Mangrove extent and canopy-structure mapping

Mapping mangrove extent demands tidal-phase-aware optical composites and L-band SAR that sees through canopy to woody structure. Together, Sentinel-1, ALOS-2 PALSAR-2, Sentinel-2 and Landsat resolve stand boundaries and canopy architecture that single-sensor approaches routinely miss.

Sensors

ALOS-2 PALSAR-2: L-band (1.27 GHz) SAR with HH and HV polarisations at 10–25 m resolution. L-band wavelength (~23 cm) penetrates the mangrove canopy and interacts with trunks and large branches, producing a double-bounce backscatter signature that distinguishes woody mangrove from herbaceous marsh. Provides the primary biomass-sensitive layer in the JAXA Global Mangrove Watch.
Sentinel-1 SAR: C-band (5.4 GHz) SAR at 10 m resolution, 6-day repeat at the equator with both satellites active. C-band penetrates less deeply than L-band but multi-temporal coherence analysis over 6- or 12-day pairs reliably separates mangrove (low coherence from canopy motion) from bare mudflat (high coherence) and from adjacent non-forest wetland. Free, open archive from 2014.
Sentinel-2 MSI: 10 m visible and near-infrared bands, 20 m red-edge and SWIR bands, 5-day revisit with both satellites. Supplies tidal-composite optical imagery for stand-boundary delineation. The red-edge bands (B5, B6, B7 at 705–783 nm) are particularly sensitive to mangrove canopy chlorophyll and help distinguish mangrove from adjacent terrestrial forest at the landward fringe.
Landsat 8/9 OLI: 30 m multispectral imagery with a 16-day repeat per satellite (8-day combined). Provides the long-term optical archive underpinning decadal change detection; Landsat data back to 1972 (MSS) and consistently to 1984 (TM) enable loss-rate analysis over periods no other free archive can match. Used in the Global Mangrove Watch historical baseline alongside JERS-1 and ALOS PALSAR.

Why tides make optical mapping so treacherous

A mangrove stand photographed at high tide looks like open water. At low tide, the same stand exposes pneumatophores, prop roots and saturated sediment between the trees, confusing spectral classifiers trained on closed canopy. Neither image alone is wrong; both are incomplete. The solution is a tidal-phase composite: selecting only those Sentinel-2 or Landsat acquisitions captured within a defined tidal window, typically mid to low tide, and then compositing across multiple dates to suppress cloud and shadow. This is the optical foundation of the JAXA Global Mangrove Watch (GMW) baseline, which used JERS-1 and successive ALOS PALSAR mosaics from 1996 to 2020 alongside tidal-filtered optical layers.

Getting the tidal window right requires pairing image acquisition timestamps with a tidal model such as FES2014 or TPXO at each coastal pixel. The computational overhead is non-trivial, but skipping it produces boundary errors of tens to hundreds of metres in macrotidal settings, which is the difference between a legally defensible map and a misleading one.

What L-band backscatter reveals that optical sensors cannot

L-band radar at 1.27 GHz has a wavelength of roughly 23 cm. That is long enough to pass through the leafy upper canopy and scatter from trunks and structural branches. The dominant mechanism over mangrove is double-bounce: energy travels down through the canopy, reflects off the trunk, bounces off the waterlogged soil surface, and returns directly to the sensor. This produces a strong HH-polarised return that is largely absent over herbaceous marsh, open water or bare mudflat. ALOS-2 PALSAR-2 exploits this in its 25 m mosaic products, and the backscatter intensity in HV polarisation correlates with above-ground woody biomass, though the relationship saturates at high biomass densities, typically above 100–150 tonnes per hectare in published calibration studies.

C-band Sentinel-1 penetrates less deeply, so it adds less biomass information. Its value is in coherence. A 6-day repeat-pass interferogram over a mangrove canopy loses coherence rapidly because wind moves the leaves and small branches between passes. Over bare tidal flat or short-grass marsh, coherence is preserved. Thresholding temporal coherence therefore draws a clean boundary between mangrove and non-forest wetland even when their optical reflectance is similar, a distinction that matters greatly for carbon accounting and hydrological modelling.

Canopy height: where lidar benchmarks SAR estimates

Canopy height is not directly recoverable from backscatter intensity alone. The standard approach is to use SAR-derived biomass proxies, calibrated against spaceborne lidar where available. NASA's GEDI instrument on the International Space Station samples canopy height at 25 m footprints along non-contiguous transects between approximately 51.6° N and S latitude, covering most mangrove-bearing coastlines. GEDI RH98 (the height below which 98% of returned energy falls) provides a canopy-top height metric that can be used to train regression models predicting height from PALSAR-2 HV backscatter and Sentinel-2 canopy-density indices across the full spatial extent.

The honest limit here is GEDI's sampling geometry. It does not provide wall-to-wall coverage; transect spacing at any given latitude varies, and some coastal strips receive sparse sampling. Airborne lidar, where it exists in the public record, fills gaps but is rarely available at national scale for tropical coastlines. Height estimates derived purely from SAR regression carry uncertainties of roughly 2–4 m root-mean-square error in published validation studies, which is acceptable for stratified biomass estimation but not for individual-tree canopy analysis.

Separating mangrove from adjacent wetland: the four-layer logic

No single sensor solves the mangrove-versus-wetland classification problem cleanly. The practical approach stacks four layers: a tidal-composite NDVI or red-edge index from Sentinel-2 to confirm photosynthetically active canopy; PALSAR-2 HH double-bounce intensity to confirm woody structure; Sentinel-1 temporal coherence to reject bare tidal flat and herbaceous marsh; and a digital elevation model or tidal inundation model to constrain the plausible intertidal zone. Each layer eliminates a class of confusion that the others cannot.

Saltmarsh, for instance, shows moderate Sentinel-2 NDVI but low PALSAR-2 HH because it lacks woody trunks. Flooded rice paddies show high PALSAR-2 HH during early growth stages but are excluded by their elevation and geometry. Terrestrial forest at the landward fringe shows strong double-bounce only when the understorey is flooded, which the tidal model flags as implausible. The four-layer stack is not perfect, but its confusion matrix in published GMW validation work shows overall accuracies above 90% for mangrove versus non-mangrove at 25 m scale.

Archive depth and what it tells a government buyer

The Global Mangrove Watch provides publicly documented baseline maps at roughly five-year intervals from 1996 to 2020, with annual updates from 2015 onward. That 25-year record is long enough to attribute loss to specific pressures: aquaculture pond expansion tends to produce abrupt, geometrically regular clearance; storm damage produces irregular loss followed by partial recovery; sedimentation-driven dieback produces gradual, diffuse thinning visible in the backscatter time series before optical sensors register any change.

For a government commissioning a national mangrove inventory, the archive depth means the baseline is not negotiable. Any new mapping programme should be expressed as a delta against the GMW baseline rather than starting from scratch, both for methodological consistency and because international reporting frameworks such as IPCC Tier 2 coastal carbon inventories expect continuity with established datasets. Satellize structures its mangrove analytics on this principle, as it does with the crop-estimation work it runs for the Kingdom of Tonga: open-archive continuity first, commercial tasking for gap-filling second.

One practical limit worth stating plainly: persistent cloud cover over tropical coastlines means that even a five-year Sentinel-2 composite may contain gaps in the wettest regions. SAR is cloud-immune, which is one reason the GMW baseline is SAR-primary. Any client expecting a fully cloud-free optical product in a year of anomalous rainfall should be prepared for SAR to carry more weight than usual, with corresponding uncertainty in spectral-class attribution.

Typical figures

Spatial resolution (SAR, primary)	10 m (Sentinel-1 IW mode); 10–25 m (ALOS-2 PALSAR-2 mosaic products)
Spatial resolution (optical)	10 m visible/NIR (Sentinel-2); 20 m red-edge/SWIR (Sentinel-2); 30 m (Landsat 8/9)
SAR revisit	6 days at equator (Sentinel-1 A+B combined); ALOS-2 PALSAR-2 annual mosaic cycle
Optical revisit	5 days (Sentinel-2 A+B combined); 8 days (Landsat 8+9 combined)
SAR frequency / polarisation	C-band 5.4 GHz VV/VH (Sentinel-1); L-band 1.27 GHz HH/HV (ALOS-2 PALSAR-2)
Minimum mappable mangrove patch	Approximately 0.5 ha at 10 m SAR resolution; smaller patches carry higher omission error
Canopy height uncertainty (SAR-lidar regression)	Approximately 2–4 m RMSE reported in published validation studies
Archive depth	SAR: Sentinel-1 from 2014, ALOS PALSAR from 2006, JERS-1 from 1996 (via GMW); Landsat optical from 1984
Cloud penetration	Full (SAR); zero (optical — tidal compositing mitigates but does not eliminate cloud gaps)
Typical delivery format	GeoTIFF extent mask, polygon shapefile (stand boundaries), raster canopy-height model, change-delta GeoPackage

Analytics Satellize can run

National mangrove extent baseline	Tidal-phase-filtered Sentinel-2 composite classification combined with PALSAR-2 HH/HV double-bounce thresholding, validated against GMW reference polygons	Polygon shapefile of stand boundaries with area statistics by administrative unit; GeoTIFF extent mask at 10 m
Annual change-detection layer	Bi-temporal SAR backscatter differencing and Sentinel-2 NDVI trajectory analysis against the GMW baseline; change pixels attributed to loss or gain class	Annual delta GeoPackage with loss/gain polygons, area table and attribution flags (abrupt versus gradual change)
Canopy height model	GEDI RH98 transect data used to calibrate PALSAR-2 HV backscatter regression; Sentinel-2 canopy-density index applied as a secondary predictor	25 m raster canopy-height model with per-pixel uncertainty band; summary statistics by stand stratum
Mangrove versus non-forest wetland classification	Four-layer stack: Sentinel-2 red-edge index, PALSAR-2 HH intensity, Sentinel-1 6-day coherence, tidal inundation model; random-forest classifier trained on GMW validation points	10 m classified raster with six wetland classes; confusion matrix and overall accuracy report
Biomass density map (above-ground woody)	PALSAR-2 HV backscatter-to-biomass lookup calibrated against published allometric relationships; saturation flagged above 100 t/ha	25 m above-ground biomass raster in t/ha with saturation mask; national and sub-national totals table
Disturbance early-warning alert	Sentinel-1 12-day coherence monitoring with anomaly detection against a rolling 2-year baseline; optical confirmation on cloud-free acquisitions	Monthly alert shapefile of candidate disturbance polygons above 1 ha; confidence score and acquisition-date metadata per alert

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.