Introduction

Geotechnical seabed surface properties impact different nearshore and offshore sediment processes that influence engineering and naval applications, such as erodibility, local geomorphodynamics, and scour at submerged infrastructure (Kirchner et al. 1990; Rucker 2006; Grabowski et al. 2011; Stark and Kopf 2011; Stark et al. 2014). Typically, in-situ testing and/or physical laboratory testing of seabed sediment samples is used to determine the geotechnical properties of the seabed. However, this can be challenging particularly in energetic environments and is typically limited to point measurements. Acoustic surveying methods can provide insight on the seabed topography, stratigraphy, and appearance. Additionally, seabed properties such as roughness, shear strength, and grain size have been estimated from acoustic seabed surveying (Hamilton 1980; Pratson and Edwards 1996).

Side scan sonar (SSS) is a common method for acoustic seabed surveying, offering imaging of large areas of the seafloor. SSS backscatter data have been used to map surficial seabed sediment properties relevant to sediment dynamics, including sediment grain size and seabed roughness (Irish et al. 1999; Buscombe et al. 2016; Chandrashekar et al. 2021). High-frequency SSS (in this study: 1000 kHz) has a small-to-negligible seabed penetration depth, which led to only limited previous explorations of correlations to (often depth-dependent) geotechnical seabed properties. However, the consideration of geotechnical seabed surface properties for the investigation of sediment dynamics in subaquatic environments as well as the high availability of SSS data motivates exploring the correlation between SSS measurements and geotechnical seabed surface properties for sediment dynamics-related applications (Briaud et al. 2001; Stark et al. 2011; Rahimnejad and Ooi 2016; Bilici et al. 2018; Albatal et al. 2019). The critical shear stress for the initiation of motion, τc, for clean, coarse-grained sediments (percentage fines less than 5%) has been related to the median grain size, d50 (Briaud et al. 2001). However, for fine-grained sediments, the parameters that influence τc include, but are not limited to, sediment plasticity index (PI), percentage of fines, the water content of the sediment (w), and the undrained shear strength (Su) (Briaud et al. 2001; Grabowski et al. 2011; Rahimnejad and Ooi 2016). Furthermore, benthic biogenic processes are known to alter seabed surface sediments, leading to possible strengthening or weakening of seabed sediments and impacts on the seabed roughness and appearance in SSS images (Lee et al. 2019; Dorgan et al. 2020; Consolvo et al. 2022; Martinez et al. 2022; Cox et al. 2023). Therefore, SSS may have the potential to also map and characterize changes to the seabed surface from benthic biogenic processes.

In-situ seabed properties such as surface roughness, sediment bulk density, and porosity affect acoustic backscatter data (Lyons and Orsi 1998; Williams and Jackson 1998; Williams 2001). As outlined in Williams and Jackson’s (Williams and Jackson 1998) bistatic bottom scattering model, when the acoustic signal hits the interface between the water and the sediment, the strength of the returned signal can either be scattered due to roughness, scattered due to sediment volume (inhomogeneities within the sediment), or transmit into the sediment and account as “bottom loss” (Williams and Jackson 1998). The impact of each scattering factor on the resultant backscatter strength is dependent on both the grazing angle of the sonar and the fines content of the sediment: roughness scattering is here relatively more important than sediment volume scattering in coarse-grained seabeds (Jackson et al. 1986). The model treats the sediment as a lossy fluid (i.e., effects of the porosity of the sediment and elasticity can be neglected) and assumes that the acoustic penetration of the seabed is small (Williams and Jackson 1998). However, it should be noted that those studies investigated SSS at significantly lower frequencies (10–100 kHz) than this study. Data in this study was collected at high frequencies of 1000 kHz, so the second assumption is valid, but the effects of porosity may need to be taken into account. The effective density fluid model (EDFM), as proposed by Williams (2001), follows a simplified version of Biot’s poroelastic model and proposes that the backscatter is indeed dependent on an “effective density” depending on tortuosity and porosity at higher frequencies (Williams 2001). The EDFM is focused on frequencies between 10 and 100 kHz but suggests that as frequency approaches infinity, the tortuosity of the sediment determines the relative motion of the fluid and sediment, which influences the backscatter (Williams 2001). Ivakin and Sessarego (2007) tested SSS backscatter strength at high frequencies (500–1500 kHz) and found that grain size correlated to backscatter strength when the ratio of grain size to acoustic wave length fell within a range of 0.15–1.3.

These models and studies highlight that grain size, porosity, density, and roughness affect the backscatter. The same parameters can be directly related to processes of sediment dynamics, as well as to geotechnical strength parameters. Thus, establishing relationships between geotechnical seabed testing methods and SSS has the potential to simplify and optimize the collecting of geotechnical information of seabed surface sediments for prediction and understanding of local sediment dynamics. Furthermore, it may assist with filling gaps in understanding the interaction between environmental conditions including benthic biogenic processes, geotechnical properties, and local sediment dynamics (Consolvo et al. 2020a, 2020b). This article explores initial statistical relationships of parameters related to surficial sediment properties, being roughness, in-situ shear strength, and grain size distribution, to high-frequency side scan sonar backscatter data based on field data and sediment sampling with subsequent laboratory testing from ten different sites.

Physical setting

Test sites comprised six sampling locations in the York River, Virginia; one site in the Piankatank River, Virginia; and three sites in the Great Bay Estuary, New Hampshire (Fig. 1). The York River locations represent an estuarine environment connecting to the Chesapeake Bay. Its main and secondary channels are dominated by muddy seabed sediments bordered by sandy shoals (Fall 2012). Three of the test locations form a cross-channel transect expected to be dominated by fine-grained sediments, while the three more eastern locations are near the Goodwin Islands sandy shoals. The sites were chosen to represent minor (when within one transect) to significant (when comparing the two transects) differences in seabed sediment conditions, while featuring similar environmental conditions and spatial proximity. The Piankatank River is an estuary connecting to the Chesapeake Bay and north of the York River. The seabed at our test sites was expected to be comprised primarily of sand/shell, muddy sand/shell, and shell/rock (Consolvo et al. 2020a). This location was chosen due to the general proximity to the York River. However, the nearby oyster reefs contributed a significant amount of shell hash to the seabed surface (Consolvo et al. 2020a). The Great Bay Estuary is located in New Hampshire, being spatially significantly disconnected from the York and Piankatank River sites. Seabed sediments range from mud to varying amounts of mud and coarse-grained sediments to clean sand and gravel to rocks (Wengrove et al. 2015). All three test locations in the Great Bay estuary represent edges of muddy tidal flats. However, they feature significant differences in fines content, water content, and bulk density, despite their general spatial proximity. All sites enabled the easy collection of high-quality sediment samples, in-situ geotechnical testing, and stationary rotary side scan sonar imaging without noticeable impacts from non-sedimentary environmental conditions, being a main reason for the choice of locations in this initial study.

Fig. 1
figure 1

Data collection sites: Great Bay Estuary, New Hampshire (NH); the Piankatank River, Virginia (VA); and the York River, VA. Each star represents a specific testing location, i.e., where the SSS was positioned and corresponding data collected (image source: Google 2022)

Full size image

Methods

A rotary side scan sonar (SSS) was deployed on a metal tripod sitting stationary on the seabed (Fig. 2) at the locations shown in Fig. 1. A portable free fall penetrometer (PFFP) was deployed within the SSS imaging window after conclusion of acoustic scanning. PFFP testing enables estimating geotechnical seabed surface properties (typically < 1 m below seabed surface) in a rapid manner. It has also been specifically suggested for relating geotechnical and geoacoustic data (Dayal 1980; Osler et al. 2001, 2006). PFFP typically derives seabed strength properties of surficial seabed sediments from measurements of resistance i.e., change in motion during impact and penetration into the seabed after free fall through the water column (Albatal et al. 2019). Samples for sediment characterization in the laboratory were obtained near and after the PFFP testing and within the scanning window of the SSS using either a push core, boxcore, or ponar grab sampler. The samples can be considered undisturbed (for the push core samples), lightly disturbed (for the box core samples), and disturbed (for the grab sampler samples).

Fig. 2
figure 2

a Rotary side scan sonar shown positioned as during deployment on the seafloor. b Side scan sonar rotary image taken at the Piankatank River (Fig. 1) with a frequency of 1000 kHz, a range of 15 m, and a gain of 25%. The image shows two banks of oyster reefs and sandy sediments with ripples in between

Full size image

SSS: data collection

A Kongsberg MS1000 PC rotary side scan sonar was used to collect acoustic imagery of the seabed. The sonar sends out an acoustic signal and registers the amount of energy reflected back, referred to as the backscatter, at sample points along a predetermined range (here: between 5 and 15 m) to create the acoustic image. The sonar then rotates in a counterclockwise fashion (.225° each step) to create the full rotary image as shown in Fig. 2b.

The image quality and brightness are partially dependent on different adjustable acoustic settings that change either the outgoing signal or change how much backscatter can be received. The parameters adjusted for the images taken in this study are summarized in Table 1. The SSS wavelength is on the order of millimeters, while scanned bedform features were on the order of centimeters to tens of centimeters, and scanned oysters were on the order of centimeters. The grain size of the scanned seabed sediments ranged from a median grain size of 0.1–0.4 mm, and shell hash was found in a similar and slightly coarser size range at the Piankatank River site (Consolvo et al. 2020a).

Table 1 Summary of key input parameters used for the SSS images
Full size table

SSS: data processing

The SSS datafile contains the settings data and image data and was processed using in-house Python codes (Smith 2022). The backscatter data was organized according to the acoustic settings and obvious seabed conditions (e.g., presence of oysters, bedforms, presence of rocks and pebbles) to track the properties that influence the backscatter. For example, in order to see roughness’ effects on the backscatter, acoustic properties such as gain and range were kept constant, and subsections within the data were categorized visually based on apparent roughness. For example, the red boxed subsection in Fig. 3 represents a certain range, here 2 m, and roughness category, “rocky,” indicating the presence of some surficial rocks and pebbles. This classification was designated visually from the SSS image and confirmed by complementary information like underwater cameras and local knowledge of the area. Once a subsection (e.g., red box in Fig. 3) was identified and classified visually, a statistical analysis of the backscatter was conducted for all backscatter values (i.e., pixels) within that subsection deriving the mean, standard deviation, maximum and minimum backscatter intensity of the subsection (see Fig. 4).

Fig. 3
figure 3

The top right corner shows the rotary image output from the sonar, and the bottom is the corresponding “stretched out” sonar image, where the top dark zone shows the sonar dead zone, which is in the center of the rotary image, and the y-axis is the radius of the image. The sonar dead zone is located below the sonar head, where the beam does not reach. The rectangular subsection shows an example sample space used for the statistical analysis of the backscatter. This space corresponds to a certain range and was assigned a certain roughness category

Full size image
Fig. 4
figure 4

Mean backscatter in (dB) plotted against the fines content in (%). The color marks the roughness description that was visually classified. Each data point grouping corresponds to samples of the mean backscatter taken at one site. The error bar shows the standard deviation for each of these sample areas

Full size image

Complementary data: field and laboratory testing

All sediment samples were classified according to the Unified Soil Classification System (ASTM D2487-17 2020) which considers median grain size, d50; percentage of fines; uniformity coefficient, Cu; and the coefficient of curvature, Cc. Water content was determined according to ASTM D2216 ( 2019). If applicable (i.e., for samples with > 50% fines), Atterberg limits including the liquid limit (LL), plastic limit (PL), and plasticity index (PI) were determined (ASTM D4318 2018). Push core samples were collected from the mud flats at the Great Bay Estuary (Fig. 1). Miniature lab vane shear tests were performed (ASTM D4648 2010), and results were used to supplement the strength data from the PFFP.

PFFP accelerometer data obtained during impact and penetration into the seabed were used to estimate sediment strength. Penetration depths into coarse sediments were limited to 10–20 cm. Penetration depths into fine and soft sediments reached 40–80 cm. The PFFP data was processed in accordance with the method outlined in Albatal et al. (Albatal et al. 2019) to find the dynamic bearing capacity of the sediment, qdyn at 8 cm. This qdyn value is a proxy of sediment strength and, henceforth, will be referred to as sediment strength in kPa in the rest of the paper. However, it should be noted that qdyn was not corrected for strain rate effects. As impact velocities were overall representing high strain rates and undrained conditions (2.9 to 5.4 m/s) between deployments, qdyn was directly compared within the study. However, further considerations of strain rate dependence are needed when comparing to other strength testing methods or data sets.

Results

Mean backscatter (in dB) against the percentage of fines (%) is shown in Fig. 4. The “general roughness” category represents seafloor surface sediments with no significant presence of bedforms, oysters, or rocks. For this category (blue in Fig. 4), a sharp decrease in mean backscatter is apparent with the introduction of any fines (≳ 5%). Clean sands (fines content ≈ 0%) exhibited a wide range of mean backscatter related to the observed roughness categories with the presence of ripples and oysters increasing mean backscatter significantly. Similarly, for sediment with a high fines content (>70 %), mean backscatter is relatively high when some surficial rocks and pebbles are present. Interestingly, pillow-hollow morphologies always led to a low mean backscatter, likely driven by significant acoustic shadow areas. Pillow-hollow morphologies typically represent rounded bedforms on the scale of centimeters to tens of centimeters in wavelength and with a vertical elevation on the order of centimeters that have an appearance of pillows or cloud structures (e.g., Le Dantec et al. 2013). Within the pillow-hollow groups, fine-grained sediments (fines > 80%) with small pillow-hollow features (on the order of ~5 cm) yielded on average the lowest mean backscatter. Standard deviations were high for hard roughness elements like oysters and the randomly distributed rocks, but more limited for bedforms like ripples and pillow-hollow features. Smallest standard deviations were associated with the presence of fines, while bare sands (“general roughness” category) exhibited larger standard deviations, likely related to the presence of some seabed surface features that could not be clearly associated with a bedform category, but still represented some form of roughness structure. Maximum and minimum backscatter per sample area exhibited the same trends shown for the mean backscatter (Figs. 4 and 5).

Fig. 5
figure 5

Mean backscatter (dB) plotted against qdyn (kPa). The color bar shows a the median grain size, d50 (mm), and b the water content (%). For both a and b, each data point grouping corresponds to samples of the mean backscatter within one sample area

Full size image

Figure 5 depicts the trend between qdyn from the PFFP and the mean backscatter (dB). The color bar shows how d50 (Fig. 5a) and water content (Fig. 5b ) trend with both parameters. This set of data is filtered to contain only data with less than 20% fines (i.e., sand) and excludes the “oysters” roughness classification. The plot shows that qdyn generally increases with increasing mean backscatter. As the mean backscatter increases, d50 generally increases, and water content generally decreases (Figs. 5a, b). These trends line up with expectations, but there are some larger outliers within the data: the R2 value is 0.526 for a linear trendline through the data. This scatter is likely associated with the impacts of the roughness’s influence on the backscatter. As exemplified here, the two groups of points that both are around 1.7 dB mean backscatter and at ~ 10 kPa and ~210 kPa qdyn, respectively, both classify as “pillow hollow structures,” which all featured low mean backscatter (Fig. 4). However, the penetrometer appeared more sensitive to the increase in grain size with an increase in strength. The measurements from Sara Creek 329 represented the coarsest sediments in this study but had low mean backscatter and qdyn (Fig. 5a). However, this site featured a high water content (i.e., porosity), representing a reasonable explanation for the low penetrometer strength and low mean backscatter. There are other outliers such as the Piankatank river site (Fig. 5b) with d50 ≈ 0.3 mm and a water content ≈ 35% alongside the highest mean backscatter and penetrometer strength within our set. In this case, the presence of shell hash (Consolvo et al. 2020b), increasing surface roughness (affecting mean backscatter), and friction angles (affecting qdyn) could possibly provide an explanation for this deviation. Goff et al. (Goff and Olson 2000) made similar observations related to grain size and the presence of shell hash. For SSS at a significantly lower frequency than used in this study, they found a significant correlation of backscatter strength to mean grain size as long as sediments were well sorted, but the presence of somewhat coarser shell hash disrupted the correlation, similar as found here.

Discussion

The rotary SSS provided clear acoustic seabed imagery and enabled a convenient collection of data while conducting geotechnical testing and sediment sampling in the same areas. The processing of the raw files through in-house Python codes allowed easy manipulation of the data. PFFP testing and sediment sampling was an easy solution for in-situ strength testing of the seabed surface and laboratory characterization. It should be noted that more detailed PFFP analysis could be performed towards deriving relative density or undrained shear strength which may pave the way for a quantitative correlation with acoustic theory (Williams 2001). However, this is out of the scope of this study that attempts to identify initial relationships and limitations of direct relationships. The following key observations were made: In the absence of larger roughness elements, the introduction of even a small amount of fines (~5 %) led to a noticeable decrease in mean backscatter and mean backscatter standard deviation over clean sands (Fig. 4). Roughness elements such as oysters, ripples, and rocks increased mean backscatter and the mean backscatter standard deviation. Pillow-hollow bedforms led to low mean backscatter across fine and coarse-grained sediments (Fig. 4). Sediment strength as estimated from a portable free fall penetrometer increased generally with increasing mean backscatter and median grain size. However, the low mean backscatter of pillow hollow structures disrupted this trend (Fig. 5). Furthermore, high water contents (i.e., large porosity) or the presence of angular particles (here, shell hash) appeared to override the effects of grain size for both the acoustic and the geotechnical measurements (Fig. 5).

These observations are generally in line with expectations. Harder surfaces (here, oysters and rocks) led to higher backscatter. With the high frequency used, bedform features were on scales approaching or beyond a magnitude larger than the acoustic wavelength. Thus, high backscatter intensity may be related to surface exposure to the acoustic pulse, as for the “perpendicular ripples,” and low backscatter for the “pillow-hollow” structure may be related to significant shadow zones. Generally, a trend of increasing backscatter with increasing d50 was observed. Previous explorations of SSS compared with mean grain size and roughness yielded similar trends to the ones found in this research (Jackson et al. 1986; Williams and Jackson 1998; Goff and Olson 2000; Collier and Brown 2005; Consolvo et al. 2020a; Wendelboe et al. 2023). However, it must be highlighted that the frequency used in most of these studies was significantly lower than the frequency used here (often < 500 kHz versus 1000 kHz used here), and thus, the acoustic wavelength was significantly larger than in this study. Snellen et al. (2018) and Wendelboe et al. (2023) suggested that a linear relationship between grain size and backscatter intensity is breaking down when the ratio of grain size to wavelength exceeds 0.1 using data sets mostly but not exclusively < 500 kHz. In this study, the ratio of grain size to wavelength ranged between 0.25 and 0.7. However, Ivakin and Sessaregi (Ivakin and Sessarego 2007) suggested that for high-frequency SSS (500–1500 kHz) and within a range of the grain size to wavelength ratio of 0.15–1.3 (including the range of the results presented here) bulk scattering due to sediment granular structure (i.e., grain size and density) is dominating and that backscatter strength follows a unique function depending on the ratio of grain size to wavelength. This means that a backscatter strength would be related to grain size for the use of a constant frequency. Those authors only tested coarse versus fine sand. In this study, many locations fall into this range. Nevertheless, the R2 value for the data is low when comparing sediment strength, water content, or d50 with mean backscatter and outliers are apparent. The latter were in some cases successfully associated with roughness features, highlighting that a combined analysis of roughness elements and sediment properties is needed to relate the acoustic mean backscatter directly to sediment properties. Furthermore, as relative density increases and porosity decreases, sediment strength increases. According to Biot’s model and the simplified EDFM model, tortuosity determines the relative motion of the soil grains and the water which affects the outcome amount of backscatter at high frequencies (Williams 2001). Tortuosity and porosity are proposed and confirmed numerically to be proportional through an empirical relationship (Matyka et al. 2008). Water content of the soil is directly related to the porosity of the soil.

This study focused on oysters as the only benthos investigated. However, it can be hypothesized that benthic biogenic processes contributed to the formation of the pillow-hollow structures (Le Dantec et al. 2013). It may also be argued that some variability observed for the “general roughness” category may be related to benthic biogenic processes which are known to occur in these areas (Cox et al. 2023). The results suggest that the applied high-frequency side scan sonar and the presented approach may be suitable to detect and possibly identify the presence and activity of benthic biogenic processes in terms of their impacts on the seabed surface through changes in mean and variability of the backscatter strength. This was recently tested with promising initial results through co-located geotechnical data collection, infauna analysis, and side scan sonar imaging, but data analysis is still ongoing and out of the scope of this study (Cox et al. 2023).

The range of data collected on the sediment properties and the SSS imagery data suggests relationships between SSS mean backscatter and geotechnical properties such as surficial sediments strength connected through sediment fabric describing properties such as porosity, relative density, and water content. It should be noted that the latter are significantly more effort and difficult to measure as they require the collection of decent quality seabed samples compared to the quick and easy deployment of the SSS and PFFP. Therefore, the ability to relate the seabed surface strength directly to mean backscatter is highly attractive for seabed sediment characterization. Based on the results in this study, a decision tree machine learning algorithm is currently being explored to consider roughness categories and sediment type from the SSS image and, following the resulting classification, to develop relationships between PFFP and SSS. This may optimize seabed surface characterization regarding speed and effort in areas of active sediment dynamics where high-quality seabed sampling can be challenging. It may also increase available data and, by doing so, spatial coverage and reliability in terms of deriving sediment details such as relative density of sands or state of consolidation and water content for fines. This would represent a significant step forward for seabed sediment characterization from SSS. It is also shown here that the presence of bedforms as well as textural sediment properties such as water content and porosity or the presence of shell hash may bias sediment type classification from SSS. PFFP testing with more detailed PFFP data analysis may address those issues (Albatal et al. 2019; Consolvo et al. 2020a).

The results of this initial study represent a steppingstone to expand upon efforts of automated seabed sediment classification from SSS by outlining theoretically and empirically how the mean backscatter trends change with multiple sediment properties (Buscombe et al. 2016; Chandrashekar et al. 2021). However, it is also clear that surface roughness is a dominating factor and that a two-step or multi-variable algorithm is needed to account for the dependence of the backscatter intensity on surface roughness and sediment properties. It was also shown that different surface roughness elements can have diverse responses, adding further complexity.

Conclusion

Mean backscatter from a rotary side scan sonar at 1000 kHz was related to visually identified seabed roughness categories, fines content, median grain size, water content, and sediment strength estimated from a portable free fall penetrometer in the upper 10 cm of the seabed surface. Multiple test locations from the York River, Virginia; the Piankatank River, Virginia; and the Great Bay estuary, New Hampshire, were analyzed. In the absence of larger roughness elements, the introduction of even a small amount of fines (~5 %) led to a noticeable decrease in mean backscatter and mean backscatter standard deviation over clean sands. Roughness elements such as oysters, ripples, and rocks increased the mean backscatter and the mean backscatter standard deviation. Pillow-hollow bedforms led to low mean backscatter across fine and coarse-grained sediments. Sediment strength as estimated from a portable free fall penetrometer increased generally with increasing mean backscatter and median grain size. However, the low mean backscatter of pillow hollow structures disrupted this trend. Furthermore, high water contents (i.e., large porosity) or the presence of angular particles (here, shell hash) appeared to override the effects of grain size for both the acoustic and the geotechnical measurements. These observations agree well with acoustic theory and suggest pathways to connect SSS backscatter measurements directly to geotechnical properties through textural sediment properties such as relative density, fines content, water content, and porosity. However, scatter and deviations are related to complex multi-variable dependencies and changes in environment-specific textural properties beyond grain size variations and sediment types. Thus, multivariable correlations appear necessary for more accurate correlations between SSS backscatter and geotechnical sediment properties. The present study represents a step into this direction, but a larger data set is needed for the development of confident correlations.