A novel use of Stereo Baited Remote Underwater Video and Drop-Down Video for biodiversity and marine landscape mapping and prediction

PLOS ONE | https://doi.org/10.1371/journal.pone.0319355 April 3, 2025 9 / 23 PLOS ONE Marine landscape and biodiversity mapping using SBRUVs and DDV the MPA. Diversity and substratum coverage for each site was predicted using ordinary krig- ing, including standard error predictions (Fig S1), in the R package automap [62] to estimate proportional substratum coverage throughout the area of the MPA. Maps and kriged calcula- tions were conducted using a combination of the rgdal [63], rgeos [64], gstat [65], geosphere [66], and raster [67] packages. As data was irregular the package akima [68] was used to create rasters for mapping, and figures were produced using the ggplot2 package [69]. Variograms were produced and 5-fold cross validation undertaken to examine Mean Error (ME), Mean Predicted Standard Error (MPSE), Mean Square Normalised Error (MSNE), Correlation Observed and Predicted (COP), and Correlation and Predicted Observed (CPR) (Table S2). Differences in biodiversity between substratum classifications and fishing management zones within the MPA were compared using a pairwise comparison, as the data did not meet the normality and homogeneity of variance assumptions for an ANOVA test non-parametric tests were conducted including Kruskal-Wallis, Wilcox and post-hoc Dunns tests using the rstatix [70], and the multcompView [71] package for visualisation. Marine landscape analysis From the kriged predictions of proportional coverage for each substratum type, each substratum type was aggregated into distinct patches to examine patch characteristics such as patch area,

ratio of perimeter to area (RPA) and fractal dimension index using SDMTools R package [72]. Unlike remote sensed multiband layers which can only have a 0 or 1 value for presence/absence of substratum type, the results in this study have cells which could theoretically be made up of a mix of all substratum types. Therefore, the base level threshold for mapping was set to 0.2 pro- portional coverage as five substratum classifications were used. This would also reduce inflating patch size where sporadic substratum occurrences were found. Patch overlap may occur where survey sites revealed locations that had a mix of substratum types. Patch analysis was conducted using a 500 m x 500 m grid overlay on the predicted substrata maps and calculating a mean value contained within each deployment (Fig S2). Fractal dimension index (FDI) ranged from 0–2 and is a unitless measure of edge complexity calculated from the base 2 log of the propor- tional increase in information obtained when the edge of the patch is evaluated at increasing spatial resolutions, a greater value indicates more complex edge characteristics [73, 74]. Results In total 511 SBRUV deployments were conducted between June and September from 2013 to 2019 (Table S3). 57 species were identified across all sites; these included a range of demersal and epibenthic species (Table S1). The DDV surveys resulted in 333 sites carried out between 2013 and 2018. Biodiversity and seabed mapping Kriged prediction including standard error predictions were produced (Fig S1), and to exam-

ine the confidence of our predictions relating to spatial variance between data points, vario- grams of the kriged predicted maps were also produced (Fig 6, Table S2). Variogram results indicate different spatial characteristics across substratum types. Algae exhibited a short range before reaching the variogram sill, suggesting limited spatial continuity in its distribution. Sand and mud showed a larger range, reflecting a more widespread and continuous spatial pattern. Biodiversity displayed the highest nugget effect, indicating substantial small-scale variability within the sampling points (Fig 6). Predicted proportional coverage of algae ranged from 0–0.95, the largest extent of algae stretched around the southwest coast of the island (Fig 7). Gravel (which also contained some