Application of AUVs for monitoring benthic marine ecosystems has experienced a rapid increase over the past two decades. Researchers have used hover class AUVs in monitoring the impacts of invasive species (Ling et al. 2016; Perkins et al. 2015), for ecosystem-based fisheries management (Smale et al. 2012), assessing population trends in demersal fishes (Clarke et al. 2009; Seiler et al. 2012), mapping of benthic habitats (Lucieer et al. 2013), examining diversity in reef communities (Bridge et al. 2011; James et al. 2017; Monk et al. 2016), changes in structural complexity of coral reefs (Ferrari et al. 2016a, b), and mapping the spatial and depth extent of kelp forests (Marzinelli et al. 2015).

Compared to other marine imagery platforms (e.g. towed systems), hover class AUVs have several strengths applicable to marine monitoring:

  • They navigate precisely defined flight paths and the geolocation of individual images along this path. The geolocation of imagery and flight paths allows relatively precise repeat transects to be conducted, and also for the imagery to be used to ground-truth multibeam sonar (Lucieer et al. 2013) as well as for modelling the environmental factors driving species’ distributions (Hill et al. 2014).
  • The time-gain it provides over an ROV. This particularly the case if the AUV system can be left alone (i.e. that are truly autonomous).
  • An AUV will follow the set path, will not slow down or divert for something pretty, exciting or scary in the water: something that tends to happen to humans when piloting an ROV.
  • They generate spatially accurate photomosaics and finescale digital elevation models. Multibeam data which is often available with accurate georeferencing can provide important information regarding habitat types and structural complexity but is often limited to cell resolutions of 50 cm to 5 m. Finescale digital elevation models from AUV photomosaics can be done at 1-10cm cell resolution, thus enabling extremely detailed structural information to be extracted (Ferrari et al. 2016a,b). Additionally, and perhaps more importantly, the benefits of using AUV to provide digital elevation models is that the AUVs also provide colour information (via the photomosaics), which is crucial for species identification and the evaluation condition (e.g. live vs. dead coral).

The manner that data is extracted from imagery (i.e. image annotation) is context-dependent and ranges from the simple scoring of presence-absence of indicator organisms or habitats within individual images (e.g. Perkins et al. 2016) to automated habitat classification that uses sophisticated algorithms (e.g. Friedman et al. 2011). Random point count is one of the commonly employed approaches in the quantification of the cover of benthic habitats or organisms (e.g. James et al. 2017; Monk et al. 2016; Perkins et al. 2016). Whilst pattern recognition annotation has the potential to substantially speed up the image scoring process, it is not a point yet where it is accurate enough to replace manual point-counts. Accordingly, this manual will focus on point-count annotation approaches.