Reading path: Coastal field methods →

Start with the quantity the image can support

A photoquadrat is a photograph taken within a defined sampling frame, often along a transect. Analysts place points over the image and classify the benthic feature under each point. The resulting proportion is a point-sampled estimate of cover for the categories and sampled locations in that survey. NOAA reef-monitoring programs use image-based surveys for benthic cover, while their protocols also distinguish these from coral-demographic measurements such as colony counts, size and condition. [1]

That distinction matters. Cover is the share of sampled points assigned to a category, not the number of colonies, species richness, health, or total living area. A single large coral colony may cover many points; a small colony may cover none. If the question concerns colony abundance or disease, a separate demographic method is needed.

A worked point-count example

Suppose an imaginary survey samples 30 images and assigns 10 random points in each image. If 84 of the 300 points fall on hard coral, the pooled estimate is 84 ÷ 300 × 100 = 28% hard-coral cover. The calculation is simple; its meaning depends on the sampling design. NOAA’s monitoring protocol describes a Pacific example with 30 images along a transect and 10 points per image, while its restoration guide likewise defines point-based cover as the relevant category points divided by all sampled points. [1] [2]

The 300 points are not necessarily 300 independent pieces of reef. Ten points in one photograph share the same small patch of habitat, and nearby images along one transect may also resemble one another. Do not report an artificially precise uncertainty as though each point were an independently randomized reef location. Keep replicate transects, sites and strata visible; calculate uncertainty at the level supported by the design.

For other research explanations and observation methods, browse science notes.

Sampling layout defines the map behind the percentage

A fixed transect helps revisit the same location and ask how that sampled patch changes. Randomly selected sites can support broader regional estimates when the selection frame and weighting are appropriate. NOAA’s Pacific program describes stratified random sampling because coral abundance varies with depth, topography and patchiness, and it uses fixed sites for comparisons through time. These designs answer different questions; a convenient dive location cannot stand in for a probability sample of a region. [1]

The NCEI archive for a Hawaii Island survey illustrates how much context belongs beside a percentage: six locations, paired shallow and deep habitats, ten fixed transects, with 25 random points per image annotated by staff. That is a particular monitoring design, not a universal recipe. Its sample frame, depth pairing and annotation procedure tell readers what its estimates cover. [4]

For a repeat survey, preserve transect identity and endpoints, depth or habitat stratum, image spacing, frame dimensions, camera orientation, and which side of the line was sampled. If the aim is change, use the same design and season where practical, and document departures. A changed frame or image interval can change the sampled patch even when the site name remains the same.

Classification is a measurement decision

Before scoring, define the categories and their hierarchy. A broad tier might separate hard coral, algae and sediment; a finer tier might distinguish growth forms or taxa. NOAA’s image-analysis guidance describes several levels of identification, from functional groups through finer taxonomic or morphological categories. Fine labels can be informative, but only if photographs resolve the features and annotators can apply definitions consistently. [3]

Blurry edges, shadows, motion, glare, sediment, overgrowth and organisms hidden beneath others can make the point beneath a pixel genuinely uncertain. Keep an “unknown” or unresolved category where the protocol permits it instead of forcing a confident label. Record calibration images, annotator decisions and any review process. Machine-assisted labeling can speed analysis, but NOAA describes automation as dependent on sufficient data; it does not remove the need to check the output against the study’s categories and imagery. [3]

Separate repeatability from ecological explanation

If estimated coral cover falls between two surveys, the result supports a change in the sampled metric only after checking comparable locations, effort, image quality and classification. It does not identify the cause. Bleaching, storms, disease, recruitment, algal growth or changes in the sampling footprint are hypotheses that require their own observations and evidence. The NOAA restoration guide emphasizes replication and notes that transect area limits what the survey can capture; if restored colonies are missed by the sampled lines, the design may need review. [2]

A useful report therefore gives the estimate with its denominator and sampling unit: for example, “28% of classified points across 30 images on the designated transect were hard coral,” followed by how images and points were selected, how uncertain labels were handled, and what replicates support the comparison. This sentence is more modest than “the reef is 28% coral,” but it is also more reproducible.

A compact field and analysis record

Save the survey date, site and coordinates, transect or stratum, depth, frame and camera setup, image sequence, point-generation method, classification guide and version, annotators or model workflow, unresolved points, quality exclusions, and the number of usable images and points. Keep original images linked to the derived table so another analyst can inspect a disputed classification. For summaries, present site-level or replicate variation alongside pooled cover rather than letting one aggregate percentage hide patchiness.

Photoquadrats are useful because a stored image can be checked again, scored at more than one taxonomic level, and compared under a documented method. Their strength is a traceable sample of visible seafloor. The image does not automatically become a reef census: the sampling design, category definitions and independent observations determine how far the conclusion can travel.

Sources & further reading

  1. National Coral Reef Monitoring Program: Protocols ↗
  2. Coral Reef Restoration Monitoring Guide: Methods to Evaluate Restoration Success from Local to Ecosystem Scales ↗
  3. Analysis of Benthic Survey Images via CoralNet: A Summary of Standard Operating Procedures and Guidelines ↗
  4. Benthic Percent Cover Derived from Photoquadrat Images Taken Along Fixed Transect Sites around Hawaii Island (NCEI Accession 0288837) ↗