Coral reef habitat mapping for reef fisheries management

Multispectral and hyperspectral satellites can distinguish coral, algae, rubble and sand on shallow reefs down to roughly 15–20 m in clear water, producing habitat quality maps that feed directly into stock assessments and no-take zone design.

Sensors

Sentinel-2 MSI: 10 m resolution in the visible bands most useful for water-column penetration (bands 2–4, centred at 490, 560 and 665 nm); 5-day revisit at the equator with two satellites. Free and open archive from 2015. The coastal aerosol band (443 nm) aids water-column correction but sits at 60 m native resolution.
Planet SuperDove: 3 m resolution across eight bands including a dedicated coastal blue at 444 nm and a green I band at 531 nm, both valuable for benthic discrimination. Daily revisit over most reef regions. Commercial tasking; archive depth varies by site.
PRISMA (ASI): Hyperspectral imager with 30 m spatial resolution and 239 contiguous bands from 400–2500 nm, giving spectral resolution of roughly 10 nm. The dense visible sampling allows endmember unmixing that separates substrate types multispectral sensors conflate. Tasked on request; revisit is not systematic.
Landsat 8/9 OLI: 30 m resolution with a coastal/aerosol band at 443 nm. Eight-day revisit with both satellites combined. The long archive (Landsat 8 from 2013, Landsat 9 from 2021) supports change detection over decadal timescales. Spatial resolution limits utility on narrow reef structures.

What the water column does to your signal

Before any photon from the reef floor reaches the satellite sensor, it has passed through the atmosphere, reflected off or transmitted through the sea surface, and then travelled down through the water column and back up again. Water absorbs red and near-infrared light almost completely within a few metres. Blue and green wavelengths penetrate furthest, which is why reef mapping is a blue-green problem.

The practical depth limit in optically clear, oligotrophic reef water is roughly 15–20 metres for Sentinel-2 class sensors. In practice most productive reef fisheries habitat sits shallower than this, so the constraint is less severe than it sounds. But any turbidity, suspended sediment or phytoplankton bloom can cut that limit to 5 metres or less. Surface sun glint, where the sea surface acts as a mirror, is a separate problem that can saturate visible bands entirely and must be corrected before classification begins. These are not engineering failures; they are physics.

Separating coral from rubble from sand: the spectral argument

Different benthic substrates reflect differently in the blue-green window. Live coral with healthy zooxanthellae has a characteristic reflectance shape influenced by chlorophyll-a absorption near 440 nm and 670 nm. Bleached coral, which has expelled its symbiotic algae, reflects more like bare carbonate and can be confused with sand without careful spectral decomposition. Macroalgae, turf algae and crustose coralline algae each have distinct signatures, though separating them reliably requires either hyperspectral data or very high spatial resolution.

Multispectral sensors such as Sentinel-2 and SuperDove work by exploiting the shape of the reflectance curve across a limited number of broad bands. Supervised classification using field-collected ground-truth spectra or airborne validation data can typically distinguish four to six benthic classes at useful accuracy levels. Hyperspectral sensors such as PRISMA allow spectral unmixing, treating each pixel as a mixture of pure endmembers and estimating fractional cover. This is more powerful but demands more calibration data and computing. Neither approach produces a perfect map. Accuracy figures in the published literature for multispectral reef classification range widely, from around 65% to above 85% overall accuracy, depending on site clarity, class scheme and validation method.

From substrate class to fisheries relevance

A classified benthic map is only useful for fisheries management if it connects to ecological function. The link is reasonably well established. Structurally complex live coral provides refuge and foraging habitat for reef-associated fish. Rubble fields and algae-dominated zones, often the legacy of bleaching events or physical disturbance, support lower fish biomass and different species assemblages. Sand channels are important for some species but are generally lower-productivity habitat per unit area.

Habitat quality indices derived from satellite maps can be constructed by weighting benthic classes according to their known fish-biomass associations, drawing on published relationships from field ecology. These indices feed into stock assessment models as spatial covariates, improving estimates of available habitat and informing where no-take zones are likely to protect the most productive areas. The satellite data does not replace underwater surveys; it provides spatial context at scales and frequencies that in-water surveys cannot match.

Change detection: bleaching scars and recovery trajectories

A single-date map is useful. A time series is far more so. Sentinel-2's archive from 2015 and Landsat's from 2013 allow analysts to track reef condition through bleaching events, cyclone impacts and recovery periods. Mass bleaching events, such as those associated with elevated sea-surface temperatures in 2016, 2017 and 2020 across the Indo-Pacific, left detectable signatures in multispectral imagery as shifts from the spectral signature of pigmented coral toward the brighter, flatter reflectance of bleached or dead substrate.

Change detection requires careful attention to image normalisation, since sun angle, atmospheric conditions and tidal stage all affect apparent reflectance. Comparing images taken under different tidal conditions introduces apparent depth changes that can be misread as substrate change. Best practice is to build composites from images acquired near the same tidal phase, or to apply empirical water-column corrections that account for depth variation explicitly. Neither is trivial at operational scale.

Honest limits and where field data remains essential

Satellite reef mapping has real limits that no amount of processing can fully overcome. Depth penetration is the most fundamental. Deeper reef slopes, which can be important for fish refugia during thermal stress events, are invisible to passive optical sensors. Turbid inshore reefs, common near river mouths or areas with chronic sediment pressure, may be unmappable by any optical method. Night-time or subsurface habitat structure is outside scope entirely.

Classification accuracy degrades when the number of target classes increases. Distinguishing live coral from bleached coral from crustose coralline algae in a single multispectral image is at the edge of what is reliably achievable without hyperspectral data and dense ground truth. Analysts should be explicit about class definitions and validation uncertainty in any product delivered to fisheries managers. A map that overstates its own precision can lead to worse management decisions than no map at all.

Satellize applies these methods on open constellations including Sentinel-2 and Landsat, with commercial tasking added where higher resolution or hyperspectral coverage is warranted. The workflow is similar in structure to the spatial analytics approach used in the Kingdom of Tonga crop-estimation programme, adapted to the spectral physics of the marine environment.

Designing a reef mapping programme that actually gets used

The most common failure mode in reef remote sensing is producing a classified map that fisheries managers cannot integrate into their existing workflows. Useful outputs are those that arrive in formats compatible with the GIS tools already in use, carry explicit uncertainty estimates, and are updated on a schedule that matches management decision cycles.

For no-take zone design, a single high-quality baseline map with annual updates is often sufficient. For bleaching response, near-real-time monitoring during thermal stress periods requires a different data pipeline, pulling frequent Sentinel-2 acquisitions and flagging anomalous spectral shifts automatically. These are different products with different architectures. Specifying which one is needed before commissioning the analysis saves considerable rework.

Typical figures

Typical spatial resolution	3 m (Planet SuperDove), 10 m (Sentinel-2 visible bands), 30 m (Landsat 8/9 OLI, PRISMA)
Revisit frequency	Daily (Planet SuperDove); 5 days at equator (Sentinel-2, two satellites); 8 days (Landsat 8+9 combined); on-request (PRISMA)
Effective depth penetration	Up to ~15–20 m in clear oligotrophic water; typically <5 m in turbid or sediment-laden conditions
Spectral bands used	Coastal blue (~443–444 nm), blue (~490 nm), green (~531–560 nm), red (~665 nm); hyperspectral: 400–700 nm contiguous at ~10 nm resolution (PRISMA)
Minimum mappable feature	Patch reefs of roughly 10–30 m diameter at Sentinel-2 resolution; ~5–10 m at SuperDove resolution; sub-pixel unmixing can detect smaller fractional contributions
Typical classification accuracy	65–85% overall accuracy for 4–6 benthic classes (published range, multispectral); higher with hyperspectral and dense ground truth
Archive depth	Sentinel-2: from 2015; Landsat 8: from 2013; Landsat 9: from 2021; Planet: variable by site
Cloud and glint sensitivity	Cloud cover renders acquisitions unusable; sun glint requires correction or scene exclusion; tropical reef regions often have persistent cloud in wet seasons
Delivery formats	GeoTIFF classified map, GeoPackage or Shapefile habitat polygons, CSV habitat quality index by management zone, PDF technical report with validation statistics

Analytics Satellize can run

Benthic habitat classification map	Supervised maximum-likelihood or random-forest classification on atmospherically and water-column corrected reflectance; trained on field spectra or airborne validation data	GeoTIFF raster and polygon GIS layer with 4–6 benthic classes and per-class confidence scores
Habitat quality index by management zone	Weighted summation of benthic class fractions using published fish-biomass association weights; aggregated to user-defined zone boundaries	CSV and GIS layer with index scores per zone, suitable for direct input to stock assessment spatial covariates
Bleaching and disturbance change detection	Bi-temporal or multi-temporal differencing of normalised reflectance in coral-sensitive bands; anomaly flagging against baseline composite	Change map GeoTIFF showing areas of spectral shift consistent with bleaching or substrate loss, with area statistics by reef zone
No-take zone design support layer	Overlay of habitat quality index with existing fishing pressure data and bathymetry; spatial optimisation to maximise high-quality habitat within candidate zone boundaries	GIS polygon layer of candidate no-take zone boundaries with habitat quality summary table and PDF briefing note
Spectral unmixing fractional cover (hyperspectral)	Linear or non-linear spectral unmixing of PRISMA imagery against a library of benthic endmember spectra; produces fractional cover per pixel for coral, algae, sand and rubble	Multi-band GeoTIFF of fractional cover per class with uncertainty estimates; suitable for sites where multispectral classification is insufficient
Annual reef condition time series	Consistent re-classification of Sentinel-2 or Landsat archive at annual intervals; tidal-phase normalisation applied to reduce apparent depth artefacts	Time-series GIS stack with summary statistics per reef zone and trend plots in PDF report

Sources

Who does the work

We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.