Mapping seabed sediments: Comparison of manual, geostatistical, object-based image analysis and machine learning approaches

Abstract

Marine spatial planning and conservation need underpinning with sufficiently detailed and accurate seabed substrate and habitat maps. Although multibeam echosounders enable us to map the seabed with high resolution and spatial accuracy, there is still a lack of fit-for-purpose seabed maps. This is due to the high costs involved in carrying out systematic seabed mapping programmes and the fact that the development of validated, repeatable, quantitative and objective methods of swath acoustic data interpretation is still in its infancy. We compared a wide spectrum of approaches including manual interpretation, geostatistics, object-based image analysis and machine-learning to gain further insights into the accuracy and comparability of acoustic data interpretation approaches based on multibeam echosounder data (bathymetry, backscatter and derivatives) and seabed samples with the aim to derive seabed substrate maps. Sample data were split into a training and validation data set to allow us to carry out an accuracy assessment. Overall thematic classification accuracy ranged from 67% to 76% and Cohen's kappa varied between 0.34 and 0.52. However, these differences were not statistically significant at the 5% level. Misclassifications were mainly associated with uncommon classes, which were rarely sampled. Map outputs were between 68% and 87% identical. To improve classification accuracy in seabed mapping, we suggest that more studies on the effects of factors affecting the classification performance as well as comparative studies testing the performance of different approaches need to be carried out with a view to developing guidelines for selecting an appropriate method for a given dataset. In the meantime, classification accuracy might be improved by combining different techniques to hybrid approaches and multi-method ensembles.

On the Use of Maps and Models in Conservation and Resource Management (Warning: Results May Vary)

“…Boosted regression trees Supervised Costa et al, 2014;Hewitt et al, 2015 Classification rule with unbiased interaction selection and estimation Supervised Ierodiaconou et al, 2011 Discriminant function analysis Supervised Degraer et al, 2008 Ecological niche factor analysis Supervised Tong et al, 2012;Sánchez-Carnero et al, 2016 Fuzzy k-means Unsupervised Falace et al, 2015 Generalized additive model Supervised Schmiing et al, 2013;Touria et al, 2015 Generalized Quick, unbiased, efficient tree Supervised Ierodiaconou et al, 2011;Hasan et al, 2012 Random forest Both Hasan et al, 2012;Diesing et al, 2014;Piechaud et al, 2015 Support vector machine Supervised Hasan et al, 2012 Frontiers in Marine Science | www.frontiersin.orgFIGURE 2 | Example of how different methods can produce different outcomes. The input data were bathymetric data, backscatter data, and topographic data (i.e., slope, easterness, northerness, and relative deviation from mean value) (see Lecours et al, 2016b).…”

Advances in assessing Sabellaria spinulosa reefs for ongoing monitoring

“…A truly unique “mottled” signature, that could be used with sufficient confidence to inform reef delineation, was only found in study area A. Expert judgment is a key component of any such analyses and can introduce a large level of subjectivity when drawing boundaries (Diesing et al., 2014). Due to the nature of the products derived from sidescan sonar data, in this instance, the most appropriate method for assessing reef extent was deemed to be expert manual visual interpretation rather than an automated approach, such as object‐based image analysis (Drăguţ, Tiede, & Levick, 2010).…”

“…(2007) and Diesing et al. (2014) highlighted the potential sources of error for mapping habitat extent from acoustic sources. Where feasible, data acquisition should utilize the same gear type to ensure consistency of data quality, and postprocessing should similarly follow standardized guidelines to limit inconsistencies.…”

“…These outcomes were supported by the studies of both Coggan, Mitchell, White, and Golding (2007) and Diesing et al. (2014) who also emphasize the importance of repeatable methodologies for data acquisition and habitat mapping which also adopt an element of automation, thereby rendering the mapping outputs less influenced by the issues of subjectivity which arise as a result of manual interpretation.…”

Application of Geobia to map the seafloor

“…Since Lucieer's (2008) paper on object-based mapping of benthic marine habitats, uptake of GEOBIA methods has been slow but is currently accelerating with studies published in the peer-review literature (e.g. Diesing et al, 2014;Lucieer and Lamarche, 2011;Montereale Gavazzi et al, 2016) and presented at conferences. This contribution will showcase a generic workflow for categorical seafloor mapping based on acoustic data sets and ground-truth observations.…”