Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Figure 1.

Location of the study area along Bonaventure's nearshore area (South of Gapsesia Peninsula, Quebec, Canada) and the underlying LiDAR bathymetric map based upon a 2 m grid.

Water depths ranged from 2.05 to 10.91 m.

More »

Figure 1 Expand

Figure 2.

Chart of the bathymetric LiDAR full-waveform monitored by the green (532 nm) channel.

This signal was acquired at 4.50 m depth and the oblique dashed line consisted of a linear fit of the water column return.

More »

Figure 2 Expand

Table 1.

Description of the LiDAR-derived environmental predictors.

More »

Table 1 Expand

Figure 3.

Workflow summarizing the statistical analysis.

Blue tabs indicated initial datasets and the discretization procedure; orange tabs highlighted machine learners used; green tab represented the models' evaluators and red tabs represented analytical evaluators.

More »

Figure 3 Expand

Figure 4.

Distributions of the eight biotic indices in the form of shadowgrams statistically analyzed by quantile box plots, and augmented by the photograph of the station corresponding to the maximum of the related index.

More »

Figure 4 Expand

Table 2.

Descriptive statistics of the eight biotic indices.

More »

Table 2 Expand

Figure 5.

Three-dimensional scatterplots of the eight biotic indices representing the values taken by four evaluators in respect to the seven machine learners and to nine numbers of classes.

The four coloured envelopes correspond to the four nonparametric density contours (each one associated with one evaluator group), drawing a 50% kernel contour shell around the points.

More »

Figure 5 Expand

Table 3.

R2 adjusted of the model describing the evolution of the eight biotic indices against the inversed number of classes as a function of the four evaluators.

More »

Table 3 Expand

Figure 6.

Random Forest tree model for Simpson index (D) discretized into three classes across a neritic benthoscape located north of the Baie des Chaleurs (Québec, Canada).

Within each node are mentioned the label of the class (1st line), the probability of belonging to the target class (2nd line) and the splitting variable. Above the node is indicated the threshold value related to the splitting variable inherent to the previous node. Pie plots associated with each node show the number of training samples belonging to the <0.3334 (green), [0.3334, 0.6667) (yellow), and >0.6667 (red) classes.

More »

Figure 6 Expand

Figure 7.

Receiver Operating Characteristics curves of the three final Simpson index (D) classes.

The diagonal line (black thin) represents the behaviour of a random classifier. The iso-performance line (black bold) embodies all the points subject to trade-off between true positive (benefits) and false positive (costs), in the ROC space. The confusion matrix is depicted on the bottom right corner.

More »

Figure 7 Expand

Figure 8.

20 stations-averaged Simpson index in respect to the three 20 stations-averaged predictors highlighted by the Random Forest learner: bathymetry (light blue), time range (light green) and skewness (light red).

Trendlines correspond to 3-degree polynomial fitting models and R2 stand for the coefficient of determination.

More »

Figure 8 Expand

Figure 9.

Map predicted epi-macrobenthic Simspon index (D) model for a neritic benthoscape of north of the Baie des Chaleurs (Québec, Canada) derived from the selected Random Forest Tree.

More »

Figure 9 Expand

Figure 10.

Hypothetical scenario explaining the evolution of the shape of the benthic waveform against the increase of Simpson diversity over seabed.

More »

Figure 10 Expand