Kelp forest canopy extent and seasonal dynamics mapping

Surface-canopy kelp produces a strong near-infrared and red-edge reflectance signal that open water cannot replicate. Sentinel-2 and Landsat imagery can resolve this contrast down to roughly 10–30 m, but cloud cover, wave submergence and turbid plumes each impose real detection limits that any honest monitoring programme must account for.

Sensors

Sentinel-2 MSI: 10 m resolution in the red and near-infrared bands (B4, B8), 20 m in the red-edge bands (B5, B6, B7). Five-day revisit at the equator, shorter at higher latitudes. The red-edge bands are particularly useful for separating kelp from other floating vegetation and for detecting partial-canopy conditions where NIR alone saturates.
Landsat 8/9 OLI and OLI-2: 30 m multispectral resolution with a 16-day single-satellite revisit (8-day combined for Landsat 8 and 9). The coastal-aerosol band (Band 1) aids water-column correction in turbid nearshore environments. Archive depth back to 1984 (Landsat 5 TM) enables multi-decadal trend analysis of canopy extent.
Planet SuperDove: 3–5 m resolution with eight spectral bands including red-edge and NIR. Near-daily revisit over most coastal areas. Useful for resolving fine-scale canopy patch structure and for capturing short-lived recovery or loss events between Sentinel-2 acquisitions, though commercial licensing is required.
Copernicus Marine Service ocean-colour products: Derived from OLCI (Sentinel-3) and merged multi-sensor products at 300 m to 1 km resolution. Not useful for mapping individual kelp patches, but provides chlorophyll concentration, sea-surface temperature and turbidity fields that contextualise spectral anomalies and flag conditions likely to cause false positives.

Why kelp is visible from orbit at all

Macroalgae such as Macrocystis pyrifera and Ecklonia radiata concentrate at the sea surface when conditions are calm. The canopy fronds contain chlorophyll a and c, fucoxanthin and other pigments that absorb strongly in the red (around 670 nm) and blue (around 440 nm) portions of the spectrum while reflecting in the near-infrared. Open ocean water, by contrast, absorbs NIR almost completely. The resulting spectral contrast is large enough that even moderate-resolution sensors can detect canopy patches with reasonable confidence.

The normalised difference vegetation index (NDVI) and the floating algae index (FAI) both exploit this contrast. FAI, defined using the NIR, red and shortwave-infrared bands, was originally developed for Sargassum detection but transfers well to surface-canopy kelp. Published work using Landsat and Sentinel-2 imagery has mapped kelp extents across California, South Africa, Tasmania and the Falkland Islands, with reported minimum detectable patch sizes in the range of one to a few Sentinel-2 pixels (roughly 100–900 m²) under good atmospheric and sea-state conditions.

Where the physics starts to lie

Canopy reflectance saturates at high biomass densities. Once frond coverage exceeds roughly 80–100% of the water surface within a pixel, additional biomass produces no further increase in NIR reflectance. This means spectral indices can distinguish presence from absence but struggle to estimate biomass density once the canopy is dense. Wet weight or stipe density cannot be read from a spectral index alone.

Turbid plumes introduce a separate problem. Suspended sediment raises NIR reflectance in open water, shrinking the spectral gap between kelp and non-kelp pixels. River outflows, storm resuspension and upwelling fronts can all push turbidity high enough to generate false positives or, conversely, to mask real canopy signal. The Copernicus Marine Service turbidity products provide a practical screening layer: pixels where turbidity exceeds a site-specific threshold are better flagged as uncertain than classified.

Wave action submerges the canopy. Even modest swell (1–2 m significant wave height) can push surface fronds below the water column during acquisition, reducing apparent NIR reflectance by a factor that depends on water clarity and submergence depth. There is no reliable spectral correction for this. It is a missed detection, not a correctable artefact.

Cloud cover and the revisit arithmetic

Kelp forests cluster in temperate and sub-polar coastlines, many of which are persistently cloudy. The Aleutian Islands, southern Chile, western Ireland and the sub-Antarctic islands all have cloud frequencies above 70% in winter months. A five-day Sentinel-2 revisit sounds adequate until cloud screening removes three-quarters of acquisitions, stretching effective revisit to three or four weeks.

The practical response is temporal compositing: selecting the lowest-cloud pixel from a rolling window of acquisitions. A 30-day composite retains spatial detail while averaging out transient cloud and wave effects. The cost is temporal resolution. A rapid canopy-loss event caused by a warm-water anomaly or storm can be missed or mis-dated by several weeks. For early-warning applications, this matters. For annual extent reporting, it is usually acceptable.

Planet SuperDove's near-daily cadence reduces the compositing window substantially, but cloud persistence still sets a hard floor. No optical sensor resolves cloud. SAR (Sentinel-1 C-band) can image through cloud but lacks the spectral bands needed to distinguish kelp from other surface features; it is useful for detecting large floating mats via backscatter anomalies but not for routine canopy mapping.

Building a seasonal dynamics time series

Kelp canopy area follows a seasonal cycle driven by water temperature, nutrient availability and storm frequency. In the northern hemisphere, canopy typically peaks in late summer to autumn and contracts in winter. In the southern hemisphere the cycle is offset by six months, though local oceanography introduces considerable site-level variation.

A useful time series requires consistent atmospheric correction across all acquisitions. The Sen2Cor processor (ESA) and the LaSRC algorithm (USGS) are the standard choices for Sentinel-2 and Landsat respectively. Both are well-documented and freely available. Applying different correction versions across a long archive introduces step-changes that mimic biological trends; version consistency matters more than most analysts acknowledge.

Once a corrected, cloud-screened stack is assembled, change detection can be framed as a simple threshold on a spectral index or as a more sophisticated pixel-wise time-series model. The BFAST (Breaks For Additive Season and Trend) algorithm, developed for forest monitoring, has been applied to kelp time series to separate seasonal oscillation from structural decline. It requires at least two to three years of data to fit the seasonal component reliably.

What the method cannot tell you

Surface-canopy mapping captures only the top layer of a three-dimensional forest. Juvenile kelp, understory species and subcanopy biomass are invisible to any optical sensor operating at these wavelengths. A site can show a healthy canopy signature while the stipe density and age structure beneath are severely degraded. Diver surveys or acoustic methods remain necessary to characterise forest condition below the surface.

Species discrimination is also limited. Sentinel-2 and Landsat cannot reliably separate Macrocystis from Ecklonia or from other large brown algae on spectral grounds alone. Where multiple canopy-forming species co-occur, the map reports total surface macroalgal extent, not species composition. Hyperspectral sensors (such as DESIS on the ISS, or future CHIME) offer finer spectral resolution that may eventually allow species-level discrimination, but published validation at operational scale is still sparse.

Satellize runs kelp-canopy analytics on open Sentinel-2 and Landsat archives, applying consistent atmospheric correction and index thresholds calibrated against available in-situ reference data. The workflow is the same one used to build time-series analytics for the Kingdom of Tonga crop-estimation programme: archive ingestion, correction, compositing and change-layer delivery as GIS-ready outputs.

Matching the sensor to the question

Annual extent reporting for a marine protected area needs long archive depth and consistent methodology more than it needs fine spatial resolution. Landsat is the right backbone: free, globally consistent back to 1984, and well-validated for kelp mapping in published literature.

Detecting a rapid canopy-loss event following a marine heatwave demands short revisit and fine resolution. Planet SuperDove is the appropriate primary sensor, with Sentinel-2 as a secondary check. Expect gaps. Expect uncertainty in the loss date. Report both.

Turbidity screening using Copernicus Marine Service products should be standard practice at any site with significant freshwater input or upwelling. Skipping it inflates apparent canopy area in ways that are difficult to detect without ground truth.

Typical figures

Typical spatial resolution	10 m (Sentinel-2 NIR/red), 30 m (Landsat OLI), 3–5 m (Planet SuperDove)
Revisit cadence	5 days (Sentinel-2, equatorial); 8 days (Landsat 8+9 combined); near-daily (Planet SuperDove)
Effective revisit after cloud screening	2–6 weeks in persistently cloudy temperate and sub-polar coastal zones
Key spectral bands	NIR (Sentinel-2 B8, ~833 nm; Landsat B5, ~865 nm), red-edge (Sentinel-2 B5–B7, 705–783 nm), red (B4/B4, ~665 nm), SWIR for FAI computation
Minimum detectable canopy patch	Approximately 100–900 m² under low-turbidity, calm sea-state conditions at Sentinel-2 resolution; larger patches reliably detectable at Landsat 30 m
Biomass density sensitivity	Spectral indices saturate above roughly 80–100% fractional canopy cover per pixel; sub-canopy biomass not detectable
Archive depth	Landsat: 1984 to present (TM, ETM+, OLI, OLI-2); Sentinel-2: 2015 to present
Atmospheric correction	Sen2Cor (ESA) for Sentinel-2; LaSRC (USGS) for Landsat; version consistency across archive is critical for trend analysis
Primary confounders	Cloud cover, wave-induced submergence, turbid plumes, floating Sargassum or foam
Typical delivery formats	GeoTIFF canopy-extent rasters, vector patch polygons (GeoJSON/Shapefile), time-series CSV of area statistics

Analytics Satellize can run

Annual canopy-extent map	Cloud-screened seasonal composite with NDVI or FAI threshold, calibrated against available in-situ or high-resolution reference data	GeoTIFF raster and vector polygon layer of classified canopy area with per-patch area statistics
Multi-decadal trend report	Landsat archive time series (1984 onwards) with consistent LaSRC atmospheric correction and linear or piecewise trend fitting	Annual area statistics table and trend-line report (PDF + CSV), suitable for MPA management reporting
Seasonal phenology profile	Monthly composites from Sentinel-2 stack; harmonic or BFAST decomposition to separate seasonal cycle from inter-annual change	Site-level phenology curves and anomaly flags where canopy area deviates from expected seasonal range
Rapid-loss alert	Near-real-time Sentinel-2 acquisition monitoring with threshold-based change detection against rolling 30-day baseline	Alert notification with mapped loss extent and estimated area, triggered within 5 days of a cloud-free acquisition showing significant canopy reduction
Turbidity-screened uncertainty layer	Co-registration of Copernicus Marine Service turbidity products with canopy classification; pixels above site-specific turbidity threshold flagged as uncertain	Canopy map with three-class confidence layer (confident kelp, uncertain, confident non-kelp)
Patch-connectivity metrics	Vector patch analysis using classified canopy polygons; nearest-neighbour distance, patch area distribution and fragmentation indices computed per season	GIS layer with per-patch attributes and summary statistics table for each annual epoch

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.