How much sampling does it take to detect trends in coral‐reef habitat using photoquadrat surveys?

Abstract

1. Coral-reef managers must detect and reverse collapses in habitat and evaluate the success of such interventions. Since these responsibilities must be met with limited time and resources, methods used should balance statistical power with practical and logistical constraints.2. Photoquadrat analysis is a commonly used method to survey coral habitats. This method, which involves photographing substratum along transects and digitally analysing habitat at points on the 'photoquadrats', affords efficiency in the field but is costly and requires extensive desk-based analysis. It remains unclear what is the optimal combination of sampling units (points, photoquadrats and transects) needed to detect important trends in coral habitat.3. Here, a dataset on Philippine coral-reef habitats, collected using intensive photoquadrat surveys, was used to explore the reliability of using different numbers of points per photoquadrat, photoquadrats per transect and transects per site to detect spatial differences in habitat.4. Results of leave-some-out analyses were compared with analysis of the complete dataset. Using fewer points per photoquadrat and fewer photoquadrats per transect caused little decline in ability to detect key trends, and lessened desk-based time; reducing the number of photoquadrats also lessened field time. Using fewer transects reduced time requirements but at the expense of statistical reliability. 5. Prospective power analyses revealed that common rates of coral recovery could not be detected using even the most intensive photoquadrat protocols. This result implies that coral recoveries within protected areas might go undetected using standard surveying techniques. 6. Using fixed rather than randomly placed photoquadrats, or more sensitive indicators of habitat recovery than coral cover (e.g. coral surface area) may improve power to detect coral recoveries. Finally, protocols that minimize desk time rarely also minimize field time and vice versa, which highlights the need to prioritize different logistical constraints when designing methods.

Transects, quadrats, or points? What is the best combination to get a precise estimation of a coral community?

“…Statistical power has received particular attention in this matter as a tool to decide when a set of conditions allowed the researchers to detect appropriate effect sizes (Aronson et al, 1994; Brown et al, 2004; Lam et al, 2006; Leujak and Ormond, 2007; Molloy et al, 2013; Houk and Van Woesik, 2006); which typically have been studied as univariate analysis of total or mean coral cover (or other particular substrates). This kind of criteria are used even if the research question is related to multivariate cases, which are particularly relevant because it is also important to understand changes occurring not only in the cover of the substrate, but also in the assemblage and functional structure of corals and other reef organisms (Wulff, 2001; Alvarez-Filip et al, 2013).…”

Spatial properties of sessile benthic organisms and the design of repeat visual survey transects

“…Although monitoring and process-based understanding of dynamics in shallow waters is reasonably well established via scubabased surveys and manipulative studies (e.g. Babcock et al, 2010;Molloy et al, 2013), the quantitative assessment of benthic ecosystems at greater depths (i.e. >30 m) is still in its infancy (Boavida, Assis, Reed, Serrão, & Gonçalves, 2015;Schlacher, Williams, Althaus, & Schlacher-Hoenlinger, 2010).…”

“…At a site level, transectbased surveys must account for the interaction between sampling effort, the spatial accuracy of repeat surveys, and the distributional properties of species (e.g. Molloy et al, 2013;Perkins, Foster, Hill, & Barrett, 2016;Ryan & Heyward, 2003). Sampling design choices for benthic image-based deployments, and for all taxonomic units, include transect layout (Foster et al, 2014), the method and number of images selected, and the number of points used to score individual images (Brown et al, 2004;Leujak & Ormond, 2007).…”

Automated annotation of coral reef survey images

“…Instead, most monitoring efforts rely on in-situ surveys [56,90]. Advances in underwater robotics such as Autonomous Underwater Vehicles [95], Remotely Operated Vehicles [52], and underwater scooters [51], have enabled collection of vast amounts of reef-survey imagery.…”

“…Statistically valid sampling of benthic habitats with image-and point-based tools requires careful attention to the primary purpose of the study, choice of the experimental and statistical approaches necessary to answer the questions being addressed, and the statistical power (i.e., a function of sample size, variance, and desired difference to be detected) required to test the relevant hypotheses. These experimental design components are not the focus of this paper, and interested readers are referred to the many excellent texts on these subjects [104,90]. Instead, we focus on the means and accuracy by which quantitative information can be manually and automatically extracted from underwater images using random point annotation, specifically for near-shore tropical marine environments.…”