Content uploaded by Vincent Lecours
Author content
All content in this area was uploaded by Vincent Lecours on Jul 21, 2023
Content may be subject to copyright.
1
Exploring the Effects of Acoustic Frequency on
Terrain Attributes and Classifications Derived from
Digital Bathymetric Models at Multiple Scales
Vincent Lecours
Université du Québec à Chicoutimi
Chicoutimi, Québec, Canada
vlecours@uqac.ca
Benjamin Misiuk
Department of Oceanography
Dalhousie University
Halifax, Nova Scotia, Canada
ben.misiuk@dal.ca
Craig J. Brown
Department of Oceanography
Dalhousie University
Halifax, Nova Scotia, Canada
craig.brown@dal.ca
Abstract—In recent years, new multibeam echosounders that can
simultaneously collect data at multiple frequencies have become
available. However, the effects of acoustic frequency on
bathymetric data have yet to be characterized, as early research on
these new systems has instead focused on backscatter data. Here we
explore such effects by deriving terrain attributes and
classifications from bathymetric data from Head Harbour, Nova
Scotia, Canada, that were collected at five different operating
frequencies. The geomorphometric analyses were conducted on
bathymetric surfaces generated from data collected at each
operating frequency using four scales of analysis. Results show that
bathymetry, its derived terrain attributes, and terrain
classifications produced with them are all dependent on the
acoustic frequency used to collect bathymetric data. While the
observed effects on the regional bathymetry were relatively minor,
local bathymetry, terrain attributes and terrain classifications were
highly impacted by the frequency used when collecting data. The
impacts were less important when the terrain attributes and
classifications were generated using broader scales of analysis.
These results raise questions about how bathymetry is measured
and defined and how we should interpret the outcomes of marine
geomorphometric analyses. This is particularly relevant as such
analyses have become a key component of marine habitat mapping
and submarine geomorphology mapping.
I. INTRODUCTION
Multibeam echosounders (MBES) remain the most effective
instruments for collecting continuous, high-resolution
bathymetric data in waters where light cannot penetrate. They are
active remote sensors that function by transmitting acoustic
waves at a given frequency and measuring the return time of the
signal after reflecting off the seafloor – much like LiDAR
systems do with light. This produces a point cloud that can be
used to generate a digital surface model (DSM) of the seafloor,
commonly referred to as a bathymetric surface. The intensity of
the return is also measured to produce backscatter data that can
be used as a proxy for surficial geology. MBES typically produce
sound at a single frequency, which is often less than 70 kHz for
deep-water systems (deeper than 200 m) and up to 500 kHz for
shallow-water systems (less than 200 m). The frequency at which
a sonar operates is at the discretion of its manufacturer, and no
standards exist to guide that choice. Bathymetric data collected
by different systems at different frequencies are almost always
considered comparable as it is assumed that all systems capture
the top of the seafloor accurately regardless of frequency.
In the past few years, however, a new generation of MBES
systems that can collect data using multiple frequencies
simultaneously has become available [1]. Research on new
opportunities provided by this technology is still in its infancy,
but to date, has focused mostly on backscatter data; backscatter
was shown to vary significantly with frequency as a function of
the interactions between acoustic wavelength and the
substrate [2,3]. This may have potential implications for how
bathymetry is interpreted, but they have yet to be explored. These
implications can be far-reaching as bathymetry is the primary
input for marine geomorphometric analyses that have become
common in disciplines like marine habitat mapping,
hydrodynamic modelling, and submarine geomorphology [4].
The goal of this work was to evaluate the effects of acoustic
frequency on bathymetric data, and on terrain attributes and
classifications that can be derived from them at multiple scales of
analysis.
2
II. METHODS
Acoustic data were collected from Head Harbour, Nova
Scotia, Canada, in October 2021, using an R2Sonic 2026 MBES.
The sonar was mounted on the M/V Eastcom and operated at 90,
180, 270, 360, and 450 kHz frequencies simultaneously.
Positioning and motion compensations were recorded with an
Applanix WaveMaster GPS and IMU, respectively. Raw
soundings from each frequency were processed separately in
QPS Qimera, including data cleaning, and sound velocity,
motion, and tidal corrections, to produce five bathymetric
surfaces at 1 m horizontal resolution (Fig. 1). The cleaned
acoustic data for each frequency were also imported to QPS
FMGT for backscatter processing, and five surfaces were
produced, also at 1 m horizontal resolution.
Figure 1. Multibeam bathymetric data in Head Harbour, Nova Scotia, Canada,
collected at a 450 kHz frequency.
The five bathymetric datasets were used to generate 15 terrain
attributes: slope, easterness and northerness, maximum,
minimum and mean curvatures, planform and profile curvatures,
twisting curvature [5], a topographic position index [6], the
relative difference from mean value [7], a vector ruggedness
measure [8], a surface area to planar area ratio [9], an adjusted
standard deviation metric [10], and the roughness index-
elevation [11]. In addition, the datasets were used to classify the
study area into seven morphometric features [12]: channels,
passes, peaks, pits, planar flat areas, planar slope areas, and
ridges. All terrain attributes and classifications were performed
using 3x3, 9x9, 27x27, and 81x81 windows of analysis, resulting
in 325 surfaces to compare. All calculations were performed
using the “MultiscaleDTM” package [10] in R v. 4.2.2. The
surfaces produced were first compared in terms of their
descriptive statistics (minimum, maximum, mean, standard
deviation, range, excess kurtosis, and skewness). Correlation
matrices (Pearson’s coefficients) and difference maps were also
built to explore variations in spatial distribution, and frequency
distributions were compared for the terrain classifications.
III. RESULTS
A. Bathymetric data
The descriptive statistics of the five bathymetric datasets were
very similar, yet all difference maps between pairs of datasets
show that less than 1% of pixels had identical depth values
(Table I). As expected, the closer the frequencies of two datasets,
the smaller their differences. In general, DSMs were shallower,
less variable, and had a narrower range of depths with increasing
operating frequency. Correlations between all DSMs were very
high, the lowest being 0.989 between the 90 kHz and the 360
kHz DSMs. The absolute maximum difference in depths between
two DSMs ranged from 42 cm (between the 270 and 450 kHz
DSMs) to 1 m (between the 90 and 360 kHz DSMs). The
absolute average differences between pairs of DSMs ranged from
2 to 21 cm. The absolute differences in depths, summarized in
Table I, were highly spatially correlated with backscatter. Areas
with higher backscatter values usually had smaller differences.
TABLE I. STATISTICAL DISTRIBUTION OF ABSOLUTE DIFFERENCES IN
DEPTH BETWEEN PAIRS OF DSMS FROM DIFFERENT FREQUENCIES
DSMs
Compared
Absolute Differences (cm)
[0]
]0-1]
]1-2]
]2-5]
]5-10]
> 10
90-180 kHz
0.05%
1.99%
1.89%
5.49%
14.88%
75.71%
90-270 kHz
0.03%
1.18%
1.28%
4.26%
5.24%
88.01%
90-360 kHz
0.02%
0.71%
0.88%
3.87%
4.77%
89.75%
90-450 kHz
0.02%
0.71%
0.86%
3.57%
4.70%
90.14%
180-270 kHz
0.16%
6.87%
9.20%
50.68%
32.01%
1.08%
180-360 kHz
0.06%
2.48%
3.28%
19.60%
64.21%
10.38%
180-450 kHz
0.05%
2.10%
2.59%
12.57%
60.32%
22.38%
270-360 kHz
0.36%
15.33%
20.96%
54.98%
8.19%
0.17%
270-450 kHz
0.19%
8.11%
11.03%
54.73%
24.89%
1.05%
360-450 kHz
0.94%
36.46%
30.47%
29.82%
2.19%
0.12%
B. Terrain Attributes
Terrain attributes demonstrated much greater differences
among frequencies than bathymetry. For example, the average of
the pairwise coefficients of correlation for twisting curvature
generated using a 3x3 window of analysis was 0.151 (Table II).
3
Slope was the least affected variable with an average correlation
of 0.894 among all frequencies (3x3 window). Others at this
window size ranged from 0.204 (relative deviation from mean
value) to 0.655 (vector ruggedness measure). Descriptive
statistics confirm the high variability of terrain attributes at
different frequencies. Roughness metrics were generally higher
on average at lower frequencies. This was also true for slope but
only when it was computed with a 3x3 or a 9x9 window of
analysis; at broader scales, the average of slope was higher at
higher frequencies (270 kHz with a 27x27 window, and 360 kHz
with a 81x81 window).
In general, using a broader window of analysis to generate the
terrain attributes increased the correlations among the different
frequencies and thus reduced the differences (Table II), yet there
was no consistent linear relationship observed between frequency
and correlation strength. For example, many measures of
curvatures had lower average correlations at a window of 27x27
than at windows of 9x9 and 81x81.
TABLE II. RANGE OF AVERAGE PAIRWISE CORRELATIONS AMONG THE
SAME TERRAIN ATTRIBUTE GENERATED FROM BATHYMETRIC DATASETS OF
DIFFERENT FREQUENCY, ACROSS SCALES OF ANALYSIS
Scale
Lowest x
Correlation
Attribute
Highest x
Correlation
Attribute
3x3
0.151
Twisting
Curvature
0.894
Slope
9x9
0.333
Relative
Deviation from
Mean Value
0.984
27x27
0.553
0.944
Adjusted
Standard
Deviation
81x81
0.410
Roughness
Index-Elevation
0.953
Northerness
C. Terrain Classification
As with the terrain attributes, the terrain classifications were
considerably affected by the frequency of the input data. At the
3x3 scale of analysis, the morphological maps produced with 360
and 450 kHz data were the most similar, yet 53% of the study
area was classified differently in the two maps. This increased to
71% between the maps produced with 90 kHz and 360 kHz.
These differences among frequencies decreased when broader
scales of analysis were used, ranging from 42 to 57% at 9x9,
from 22 to 78% at 27x27, and between 1 and 6% at 81x81.
The differences are also reflected in the relative coverage of
the different morphological features. For example, the finer-scale
analysis of the 90 kHz data indicated that 30% of the study area
was channels, 29% ridges, and 25% passes. When the analysis
was repeated using the 450 kHz bathymetric data as input, these
numbers changed respectively to 16%, 16%, and 42%. These
three feature types were the most variable with changing
frequency, followed by pits and peaks. The relative coverage
confirms the previous observation that the variability is smaller
when the analysis is performed at broader spatial scales.
IV. DISCUSSION
While the differences among bathymetric surfaces were
widespread and could reach up to 1 m in places (Table 1), the
high correlations among DSMs indicate that the spatial
distribution of depth values remains relatively similar. This
suggests that the effects of frequency on bathymetric data are
primarily local, with low impact on the regional representation of
the seafloor. The fact that higher frequencies produced generally
shallower depths aligns with the theory behind signal penetration.
Terrain attributes were highly frequency-dependent, yet little
consistency was observed in the relationships between
frequencies and the statistical and spatial distributions of terrain
attribute values. The results for both bathymetry and the terrain
attributes raise questions about what is measured by the acoustic
signal, and about how fine-scale interactions between the
acoustic signal and the composition of the seafloor affect the
response. This has implications for how these datasets can be
used in subsequent analyses in various contexts. While additional
field experiments are necessary to answer these questions, the
seafloor is largely inaccessible to broad-scale and detailed
ground-truthing. Additionally, it is virtually impossible to fully
replicate complex field conditions in experimental tanks due to
spatiotemporal variability in oceanographic conditions that may
affect the interactions between the mechanical acoustic waves
and the components of the transmission medium and the target. A
bathymetric lidar dataset could be used as a comparison, but
water conditions in this study area (i.e., turbidity) are not suitable
for bathymetric lidar data collection. In addition, it may be
difficult to align what is measured by sound at given frequencies
with what is measured by light at a given frequency.
While this work does not address the theoretical questions
surrounding marine acoustic-sediment interactions, it
demonstrates important effects of acoustic frequency on recorded
depth values. This has implications for multi-source datasets of
multiple frequencies. Artefacts are likely where datasets of
differing frequency overlap, which can then influence subsequent
analyses like terrain classification. We also found that the effects
on terrain attributes were amplified compared to the bathymetry.
Terrain attributes may vary considerably based on the frequency
at which the bathymetric data were collected, however these
impacts are unpredictable as no consistent patterns could be
established. A similar conclusion was reached in previous work
when looking at how various artefacts in digital bathymetric
models impact terrain attributes [13], raising questions about
whether artefacts in the data are influencing the
geomorphometric analyses. The observed frequency dependence
highlights the importance of critically evaluating the fitness for
use of datasets. However, despite increased awareness in the
community of how sensitive bathymetry can be to survey
parameters and conditions, most marine research does not have
the benefit of bathymetric data at multiple frequencies, which
enables comparison, validation, and multiple redundancies (and
therefore, reduces data uncertainty). It also remains unclear
4
whether generalizations can be made, as results may be
dependent on the characteristics of the local seafloor.
The variability of terrain classification with acoustic
frequency was higher than initially expected, but this result aligns
with the observation that frequency impacts local variability
more than regional variability. When the morphological features
were delineated using terrain attributes that were computed over
analysis distances of 3 and 9 m, they were thus more affected by
the local variability of the bathymetry. The effect decreased when
characterizing morphological features at greater analysis
distances (i.e., 27 and 81 m). This highlights the need to match
the scale of analysis with the scale of local morphological
features and indicates that the quantification of finer-scale
morphological patterns may be highly variable and inconsistent,
especially if the geomorphometric analyses capture artefacts in
the bathymetric data caused by the operating frequencies, the
angular dependence of seafloor detection, or platform motion.
Future work should continue looking for patterns between
acoustic frequency, bathymetry, and terrain attributes – for
example by quantifying variations in spatial autocorrelation.
Testing multiscale terrain attributes to identify whether they can
help differentiate artefacts from real fine-scale variability, and
therefore capture the relevant characteristics at broader scales
regardless of frequency would also be interesting. In terms of
terrain classification, we should explore how combining terrain
attributes with backscatter data, which are also known to vary
significantly with acoustic frequency, may impact seafloor
classification. Given the low correlations observed between some
of the terrain attributes that were calculated at different
frequencies, it may be fair to assume that they provide different
information on the first few centimetres of seafloor, and they
could be considered as independent variables. This may enable
testing whether combinations of terrain attributes from different
frequencies allows for a better discrimination of seafloor features
and characteristics than single-frequency data.
V. CONCLUSION
The ability to collect multibeam bathymetric data
simultaneously at different acoustic frequencies is new, and much
remains to be understood about these data’s characteristics and
uses. Here we compared five bathymetric datasets collected
between 90 and 450 kHz, and derived 15 terrain attributes and a
classification of morphological features from each. This analysis
was repeated at four different spatial scales of analysis. Results
show that acoustic frequency alters measured bathymetry, and
that the effect is amplified for all derivatives of bathymetry. Few
consistent patterns between frequency, terrain attributes, and
spatial scale could be established. This work is a first step in
trying to understand the utility, differences, and potential pitfalls
of quantitatively characterizing the morphology of the seafloor at
different acoustic frequencies.
V. ACKNOWLEDGMENTS
This research is part of the Ocean Frontier Institute/Canada
First Research Excellence Fund - Benthic Ecosystem Mapping
and Engagement (BEcoME) Project. Data collection was
completed in partnership with R2Sonic LLC, USA, and Quality
Positioning Services (QPS) Canada. Funding through
institutional support programs was also provided by the
Université du Québec à Chicoutimi.
REFERENCES
[1] Brown, C.J., J. Beaudoin, M. Brissette, and V. Gazzola, 2017. “Setting the
stage for multispectral acoustic backscatter research”. Proceedings of the
United States Hydrographic Conference, Galveston, USA, 20 March 2017.
[2] Brown, C.J., J. Beaudoin, M. Brissette, and V. Gazzola, 2019.
“Multispectral multibeam echo sounder backscatter as a tool for improved
seafloor characterization”. Geosciences, 9(3), 126, 1-19.
https://doi.org/10.3390/geosciences9030126
[3] Gaida, T.C., T.A.T. Ali, M. Snellen, A. Amiri-Simkooei, T.A.G.P. van
Dijk, and D.G. Simons, 2018. “A multispectral Bayesian classification
method for increased acoustic discrimination of seabed sediments using
multi-frequency multibeam backscatter data”. Geosciences, 8, 455, 1-25.
https://doi.org/10.3390/geosciences8120455
[4] Lecours, V., M.F.J. Dolan, A. Micallef, and V.L. Lucieer, 2016. “A review
of marine geomorphometry, the quantitative study of the seafloor”.
Hydrology and Earth System Sciences, 20, 3207-3244.
https://doi.org/10.5194/hess-20-3207-2016
[5] Minár, J. I.S. Evans, and M. Jenčo, 2020. “A comprehensive system of
definitions of land surface (topographic) curvatures, with implications for
their application in geoscience modelling and prediction”. Earth-Science
Reviews, 211, 103414, 1-24.
https://doi.org/10.1016/j.earscirev.2020.103414
[6] Weiss, A., 2001. “Topographic position and landforms analysis”. ESRI
User Conference, San Diego CA, USA.
[7] Lecours, V., R. Devillers, A.E. Simms, V.L. Lucieer, and C.J. Brown,
2017. “Towards a framework for terrain attribute selection in
environmental studies”. Environmental Modelling & Software, 89, 19-30.
https://doi.org/10.1016/j.envsoft.2016.11.027
[8] Sappington, J.M., K.M. Longshore, and D.B. Thompson, 2007.
“Quantifying landscape ruggedness for animal habitat analysis: a case
study using bighorn sheep in the Mojave desert”. The Journal of Wildlife
Management, 71, 1419-1426. https://doi.org/10.2193/2005-723
[9] Du Preez, C., 2015. “A new arc-chord ratio (ACR) rugosity index for
quantifying three-dimensional landscape structural complexity”. Landscape
Ecology, 30, 181-192. https://doi.org/10.1007/s10980-014-0118-8
[10] Ilich, A.R., B. Misiuk, V. Lecours, and S.A. Murawski, 2023.
“MultiscaleDTM: An open-source R package for multiscale
geomorphometric analysis”. Transactions in GIS, in press.
https://doi.org/10.1111/tgis.13067
[11] Cavalli, M., P. tarolli, L. Marchi, and G.D. Fontana, 2008. “The
effectiveness of airborne LiDAR data in the recognition of channel-bed
morphology” CATENA, 73(3), 249-260.
https://doi.org/10.1016/j.catena.2007.11.001
[12] Wood, J., 1996. “The geomorphological characterization of digital
elevation models”. PhD thesis, University of Leicester.
[13] Lecours, V., R. Devillers, V.L. Lucieer, and C.J. Brown, 2017. “Artefacts
in marine digital terrain models: a multiscale analysis of their impact on the
derivation of terrain attributes”. IEEE Transactions on Geoscience and
Remote Sensing, 55(9), 5391-5406.
https://doi.org/10.1109/TGRS.2017.2707303
