Acoustic seabed classification of marine habitats: studies in the western coastal-shelf area of Portugal

Abstract

Two single-beam, seabed-classification systems, QTC VIEW Series IV and QTC VIEW Series V, were used to identify and map biosedimentary gradients in a mid-shelf area off Western Portugal. The survey area has a moderate slope, a depth ranging from 30 to 90 m along a 3.5-km axis perpendicular to the shoreline, and is characterized by smooth sedimentary and biological gradients. Ground truth for sediment grain size and macrofaunal communities was based on grab sampling at 20 sites. The sedimentary and biological data were analysed using classification and ordination techniques. The acoustic data were analysed with qtc impact software and classified into acoustic classes. The affinity groups obtained in each data set were mapped using a Geographics Information System. All showed good agreement and identified prevailing gradients along a northwest–southeast direction. Three acoustic classes were identified, corresponding to the predominant sediment types, namely fine sand with low silt and clay content, silty, very fine sand, and mud. A close relationship with benthic communities was also verified, although less marked because benthic communities continuously change along the northwest–southeast gradient. Overall, the acoustic system coupled with ground-truthing data was able to discriminate and characterize the various benthic biotopes in the survey area.

Introduction

Recent progress in acoustic technology offers new opportunities for describing the marine environment. Echosounders and sidescan sonar are commonly used for remote characterization of the seafloor, including, recently, the discrimination of benthic biotopes (Kenny et al., 2003). Tools such as QTC VIEW and RoxAnn process the acoustic signals from single-beam echosounders and output data to Geographic Information Systems (GIS) to map differences in seafloor characteristics (Greenstreet et al., 1997; Hamilton et al., 1999; Kloser et al., 2001; Anderson et al., 2002).

The QTC VIEW Series IV and Series V seabed-classification systems used in this study are powerful tools for the discrimination of marine benthic habitats. Several studies have shown their response to bottom features such as sediment grain size and compactness, seabed roughness, bedrock, benthic organisms, and bottom slope (Collins et al., 1996; Hamilton et al., 1999; Preston et al., 1999; Preston, 2001; Anderson et al., 2002; Ellingsen et al., 2002; von Szalay and McConnaughey, 2002). Most of these studies covered areas with a variety of contrasting bottom features with sharp discontinuities. Recently, their efficiency was assessed in an area of relative seascape monotony, viz. in a sand and gravel, nearshore shelf area, with very low silt content (Freitas et al., 2003). In this present study, both acoustic systems were used in a mid-shelf area with a smooth biological gradient and sediment grain size ranging from clean, fine sand to mud with silt and clay content above 75%, with a view to comparing the results of the QTC VIEW Series IV and Series V systems.

Material and methods

Sampling

The QTC VIEW Series V is an advance in signal acquisition by faster sample digitization and better sample resolution, and dynamic range (Table 1). These have resulted in greater operating water depths and an advanced compensation method for echo-length changes. The Series V acquires and logs the waveform as raw data, in contrast to the pre-processed set of echo descriptors in Series IV. A mid-shelf area approximately 20 km² with depth ranging from 30 to 90 m was surveyed in April 2001 using QTC VIEW IV. Survey lines at 500-m spacing were run aboard “N.R.P. Andrómeda” (Figure 1). The QTC VIEW V was used in April 2002 over that part of the area closer to the outfall branches, with the survey lines approximately 100 m apart (see Figure 1), aboard the “N.R.P. Auriga”, a twin vessel to “N.R.P. Andrómeda” of similar size, design, and engine size. In both surveys the transducer was fixed to the side of the vessel being used and the speed was close to 6 knots. Positions were confirmed with a Global Positioning System (GPS). Both acoustic systems include a computer for the acquisition, display, and storage of the data collected. Table 2 summarizes the echosounder and QTC VIEW base settings for both surveys.

Figure 1

The study area showing the acoustic-survey lines from QTC VIEW Series IV (larger area) and Series V, the 20 sampling sites for the study of benthic communities and superficial sediments along with the sewage-outfall branches.

Table 1

The QTC VIEW Series IV and V systems compared.

	Setting
Parameter	QTC VIEW Series IV	QTC VIEW Series V
Sample rate	20 kHz	5000 kHz
Resolution	8 bits	12 bits
Dynamic range	60 dB (manual gain)	+80 dB (automatic gain control)
Depth range	10–500 m	0.75–2000 m
Depth compensation	Manual reference depth selection	Automatic standard echo length
Raw data	Feature vectors	Full bipolar waveform, interpolated envelope
GPS input	GGA or GLL, 4800 baud	GGA, GLL, RMC custom unlimited baud
Acoustic classification	Real time and post-processing	Post-processing
Quality assurance/quality control during acquisition	Off-line waveforms, real-time manual water-depth check	Real-time waveform visualization and depth pick

Table 2

Survey base-settings for both echosounders and acoustic systems. (AGC, automatic gain control.)

		Setting
	Parameter	QTC VIEW Series IV (NRP Andrómeda)	QTC VIEW Series V (NRP Auriga)
Echo sounder	Beam width	44°	19°
	Transmit power	150 W	100 W
	Pulse duration	625 μs	300 μs
	Ping rate	5 per s	5 per s
	Frequency	50 kHz	50 kHz
QTC VIEW	Base gain	5 dB	AGC

In April 2001, five ground-truth 0.1 m² Smith–McIntyre grab samples were taken at each of 20 sites (see Figure 1), two for sediment and three for macrofaunal analysis. These were washed over a 1-mm mesh screen and the remaining material fixed in 4% buffered formalin.

Acoustic classification

QTC VIEW applies a series of algorithms to the shape of the first returning echo, translating it to an array of 166 elements (Collins et al., 1996). Through Principal Component Analysis (PCA), a reduced description comprising three values (Q1, Q2, Q3) is obtained. The Q-values correspond to the first three PCA axes (Collins and McConnaughey, 1998). This matrix was classified using a K-means algorithm, with the software qtc impact v3.00. This non-hierarchical, divisive method promotes a progressive splitting process. At each split, a series of statistical measures are provided, namely the total score and the Cluster Performance Index (CPI) rate. The total score is the sum of scores of the individual classes and the CPI measures the ratio of the distance between cluster centres and the extent of the clusters in the Q-space. They were used as indicators of the optimal split level. Initially, the total score decreases rapidly, and further splits lead to smaller changes in this descriptor. Plotting the number of splits against total score, the inflection point of the resulting curve gives an indication of the optimal split level (QTC, 2002). CPI rate, defined as CPIr=(CPI(n)−CPI(n−1))/CPI(n−1), tends to be maximum at the optimal split level (Kirlin and Dizaji, 2000), and was also used as an indicator of the optimal number of acoustic classes to retain (Freitas et al., 2003). Recently, Legendre et al. (2002) proposed a method by which to analyse QTC VIEW data, a method that also combined PCA and K-means but used a different evaluation for the best number of clusters to retain.

Laboratory analysis

Sedimentary and biological descriptors for the 20 sites included sediment grain size, total volatile solids, and redox potential and macrofauna species composition and abundance. Grain size was analysed by wet and dry sieving. The silt and clay fraction, i.e. fine particles, with diameters less than 0.063 mm, and the gravel fraction, particles with diameters above 2 mm, were expressed as a percentage of the total sediment (dry weight). The sand fraction (0.063–2.0 mm) was sieved through a battery of meshes to sort the particles into the size ranges given in Table 3. The sediment was classified according to the median value of φ=−log₂, particle size in mm, and the Wentworth scale (Buchanan, 1984). Total volatile solids were determined by loss on ignition at 450°C (Byers et al., 1978). Redox potential was measured on board at −4 cm from the sediment surface with specific probes (Pearson and Stanley, 1979). The three replicate samples per site for the study of macrofauna were processed individually. In the laboratory, the animals were sorted and identified to the lowest possible taxonomic level, and for each sample a species list with the respective abundance was determined.

Table 3

The mean values for the sedimentary data in each of the affinity groups identified by classification and ordination analysis.

	Groups
	A	B1	B2	C
Sampling sites	13	1–12,14,17,18	15,16,20	19
Total volatile solids (%)	0.73	1.05	2.33	5.48
Redox potential (mV)	390.50	129.70	79.20	−7.50
Gravel (%)
>2.0 mm	0.13	0.31	0.63	0.03
1.0–2.0 mm	3.77	0.76	0.53	0.08
0.5–1.0 mm	25.86	2.25	0.53	0.18
0.25–0.5 mm	37.00	4.92	0.77	0.27
Sand (%)
0.125–0.25 mm	23.01	65.68	37.57	2.90
0.063–0.125 mm	9.30	23.76	40.70	15.33
Fines (%)
<0.063 mm	0.94	2.37	19.36	81.26
Median (Φ)	1.55	2.68	3.20	>4.00
Sediment classification	Medium sand	Fine sand	Silty very fine sand	Mud

Data analysis

For each site, the environmental data matrix includes the seven grain-size classes, the median, the total volatile solids content, and the redox potential. The normalized Euclidean distance was used to produce a [sites×sites] distance matrix submitted to classification analysis using the average-clustering algorithm and to ordination analysis using non-metric, multidimensional scaling (MDS). Both used the software primer v5 (Clarke and Gorley, 2001).

The biological data were represented by a matrix of 20 sites per 119 variables, corresponding to the species abundances. After square-root transformation, the [sites×sites] Bray–Curtis similarity matrix was classified with the average-clustering algorithm. Ordination was done by correspondence analysis using the software mvsp v3.12d (Kovach, 1999).

The classification output files representing the acoustic diversity were analysed in Arc View 8.1. For this, the final output files from both surveys were opened separately in a spreadsheet and the echo description, latitude and longitude, class name, class confidence, and class probability were selected from the appropriate fields. The data were sorted first by confidence level, and those under 98% were deleted. The confidence value is the probability that a record belongs to the class to which it has been assigned, rather than to any other class. Based on Bayes' theorem, this value is a measure of the covariance-weighted distances between the position of the record in Q-space and the positions of all cluster centres (QTC, 2002). The resulting file was further sorted by the probability and values under 1% were ignored. The probability value of a record is based on the position of that record in the Q-space and the characteristics of the class to which it has been assigned. This is a measure of the closeness of the record to the cluster centre, weighted by the covariance of the cluster in the direction of the record. Probability and confidence calculations are based on Bayes' theorem and the assumption that the underlying distribution in Q-space is Gaussian (QTC, 2002). The acoustic, sediment, and macrofauna plots were overlapped to facilitate comparison.

Results

Sedimentary gradients

The classification and ordination analysis of the environmental data is displayed in Figure 2, and a summary characterization of each group is given in Table 3. When including all the sampling sites in the analysis, three groups were separated in the ordination diagram (Figure 2a): group A (site 13), group C (site 19), and group B (the remaining sites). Sites 13 and 19 over-dominate the ordination pattern because of their particular grain size. The coarser sediment was observed at site 13, the only site classified as medium sand, and site 19 had a silt/clay fraction much higher than elsewhere (Table 3). Excluding these two sites from the analysis, group B is further subdivided into subgroups B1 and B2, as shown in the classification and ordination diagrams (Figure 2b, c). The spatial distribution of the major affinity groups (A, B1, B2, C) is shown in Figure 2d. Along the axis A→B1→B2→C, the superficial sediments show gradual increases in the median value, the silt and clay content, and the total volatile solids, while the redox potential decreases (Table 3). Most of the superficial sediments in the study area correspond to fine sand with low silt and clay content (subgroup B1). With increasing depth (inshore–offshore axis, cf. Figure 1) and towards the estuary (shelf–estuary axis, cf. Figure 1), the superficial sediment becomes silty, very fine sand (subgroup B2), and finally mud (group C) (cf. Table 3).

Figure 2

Sedimentary affinity groups (A, B1, B2, and C) identified among the sampling sites. (a) MDS with all sampling sites; (b) classification analysis, excluding sites 13 and 19; (c) MDS excluding sites 13 and 19; and (d) spatial distribution of the affinity groups.

Biological gradients

The ordination and classification diagrams of the biological data are shown in Figure 3. Sites 13 (Group A) and 19 (Group C) tend to over-dominate the ordination pattern (Figure 3a), as seen previously with the sedimentary data. Excluding them, the analyses show the subdivision of Group B into B1 and B2, and a further split into B21 and B22 (Figure 3b, c). Their distribution in plane 1–2 of the correspondence analysis (Figure 3c) indicates continuous change rather than sharp discontinuities between the groups. This is confirmed in Table 4, where the species succession along the biological gradient and the mean species richness and abundance in each affinity group are summarized.

Figure 3

Biological affinity groups (A, B1, B21, B22, and C) identified among sampling sites. (a) Correspondence analysis with all sampling sites; (b) classification analysis, excluding sites 13 and 19; (c) correspondence analysis, excluding sites 13 and 19; and (d) spatial distribution of the affinity groups.

Table 4

The biological succession in the affinity groups obtained by classification and ordination analysis. The taxa are represented by their mean abundance per unit sample (0.1 m²) and include only the species whose abundance per site is higher than 3% of the site total. Highlighted values indicate the highest mean abundances by group.

	Groups
	A	B1	B21	B22	C
Sampling sites	13	1–7,10–12	8,9,14,17,18	15,16,20	19
Mean abundance (A/0.1 m2)	96.7	102.1	143.1	243.4	109.3
Mean species richness (S/0.3 m2)	38.0	41.3	46.0	52.0	31.0
Mean species richness (S/0.1 m2)	16.3	25.0	26.9	33.5	19.0
Species succession
Pisione remota	72.0
Glycera oxycephala	9.0				2.0
Mediomastus capensis	50.0	4.7	0.8	1.0
Atylus falcatus	3.0	2.6
Spionidae n. det.	10.0	0.4	2.4	1.7
Tellina fabula	18.0	41.3	5.0
Chaetozone setosa	52.0	61.0	2.2
Urothoe pulchella	1.0	6.2	0.8	0.7
Capitella spp.		27.3
Aora typica		0.8
Mactra corallina		9.3	1.8
Sigalion mathildae		2.8	1.2
Mysella bidentata		7.8	7.4	4.3
Atylus swammerdami		4.0	3.4	0.3
Anomura n. det.		4.7	3.0	2.0
Photis longicaudata		8.2	1.8	3.0
Glycera tridactyla	1.0	5.4	4.8	3.0	3.0
Spio decoratus	1.0	4.2	1.6	1.7
Nassarius reticulatus	3.0	15.0	2.0	4.7
Ampelisca brevicornis		5.0	3.0	3.0	2.0
Sabellaria alveolata		7.2		0.3
Magellona filiformis	8.0	4.7	67.6	0.3
Paraonidae n. det.	3.0	6.8	10.2	4.0	10.0
Aoridae n. det.	3.0	0.3	7.8	3.0
Spiophanes bombix	1.0	6.8	23.0	10.0
Hyalinoecia bilineata	8.0	11.3	89.6	125.7
Prionospio spp.	5.0	11.4	31.0	32.0	4.0
Ampelisca spp.	1.0	1.6	13.2	40.3	24.0
Nucula spA		0.8	2.4	12.0	1.0
Tellina pulchella		0.3	1.6	28.3	1.0
Maldanidae spA		0.2	10.8	150.7
Spiophanes kroeyeri		0.4	9.4	31.7	6.0
Lumbrinereis cf. latrelli		1.9	69.8	98.3	80.0
Chaetopteridae n. det.		0.1	0.2	17.7	1.0
Abra alba		2.7	1.2	19.0	2.0
Thyasira flexuosa			3.8	23.0	16.0
Maldanidae spB			0.2	18.0
Terebellidae n. det.		0.9	0.2	1.0	46.0
Hydrobia ulvae					17.0
Thyasira spA					19.0

The spatial distribution of the benthic-affinity groups identifies the same dominant patterns along the inshore–offshore and shelf–estuary directions as observed in the spread of the sedimentary gradient (Figure 3d). This pattern has been consistently reported in this coastal region in the period 1994–1998 (Quintino et al., 2001). The succession represented by groups A→B1→B21→B22→C is similar to that obtained with the sedimentary data (Figure 2d). At the northwest extremity, site 13 (group A) is characterized by interstitial polychaetes (Table 4). At the southeast extremity, site 19 (group C) is characterized by faunal impoverishment (Table 4) due to the high fines content and chronic hydrocarbon contamination of the superficial sediments (Quintino et al., 2001). Between these two groups, the faunal succession corresponds to a gradual replacement of the dominant species (Table 4). Apart from site 19, the overall tendency along this succession is a slight increase in both species richness and abundance. Within the succession, the subgroup B21, spatially located between B1 and B22 (Figure 3d), is the less well characterized, with the smaller number of dominant species. This agrees with its position in the ordination, i.e. closer to the origin and between B1 and B22 (Figure 3c).

Acoustic gradients

The results of the acoustic classification by both QTC VIEW systems are given in Table 5. In both cases the optimal-classification solution corresponds to three acoustic classes, A, B, and C. These classes were obtained at the second split, when total score tended to stabilize (QTC VIEW Series IV) or reached the minimum value (QTC VIEW Series V), and the CPI rate was at the maximum value (Table 5). The acoustic pattern identified in both surveys is similar (Figure 4). The acoustic classes from the Series IV survey change along the inshore–offshore and shelf–estuary directions. Those of the Series V survey detail the inshore–offshore succession using a finer spatial grid.

Figure 4

GIS mapping of the acoustic classes A, B, and C identified with the QTC VIEW Series IV (larger area) and Series V (smaller area, closer to the outfall branches).

Table 5

Classification statistics for the QTC VIEW surveys.

System	Split	Total score	CPI	Class	Members	Chi2	Score	CPI rate
	0	246178.28	–	–	15517	15.87	246178	–
	1	178679.63	1.43	A	9704	16.62	161294	–
				B	5813	2.99	17386
				A	5498	4.90	26916
QTC View Series 4	2	88535.14	5.42	B	4751	9.21	43741	2.79
				C	5268	3.39	17878
				A	4829	8.84	42695
	3	85539.83	13.61	B	4243	6.61	28060	1.51
				C	3345	2.66	8906
				D	3100	1.90	5879
	0	17216.63	–	–	3921	4.39	17217	–
	1	11105.64	1.04	A	2456	4.02	9870	–
				B	1465	0.84	1236
				A	1551	1.09	1692
QTC View Series 5	2	5119.34	4.06	B	1259	1.37	1726	2.90
				C	1111	1.53	1701
				A	1100	2.09	2304
	3	6625.07	10.29	B	916	1.14	1043	1.53
				C	949	0.81	770
				D	956	2.62	2508

Total score=sum of the scores of the individual classes; CPI=cluster performance index; members=number of data in each class; Chi²=measure of clumpiness of each cluster in Q-space; score=members×Chi²; CPIr=[CPI(n)−CPI(n−1)]/CPI(n−1), where n is the split number (see text).

The joint geographical distribution of the acoustic classes and the sedimentary and biological affinity groups, shown in Figure 5, indicates close correspondence between the acoustic patterns and the main sedimentary and biological assemblages. Acoustic class A is predominant in the survey area and corresponds to the region occupied by fine sand with low silt content (sedimentary group B1, Figure 5a, and Table 3; biological group B1, Figure 5b, and Table 4). Acoustic class B corresponds well with the area of silty, very fine sand (sedimentary group B2, Figure 5a, and Table 3; biological group B22, Figure 5b, and Table 4). Finally, acoustic class C corresponds to the area of mud with high silt content (sedimentary group C, Figure 5a, and Table 3; biological group C, Figure 5b, and Table 4). A single ground-truth sample was taken inside acoustic class C (site 19). During a recent survey (October 2002, unpublished data), several other samples were taken within this area, confirming that the superficial sediment is similar to that described in this article for site 19.

Figure 5

GIS representation of the acoustic classes A, B, and C, jointly displayed with the (a) sedimentary affinity groups and (b) the biological affinity groups.

Discussion

Using acoustic methods, Collins et al. (1996) were able to distinguish habitats suitable for different age classes of juvenile Atlantic cod, habitats characterized by specific combinations of sediment grain size, bathymetric relief, water depth, and the presence or absence of algae. Collins and Galloway (1998) showed that acoustic diversity successfully captured a high variety of seabed types based on sediment grain size and the presence or absence of shell debris. Preston et al. (1999) reported comparable results, showing that sediment porosity and grain size influence the acoustic response. Hamilton et al. (1999) found that the bottom classes suggested by the acoustic system had consistent grain size and texture properties and followed grain-size trends. The work of Ellingsen et al. (2002) showed that the acoustic variety was generally in accordance with sediment grain size. In an area on the Portuguese coastal shelf dominated by a range of sandy sediments, all with very low silt and clay content, Freitas et al. (2003) showed close agreement between sediment grain size and the acoustic variability.

All these applications show that the QTC VIEW seabed-classification system is responsive to sediment grain size. The present study agrees with this finding. In fact, the spatial distribution of our three acoustic classes, A, B, and C, follows the same pattern as the sedimentary and biological descriptors, along inshore–offshore and shelf–estuary directions. Class A, to the northwest, corresponds to fine sand with low silt and clay content, and class C, to the southeast, to mud with silt and clay content above 75%. Class B, located between classes A and C, corresponds neatly with the distribution of the silty, very fine sand. The acoustic pattern was thus effective in identifying gradual fines increase of the superficial sediments. Following this sediment succession, the macrofauna exhibit a gradual change of the dominant species.

Some exceptions were noticed in the overall agreement between the acoustic classes and the prevailing sediment and biological affinity groups. The most important concerns the coarser sediment locally observed at site 13, corresponding to a particular biological assemblage dominated by small interstitial annelids. This area has no corresponding acoustic class. A recent sedimentary survey (October 2002, unpublished data), confirmed that there is a coarser sediment area extending westward of site 13. This coarser sediment is probably associated with the stronger currents along the western coast, as the protection of the cape located to the north of the study area is left (Figure 1). This apparent lack of correspondence could be due to the fact that the area of coarser sediment is of small spatial extent and hence has limited influence in establishing a separate acoustic class. The second apparent exception concerns the biological assemblage B21, which does not have a direct corresponding group, either in the sedimentary or in the acoustic data (Figure 5). This group establishes the transition between the biological assemblages B1 and B22 (Figure 5b), better characterized than B21, given the distribution of the dominant species among the affinity groups (Table 4). As such, the fact that the detailed biological succession has no counterpart in the sedimentary and the acoustic data should not be regarded as a case of acoustic misclassification. In fact, the transition group identified as B21 is not always detected through data treatment, whereas groups B1 and B22 are recognized consistently in this area in surveys undertaken since 1994 (Quintino et al., 2001).

The final exception concerns the QTC VIEW Series V survey results. Although the two surveys show a consistent acoustic-diversity pattern overall, within the Series V class A there are several records classified as class C. These records are not randomly distributed but rather located close to the outfall branches (Figure 4). The acoustic class A was shown to correspond with the distribution of fine sand with very low silt content. Previous surveys have occasionally identified coarser sediment in sites located between the outfall branches (Quintino et al., 2001). This was recently confirmed with a finer spatial sampling grid (October 2002, unpublished data). Although the acoustic system detected differences in that area (Figure 4), these could not be assigned to a new acoustic class, perhaps as a result of the relatively low number of echoes sampled between the branches.

Given these results, we conclude that both seabed-classification systems present high potential for the remote assessment of benthic patchiness, although ground truth will be needed to interpret the acoustic classifications. It was also shown that the information acquired by the two seabed-classification systems was consistent using different equipment and different base settings, and identified the same benthic biotopes. Such agreement between two surveys taken a year apart, April 2001 and April 2002, supports the idea that a more general application of acoustics as a remote-sensing tool to identify and interpret soft-bottom heterogeneity is possible.

The first two authors benefited from grants (Rosa Freitas—SFRH/BD/769/2000; Susana Silva—PRAXIS XXI/BD/21298/99) from the Portuguese FCT (Fundação para a Ciência e Tecnologia). This work was partially financed by SANEST, S.A. (“Estudo de Monitorização Ambiental da Descarga no Mar do Efluente do Sistema de Saneamento Multimunicipal da Costa do Estoril”) and by the FCT and POCTI (FEDER) (“ACOBIOS, POCTI/38203/BSE/2001, Acoustic and biological methods in the assessment of subtidal benthic biotopes in coastal ecosystems”). Rui Marques assisted in the preparation of the acoustic system and data collection. We acknowledge the helpful comments of two referees.