Comparing Coral Reef Survey Methods Reveals Effects on Benthic and Substrate Composition

Coral Reef Surveys Can Give Conflicting Answers—Unless Scientists Calibrate the Methods

A coral reef can appear healthier or more degraded depending on how scientists measure it. A diver tracing a line across the seafloor, recording organisms at fixed points, may produce a substantially different picture from a diver photographing the same reef for later computer-assisted analysis. Now, a study spanning 56 sites across Indonesia has quantified those discrepancies and developed a statistical way to translate older survey data into the digital format increasingly used today. The approach could help researchers recover decades of valuable reef-monitoring records that would otherwise be difficult to compare with modern observations.

The problem is more than a technical inconvenience. Coral reefs are declining under pressure from marine heatwaves, pollution, sedimentation, overfishing and coastal development, making long-term monitoring essential for conservation decisions. Detecting whether a reef is recovering or deteriorating requires measurements that can be compared through time. Yet monitoring programmes often change methods as cameras become cheaper, image-analysis software improves and national agencies adopt new protocols. If one method systematically records more hard coral, algae or rubble than another, an apparent ecological trend may partly reflect a change in measurement rather than a change underwater. The researchers therefore compared three widely used approaches: line-intercept transects, point-intercept transects and underwater photo transects.

The surveys were conducted by scuba divers between 2013 and 2014 at four Indonesian locations: Buton, Gili Matra, Natuna and Bintan. Together, the sites represented reefs inside and outside the Coral Triangle and exposed the analysis to a wide range of environmental conditions. At each site, the three methods were applied concurrently along reef slopes at approximately 5 metres depth, with transects laid parallel to the coastline. The researchers classified 11 benthic and substrate categories, including hard coral, dead coral, dead coral overgrown with algae, soft coral, sponges, macroalgae, rubble, sand, silt and rock. Each method was reduced to a single composition for each site, leaving 56 comparable observations per method rather than treating individual points or images as independent samples.

The line-intercept method records the length of each organism or substrate type directly beneath a transect tape. In this study, divers surveyed three 10-metre sections, producing 30 metres of total line data. Point-intercept transects, by contrast, record whatever lies beneath the tape at fixed intervals; here, divers sampled 100 points spaced 50 centimetres apart along a 50-metre transect. The underwater photo method covered a similar 50-metre transect but used a 44-by-58-centimetre frame placed every metre, alternately on either side of the line. Fifty images were collected from each transect, and 30 randomly positioned points were analysed in each image using Coral Point Count with Excel extensions, or CPCe. The resulting 1,500 image points gave the photo method a much denser two-dimensional sample than either diver-recorded approach.

That difference in sampling geometry left a clear statistical fingerprint. Underwater photo transects generally produced the highest estimated cover for most categories, point-intercept transects gave intermediate values and line-intercept transects produced the lowest. Photoquadrats can capture small, patchy organisms scattered across an area, whereas a line samples only the narrow path directly beneath it. A rare sponge or fragment of rubble may occupy several points in an image while never touching the transect tape. The strongest reversal involved hard coral. The model estimated average hard-coral cover at 43.65 per cent for line-intercept transects, 36.71 per cent for point-intercept transects and 28.18 per cent for underwater photo transects. The researchers suggest that a line often follows the physical relief of a reef, which is itself shaped by hard corals, increasing the chance that the tape intersects large coral surfaces. Observer choices in laying the line may reinforce that bias.

To analyse these differences, the team used Bayesian Dirichlet regression, a model designed for compositions whose components must add up to a fixed total. Reef-cover data are compositional: if sand occupies a smaller percentage of a quadrat, the percentages of other categories must rise even if their absolute areas have not changed. Ordinary statistical models can mistake this mathematical constraint for a biological relationship and may generate impossible predictions, such as negative cover. The Dirichlet distribution instead models proportions on the appropriate “simplex,” the mathematical space in which all components are non-negative and sum to one. The researchers also included location and site nested within location as hierarchical effects, allowing the expected composition to vary with environmental setting while borrowing information across sites. A leave-one-out cross-validation analysis showed that the complete model, including both survey method and spatial effects, predicted the data best. A more elaborate zero-and-one-inflated model performed worse, suggesting that most zeros represented sampling scarcity rather than true, structural absence.

The researchers then built conversion models that used line-intercept or point-intercept compositions to predict what the same observations would have looked like under the underwater photo protocol. The results were strongest for the line-to-photo conversion, where errors fell sharply across nearly all categories. For hard coral, mean absolute error declined from 13.4 percentage points in the original line-intercept data to 1.39 points after conversion, while root mean squared error fell from 15.0 to 1.79. The squared correlation between predicted and observed values rose from 0.84 to 0.99. Dead coral with algae showed a similar improvement, with mean absolute error falling from 9.79 to 1.02. Point-to-photo conversion also substantially improved agreement. For hard coral, mean absolute error dropped from 7.76 to 2.03, root mean squared error from 9.55 to 2.46 and squared correlation from 0.78 to 0.97. Categories present at very low percentages, particularly silt, dead coral and rock, remained less certain because many observations were zero or near zero.

The most important test asked whether the converted data matched the overall reef community, not merely individual categories. A multivariate PERMANOVA found a significant difference between the original line-intercept and photo-transect compositions, confirming that the methods produced measurably different ecological portraits. After conversion, however, the line-derived predictions were statistically indistinguishable from the observed photo data. The same pattern appeared for point-intercept data: the original comparison was marginally different, while the converted composition closely matched the photo-based composition. The researchers emphasize that this does not mean photographs eliminate human judgement. Divers still classify organisms, and difficult images can be ambiguous. Instead, the model estimates and corrects systematic differences associated with sampling design, while retaining uncertainty around every predicted proportion.

The framework could allow conservation agencies to combine historical field records with newer digital surveys without discarding the past or pretending that all methods are equivalent. Indonesia is an especially valuable test case because its national reef-monitoring archive contains data collected under multiple protocols, while its reefs span distinct ecological regions and human-pressure gradients. The conversion models can be refitted as observations from additional sites and locations become available, potentially improving their regional reliability. They can also be applied to entirely new locations by omitting the spatial random effects, although predictions will be less precise. The method is not a substitute for consistent monitoring, and the authors caution that low-abundance categories remain difficult to estimate. But by translating dive-slate measurements into a common, photo-equivalent composition, the approach offers a practical bridge between old and new surveys—one that could make subtle changes in reef condition easier to detect before they become ecological crises.