Detection of deep water benthic macroalgae using image-based classification techniques on multibeam backscatter at Cashes Ledge, Gulf of Maine, USA

C McGonigle, JH Grabowski, CJ Brown, Tom Weber, R Quinn

Research output: Contribution to journalArticle

31 Citations (Scopus)

Abstract

Benthic macroalgae form an important part of temperate marine ecosystems, exhibiting a complex three-dimensional character which represents a vital foraging and spawning ground for many juvenile fish species. In this research, image-based techniques for classification of multibeam backscatter are explored for the detection of benthic macroalgae at Cashes Ledge in the Gulf of Maine, USA. Two classifications were performed using QTC-Multiview, differentiated by application of a threshold filter, and macroalgal signatures were independently extracted from the raw sonar datagrams in Matlab. All classifications were validated by comparison with video ground-truth data. The unfiltered classification shows a high degree of complexity in the shallowest areas within the study site; the filtered demonstrates markedly less variation by depth. The unfiltered classification shows a positive agreement with the video ground-truth data; 82.6% of observations recording Laminaria sp., 39.1% of Agarum cribrosum and 100.0% (n = 3) of mixed macroalgae occur within the same acoustically distinct group of classes. These are discrete from the 8.1% recorded agreement with absences and nulls (>40 m) of macrophytes (n = 32) from a total of 86 ground-truth locations. The results of the water column data extraction (WCDE) show similar success, accurately predicting 78.3% of Laminaria sp. and 30.4% of A. cribrosum observations. The unfiltered classes which showed agreement with the ground-truth data were then compared to the WCDE results. Comparison of surface areas reveals the overall percentage agreement is relatively constant with depth (67.0–70.0%), with Kappa coefficient increasing from k = 0.17–0.35 as depth (and surface area) increases. The results have demonstrated that both methods were more effective at detecting the presence of Laminaria sp. (82.6–77.3%) than Agarum cribrosum, (66.6–30.4%), and that the efficiency of prediction decreased with depth. Canopy volume derived from the WCDE analysis was between 1.21 × 106 m3 at <24 m water depth, 1.82 × 106 m3 at <30 m and 2.45 × 106 m3 at <40 m. These results suggest that the presence of benthic macrophytes has a significant capacity to affect image-based classification of acoustic data, and highlights the fact that multibeam backscatter and image-based classification have significant potential for benthic macroalgal research. This is beneficial to help refine segmentations of substrates, adding valuable contextual information about biological characteristics of infaunal and epifaunal benthic communities.
LanguageEnglish
Pages87 - 101
Journalestuarine, coastal and shelf science
Volume91
Issue number1
Early online date22 Oct 2010
DOIs
Publication statusPublished - 1 Jan 2011

Fingerprint

backscatter
deep water
water column
surface area
acoustic data
spawning ground
biological characteristics
sonar
gulf
detection
marine ecosystem
segmentation
benthos
water depth
canopy
filter
substrate
fish
prediction

Keywords

  • classification

Cite this

@article{ff1932aa5ccd48e28e438d8a86a3cc7c,
title = "Detection of deep water benthic macroalgae using image-based classification techniques on multibeam backscatter at Cashes Ledge, Gulf of Maine, USA",
abstract = "Benthic macroalgae form an important part of temperate marine ecosystems, exhibiting a complex three-dimensional character which represents a vital foraging and spawning ground for many juvenile fish species. In this research, image-based techniques for classification of multibeam backscatter are explored for the detection of benthic macroalgae at Cashes Ledge in the Gulf of Maine, USA. Two classifications were performed using QTC-Multiview, differentiated by application of a threshold filter, and macroalgal signatures were independently extracted from the raw sonar datagrams in Matlab. All classifications were validated by comparison with video ground-truth data. The unfiltered classification shows a high degree of complexity in the shallowest areas within the study site; the filtered demonstrates markedly less variation by depth. The unfiltered classification shows a positive agreement with the video ground-truth data; 82.6{\%} of observations recording Laminaria sp., 39.1{\%} of Agarum cribrosum and 100.0{\%} (n = 3) of mixed macroalgae occur within the same acoustically distinct group of classes. These are discrete from the 8.1{\%} recorded agreement with absences and nulls (>40 m) of macrophytes (n = 32) from a total of 86 ground-truth locations. The results of the water column data extraction (WCDE) show similar success, accurately predicting 78.3{\%} of Laminaria sp. and 30.4{\%} of A. cribrosum observations. The unfiltered classes which showed agreement with the ground-truth data were then compared to the WCDE results. Comparison of surface areas reveals the overall percentage agreement is relatively constant with depth (67.0–70.0{\%}), with Kappa coefficient increasing from k = 0.17–0.35 as depth (and surface area) increases. The results have demonstrated that both methods were more effective at detecting the presence of Laminaria sp. (82.6–77.3{\%}) than Agarum cribrosum, (66.6–30.4{\%}), and that the efficiency of prediction decreased with depth. Canopy volume derived from the WCDE analysis was between 1.21 × 106 m3 at <24 m water depth, 1.82 × 106 m3 at <30 m and 2.45 × 106 m3 at <40 m. These results suggest that the presence of benthic macrophytes has a significant capacity to affect image-based classification of acoustic data, and highlights the fact that multibeam backscatter and image-based classification have significant potential for benthic macroalgal research. This is beneficial to help refine segmentations of substrates, adding valuable contextual information about biological characteristics of infaunal and epifaunal benthic communities.",
keywords = "classification",
author = "C McGonigle and JH Grabowski and CJ Brown and Tom Weber and R Quinn",
year = "2011",
month = "1",
day = "1",
doi = "10.1016/j.ecss.2010.10.016",
language = "English",
volume = "91",
pages = "87 -- 101",
journal = "Estuarine Coastal and Shelf Science",
issn = "0272-7714",
publisher = "Elsevier",
number = "1",

}

Detection of deep water benthic macroalgae using image-based classification techniques on multibeam backscatter at Cashes Ledge, Gulf of Maine, USA. / McGonigle, C; Grabowski, JH; Brown, CJ; Weber, Tom; Quinn, R.

In: estuarine, coastal and shelf science, Vol. 91, No. 1, 01.01.2011, p. 87 - 101.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Detection of deep water benthic macroalgae using image-based classification techniques on multibeam backscatter at Cashes Ledge, Gulf of Maine, USA

AU - McGonigle, C

AU - Grabowski, JH

AU - Brown, CJ

AU - Weber, Tom

AU - Quinn, R

PY - 2011/1/1

Y1 - 2011/1/1

N2 - Benthic macroalgae form an important part of temperate marine ecosystems, exhibiting a complex three-dimensional character which represents a vital foraging and spawning ground for many juvenile fish species. In this research, image-based techniques for classification of multibeam backscatter are explored for the detection of benthic macroalgae at Cashes Ledge in the Gulf of Maine, USA. Two classifications were performed using QTC-Multiview, differentiated by application of a threshold filter, and macroalgal signatures were independently extracted from the raw sonar datagrams in Matlab. All classifications were validated by comparison with video ground-truth data. The unfiltered classification shows a high degree of complexity in the shallowest areas within the study site; the filtered demonstrates markedly less variation by depth. The unfiltered classification shows a positive agreement with the video ground-truth data; 82.6% of observations recording Laminaria sp., 39.1% of Agarum cribrosum and 100.0% (n = 3) of mixed macroalgae occur within the same acoustically distinct group of classes. These are discrete from the 8.1% recorded agreement with absences and nulls (>40 m) of macrophytes (n = 32) from a total of 86 ground-truth locations. The results of the water column data extraction (WCDE) show similar success, accurately predicting 78.3% of Laminaria sp. and 30.4% of A. cribrosum observations. The unfiltered classes which showed agreement with the ground-truth data were then compared to the WCDE results. Comparison of surface areas reveals the overall percentage agreement is relatively constant with depth (67.0–70.0%), with Kappa coefficient increasing from k = 0.17–0.35 as depth (and surface area) increases. The results have demonstrated that both methods were more effective at detecting the presence of Laminaria sp. (82.6–77.3%) than Agarum cribrosum, (66.6–30.4%), and that the efficiency of prediction decreased with depth. Canopy volume derived from the WCDE analysis was between 1.21 × 106 m3 at <24 m water depth, 1.82 × 106 m3 at <30 m and 2.45 × 106 m3 at <40 m. These results suggest that the presence of benthic macrophytes has a significant capacity to affect image-based classification of acoustic data, and highlights the fact that multibeam backscatter and image-based classification have significant potential for benthic macroalgal research. This is beneficial to help refine segmentations of substrates, adding valuable contextual information about biological characteristics of infaunal and epifaunal benthic communities.

AB - Benthic macroalgae form an important part of temperate marine ecosystems, exhibiting a complex three-dimensional character which represents a vital foraging and spawning ground for many juvenile fish species. In this research, image-based techniques for classification of multibeam backscatter are explored for the detection of benthic macroalgae at Cashes Ledge in the Gulf of Maine, USA. Two classifications were performed using QTC-Multiview, differentiated by application of a threshold filter, and macroalgal signatures were independently extracted from the raw sonar datagrams in Matlab. All classifications were validated by comparison with video ground-truth data. The unfiltered classification shows a high degree of complexity in the shallowest areas within the study site; the filtered demonstrates markedly less variation by depth. The unfiltered classification shows a positive agreement with the video ground-truth data; 82.6% of observations recording Laminaria sp., 39.1% of Agarum cribrosum and 100.0% (n = 3) of mixed macroalgae occur within the same acoustically distinct group of classes. These are discrete from the 8.1% recorded agreement with absences and nulls (>40 m) of macrophytes (n = 32) from a total of 86 ground-truth locations. The results of the water column data extraction (WCDE) show similar success, accurately predicting 78.3% of Laminaria sp. and 30.4% of A. cribrosum observations. The unfiltered classes which showed agreement with the ground-truth data were then compared to the WCDE results. Comparison of surface areas reveals the overall percentage agreement is relatively constant with depth (67.0–70.0%), with Kappa coefficient increasing from k = 0.17–0.35 as depth (and surface area) increases. The results have demonstrated that both methods were more effective at detecting the presence of Laminaria sp. (82.6–77.3%) than Agarum cribrosum, (66.6–30.4%), and that the efficiency of prediction decreased with depth. Canopy volume derived from the WCDE analysis was between 1.21 × 106 m3 at <24 m water depth, 1.82 × 106 m3 at <30 m and 2.45 × 106 m3 at <40 m. These results suggest that the presence of benthic macrophytes has a significant capacity to affect image-based classification of acoustic data, and highlights the fact that multibeam backscatter and image-based classification have significant potential for benthic macroalgal research. This is beneficial to help refine segmentations of substrates, adding valuable contextual information about biological characteristics of infaunal and epifaunal benthic communities.

KW - classification

U2 - 10.1016/j.ecss.2010.10.016

DO - 10.1016/j.ecss.2010.10.016

M3 - Article

VL - 91

SP - 87

EP - 101

JO - Estuarine Coastal and Shelf Science

T2 - Estuarine Coastal and Shelf Science

JF - Estuarine Coastal and Shelf Science

SN - 0272-7714

IS - 1

ER -