data clipping at short ranges.In total, 40 hydrophones (10 EARS buoys) were deployed on 3 different moorings at 20 different depths.
The acoustic elements mooring design used during the experiment are given in Fig. 2. At the top is the floatation (Benthos balls) followed by a short (5 m) length of wire rope.The wire rope aids in deployment of the mooring and gives more isolation of the floatation from the array to reduce any noise (mechanical or flow) induced at the floatation.The EARS instrumentation package, frame, and acoustic array are directly below the wire rope.A short (10 m) section of the wire rope is below the array and above a secondary set of floatation.Additional sets of floatation support additional mooring segments, the acoustic releases, and aid in deployment and recovery.Below the acoustic releases, there is a short 1-cm-wide chain, 0.6 cm in diameter Kevlar mini-line, and another short length of 10-cm-wide chain.The length of the mini-line is determined by the experimental parameters to place hydrophones at the desired depths in the water column.At the bottom is the anchor whose weight is chosen with the amount of floatation for the desired tension in the mooring.
The three receiving arrays (moorings) were deployed in approximately 1500 m of water.In the following discussion, the moorings are referred to as A1 (east mooring, 26 57 0 22.68 00 N, 95 18 0 5.39 00 W), A2 (center mooring, 26 57 0 25.56 00 N, 95 18 0 35.64 00 W), and A3 (west mooring, 26 57 0 24.48 00 N, 95 19 0 17.4 00 W).The post-deployment measured horizontal distance between moorings A1 and A2 was 964 m; the horizontal distance between moorings A1 and A3 was 2004 m; the horizontal distance between moorings A2 and A3 was 1064 m.The horizontal distances between the moorings and hydrophone placement depths were determined based on the comprehensive experimental design to achieve the source array characterization experiment measurement objectives and assure redundancy of the data to minimize the risks and potential losses.
Sixteen hydrophones were on the east and west moorings and eight hydrophones were on the center mooring.The EARS mooring configuration for one of the two long moorings (east/west) is shown in Fig. 2. The center mooring was the same as the lower part of the mooring in Fig. 2 except that the hydrophone spacing was 7 m instead of 14 m.In addition, the east and west moorings contained a current meter and four ultra short base line (USBL) acoustic transponders and the center array contained an acoustic Doppler current profiler (ADCP) and two USBL transponders to monitor the mooring's profiles and provide accurate dynamically updated depths of the hydrophones during the entire study.
The average hydrophone depths and operational status (determined post-experiment) are detailed in Fig. 3

(top).
Sensitive and desensitized hydrophones were collocated at each nominal depth to record the acoustic field without clipping and to provide an extended dynamic range for the recording system.In processing, the desensitized channels were only used when a signal from a sensitive hydrophone was clipped.The data quality of each channel was assessed.The data from the hydrophones depicted with yellow triangles (as shown in Fig. 3, top) were not used in the processing presented here due to data recording issues at some stages during the deployment.The depth of each hydrophone pair was updated every 5 min as shown in Fig. 3 (bottom plots).The bottom left plot of Fig. 3 details the dynamic depth variations for the hydrophone at the nominal depth of 122 m on the east mooring.The bottom right plot of Fig. 3 shows the dynamic depth variations for the hydrophone at the nominal depth of 579 m on the west mooring.The diurnal tidal pattern typical for the GOM is the cause of the oscillatory pattern observed on the plots.The resolved position of each hydrophone at the closest to the array pulse emission time was used in processing.
The individual hydrophones were calibrated and had a flat omnidirectional frequency response between 10 Hz and 23 kHz.Table I summarizes all operational hydrophone sensitivities and average depths.The mean hydrophone sensitivity was about -191.8 dB re 1 V/lPa.The attenuation of about 60 dB was added to desensitize one hydrophone at each depth, and the net sensitivity after attenuation is shown in Table I.
The individual hydrophone sensitivity levels were used in processing.The EARS buoys were calibrated using the frequency domain method by injecting a set of predetermined harmonic signals as an input and measuring the system's output.Figure 4 (bottom) shows a sample of the frequency response curves from the east mooring.The desensitized channel curves include attenuation.The frequency response is nearly flat from 10 Hz to 23 kHz.The ADC was 16-bit with the full voltage scale (FVS) measuring range between -0.7 and 0.7 V (FVS ¼ 1.4 V).The ADC resolution factor is FVS/2 16 .Figure 4 (top) details the data calibration process to convert the recorded 16-bit integer values to the measurements of received pressure in micropascal (lPa).The first step (conversion to voltage) implies the multiplication of the integer data value by the ADC resolution factor.This step is not presented in Fig. 4

(top).
Two marine vessels (M/Vs) were designated for the study.The source array was towed by M/V FAIRFIELD ENDEAVOR.M/V CAPE HATTERAS was used to deploy and recover the recording moorings, collect oceanographic data [expendable bathythermograph (XBT), and conductivity, temperature, and pressure (CTD) profiles] and conduct active positioning transmissions.The 21-element 4140 in. 3 (67.84liter) source array configuration is shown in Fig. 5 (top).The array operating pressure was 2000 lbf/in. 2 (13.79 MPa).This is a typical source array used for deep water seismic exploration surveys or seismic research studies.The average towed depth of the array was 7 m.The dynamic depth of the array center, used in the calculations, is shown in Fig. 5 (bottom).
The M/V FAIRFIELD ENDEAVOR followed a pattern of parallel survey lines over the 12 km Â 12 km survey area with variable spacing between lines as shown with the different colors in Fig. 6 (the post-experiment grid based on the array positioning data is plotted).Four separate grids (could be visualized as array center locations in the horizontal plane at which the array was activated to produce an acoustic pulse) were acquired into the receivers to assure the near uniform coverage of the received emission angles.Each grid had variable ship track line spacing corresponding to 6 , 3 , 1 , and 0.5 changes in the emission angle relative to a fixed hydrophone at a selected design depth of 920 m as pulses were acquired in the fixed azimuthal plane from different lines (refer to the bottom diagram in Fig. 6).The angular reference terminology for the line spacing used in the design is illustrated from two perspectives: that of an observer fixed at the receiver location and an observer fixed at the source array center.The line lengths were reduced for the more tightly sampled grids as the grids had been optimized during the design stage to acquire the required data as  efficiently as possible, minimizing the source vessel effort.When the 6 and 3 areas are combined, they define a rectangle which is 11 km Â 9 km and which has 3 net spacing.The 3 grids are separate from the 6 grids because of the trapezoidal shape shown in Fig. 6.This shape was chosen to minimize redundant shots.Before any consideration is given to the grid geometry and hydrophone placement depths, one must realize that the variable sound speed profile (SSP) in the water column leads the acoustic rays to be refracted and not follow a straight-line trajectory.The main factor influencing the size of the largest experiment grid was the limited range of the direct refracted arrival.For the reference depth of 920 m during the design stage, this range was determined to be approximately 4.5 km.Therefore, it was decided to make the largest grid a 9 km Â 9 km square.After the decision was made to deploy three moorings (for data redundancy after the risk assessment study) in a line separated by 1 km along the acquisition line direction, the side parallel to the line direction was extended by 2 km to give a rectangle of size 11 km Â 9 km for the largest grid.
The array was activated to produce a sound pulse every 25 m, approximately every 12 s.In summary, the acquisition pattern was designed to provide uniform coverage and redundancy of emission and azimuthal angles at different distances between the source and receivers.The locations of the EARS moorings are also shown in Fig. 6.
The azimuthal angle, u, and emission angle, h, are used in Secs.III A-III C to characterize the direction pattern of the acoustic field produced by the source array.The azimuthal angle, u, describes the angle in the horizontal plane of the array between the sailing direction and direction toward the receiver.The vertical emission angle, h, describes the angle between the vertical line passing through the array center and the direction towards a receiver from the array center.0 corresponds to the center array being directly above a receiver.90 corresponds to the emission in the horizontal direction.Figure 7 shows the reference coordinate system, which is used to characterize the array directionality.
Given that the hydrophone arrays were to consist of eight pairs of hydrophones (one sensitive and one desensitized) deployed on a single vertical mooring (with the plan to deploy three moorings), the question to be answered, next, during the design stage was which depth placements of those hydrophones could achieve source characterization objectives, i.e., to fully characterize the 3D acoustic pressure field of the source array, p(r,h,u), where r is the distance between the array center and receiver.To understand the choice of the hydrophone placement depths, the geometry of the source ship tracks and their relation to the vertical moorings of the hydrophones must be known.The iterative optimization study was conducted at the design stage to lead to the hydrophone placement depths presented in Fig. 3.
The following requirements were to be satisfied: at least 50 pulses of direct refracted arrival are to be received in each angular bin (bin size of 3 emission angle Â 10 azimuthal angle) and these pulses have to pass the time separation criteria of at least 0.3 s from any bottom reflected or bottom refracted arrivals to exclude the interference effects in the field characterization.Figure 8 (bottom left) shows the results of the design study for the refraction boundary limitation.The emission angles for the direct arrivals (colors) versus receiver depth and source-receiver horizontal range are shown.The dark red area to the right represents points (depth of hydrophone and horizontal range to the source center) for which no direct rays would reach the hydrophone because of refraction.If the 0.3 s minimum time separation between the direct arrival and first bottom reflected arrival is imposed for the inclusion of pulses, and then a similar plot can be generated (Fig. 8, bottom right).In this plot, the refraction limitation and time separation limit are imposed on a plot of hydrophone depth versus horizontal distance.The deep blue area marks the depths and ranges for which no direct arrivals are counted either because none is received due to refraction or the received pulses do not satisfy the time separation requirement between direct and bottom reflected arrivals.The color in the bottom right plot in Fig. 8 shows the time separation between the direct and first bottom reflected arrivals for the received pulses.The largest horizontal range to contribute to the direct pulse arrival characterization, based on two limitations, was estimated to be about 4 km.

III. METHODS: ACOUSTIC METRICS OF THE SOURCE ARRAY PRIMARY FIELD A. Angular coverage and range estimation
After completing the field experiment, quality control (QC) was conducted for collected positioning and acoustic data.The usable recorded direct acoustic pulses with associated positioning data were identified.Based on the sourcereceiver geometry, all usable recorded pulses were binned into the 3 -wide bins for emission angles and 10 -wide bins for azimuthal angles.The angular coverage (the number of QC-passed pulses collected in each angular bin) is summarized in Fig. 8 (top right).The experiment design aimed to provide the uniform coverage of at least 50 shots in each angular bin with 40 hydrophones deployed at different depths on 3 moorings.However, based on the postexperiment QC, we choose only 19 hydrophones for processing with 12 of them existing in depths of less than 600 m.That caused the biased coverage observed in Fig. 8.
As a result of the SSP, the direct acoustic paths between source and receiver are refracted and do not follow straight lines.As a consequence, the emission angle differs from the angle at which acoustic energy arrives at a receiver, and the propagation path is longer than the straight-line distance between the array acoustic center and receiver.The average SSP during the study is shown in Fig. 9 (top).
The BELLHOP propagation model (Jensen et al., 2011;Porter and Bucker, 1987) was used to calculate the array emission angles and distance along the ray for direct refracted paths between the source array and receivers.No bottom properties information was collected during the study.Because only the direct refracted path arrival and shortly followed surface reflection (ghost) were analyzed, the local seabed properties were not critical for the propagation model to derive the source emission angle and distance along the true acoustic path.Figure 9 (bottom) illustrates how the range along the ray and array emission angles were calculated based on the BELLHOP ray trajectory output.

B. Temporal acoustic metrics
There are several common metrics which are used to characterize impulsive sounds and their impact on marine environments (Chapman and Ellis, 1998;Madsen, 2005;Keen et al., 2018;Southall et al., 2009;ISO, 2017): the peak compressional pressure, p pk;c , when p t ð Þ > 0; and The L p;pk;c and L p;pkÀpk are the attributes to characterize the rapid change in pressure field due to the direct arrival and its shortly followed ghost (surface reflected arrival), respectively.The peak compressional pressure is the maximum positive value of the received acoustic pressure time series, p(t), from an individual pulse produced by the array.The associated peak compressional SPL is obtained as where the reference pressure, p 0 ¼ 1 lPa.
The peak-to-peak SPL, which is determined by the difference between the most positive and most negative pressures of the received pulse is given by L p;pk;c and L p;pk-pk are calculated and presented in Sec.IV for each direct þ surface reflected arrival after the system calibration is applied, using an automated peak pressure detection algorithm.The automatic algorithm estimates an approximate straight-line distance and the earliest arrival time of the direct pulse by using the minimum of the measured SSP, the Global Positioning System (GPS) array location (when the source is activated), and the receiver positioning information.The 1-s data segment forward from the earliest arrival time estimation is searched to find the global maximum (the peak compressional sound pressure) and global minimum (the peak rarefactional sound pressure) and their arrival times (t max ; t min Þ; followed by calculating the L p;pk;c and L p;pk-pk . The peak compressional SPL and peak-to-peak SPL are instantaneous characteristics of the received pulse because they do not take the duration of a transient pulse into account.Therefore, the RMS sound pressure and, more recently, sound exposure function have been proposed as metrics to account for the mean of the squared sound pressure and the energy received within a given duration.The RMS SPL is given by (Madsen, 2005;Erbe, 2011;ISO, 2017) L p;rms ¼ 20 log 10 ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi where T ¼ Dt Ã N is the averaging time interval of a given transient, and t 1 is the beginning of the averaging time.To measure L p;rms and L E;p , one needs to select a time window, T.An example of the signal recorded at close range is shown in Fig. 10(a).The typical features include the fast positive (compressional) pressure rise associated with the compressed air release, the negative peak associated with the surface reflection, and residual air bubble pulsations [close-up in the inset of Fig. 10(a)].In our study, the RMS SPLs were calculated by using two different averaging windows.The first averaging window only includes the primary pulse and surface reflected arrival [shown in Fig. 10(a L p;rms for each extracted received pulse without taking into account bubble oscillations are outlined in Algorithm 1.
The algorithm for computing L E;p and L p;rms ; when taking the residual bubble oscillations into account is as follows.
It should be noted that the sound exposure is integrated only over 90% of the received pulse energy and not over 100% as one might expect.It leads to the SEL levels calculated by using Algorithms 1 and 2 to be 0.46 dB lower than the SEL levels calculated for the entire pulse duration.
It is practically and computationally desirable to estimate the source array sound at a receiver location quantified by L p;pk;c ; L p;pk-pk ; L p;rms ; and L E;p from the source waveform, which is expected to be direction-only dependent for the far-field of the array (Ziolkowski et al., 1982;ISO, 2017).Because the presented metrics are calculated for the pulses that are a combination of the direct arrival and its surface reflection, we are characterizing the surface-affected source waveform, s 0 t ð Þ; in this paper, which is often referred to as a source "far-field signature" in reflection seismology.Neglecting absorption (that is a reasonable assumption for the distances used in the analysis at low frequency), the relationship between the acoustic field metrics and surfaceaffected source level (SASL) and surface -affected energy source level in the specified direction, L 0 where R is a distance in the specified direction from the array center to a receiver.The reference source levels are 1 lPa 2 m 2 for L 0 S and 1 lPa 2 m 2 s for L 0 S 0 ;E : Once distances and received levels are calculated, the SASLs can be estimated by linear fitting the received levels to a function of the logarithm of the distance and extrapolating the determined function back to the array center as a convention (ISO, 2017).Such a simplified source model can be used to estimate the SPLs of direct þ surface reflected arrivals in the Fraunhofer zone of the array for a fixed propagation direction.The SASLs derived in such a way vary depending on the angular orientation of the receiver relative to the array center.It was highlighted in several publications that (1) Determine the time of the positive peak pressure value, t max , and the time of the first break (the first recorded pressure attributable to energy arrived from the source array, the onset of significant deviation from the baseline), t s , by searching back in time from t max to the first zero crossing [refer to Fig. 10(b)], define t start ¼ t s as the start of the direct pulse arrival; (2) determine the end of the primary þ ghost arrival, t end , by searching forward in time for the first zero crossing after reaching the global peak negative (rarefactional) pressure value at t min ; (3) calculate the cumulative sound exposure (sum of squared pressures) of the received signal between t start and t end [Fig. 10(b), bottom] such that (4) determine the start, t 1 , and duration of the averaging window, (5) (5) and calculate L E;p and L p;rms by summing squared pressure values over the determined window, T, per Eq. ( 3), the acoustic field levels in the near-field of the array cannot be extrapolated from this model (Fontana and Boukhanfra, 2018;Carey, 2009).
Because we use the angular bin statistics for the source characterization in this study, it is important to understand which primary factors are responsible for the variance of data in a particular bin and how accurately the SASL model predicts the acoustic metrics in a bin.For this purpose, we define the cost function to provide a sensitivity measure for the metrics value spread to the sizes of the azimuthal and emission bins.
The cost function, J, is defined as where F Si is the estimated source factor for bin i, R ni is a distance between the array center and a receiving hydrophone for a given shot ni, x ni is a pressure metrics measured from the experimental records (e.g., the peak compressional pressure of the arrived pulse), N is the total number of received pulses in bin i, M is the total number of bins, and Q is the total number of emissions that were binned.The cost functions are calculated for the bins corresponded to the straight-line geometry, presented in Fig. 7, and the curved ray geometry, explained in Fig. 9.

C. Frequency domain metrics
The concept of loudness is very important when characterizing physical and physiological disturbance and communication masking introduced by anthropogenic sounds (Finneran and Schlundt, 2011).Marine organisms have different hearing sensitivities in different frequency bands, therefore, the frequency content of the received sound should be taken into account.The frequency domain auditory weighting functions for various groups of marine mammals should be used to address the perceived loudness, which varies with frequency (Finneran, 2015;Hawkins et al., 2014a,b;Hawkins et al., 2014c;National Marine Fisheries Service, 2018).In the context of this paper, we present several frequency domain metrics to characterize the source array, associated with the frequency spectrum of the received signals: (1) the decidecade band received sound levels as functions of range and emitted angular bin, (2) the surface-affected energy source spectral density levels (SAESSDL) for a fixed angular bin, and (3) the directivity patterns of the source array for different frequencies.
The decidecade band received sound exposure levels (band level) are obtained as (Erbe, 2011;IEC, 2014) where E f i ð Þ is the mean value of the sound exposure spectral density, averaged across the ith decidecade band, and f ctr i ð Þ is the central frequency of band i.The sound exposure spectral density, E f i ð Þ (ISO, 2017), is calculated over the temporal window, which includes the residual bubble pulsations as described above in Algorithm 2.
The SAESSDL is obtained as where p t À t 0 þ r=c ð Þis a received far-field sound pressure from a particular array activation, G À1 ¼ r is the backpropagation operator to estimate the surface-affected source waveform, s 0 t ð Þ; r measures the distance in a specified direction along the refracted ray (unless mentioned otherwise).FT stands for the Fourier transform, and h; u are the emission and azimuthal angles calculated using ray tracing.In this paper, we only analyze the direct arrivals and associated surface reflected ghosts, thus, the backpropagation operator is approximated by a simple form, resulting in multiplication by the distance along the ray connecting the array center and a receiver.The SAESSDLs are calculated for each received seismic pulse and then grouped into angular bins to account for the directional properties of the SAESSDL.In Sec.IV, we present the statistics of the SAESSDLs in each angular bin.The preferred SAESSDL in each bin, obtained from the shortest r to minimize the propagation effects, is ALGORITHM 2. L E;p and L p;rms calculations for the window encompassing the direct arrival, surface reflection, and decaying bubble pulsations.
(1) Determine the time of the positive compressional peak pressure value, t max , and the first break, t s ; by searching back in time from t max to the first zero crossing; (2) define Dt ¼ t max À t s , and t start ¼ t max À 1:5Dt as the start point of the calculation; (3) calculate the absolute values of the derivatives (rate of changes) of the cumulative sound exposure curve, E p t ð Þ [defined by Eq. ( 4) and calculated between t start and t start þ 1 s such that and set the baseline threshold by calculating the average of the derivative curve between t start and t s [the derivative curve in dB re 1 lPa 2 , 10 log 10 p 2 ðtÞ=p 2 0 À Á ; is shown in Fig. 10(c)]; (4) calculate the derivatives of the cumulative sound exposure curve, i.e., p 2 ðtÞ, over extended temporal window (between t start and t start þ 1 sÞ and obtain the medians of the derivatives in the temporal intervals of jt start À t s j duration [median's curve (smoothed curve) and the baseline threshold in dB re 1 lPa 2 are shown in Fig. 10(d)]; (5) determine the terminal temporal point, t end , by searching for the last temporal point in the 1-s window at which the median value is greater than the baseline threshold; (6) calculate the cumulative sound pressure exposure for the pulse between t start and t end (similar to step 3 of Algorithm 1); (7) determine the analysis window, T, by taking the temporal interval over which the cumulative sound pressure exposure rises from 5% to 95% of the total sound pressure exposure, calculated in step (6) [similar to step (4) of Algorithm 1]; and (8) calculate L E;p and L p;rms by summing the squared pressure values over the determined window, T, found in step (7) per Eq.(3).specified.The source direction pattern is estimated by taking the median of estimated SAESSDLs in each bin for a fixed frequency and tracing its values across all of the bins to obtain angular dependences.
The measured source array vertical direction pattern is compared to the dipole direction pattern (to model the array as a point source and account for the surface reflection) and the direction pattern of the point-source element array of the same geometry near pressure-released flat boundary (ideal water-air interface).The dipole directivity is described by (Rossing and Fletcher, 2013) where k ¼ 2pf =c represents the wavenumber, h is the nominal array depth (7 m), and h is the emission angle in degrees measured from the vertical.
The specified frequency directivity of the source array þ its ghost is modeled via the superposition of spherical waves generated by the point-source elements (replacing compressed air sources) at a receiver location.The specified frequency direction pattern for the model array, D a h; u ð Þ, which includes the contribution from the surface reflection using the image theory, can be obtained as (Landau and Lifshitz, 1987) where k is the wavenumber corresponding to the frequency of interest, r i is the distance between the ith point source [at the location of a compressed air source (airgun)] and a hypothetical receiver, r is is the distance between the surface reflected image of the ith source in the array and a hypothetical receiver, A i represents the ith amplitude, N ¼ 21 is the total number of the array elements, "*" indicates the complex conjugate operation.Some elements of the source array included two collocated compressed air sources (referred to as the cluster elements in Fig. 5).Two collocated sources are modeled by doubling the pointsource amplitude, A i .The relative point-source amplitudes, A i s, are modeled by using the fact that they are proportional to the cubic root of the volume of an individual compressed air source, indicated in cubic inches in Fig.

IV. RESULTS
This section is divided into two subsections.The analysis of temporal metrics and frequency domain metrics are shown in Secs.IV A and IV B, respectively.

A. Temporal metrics calculations
Figure 11 shows the typical examples of the recorded temporal waveforms and Power Spectral Density (PSD) levels of two seismic pulses with the emission angles of 3.4 (near vertical) and 73.2 (near horizontal).The distances along the ray between the array center and receiving hydrophone are 310 and 2068 m with the azimuthal angles of 7.4 and 7.0 , respectively.The main takeaway points are changes in the temporal pulse structure as the distance and direction from the source changes, more obvious bubble pulsation profile for near horizontal propagation directions, pressure value range, and a typical spectral feature of the surface ghost notch.Here, PSD is calculated over the 1-s temporal window (as in many ambient noise studies) with the same time origin as for the temporal pulse function.It can be deduced that the main pulse (direct þ surface ghost) duration is around 0.01 s and the peak compressional pressure levels, L p;pk;c , are 207.4 and 175.3 dB re 1 lPa, respectively.The PSD levels show the frequency distribution of acoustic intensity in the received pulses.The maximum PSD levels are 161.8 and 131.7 dB re 1 lPa 2 /Hz.The surface reflected ghost pulse is responsible for the frequency specific interference pattern ("ghost notches") controlled by the array towing depth (Gisiner, 2016) and a receiver position.The first notch is indicated by the arrows in Fig. 11.
First, all recorded pulses (direct path and surface reflected ghost) from the source array moving along the grid (Fig. 6) and associated positioning data were processed to compute L p;pk;c , L p;pk-pk ; L p;rms , and L E;p as functions of the range along the ray between the array center and a receiver and emission and azimuthal angles.Figure 12 shows the four temporal metrics for all collected pulses on the eastern mooring as a function of the range only (distance along the ray) with colors indicating the nominal hydrophone depths.The hydrophone depths, dynamically estimated from the positioning data, were used in the calculations presented here.The hydrophone positions were dynamically updated every 5 min.Because all of the pulses are used regardless of their emission angles relative to a receiver, the dense (perceived as continuous) range coverage is seen in Fig. 12.
The critical observation from Fig. 12 is that the received levels exhibit a large variability (up to 40 dB) for the fixed range between the source and receiver.Thus, the distance cannot be used as a single parameter when assessing the sound levels for a moving array source.The variability is due to the combined effect of the source array directionality (as the orientation of the source array relative to the hydrophone changes over time for a moving source), reflecting sea-surface interface forming the phase inverted ghost, and waveguide propagation (Eliseyevnin, 1991;Sidorovskaia, 2004).The variable decay rates, dependent on the range and a receiver depth, are seen in Fig. 12.The decay rate laws for the transient signal metrics in waveguides are complex and depend on many factors, including water depth, signal frequency content, sub-bottom structure, temporal analysis window, shadow zone formation, and other factors, in addition to the transition between spherical (at close ranges) and cylindrical (at ranges much larger than the local water depth) geometrical spreading laws.The paper focuses on the analysis of the metrics of the surface-affected direct arrivals only.The ranges up to 4 km were analyzed further to allow the building of a simple prediction model to characterize the measured sound levels.The spherical spreading propagation losses will dominate at these ranges.The spherical spreading model, used here, is not applicable in shallow-water environments, where it will result in a significant underestimation of received levels even for close ranges and short temporal windows (Ainslie et al., 2014).
To remove the variability due to array directionality, the calculated metrics were binned into the 3 -wide bins in emission angles by the 10 -wide bins in azimuthal angles.Due to the limitations imposed by the refraction of the direct arrival, the propagation ranges for the metrics calculations were also limited.Figure 13 shows the four temporal metrics calculated for the selected bin of 72 -75 for the emission angles and 0 -10 for the azimuthal angles.It includes data collected on all of the operational hydrophones.Figure 13 (top left) shows the binned data for the curved ray geometry and associated 1/R fitting curves.Figure 13 (top right) shows the data in the bin based on the straight-line geometry.The cost functions for different sizes of the azimuthal and emission bins are shown at the bottom panel of Fig. 13 for the sensitivity analysis to address the remaining data spread.The yellow histograms represent the straight-line propagation geometry and the blue histograms represent the curved ray path geometry.The observed data segmentation in the range is due to binning.The variability of the metrics at a fixed range is considerably decreased (below 5 dB) and the 1/R (spherical spreading) fit curves, shown as solid lines, indicate that the binned metrics attenuation with the range can be approximated by the spherical spreading law.The accuracy of approximation is increased if the distance is calculated along the ray in a specified direction (particularly for longer ranges).The data spreading inside the bin are mostly due to the size of the emission angle bin.
The measured SPLs and SELs for two bins (near vertical and near horizontal propagation) are shown in Fig. 14.The SPL rms and SELs are calculated for two temporal windows (primary pulse þ surface reflected ghost and the extended window that encompasses the residual bubble pulsations).The histograms showing the distribution of window sizes for each bin are included as insets on the graphs.Figure 14 illustrates that the SPLs and SELs of direct arrivals and associated surface reflections decay with increasing range along the ray following the spherical spreading loss approximation.The high values of the Pearson correlation coefficient (PCC) indicate the existence of a linear relationship between SPLs and 20 log R fitting curves (dotted lines) that supports the validity of the hypothesis that in deep water, the spherical spreading loss is the dominant propagation loss in a specified direction for the surface-affected direct pulse energy in the far-field of the source array for the ranges comparable to the water depth.The bin SASLs or surface-affected energy source levels (SAESLs) for each metrics are also listed in Fig. 14.
As the distance between the source array and receiver increases and the acoustic energy propagates away from the vertical, the residual bubble signal becomes a more prominent feature of the signal and leads to the elevated levels of the array sound against the baseline for a longer time.It manifests itself in the overlap amount for the distributions of window sizes, which are shown in the insets and calculated in accordance with Algorithms 1 and 2 described in Sec.III.The blue distribution (referred to in the insets as "with bubbles") for the windows associated with the return time to baseline noise level after the primary arrival (Algorithm 2) has no significant overlap with the distribution associated with the arrivals of the compressional pressure rise and surface reflected ghost (pink distribution) for near horizontal propagation [Fig.14(b) inset].The two distributions fully overlap for near vertical propagation [Fig.14(a) inset].However, if one compares the RMS SPLs and SELs calculated over the temporal windows that either include or exclude the residual bubble pulsations, the difference is insignificant.This indicates that the main bulk of the energy (90%) of the surface-affected direct arrival is concentrated in the time interval containing the initial pressure rise and its surface reflection.The insignificant impact of the bubble pulsations also indicates that the firing sequence of the source array was well-designed to suppress the collective bubble pulsations.
Modeling of the sound field of the broadband source array at an arbitrary location in an oceanic waveguide is challenging and computationally intensive.It requires reliable broadband models of the source functions for each array element, such as Gundalf (Hatton, 2008;Laws et al., 1990), Nucleus (Goertz et al., 2013), or AAMS (MacGillivray and Chapman, 2012); models describing waveguide transfer functions between each element of the source array and a receiving location (Jensen et al., 2011); and knowledge of the environmental parameters (sub-bottom structure, SSP, bathymetry, etc.) (Tashmukhambetov et al., 2008;Ziolkowski et al., 1982).However, the field measurements described here support the feasibility of deriving all surface-affected far-field source metrics in fixed directions from the single surface-affected source waveform (Caldwell and Dragoset, 2000;Gisiner, 2016;Laws, 2012).The detailed surface-affected source waveform and source spectrum are desirable if one plans to model far-field waveforms and associated metrics at an arbitrary receiver point and possibly in different environments.For this purpose, the sSAESSDLs are calculated and investigated.The 1-s temporal window was used to obtain SAESSDL to encompass the entire source waveform with the same time origin as for a temporal source signature.Figure 15 shows the estimated surface-affected source waveforms and SAESSDLs for the two angular bins [refer to Eq. ( 9)].The surface-affected source waveform obtained from the closest distance along the ray for a given bin is referred to as the preferred in-bin source waveform (dark red color).The preferred in-bin source waveform is the most accurate one in the framework of this model due to the fact that it is reconstructed from the signal, which is the least impacted by the attenuation and other propagation effects in addition to the spherical spreading.The use of received pulses at greater distances to reconstruct the in-bin source waveform will tend to underestimate the associated source level values.The median values of source spectra at each frequency in the bin (middle green curve) are further used to investigate the source directionality presented in Fig. 19.
In addition to the temporal and spectral properties of the source in-bin far-field signature (illustrated by Fig. 15), the angular dependence of the source array radiated energy also matters.The angular dependence of surface-affected peak compressional source level and SAESL, L 0 s 0 h; u ð Þ and L 0 s 0 ;E h; u ð Þ, respectively, are presented in Fig. 16 [refer to Eq. ( 6)].The source levels in specified directions (bins) were estimated from the parameters of the fitted straight lines (similar to the bin examples in Fig. 14).The SAESL angular-dependence surface was calculated using the temporal window which encompasses bubble pulsations (Algorithm 2).L 0 s 0 h; u ð Þ and L 0 s 0 ;E h; u ð Þ are the parameters of the model, inferred from measurements (as in this study) or existing source models.They can be used to estimate farfield acoustic metrics using Eq. ( 6).One is required to use different source levels for different propagation directions.
From the examples in Fig. 14 for the low emission angle bin of 3 -6 (near vertical propagation) and the azimuthal angle bin of 0 -10 , the estimated surface-affected peak compressional source level is 256.5 dB re 1 lPa 2 m 2 , which is about 4.0 dB lower than the source level to model peak-to-peak SPL, estimated as 260.5 dB re 1 lPa 2 m 2 .For the high emission angle bin of 72 -75 (near horizontal propagation), the estimated surface-affected peak compressional source level is 246.1 dB re 1 lPa 2 m 2 , which is about 10.4 dB lower than the peak compressional source level estimated for near vertical propagation.
As the results based on the data in Fig. 16 show, the estimated SASLs vary depending on the propagation direction (angular bin).Table II summarizes the estimated (across all of the emission bins) average SASL and source levels in the 5th and 95th percentiles for each acoustic metric.The temporal window which encompasses the bubble pulsations is used for the RMS and SEL metrics calculations.
The analytical relationships among the SASLs for the four acoustic metrics measured in a specified direction could be estimated based on the following assumptions.If the pulse shape of the initial pressure rise and its surface reflection are perfectly anti-symmetrical (ideal flat ocean surface), the expected difference between SASL pk-pk and SASL pk,c is 6 dB.The SASL rms will be about 6 dB lower than SASL pk,c if the averaging window only includes the direct arrival and its anti-symmetrical surface reflection.The difference between the SASL rms and SAESL will depend on the temporal integration window (its statistics in the bin).For example, for the constant 0.015-s integration window, the SAESL is 18 dB below the SASL rms as can be seen from Eq. (3).Due to variability of the environment and binning, these relationships are only approximately satisfied.The average (across all of the bins) difference between SASL pk-pk and SASL pk,c is about 5.52 dB with 5th and 95th percentiles of 4.08 and 6.98 dB, respectively; between SASL pk,c and SASL RMS , it is about 4.46 dB with 5th and 95th percentiles of 2.13 and 6.90 dB, respectively; and between SASL RMS and SAESL, it is about 20.67 dB with 5th and 95th percentiles of 19.41 and 22.82 dB, respectively.

B. Frequency domain metrics calculations
Figure 17 shows the decidecade band received sound exposure levels [Eq.( 8)] for two angular bins of 3 -6 and 72 -75 in emission angles (and the same 0 -10 bin in The fit supports the hypothesis that the spherical spreading loss can account for the propagation loss in a specified direction (bin) for the surface-affected direct arrival in the narrow frequency band.The spherical spreading loss fit will worsen and will not be applicable at high frequencies and ranges much greater than the local water depth.
Figure 17(e) shows the comparison among the band levels extracted from Figs. 17(a) and 17(b) for the closest and furthest arrivals in the bins, respectively, and the median of the local ambient noise band levels, calculated during the times when the array was not in operation.The desensitized channel noise floor is also shown.The high frequency plateau at closer ranges, shown in Fig. 17(b), for near horizontal propagation seems to correspond to the high frequency output of the array and is not due to internal system noise.This hypothesis is also corroborated by the directionality modeling presented next.
All of the processed data were binned and fitted and the SAESLs for the specified bands were estimated.The examples, based on the data in Fig. 17, show that the SAESL for the decidecade band with central frequency of 50.12 Hz for the near vertical emission bin (3 -6 in emission angle and 0 -10 in azimuthal angle) is 221.9 dB (re 1 lPa 2 m 2 s); the SAESL for the decidecade band with central frequency of 50.12 Hz for the near horizontal emission bin (72 -75 in emission angle and 0 -10 in azimuthal angle) is 210.3 dB (re 1 lPa 2 m 2 s).The SAESLs were estimated for all 1080 bins and 2 selected bands (with the central frequencies of 50.12 and 251.19 Hz).
The source directivity was further investigated by tracing the SAESLs across all of the bins for the band of interest.Figure 18 15).The comparison between the direction pattern reconstructed from the measured data and dipole and array models is presented in Fig. 19 for two selected frequencies.
Figure 19 suggests that the low frequency directivity of the studied source array could be well described by the surface-affected model array directivity and the high frequency components of the array signal propagate more   rigorous array model is needed to predict the measured directivity at the intermediate emission angles with higher accuracy.The mismatch may be due to the facts that the surface reflected signals may not exhibit full symmetry in reference to the direct pulse and the sea-surface interaction with the high frequency pulse components requires more complex model than perfect reflection from the pressure release surface.The small displacements of the towed array elements, connected by a flexible long cable, and finite size of each array element may also play a role.

V. DISCUSSION
This paper discusses the acoustic metrics derived from the unique field dataset collected in the Northern GOM to fully characterize the primary 3D acoustic field of a standard compressed air source array, commonly used during offshore seismic exploration and research surveys in deep water.The acoustic metrics of SPLs, SELs, received pulse spectra, and decidecade band levels are used to quantify the acoustic contributions of the source array surface-affected direct arrivals into deep water oceanic soundscapes.The results presented in Fig. 12 show that the SPLs of direct surface-affected arrivals from a moving source array exhibit large variability (sometimes exceeding 40 dB) for the specified distance between the source array center and receiver due to the waveguide propagation effects and array direction pattern (as orientation of the source array changes over time as the source moves along the designed grid lines relative to the fixed hydrophone locations).Therefore, the distance between a moving array center and a static receiver cannot be used as a single parameter to predict received sound levels.However, the further analysis of the acoustic metrics in narrow angular bins (Figs. 13 and 14) supports the validity of the simplified approach for predicting the source array far-field.The surface-affected direct refracted arrival in narrow angular bins in deep water can be predicted from the direction-dependent surface-affected point-source waveform.The validity of the approach is limited to the ranges that do not greatly exceed local water depths [<O(10) water depths from the array].The binned data suggest that the SPLs of surface-affected direct arrivals in a specified direction decay with distance, following the spherical spreading law.The accuracy in predicting the acoustic level's decay with distance in a narrow bin increases if the distance between the source array center and a receiver is calculated along the true curved acoustic path using a ray propagation model (Fig. 13).The sensitivity analysis based on the variation of the angular bin size indicates that the accuracy of the model fit strongly depends on the size of the bin for the emission angle and is much less sensitive to the size of the azimuthal angle bin (Fig. 13).The size of the emission angle bin (3 ) is primarily responsible for the spread in the received acoustic levels at the fixed distance between the source and a receiver in the specified angular bin (Fig. 14).
The study of the measured and modeled array directivity patterns (Figs. 15,18,and 19) suggests that the beampattern for high frequencies has two peaks in the plane perpendicular to the sailing direction, indicating that high frequencies will propagate in this plane more efficiently in the directions close to vertical and horizontal.Figure 17(e) demonstrates that the source array effectively generated the broadband acoustic signal, which brought a rise to the ambient noise band levels across the entire measured frequency band (up to 80 dB in the selected bands, ranges, and directions) during the arrival of the direct surface-affected pulse.The smaller gap between the dashed (near horizontal propagation) and solid lines (near vertical propagation) for the frequencies above 1 kHz indicates that the high frequencies propagate more effectively in the lateral directions [Fig.17(e)].The comparison of the quoted percentiles in Fig. 16 at 1 and 10 kHz also supports the hypothesis that the high frequency source components (up to 23 kHz) are critical for modeling the primary acoustic field of the source array, particularly for the near horizontal propagation directions.It is important to note that this study only addresses the analysis of direct and ghost (first surface reflected) arrival metrics, thus, the directivity is driven by the array design and presence of the sea-surface interface.The largest range along the ray in analyzed arrivals is limited to 4.5 km.Up to this range, the direct arrival is present at the depths corresponding to the deployed hydrophones, does not interfere with bottom reflection, and the impact of the waveguide propagation on the directivity is easier to take into account via the refraction of the direct and surface reflected arrivals.Surface-affected direct arrival data are the most suitable for benchmarking the algorithms for modeling source broadband far-field signatures, which can then be used to predict received acoustic levels in the far-field of the array, including the long range propagation effects.
In 2018, the NMFS of National Oceanic and Atmospheric Administration (NOAA) published a revised technical guidance for assessing the effects of anthropogenic sounds on marine mammal hearing (National Marine Fisheries Service, 2018).Two metrics were recommended for impulsive sounds in calculating the permanent threshold shift (PTS; permanent damage to marine mammal hearing system) onset thresholds: zero-to-peak unweighted SPL and weighted cumulative (over 24 h) sound exposure level.
The updated auditory weighting functions are compiled in the report for the five hearing groups of marine mammals.The dual metric thresholds are different and defined for five hearing groups.Our data could be viewed as informative in the framework of such an assessment but should be interpreted with caution.
Mid-frequency cetaceans (dolphins, sperm whales, and beaked whales) are broadly present in the Northern GOM.The NMFS recommendation for the PTS onset threshold in received level for the zero-to-peak unweighted (flat) SPL for this hearing class is 230 dB.Referring to Table II and selecting the 95% percentile for the SASL for the peak compressional pressure level, L 0 S 0 ;pk;c , of 256.75 dB (worst case scenario), the distance from the seismic array would be 21.75 m to reach 230 dB level in the spherical spreading loss approximation.However, the largest array size is about 20 m, therefore, the far-field approximation is not valid at the 21.75 m range, where the near-field structure is more complex and has not been measured in the experiment described in the paper.The far-field spherical propagation loss law cannot be employed until a receiver is at ranges much larger than the source size (at least five times larger and at least 100 m away from the source).In the context of the presented data, we hypothesize that the mid-frequency cetaceans at distances greater than 100 m away from the studied source array will not experience the PTS onset threshold in the received level for the zero-to-peak unweighted (flat) SPL.However, the zero-to-peak SPL metrics cannot be used alone without the assessment of the 24-h cumulative SEL to assure the environmental safety protocol.No direct assessment of the NMFS risk-of-injury range can be conducted using this dataset if the threshold risk levels are investigated in the near-field of the source array (closer than 100 m to the array).

VI. CONCLUSIONS
We have presented the far-field metrics of the primary surface-affected acoustic field of the industrial compressed air (airgun) source array in the water column.The metrics were derived from the fully controlled source array characterization dataset.Each array generated pulse was recorded by the calibrated acoustic system, and the source and receiving acoustic moorings positioning were monitored during the entire study.As far as we know, it is the first study of the primary field acoustic metrics for a compressed air source array derived from the controlled at-sea measurements.The availability of such metrics to the broad community of scientists and regulators is intended to assist in developing more accurate modeling tools for the environmental impact assessment.The calibrated recordings of the far-field output from the standard source array in the broad range of source-receiver distances and directions and the calculation of the broadband acoustic metrics, presented in the paper, realistically describe the sound field structure and sound levels that were observed in the deep water seismic survey.The data are intended to be widely used for benchmarking the existing and newly developed source/ propagation models for predicting acoustic field levels in the water column for seismic exploration surveys in different regions and developing simplified data-supported models for the environmental impact assessment.The study discusses and reinforces (based on the comprehensive field data) the basic scientific principles to be used in a risk-based approach to seismic operations.The dataset and analysis also contribute to advancing knowledge related to the broadband output of the industrial source arrays and understanding the high frequency sound energy levels and propagation at near horizontal angles associated with such source arrays.The collected data and calculated metrics will give policymakers (e.g., NMFS) access to the field measured metrics to aid in decision-making and development of the future environmental regulations.
FIG. 1. (Color online) The data acquisition area.The local average water depth is 1500 m.
FIG. 3. (Color online) (Top) The average hydrophone depths and operational status (determined post-experiment), where the sensitive hydrophones are shown as blue triangles and the desensitized hydrophones are shown as red triangles.(Bottom) Examples of the dynamic hydrophone depths are depicted.

FIG. 4
FIG. 4. (Color online) The EARS buoys frequency response curves (East mooring) and data calibration workflow, showing The (top) data calibration workflow, (bottom-left) low frequency response, and (bottom-right) high frequency response.

FIG. 5
FIG. 5. (Color online) The source array configuration and array center dynamic depth data.The number inside each array element is the individual volume of compressed air source in cubic inches.

FIG. 6
FIG. 6. (Color online) (Top) The combined survey grids and (bottom) an illustration of angular nomenclature for the survey grid are depicted.

FIG. 8
FIG. 8. (Color online) The (Top left) coordinate system used for the color display (view on the array from below) and (top right) angular coverage statistics (number of pulses collected into the bin) for all pulses passing the quality control (QC) are shown.The angular bin size is 3 in emission angle and 10 in azimuthal angle.The color corresponds to the number of pulses received in each bin.(Bottom) The refraction (left) and time limitation (right) boundaries versus hydrophone depth and horizontal source-receiver range are depicted.(Reproduced with permission from the design report submitted to IOGP.) FIG. 9. (Color online) The (Top) mean SSP measured during the study and (Bottom) schematics explaining the calculation of the range along the ray and array far-field emission angle are shown.
FIG. 10. (Color online) An illustration of the SEL and RMS sound pressure level calculation algorithm.(a) The waveform of a source array pulse on a desensitized hydrophone; the time interval, which only includes direct arrival and surface ghost, is also shown between two dashed lines.(b) The waveform of the surface-affected direct arrival, cumulative sound exposure curve, and temporal window calculation is illustrated.The (c) derivatives of the cumulative exposure curve over 1 s in dB scale and (d) smoothed curve (median over short sliding window) of the derivatives of the cumulative exposure curve over 1 s in dB scale are shown.The time interval, which includes direct arrival, surface ghost, and residual bubble pulsations, is shown between two vertical (red) dashed lines.The width of the small rectangle is the window with 5%-95% of the cumulative exposure.

FIG. 11 .
FIG. 11.The representative temporal and frequency signatures of received pulses propagated near the vertical and near horizontal directions.(Top two) The temporal waveform and PSD (over 1-s window) levels of the signal received at the distance of 310 m from the array center at the emission angle of 3.4 and the azimuthal angle of 7.4 , as well as (bottom two) the temporal waveform and PSD (over 1-s window) levels of the signal received at the distance of 2.68 km from the array center at the emission angle of 73.2 and the azimuthal angle of 7.0 are shown.
FIG. 12. (Color online) The acoustic metrics of the received pulses versus range along the ray recorded on the eastern mooring.(Top left) L p;pk-pk , (bottom left) L p;pk;c , (top right) L p;rms , and (bottom right) L E;p , where red, blue, black, and magenta represent the nominal hydrophone depths of 122 m, 321 m, 532 m, and 934 m, respectively.The temporal windows for the L E;p and L p;rms calculations encompass the bubble pulsations (Algorithm 2).
FIG. 13. (Color online) The binned received L p;pk-pk (red), L p;pk;c (blue), L p;rms (black and green), and L E;p (magenta and yellow) versus range for the angular bin of 72 -75 in emission angle and 0 -10 azimuthal angle for the curved ray geometry (top left) and the straight-line geometry (top right).The solid curves are the spherical spreading fitting curves.(Bottom panel) The cost function as a function of the emission angle bin size (bottom left) and azimuthal angle bin size (bottom right) are shown.Blue respresents the curved ray geometry and yellow depicts the straight-line geometry.
FIG. 14. (Color online) The received SPLs and SELs versus range (distance measured along the ray) for the angular bin of 3 -6 in emission angle and 0 -10 in azimuthal angle (top) and the angular bin of 72 -75 in emission angle and 0 -10 azimuthal angle (bottom; the horizontal axis is in a logarithmic scale).The distributions of temporal window sizes for calculating L p;rms and L E;p are shown in the insets.The dotted lines represent the linear fit with the Pearson correlation coefficient (PCC) listed for each metrics.The bin SASLs (or SAESLs) for each of the metrics are also listed.
FIG. 15. (Color online)The surface-affected source waveforms and SAESSDLs for all of the binned data for the angular bin of 3 -6 in emission angle and 0 -10 in azimuthal angle (top two plots) and the angular bin of 72 -75 in emission angle and 0 -10 in azimuthal angle (bottom two plots).Each pink-colored waveform represents an individual surfaceaffected source waveform.The preferred in-bin source waveform for the shortest propagation distance along the ray is shown in dark red.The green curves are the SAESSDLs for th e16th, 50th, and 84th centiles.The preferred in-bin spectral signature is shown in dark red.
summarizes the estimated vertical source directivities for the central frequencies of 50.12 and 251.19 Hz in the four azimuthal angle bins of 0 -10 (in front of the array), 90 -100 (the starboard side of the array), 180 -190 (behind the array), and 270 -280 (the port side of the array).The zero-dB level corresponds to the global maximum across all of the angular bins in the SAESLs at the specified central frequency.As is shown in Fig. 18, the low frequency component of the energy emitted by the source array is primarily directed toward the ocean bottom as desired for seismic exploration projects.The high frequency component effectively propagates in the near vertical and near horizontal directions as indicated by the multilobe structure of the directivity in the 251.19-Hz central frequency band.The measured directivity was determined by tracing the bin-median values of the SAESSDL at 50 and 250 Hz across all 1080 bins (the examples for 2 bins are shown in Fig. FIG. 16. (Color online) The (left) estimated SAESL as a function of the emission and azimuthal angles and (right) estimated surface-affected peak compressional source level as a function of the emission and azimuthal angles are shown.The different colored lines indicate contours with equal estimated SASLs.The reference levels are 1 lPa 2 m 2 for SASL and 1 lPa 2 m 2 s for SAESL.The temporal windows for the SAESL calculations encompass bubble pulsations (Algorithm 2).

FIG. 17
FIG. 17. (Color online) The waterfall displays of the decidecade band sound exposure levels as a function of the decidecade band central frequency and distance along the ray are shown for the (a) angular bin of 3 -6 in emission angle and 0 -10 in azimuthal angle and (b) angular bin of 72 -75 in emission angle and 0 -10 in azimuthal angle.The extracted decidecade band levels versus distance along the ray for the central frequency of 50.12 Hz are shown for the (c) angular bin of 3 -6 in emission angle and 0 -10 in azimuthal angle and (d) the angular bin of 72 -75 in emission angle and 0 -10 in azimuthal angle.The blue lines represent the least squares regression to À20 log 10(R)-curve.The upper (95%) and lower (5%) regression bounds are shown as dashed yellow lines, and (e) the comparison of the band levels with the regional ambient noise baseline are depicted.The internal system noise floor for the desensitized channels is shown as a black curve.The temporal window for spectral analysis encompasses bubble oscillations (Algorithm 2), and the displayed range of decidecade frequency bands is between 10 Hz and 19.9526 kHz, corresponding to the flat region of the system calibration curve.
FIG. 19.(Color online) A study of the surface-affected source array directivity at frequencies of 50 Hz (top) and 250 Hz (bottom).The red curve represents the theoretical dipole directivity.The thin-line curves represent the modeled directivities of the point-source array and its image due to the surface reflection in specified planes.The dashed thin lines correspond to the directivities in the plane perpendicular to the sail direction, and the solid thin lines represent the directivities in the plane passing through the array center and parallel to the sail direction.

TABLE I .
The summary of the operational hydrophone sensitivities and average depths.The abbreviations for sensitive channels and desensitized channels are "S" and "D," respectively.
by taking the interval in which the cumulative sound pressure exposure rises from 5% to 95% of the total sound pressure exposure, calculated in step (3) [refer to Fig 10(b)] such that

TABLE II .
The estimated average SASLs (SAESLs) and 5th and 95th percentiles for the peak-to-peak, peak compressional, and RMS SPLs and SEL metrics for all 1080 bins.The temporal windows for calculation encompass bubble pulsations (Algorithm 2).
effectively in the directions close to horizontal.For the higher frequency, the array and model array direction patterns are in a good agreement for the mainlobe (near vertical propagation) and the propagation directions close to horizontal.However, the measured directivity does not have the deep notches predicted by the model array.The individual compressed air sources in the array are approximated as noninteracting point sources in Eq. (11), therefore, the more