The annotation approach used for marine imagery impacts the detection of temporal trends in seafloor biota

Abstract

Image-based surveys of the marine benthos are being increasingly adopted as a monitoring tool for habitats and biota, particularly in mesophotic depths (∼30–150 m) which are technically difficult to survey. Many modern tools for these surveys, such as remotely operated vehicles and autonomous underwater vehicles, can capture thousands of images in a single deployment. Turning this into quantitative data typically involves human annotation which is often time-consuming and costly. Percent cover of organisms is one of the most common metrics for monitoring changes in abundance, which may be attained through visual estimation, digitization, or point-count approaches. However, alternative metrics of abundance such as density (direct counts) or presence-absence, as well as metrics that quantify condition (e.g., bleaching) and size-structure can also provide quantitative information for tracking change. Understanding the statistical power of different approaches is critical to designing effective image-based monitoring programs. Given the differing statistical power and time taken using different approaches, program managers need to decide where to allocate resources. Here we use benthic imagery from two long-term monitoring sites in south-eastern Australia to annotate three example morphospecies (morphologically distinct organisms) using three approaches: point-count (percent cover), full count and presence-absence within imagery. Also, we compare the performance of the point count and full count approaches for monitoring bleaching in one of our morphospecies. We use spatio-temporal models to quantify trends in the empirical data and simulations to quantify the power of these approaches to detect different levels of temporal change using different sampling efforts (either 100 or 200 images). Additionally, we examine the additional insights that size-structure information can provide for two morphospecies. We find that the full count approach provides a higher statistical power to detect change than the other approaches for our example morphospecies, including tracking bleaching status. Size-structure information can provide additional insights such as the occurrence of recruitment or mortality events or growth of individuals. We recommend that monitoring programs using benthic imagery should consider the choice of annotation approach as this is likely to impact observed temporal patterns, particularly when the focus is specific indicator species rather than total biodiversity.