Annotation software
There is a range of readily available image annotation software and platforms available such as TransectMeasure (seagis.com.au/transect.html), Squidle+ (squidle.org), CoralNet (coralnet.ucsd.edu), BenthoBox (benthobox.com), and ReefCloud (reefcloud.ai), all of which are suitable for mono video annotation. For stereo-video annotation, we have used SeaGIS EventMeasure (seagis.com.au/event.html) and recommend this as a widely used and well-supported software workflow for stereo-annotation and measurement.
Image annotation
For horizontally facing wide-field imagery, we recommend annotating 20 random points assigned to the lower 50% of each image. We provide example annotation and quality control workflows (globalarchivemanual.github.io/CheckEM/). A simulation study of point annotation of downward-facing imagery found that 20 points would provide an adequate estimate of variance in benthic assemblage composition whereas 80 points would provide a highly consistent estimate (Dumas et al. 2009). Similarly, for the horizontal-facing images collected by the BOSS, we explored the implication of annotating one field of view, using 20 points, to up to four fields of view, a total of 80 points, across multiple independent tropical, subtropical, and temperate locations (Effect of increasing number of fields of view on habitat observations and cost of data collection, code publicly available at github.com/UWA-Marine-Ecology-Group-projects/paper-boss-habitat/). We found generally more precise estimates of habitat composition using 40 to 80 points, annotating two to four fields of view, justifying our recommendation to annotate the combined field of view (~270o) of the four cameras, to characterise benthic composition (See Effect of increasing number of fields of view on habitat observations and cost of data collection).
For annotation of benthic composition, we recommend the CATAMI classification schema (Althaus et al., 2015), which classifies habitats into morphological groups. This schema is also recommended for similar marine sampling protocols for towed video, ROVs, AUVs (Przeslawski et al., 2023) and benthic composition from BRUV (Langlois et al., 2020). We provide a controlled repository of CATAMI formatted for use in TransectMeasure available at github.com/GlobalArchiveManual/CheckEM/tree/main/annotation-schema, which also includes species-specific annotation for certain common and easily identifiable taxa from the CAAB classification schema relevant to Australia (Rees et al., 1999). Also included is an annotation schema for visual estimates of structural complexity or relief (see Langlois et al., 2020).
Effect of increasing number of fields of view on habitat observations and cost of data collection
The benefits of using wide-field or 360-degree cameras has been demonstrated when quantifying fish assemblages (Whitmarsh, Huveneers, and Fairweather 2018; Pelletier et al. 2021), but is also beneficial when characterising benthic habitats (Mallet et al. 2021; Pelletier et al. 2020). A wide-field of view, made up of multiple composited views, reduces issues with observation direction in single view systems, where at habitat edges or in high-relief environments where the dominant seascape feature may be missed (Mallet et al. 2021). We further demonstrate the value of additional fields of view (Fig. 9), by showing that habitat heterogeneity per sample increases as the number of fields of view increases, each of approximately 70o wide, across seven locations from the subtropical Abrolhos Marine Park (Western Australia) to the temperate Franklin Marine Park (Tasmania). The code and data are publicly available at github.com/UWA-Marine-Ecology-Group-projects/paper-boss-habitat/. The increase in effort to annotate this additional imagery is minor (1-2 minutes) and the field time required to deploy a single view versus multiple view system is the same. The relative increase in habitat classes varied across marine parks surveyed from a limited increase (e.g. Eastern Recherche Marine Park) to a two-fold increase in the number of habitats identified (South-west Corner Marine Park, Fig. 9). At least two fields of view provide a consistent benefit to sampling habitat heterogeneity, and up to four fields of view can be beneficial at some locations, with minimum increases in annotation costs. Having more information on habitat heterogeneity better informs any habitat distribution modelling and mapping, thereby justifying the use of additional fields of view.
Figure 9. Relationship between number of fields of view and number of habitat classes detected across seven continental shelf locations within the Australian Marine Parks..