Coral Reef Benthic Video Surveys Facilitate Long-Term Monitoring in the Commonwealth of the Northern Mariana Islands: Toward an Optimal Sampling Strategy

This study describes a step-by-step process used to design an effective benthic video survey component of the Commonwealth of the Northern Mariana Islands long-term monitoring program. Documenting abundance of major benthic groups at relatively large spatial scales, at the appropriate localities, can empower monitoring programs with the capacity to detect changes over time and assess whether management practices are working. Most pertinent to any long-term monitoring program is the overriding question: do we have enough information, or statistical power, to detect changes if changes occur? To assess the power of our benthic video surveys to detect change in coral cover and diversity we varied (1) transect lengths, (2) number of transects, (3) number of frames per transect, and (4) number of data points per frame. Five replicated 50-m transects yielded the most consistent estimates with the highest statistical power, compared with more numerous replicates of shorter (35-m and 15-m) transects. Increasing the number of frames analyzed per 50-m transect yielded greater power than increasing the number of data points per frame, but increasing the number of data points was more effective at estimating species richness. The greatest power of detecting a change in the benthos at each site, within a feasible sampling period, was evident using 5 by 50 m random transects, extracting 60 frames per transect, and analyzing five data points on each frame. This optimal sampling strategy was tested at 23 other long-term monitoring sites and yielded 90% power to detect a 20–30% relative change in dominant benthos abundance estimates (benthos >20% coverage). Our study addresses the sampling unit, accuracy, and ways to improve estimates, but this does not remove the onus of concisely stated questions for monitoring programs pertaining to management. Long-term monitoring of coral reefs requires appropriate strategies that account for the spatial arrangement of organisms and should yield accurate abundance estimates with sufficient statistical power to detect (a desired level of ) change (Aronson et al. 1994, Green and Smith 1997, Page et al. 2001). Many techniques are available to aid researchers in these assessments. Some in situ methods include line-intercept transects (Loya 1978), point quadrats (Chiappone et al. 2001), and belt transects (Littler et al. 1997). In comparison with in situ methods, capturing images underwater increases field efficiency (Foster et al. 1991) and increases the spatial extent of reef coverage, while maintaining a useful taxonomic resolution (Carleton and Done 1995, Vogt et al. 1997). Because of their accuracy, efficiency, and simplicity, video surveys are used in many coral reef monitoring programs (Aronson et al. 1994, Vogt et al. 1997, Page et al. 2001, Rogers and Miller 2001, Brown et al. 2004). Pacific Science (2006), vol. 60, no. 2:177–189 : 2006 by University of Hawai‘i Press All rights reserved 1 Manuscript accepted 30 June 2005. 2 Department of Biological Sciences, Florida Institute of Technology, 150 West University Boulevard, Melbourne, Florida 32901. 3 Commonwealth of the Northern Mariana Islands, Division of Environmental Quality, P.O. Box 501304, Saipan, Mariana Islands 96950. 4 Author for correspondence (e-mail: [email protected]). Coral cover, diversity, and population density estimates will vary depending on the methodology used for data collection (Weinberg 1981, Rogers et al. 2002, Brown et al. 2004). Similarly, for benthic video surveys, relative abundance estimates and the statistical power to detect change will vary according to (1) the number of data points analyzed in each paused (video) frame, (2) the number of frames analyzed in each transect, (3) the number of replicate transects used, and (4) the length of transect. Studies and monitoring programs vary with respect to these criteria but vary also with respect to the question being addressed. For example, it may be useful to analyze a large number of points on any one frame for estimates of coral cover but problematic using the same high number of points to examine the number of colonies because of the increased likelihood of autocorrelation (i.e., counting the same colonies more than once) (Carleton and Done 1995). Similarly, rapid ecological assessments are less concerned about the statistical power of detecting a change than long-term monitoring programs (Andrew and Mapstone 1987). Besides the various techniques available for data collection, monitoring programs must also decide on the degree of site permanency and at what level to randomize. A time series of permanent photoquadrats provides valuable insight into population dynamics but lacks general information regarding the spatial arrangement of organisms within the community as a whole (Green and Smith 1997). One benefit of permanent, repeatedmeasure sampling is the elucidation of recruitment, death, and survival rates of the organisms under question (Wilson and Bossert 1971). However, no confidence measure is available for upscaling the results from a fixed location to an entire reef community, which may bias a spatial interpretation (Davidson 1997). Fully randomized designs provide community characterizations that account for the spatial arrangement of individuals in the communities (Green and Smith 1997). High statistical power achieved through randomized designs means that any measured change is indicative of a change in the community, yet sampling effort may need to be tremendously intensive (Aronson et al. 1994, Green and Smith 1997). At each study site the Australian Institute of Marine Science’s (AIMS) Long-Term Monitoring Program uses a fixed, unstratified sampling design consisting of 5 by 50 m transects, 40 frames within each transect, and five data points from each frame (Page et al. 2001). The Hawai‘i Coral Reef Assessment and Monitoring Program (CRAMP) uses a fixed, stratified sampling design consisting of 10 by 10 m transects, 20 frames per transect, and 50 data points per frame (Brown et al. 2004). Carleton and Done (1995) recognized the potential implications of autocorrelation within each frame and therefore increased the number of frames while analyzing only one data point per frame. Aronson et al. (1994) used a random, unstratified design consisting of 10 by 25 m transects, 50 frames within each transect, and 10 data points for each frame for Caribbean forereefs. Clearly, different reef systems and studies may require different sampling designs, transect lengths, transect replicates, numbers of frames, and numbers of data points to achieve a desired power to detect change of different benthic organisms (Table 1). Here, we present a study that was used to determine an appropriate benthic video survey design for the Commonwealth of the Northern Mariana Islands long-term monitoring program. The goal of this study was to design a scheme that will elucidate a useful sampling unit with sufficient power (90%) to detect a (relative) 20–30% change in dominant benthos cover over time. This study focused on a general approach used to select appropriate transect lengths, transect replicates, number of frames, and number of data points for benthic video surveys at each site. It did not address the need to replicate sites within each combination of time, location, and management regime. Because the study was primarily concerned with the spatial distribution and variance in the cover of dominant organisms, benthic video surveys were complemented with point quadrat coral community surveys (Houk et al. 2005), coral recruitment belt transects, overall diversity assessments, and permanent photo178 PACIFIC SCIENCE . April 2006