Saltmarsh extent and blue-carbon stock mapping
Saltmarshes store disproportionate carbon for their area, yet their extent is poorly mapped and their below-ground stocks invisible from orbit. Multispectral and SAR data together can delineate marsh communities and estimate above-ground biomass, but allometric field work remains non-negotiable for full carbon accounting.
Sensors
- Sentinel-1 C-band SAR (ESA): 10 m ground range resolution in IW mode, 6-day repeat at mid-latitudes with both satellites active. C-band backscatter (5.4 GHz) responds to canopy volume, stem moisture and surface roughness, allowing marsh to be separated from bare mudflat regardless of cloud or tidal illumination. VV/VH cross-polarisation ratios help discriminate dense Spartina swards from sparse pioneer fringe.
- Sentinel-2 MSI (ESA): 10 m visible and NIR bands, 20 m red-edge bands (B5 705 nm, B6 740 nm, B7 783 nm), 5-day revisit with both satellites. Red-edge reflectance is sensitive to chlorophyll concentration and canopy structure, enabling separation of halophyte communities such as Salicornia, Spartina and Puccinellia that have similar visible-band signatures. Cloud cover over coastal margins is a persistent constraint; temporal compositing across multiple acquisitions is standard practice.
- Landsat 8/9 OLI (USGS/NASA): 30 m multispectral, 16-day single-satellite revisit. Lower spatial resolution limits detection of narrow marsh fringes below roughly 30–60 m width, but the archive from 1984 onward (TM/ETM+/OLI) makes Landsat the primary source for long-term marsh loss and recovery trajectories. Coastal aerosol band (443 nm) aids atmospheric correction over water-adjacent pixels.
- ALOS-2 PALSAR-2 (JAXA): L-band SAR at 1.27 GHz, 3–10 m resolution depending on mode. L-band penetrates the canopy and interacts with stems and soil surface, providing a structurally different backscatter signal from Sentinel-1 C-band. The combination of C- and L-band is documented in the literature as improving biomass estimation in dense marsh canopies where C-band saturates at relatively low above-ground biomass densities.
Why the tide is the first problem to solve
Saltmarsh occupies the intertidal zone, which means any single satellite overpass captures a scene somewhere between full submersion and full exposure depending on tidal phase. A pixel classified as open water at high tide and as vegetated marsh at low tide is both correct. Without tidal-phase correction, multi-date mosaics produce systematic misclassification at the marsh-mudflat boundary, which is precisely where ecologists most need accuracy because pioneer fringe communities are both the most dynamic and the most carbon-active per unit area.
The practical fix is to filter acquisition dates to a target tidal window, typically within one hour of mean low water, using tide-gauge records or a global tidal model such as FES2014. Sentinel-1's all-weather capability makes it far easier to accumulate a stack of low-tide acquisitions than optical sensors, which may yield only a handful of cloud-free low-tide images per season. For Sentinel-2, coastal pixels flagged by the Scene Classification Layer as water at high tide but vegetated at low tide provide a useful quality-control mask.
What the red-edge bands actually discriminate
Saltmarsh halophytes are physiologically unusual. High tissue salinity, succulent morphology in Salicornia species, and the dense erect stems of Spartina alterniflora produce spectral signatures that diverge from freshwater wetland vegetation in the red-edge region. Sentinel-2 B5 (705 nm) and B6 (740 nm) capture the inflection point of the red-edge slope, which shifts with chlorophyll content and leaf water. Studies published in Remote Sensing of Environment and similar journals have used red-edge normalised difference indices to separate Spartina-dominated lower marsh from Puccinellia- or Festuca-dominated upper marsh with overall accuracies typically in the 80–90 % range, though performance degrades where communities intergrade or where canopy height is below roughly 15–20 cm.
Landsat OLI lacks dedicated red-edge bands, which is a genuine limitation for community-level mapping. It remains useful for extent mapping at the zone level and for multi-decadal change detection where spectral precision matters less than temporal depth. The two sensor families are complementary rather than interchangeable.
SAR backscatter as a structural proxy
Sentinel-1 C-band backscatter in VH polarisation increases with canopy volume and moisture. Dense Spartina marsh returns backscatter values measurably higher than adjacent mudflat or sparse pioneer vegetation, and the contrast is largest when the marsh canopy is at peak biomass in late summer. This makes SAR classification most reliable in the growing season and less discriminating in winter when many temperate marsh species senesce and the canopy collapses.
ALOS-2 PALSAR-2 L-band adds a different dimension. Because L-band wavelengths (approximately 23 cm) interact with larger structural elements, the sensor is less affected by canopy moisture variability and penetrates further into the vegetation column. Published research has demonstrated that combining C- and L-band backscatter improves above-ground biomass retrieval compared with either band alone, particularly for dense Spartina stands where C-band backscatter saturates at above-ground biomass values that may be well below the actual stock. PALSAR-2 revisit is less frequent than Sentinel-1, typically 14 days in standard observation modes, and tasking is required for specific acquisition geometries.
The ceiling on what satellites can tell you about carbon
Above-ground biomass can be estimated from spectral and backscatter proxies with reasonable confidence once a site-specific or regionally validated allometric model is in place. The more important number for carbon accounting is below-ground, and that is where remote sensing reaches a hard boundary. Saltmarsh soils accumulate organic carbon over centuries at depths that can exceed one metre. This below-ground pool typically represents 70–90 % of total ecosystem carbon stock, depending on species composition, sediment type and accretion history. No current spaceborne sensor can directly observe it.
Field calibration is therefore not optional. A credible carbon stock map requires soil core data collected at a density sufficient to capture spatial variability in organic matter content and bulk density. The satellite layer provides the spatial extrapolation framework: it tells you where each community type occurs and, through above-ground biomass proxies, something about productivity. The field data tells you what the soil beneath each community type actually holds. Skipping the field work and reporting a total carbon stock from satellite data alone would produce a number that is not defensible under any recognised carbon accounting protocol.
Change detection and the loss signal
Saltmarsh loss to erosion, sea-level rise, reclamation and invasive species replacement is detectable in multi-temporal stacks. The Landsat archive from the early 1980s onward provides a baseline that no commercial constellation can match. Vegetation index time series show characteristic decline signatures before complete loss: a drop in red-edge reflectance, reduced VH backscatter in summer, and eventual replacement by the flat, low-backscatter signature of bare mudflat or open water.
Invasion by Spartina anglica in European estuaries, or by Phragmites australis in North American marshes, produces a distinct change signal because the invader typically has higher canopy density and different phenological timing than the native community it displaces. Detecting this requires multi-date imagery across the growing season rather than a single annual snapshot. Sentinel-2's 5-day revisit, even accounting for cloud loss, generally provides enough acquisitions per season to characterise phenological trajectories at the community level.
Satellize applies this multi-sensor approach operationally on open constellations, with the Tonga crop-estimation programme demonstrating the same underlying principle of combining spectral time series with in-country calibration data to produce accountable estimates rather than indicative maps.
Honest limits and what they mean for procurement
Narrow marsh fringes below 20–30 m in width are systematically underestimated at Sentinel-2 and Landsat resolutions. Mixed pixels at the marsh edge contain contributions from both vegetation and water or mudflat, suppressing vegetation indices and causing the classification to assign the pixel to the non-marsh class. Very high resolution commercial imagery (sub-metre to 3 m) can recover some of this fringe area but at substantially higher cost and without the temporal depth of open constellations.
Cloud cover over temperate and tropical coastal zones can reduce usable optical acquisitions to fewer than 10 per year in some regions, making SAR the primary mapping tool for those environments. Even so, SAR classification accuracy for community-level discrimination is generally lower than multispectral classification in clear-sky conditions. A realistic specification for a national saltmarsh mapping programme should include explicit accuracy targets by community class, a tidal filtering protocol, a minimum mappable unit, and a plan for ground-truth collection that is sized to the spatial variability of the target estuary system rather than to budget convenience.
Typical figures
| Spatial resolution (SAR mapping) | 10 m (Sentinel-1 IW mode); 3–10 m (ALOS-2 PALSAR-2 depending on mode) |
| Spatial resolution (multispectral mapping) | 10 m visible/NIR, 20 m red-edge (Sentinel-2); 30 m (Landsat OLI) |
| Revisit (Sentinel-1 constellation) | 6 days at mid-latitudes with both satellites; single-satellite 12 days |
| Revisit (Sentinel-2 constellation) | 5 days at equator with both satellites; cloud loss typically reduces usable optical acquisitions to 10–30 per year in temperate coastal zones |
| Key spectral bands for community discrimination | Sentinel-2 B5 705 nm, B6 740 nm, B7 783 nm (red-edge); B8A 865 nm (NIR plateau); Landsat OLI Band 5 (NIR), Band 4 (red) |
| SAR frequency and polarisation | Sentinel-1: C-band 5.4 GHz, VV+VH; ALOS-2: L-band 1.27 GHz, HH+HV |
| Minimum mappable marsh unit | Approximately 0.1 ha at 10 m resolution; narrow fringes below 20–30 m width are systematically underestimated |
| Archive depth | Sentinel-1: from 2014; Sentinel-2: from 2015; Landsat: from 1984 (TM); ALOS-2: from 2014 |
| Tidal correction requirement | Acquisition filtering to within ±1 hour of mean low water recommended; FES2014 or local tide-gauge records |
| Delivery formats | GeoTIFF classification raster, vector polygon (GeoPackage or Shapefile), per-community biomass raster, change-detection summary report (PDF) |
Analytics Satellize can run
| Tidal-phase-corrected marsh extent map | Multi-date SAR and optical stack filtered to low-tide acquisitions; supervised classification using random forest or support vector machine with red-edge and backscatter features | GeoTIFF and polygon layer with per-community class labels and per-pixel confidence scores |
| Halophyte community discrimination layer | Sentinel-2 red-edge index stack (NDRE, CIre) combined with phenological metrics extracted from time series; field-validated training samples required | Community-class raster at 20 m resolution with accuracy assessment table (confusion matrix, per-class F1 scores) |
| Above-ground biomass density map | Regression of field-measured biomass against Sentinel-1 VH backscatter and Sentinel-2 red-edge indices; published allometric relationships used where site-specific calibration data are unavailable, with explicit uncertainty bounds | Biomass raster (t dry matter ha⁻¹) with uncertainty layer; summary statistics by estuary or management zone |
| Multi-decadal extent change analysis | Landsat OLI/ETM+/TM time series classification at annual or biennial intervals from 1984 to present; change vector analysis to identify gain, loss and community transition | Change map series and tabular area statistics by epoch; PDF summary report with annotated time series plots |
| Carbon stock framework layer (above-ground component) | Above-ground biomass map converted to carbon equivalent using published species-specific carbon fractions; below-ground component flagged as requiring field calibration and not estimated from satellite data alone | Above-ground carbon raster with explicit notation of below-ground data gap; methodology note suitable for submission to carbon registry scoping review |
| Seasonal SAR phenology stack | Monthly median Sentinel-1 VH composites across the growing season to characterise canopy development and senescence timing by community class | 12-band annual phenology raster and per-community seasonal backscatter profile charts |
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.