Limitations of Predicting Substrate Classes on a Sedimentary Complex but Morphologically Simple Seabed

Abstract

The ocean floor, its species and habitats are under pressure from various human activities. Marine spatial planning and nature conservation aim to address these threats but require sufficiently detailed and accurate maps of the distribution of seabed substrates and habitats. Benthic habitat mapping has markedly evolved as a discipline over the last decade, but important challenges remain. To test the adequacy of current data products and classification approaches, we carried out a comparative study based on a common dataset of multibeam echosounder bathymetry and backscatter data, supplemented with groundtruth observations. The task was to predict the spatial distribution of five substrate classes (coarse sediments, mixed sediments, mud, sand, and rock) in a highly heterogeneous area of the south-western continental shelf of the United Kingdom. Five different supervised classification methods were employed, and their accuracy estimated with a set of samples that were withheld. We found that all methods achieved overall accuracies of around 50%. Errors of commission and omission were acceptable for rocky substrates, but high for all sediment types. We predominantly attribute the low map accuracy regardless of mapping approach to inadequacies of the selected classification system, which is required to fit gradually changing substrate types into a rigid scheme, low discriminatory power of the available predictors, and high spatial complexity of the site relative to the positioning accuracy of the groundtruth equipment. Some of these issues might be alleviated by creating an ensemble map that aggregates the individual outputs into one map showing the modal substrate class and its associated confidence or by adopting a quantitative approach that models the spatial distribution of sediment fractions. We conclude that further incremental improvements to the collection, processing and analysis of remote sensing and sample data are required to improve map accuracy. To assess the progress in benthic habitat mapping we propose the creation of benchmark datasets.

Using Spatial Validity and Uncertainty Metrics to Determine the Relative Suitability of Alternative Suites of Oceanographic Data for Seabed Biotope Prediction. A Case Study from the Barents Sea, Norway

“…Results are generally compared with available field measurements (e.g., [21]) to verify that the model is able to reproduce all potential outcomes but, since observations are limited, spatial uncertainty results are not usually available for the entire model extent. Uncertainty is therefore an integral part of the oceanographic data we extract from such models, just as uncertainty is intrinsic to the bathymetric terrain models and derived variables (e.g., [22,23]), geological classifications (e.g., [24,25]), biochemical parameters (e.g., [20]), and other data used as predictor variables in HDMs. It is the knock-on spatial effects of this uncertainty on HDMs which will likely be a determining factor in selecting which oceanographic models are suitable for HDMs, or in providing the impetus for developing better suited oceanographic model data.…”

Ensemble Mapping and Change Analysis of the Seafloor Sediment Distribution in the Sylt Outer Reef, German North Sea from 2016-2018

“…The accuracy of the predicted seafloor sediments in a heterogenous area, like the Sylt Outer Reef, can be influenced by several factors that may negatively influence results [21]. These factors include (1) an inadequacy of the selected classification system, (2) a low discriminatory power of the predictors, or (3) a mismatch between the response (i.e., grab sample) and predictor variables (e.g., backscatter mosaic).In addition, an unequal number of samples between sediment classes may result in under-or over-predictions in the modelling results [46] .Discrepancies between different techniques can be very large and some models may be more sensitive to sampling bias, which might reduce model transferability and selection [15,56,68].…”

“…In addition to ensemble modelling, ensemble mapping has been suggested as another sediment mapping approach to alleviate the limitations of predicting sediment classes [21]. In ensemble mapping, predictions for each sediment class were generated using single or multiple classification techniques, and then combined the results into a single map by aggregating the modal classes.…”

“…In ensemble mapping, predictions for each sediment class were generated using single or multiple classification techniques, and then combined the results into a single map by aggregating the modal classes. This method has been utilized to develop seafloor sediment distribution maps as an alternative to the typical thematic mapping (i.e., predicting multiple classes in one row) [2,21]. However, in these studies each sediment class was predicted using only a single model and not by ensemble modelling.…”

“…Most of the seafloor mapping studies that used classification models applied the algorithms to data with appropriate number of ground-truth samples [2,6,8,21,22], which raises the question of their applicability to studies with a smaller number of data ( e.g., <50). Especially for wide-scale hydroacoustic seafloor mapping, time and budget for comprehensive ground-truth sampling is scarce.…”