Seagrass and coastal nursery habitat mapping for fisheries management
Seagrass meadows and mangrove fringes are the nursery infrastructure behind coastal fisheries, yet most nations cannot map them reliably. Satellite optical physics can reach the seabed in clear water to roughly 15–25 metres, but only after correcting for what the water column does to the signal first.
Sensors
- Sentinel-2 MSI: 10 m resolution in the visible bands (443, 490, 560, 665 nm) most useful for benthic mapping; 5-day revisit at the equator with two satellites. The coastal aerosol band at 443 nm is particularly important for water-column depth estimation. Free and open archive from 2015.
- Landsat 8/9 OLI: 30 m resolution across comparable visible bands; 16-day revisit per satellite, 8-day combined. The ultra-blue coastal/aerosol band (435–451 nm) supports the same depth-invariant index calculations. Archive extends to 1984 for Landsat 5/7, giving decades of change context.
- Maxar WorldView-3: 31 cm panchromatic, 1.24 m multispectral at nadir, with eight VNIR bands including a dedicated coastal band. At this resolution, individual seagrass patch boundaries, sparse-versus-dense canopy gradients and small clearings within meadows become distinguishable. Tasked commercially; not free.
- Planet SuperDove: 3 m resolution, eight bands including two blue bands (444 nm and 490 nm) suited to water-column correction. Near-daily revisit globally. Useful for change detection between WorldView baseline maps and for cloud-gap filling in time series.
What a seagrass meadow looks like to a satellite, and why it is complicated
A satellite sensor above a coastal lagoon does not see the seabed directly. It receives a mixed signal: sunlight reflected from the water surface, light scattered within the water column, and a small fraction that has travelled all the way down to the benthos and back. Seagrass absorbs strongly in the red and near-infrared, but water absorbs those wavelengths too, so by the time you are looking at even 3 metres of water, the seagrass NIR signal has been extinguished. Benthic mapping from orbit must therefore work almost entirely in the blue and green portions of the spectrum, where water is least absorptive.
The standard approach is the depth-invariant index, first formalised by Lyzenga in the 1970s and still the workhorse of the field. By exploiting the known ratio of attenuation coefficients between two blue-green band pairs, analysts can produce a transformed image in which brightness differences reflect bottom type rather than depth variation. The method assumes optically deep water is available nearby for calibration and that the water column is horizontally homogeneous, neither of which is always true in turbid estuaries or around river mouths. Honest practitioners treat depth-invariant indices as a first-pass classification tool, not a finished map.
The 15 to 25 metre ceiling, and what sits above it
In clear oceanic water, blue light can survive a return trip from roughly 25 metres depth with enough signal-to-noise for classification. In the more typical coastal water encountered around most seagrass systems, the practical limit is closer to 10 to 15 metres. Turbidity from river discharge, resuspended sediment or phytoplankton blooms can compress that window to just a few metres on bad days. This is not a solvable problem with better algorithms; it is physics. Managers should treat satellite seagrass maps as reliable for the intertidal to roughly 10 m zone, indicative to perhaps 15–20 m in clear-water settings, and blind beyond that.
What falls within that window is nonetheless ecologically critical. The bulk of productive seagrass habitat globally sits in shallow, clear-water coastal zones: the Indo-Pacific, the Caribbean, the Mediterranean, the Persian Gulf and the Australian coast. These are precisely the areas where juvenile fish and invertebrate densities are highest and where habitat loss translates most directly into recruitment failure for commercial stocks. The depth limit is a constraint, not a disqualification.
From raw reflectance to a habitat map that a fisheries manager can use
The processing chain has several distinct steps. Atmospheric correction comes first: converting top-of-atmosphere radiance to surface reflectance, removing aerosol and Rayleigh scattering effects. For coastal water, standard land-oriented corrections are often inadequate; processors designed for aquatic environments, such as ACOLITE (developed at the Royal Belgian Institute of Natural Sciences and published in peer-reviewed literature) or the C2RCC processor available in the ESA SNAP toolbox, handle the adjacency effects and water-leaving radiance more accurately.
After atmospheric correction, the water-column correction is applied to produce depth-invariant band ratios. These feed a supervised classification, typically a random forest or support vector machine trained on field-validated ground-truth points: snorkel or dive transects, drop-camera frames, or airborne lidar where available. The output is a benthic habitat map with classes such as dense seagrass, sparse seagrass, macroalgae, coral rubble, sand and deep water. Accuracy assessments in published studies using Sentinel-2 over clear-water sites commonly report overall accuracies of 70–85 %, with seagrass-versus-sand confusion being the dominant error source at sparse canopy densities below roughly 20 % cover.
Mangrove fringes: easier to detect, harder to quantify as nursery habitat
Mangroves sit above the waterline, which makes them spectrally straightforward. Their dense canopy produces a strong NIR response that separates cleanly from open water and bare sediment in any multispectral image. Sentinel-2 at 10 m can map mangrove fringe extent to a boundary precision of roughly one to two pixels, meaning features narrower than about 10–20 m may be missed or underestimated in area.
The harder question is functional connectivity: which mangrove patches are actually adjacent to seagrass beds, and which fish species use both habitats across their juvenile life stage? That requires combining the habitat map with hydrodynamic connectivity modelling, which satellite data alone cannot answer. What satellite data can do is provide the habitat geometry layer that such models need as input, updated annually or after disturbance events such as cyclones, to track whether the structural nursery network is intact.
Change detection and the long archive
A single-date habitat map has limited management value. The power of the satellite record is the archive. Landsat data from the 1980s and 1990s, combined with more recent Sentinel-2 and commercial imagery, can reconstruct decades of seagrass change in areas where no field surveys were conducted. This matters because many seagrass losses are slow and cumulative, driven by chronic eutrophication or gradual increases in water turbidity, rather than acute events. A manager who sees only the current state cannot distinguish a stable system from one that has already lost 60 % of its extent.
Change detection between epochs requires careful normalisation for seasonal phenology. Seagrass canopy density varies with season, and a summer-to-winter comparison will show apparent loss that is not real. Best practice is to compare same-season imagery across years, or to build a time series dense enough to separate phenological cycles from genuine trend. Planet SuperDove's near-daily cadence makes dense time series feasible for the first time at 3 m resolution, though the shorter archive (from 2016 for early Dove, with SuperDove from around 2021) limits long-term trend analysis.
Satellize runs seagrass and coastal habitat analytics on open Sentinel and Landsat archives and adds commercial tasking where boundary precision demands it. The workflow is the same one underlying the Tonga crop-estimation programme: multi-epoch classification, change attribution and delivery as GIS-ready layers with confidence intervals attached.
What the map cannot tell you, and what to do about it
Satellite mapping tells you where structured habitat exists and how its extent has changed. It does not measure seagrass shoot density, below-ground biomass, sediment carbon stock or the actual abundance of juvenile fish using the habitat. Those require field measurement. The satellite product is most defensible when used to prioritise field survey effort, to scale point measurements across a broader area, and to monitor whether management interventions are holding the habitat stable between survey campaigns.
Cloud cover is a persistent problem in tropical coastal zones, which is precisely where the most biodiverse seagrass systems occur. A single Sentinel-2 pass over a monsoon-affected coastline may be cloud-free only a handful of times per year. The practical solution is to build composites from multiple passes, accepting that the composite represents a temporal average rather than a single date. In persistently cloudy regions, the honest answer is that optical satellite mapping has a seasonal window, and managers should plan field validation campaigns to coincide with that window.
Typical figures
| Spatial resolution (open data) | 10 m (Sentinel-2 visible bands), 30 m (Landsat 8/9 OLI) |
| Spatial resolution (commercial) | 1.24 m multispectral / 0.31 m pan (WorldView-3); 3 m (Planet SuperDove) |
| Revisit cadence | 5 days (Sentinel-2, two satellites); 8 days combined (Landsat 8+9); near-daily (Planet SuperDove) |
| Effective depth limit for benthic mapping | ~10–15 m in typical coastal water; up to ~25 m in exceptionally clear oceanic water |
| Key spectral bands for water-column correction | Coastal/aerosol (~443 nm) and blue (~490 nm); green (~560 nm) for band-ratio depth invariance |
| Minimum mappable patch size | ~0.01 ha at WorldView-3 resolution; ~0.1 ha at Sentinel-2 10 m; ~1 ha at Landsat 30 m |
| Archive depth | Sentinel-2 from 2015; Landsat from 1984 (Landsat 5); WorldView series from ~2007 |
| Typical classification accuracy (published studies, clear water) | 70–85 % overall accuracy for seagrass vs. other benthic classes using Sentinel-2 |
| Delivery formats | GeoTIFF habitat classification raster, vector polygon shapefile/GeoPackage, change-detection report |
| Processing latency (archive analysis) | Days to weeks depending on scene count, cloud screening and validation requirements |
Analytics Satellize can run
| Benthic habitat classification map | Atmospheric correction (ACOLITE or C2RCC) + Lyzenga depth-invariant band ratios + random forest supervised classification trained on field-validated points | GeoTIFF and polygon GIS layer with habitat classes (dense seagrass, sparse seagrass, macroalgae, sand, rubble, deep water) and per-class confidence score |
| Multi-epoch seagrass change detection | Same-season image compositing across Sentinel-2 and Landsat archives, normalised difference classification comparison, area change statistics per habitat class | Change map showing gain, loss and stable zones per epoch pair, with tabular area statistics and trend chart |
| Mangrove fringe extent and connectivity layer | NIR-based threshold segmentation on Sentinel-2 or Planet imagery, vectorised fringe polygons, spatial adjacency analysis to seagrass polygons | Vector polygon layer of mangrove extent with adjacency flag indicating proximity to mapped seagrass, updated annually |
| Water-column turbidity screening | Satellite-derived turbidity index from red and NIR bands (published NECHAD or similar empirical relationships) to flag scenes or zones where benthic mapping is unreliable | Per-scene turbidity mask appended to habitat map, with documented reliability zones |
| High-resolution patch boundary delineation | WorldView-3 or Planet SuperDove object-based image analysis (OBIA) for fine boundary precision, merged with Sentinel-2 time series for change context | Sub-2 m precision habitat boundary polygons suitable for permit boundary definition and MPA zoning input |
| Nursery habitat connectivity index | Graph-theoretic patch connectivity analysis on classified habitat polygons, using published least-cost path or proximity-based connectivity metrics | Tabular and mapped connectivity scores per habitat patch, identifying critical stepping-stone patches for juvenile fish movement |
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.