Mangrove extent and change detection

Mangroves occupy less than 0.5% of the world's forest area yet rank among the most carbon-dense and ecologically critical coastal ecosystems. Satellite SAR and multispectral data, combined carefully, can map their extent and track losses to aquaculture and development with sub-hectare precision.

Sensors

Sentinel-1 C-band SAR (ESA): 10 m ground range detected resolution in IW mode; 6-day repeat at the equator with both satellites operating. C-band (5.405 GHz) double-bounce from prop roots over open water produces VV backscatter typically 4–8 dB above surrounding mudflat, giving a stable, tide-modulated but distinctive mangrove signature.
Sentinel-2 MSI (ESA): 10 m visible and NIR bands, 20 m red-edge and SWIR; 5-day revisit. NDVI, NDWI and red-edge indices separate mangrove canopy from adjacent terrestrial forest and bare intertidal sediment, though cloud cover over tropical coasts routinely degrades single-date utility and demands multi-month compositing.
ALOS-2 PALSAR-2 L-band SAR (JAXA): L-band (1.27 GHz) penetrates the canopy and interacts with larger woody stems and prop-root structures, making it sensitive to canopy density and biomass gradients that C-band misses. Spatial resolution 3–10 m depending on mode; revisit approximately 14 days. Particularly useful for distinguishing degraded from intact mangrove stands.
Landsat 8/9 OLI (USGS/NASA): 30 m resolution, 16-day single-satellite revisit (8-day with both). The long Landsat archive back to 1972 (with TM/ETM+ predecessors) enables multi-decade change analysis. OLI's coastal aerosol band (443 nm) aids water-column correction in shallow tidal areas. Resolution limits detection of narrow fringe stands to roughly one to two pixels wide.

What a prop root gives away to radar

Mangroves are structurally unusual. Their aerial prop roots create a lattice of vertical woody cylinders standing in open water or saturated mud. When a C-band radar pulse strikes this geometry, it bounces off the water surface and then off the root structure, returning to the sensor along almost exactly the path it came. This double-bounce mechanism produces backscatter values that are consistently and measurably higher than those from surrounding mudflat, open water, or even adjacent terrestrial forest where the ground is not free water.

The effect is well-documented in the literature and underlies the use of Sentinel-1 for global mangrove mapping. In VV polarisation, mangrove stands typically return 4–8 dB above adjacent tidal flat. The signal is not perfectly stable: it varies with tidal height, canopy closure and the proportion of woody versus pneumatophore root structure. But across a multi-date composite it is sufficiently consistent to delineate extent with high confidence, even through moderate cloud, which is the operative advantage over optical-only methods in persistently overcast tropical environments.

The tidal phase problem and how compositing addresses it

Tidal inundation is both the reason the double-bounce signal exists and the main source of commission error in SAR-based mangrove mapping. At high tide, water extends beneath and between root structures in ways that can temporarily suppress or shift the backscatter pattern. At low tide, exposed mud can produce returns that resemble degraded mangrove. A single SAR acquisition, taken at an unknown tidal phase, can therefore misclassify either direction.

The standard response is multi-date compositing: stacking 20 or more Sentinel-1 acquisitions across multiple tidal cycles and taking percentile or median backscatter values. This averaging suppresses tidal artefacts and reveals the persistent structural signature of the canopy. Sentinel-1's 6-day repeat makes this feasible within a single season. Tidal phase metadata from hydrodynamic models (such as FES2014) can be used to stratify acquisitions and weight the composite, reducing the number of scenes needed to achieve stable results. The honest caveat is that very narrow fringe mangroves, one to two tree-widths wide, remain prone to mixed-pixel contamination even after compositing.

Optical indices: what they add and where they fail

Sentinel-2 and Landsat OLI contribute spectral discrimination that SAR alone cannot provide. NDVI separates photosynthetically active canopy from bare sediment reliably. The red-edge bands on Sentinel-2 (bands 5, 6 and 7, centred near 705, 740 and 783 nm) are sensitive to chlorophyll content and canopy stress, which helps distinguish healthy closed-canopy mangrove from sparse or degraded stands. NDWI and modified NDWI variants isolate open water within the tidal zone, helping to define the seaward boundary of the forest.

The limitation is cloud. Tropical coastlines where mangroves occur, particularly in South and South-East Asia, West Africa and the Caribbean, experience persistent cloud cover for months at a time. A single Sentinel-2 scene is frequently unusable. Building a cloud-free composite over a 12-month window is standard practice, but it means the optical layer represents a seasonal average rather than a snapshot, and rapid losses can be missed if they occur entirely within the cloudy season. SAR is not subject to this constraint, which is why the two data streams are most powerful in combination rather than as alternatives.

Detecting change: aquaculture ponds, reclamation and sea-level retreat

The three dominant drivers of mangrove loss each leave a distinct signature. Aquaculture conversion produces rectangular or geometrically regular open-water features with low SAR backscatter and low NDVI, appearing abruptly between one annual composite and the next. Coastal reclamation and infrastructure development introduce high-backscatter hard targets (concrete, metal roofing, compacted fill) where canopy previously existed. Sea-level-driven retreat is subtler: the seaward fringe thins progressively, and the change signal per pixel is small, requiring careful baseline comparison across multi-year Landsat or Sentinel archives to distinguish genuine loss from inter-annual tidal variability.

Bitemporal change detection using annual composites is the workhorse method. Log-ratio of SAR backscatter between two composite periods highlights areas of significant structural change. Optical NDVI differencing flags canopy loss. The two signals cross-validated against each other reduce false positives substantially. For high-confidence loss attribution, very-high-resolution commercial imagery can be used to visually verify the land-cover type that replaced the mangrove, though that step sits outside what open-constellation data alone can deliver.

L-band for density gradients and degradation

ALOS-2 PALSAR-2 operates at L-band (1.27 GHz), a longer wavelength that penetrates the mangrove canopy and interacts with larger structural elements: trunks, major branches and the denser root masses of mature stands. Where C-band primarily responds to the surface geometry of the prop-root lattice, L-band backscatter correlates more strongly with above-ground woody biomass and canopy closure. This makes PALSAR-2 useful for distinguishing intact high-biomass mangrove from degraded or regenerating stands that may have similar extent but very different ecological value.

The practical constraint is revisit and data access. PALSAR-2 operates on a 14-day repeat and data availability for analysis outside Japan is less immediate than for Sentinel-1. Multi-temporal L-band composites are therefore typically built over longer windows, which limits their utility for near-real-time change detection. The combination of Sentinel-1 for extent and change detection with PALSAR-2 for structural characterisation represents the current state of practice in published global mangrove assessments, including the Global Mangrove Watch dataset produced under JAXA's leadership.

Putting the analysis to work

A credible mangrove monitoring programme needs three things: a defensible baseline extent map, an annual or semi-annual change layer with attributed loss drivers, and a reporting format that non-specialist government users can act on. The baseline is typically derived from a multi-sensor composite, combining Sentinel-1 double-bounce classification with Sentinel-2 NDVI masking and, where available, PALSAR-2 density stratification. Change layers are generated by comparing successive annual composites and flagging pixels that cross a validated backscatter or NDVI threshold.

Satellize runs this class of analysis on open constellations for coastal-state clients, applying the same compositing and change-detection pipeline it has developed for other small-island and tropical contexts, including the Kingdom of Tonga crop-estimation programme. The output is a GIS-ready polygon layer with per-patch change attribution, delivered annually or on a triggered basis when a monitoring alert crosses a configurable loss threshold. Buyers should be clear-eyed about what the method cannot do: it will not resolve individual trees, it will not reliably detect degradation that leaves the canopy crown intact, and it requires at least one full tidal cycle of SAR acquisitions before a new baseline is stable enough to support change detection.

Typical figures

Best spatial resolution (SAR)	10 m (Sentinel-1 IW mode); 3–10 m (ALOS-2 PALSAR-2 depending on mode)
Best spatial resolution (optical)	10 m (Sentinel-2 visible/NIR); 30 m (Landsat 8/9 OLI)
SAR revisit (Sentinel-1)	6 days at equator with two satellites; single-satellite 12 days
Optical revisit (Sentinel-2)	5 days at equator; effective cloud-free revisit in humid tropics often 30–90 days
Minimum detectable patch	Approximately 0.1 ha for closed-canopy stands in SAR composite; narrow fringe stands (<20 m wide) prone to mixed-pixel error
SAR frequency (Sentinel-1)	C-band, 5.405 GHz; VV and VH polarisations in IW mode
SAR frequency (ALOS-2 PALSAR-2)	L-band, 1.27 GHz; HH, HV, VV, VH polarisations available
Archive depth	Sentinel-1 from 2014; Landsat back to 1972 (TM/ETM+/OLI); PALSAR-2 from 2014 (PALSAR from 2006)
Composite window for stable baseline	Minimum 3–6 months of SAR acquisitions (approximately 15–30 Sentinel-1 scenes) to average tidal phase variation
Typical change-detection latency	Annual composites: 2–4 weeks post-period-end for processing; near-real-time SAR alerts: 1–3 days after acquisition

Analytics Satellize can run

Mangrove extent baseline map	Multi-date Sentinel-1 VV/VH percentile composite with double-bounce thresholding, cross-validated against Sentinel-2 NDVI mask	GIS polygon layer (GeoPackage or Shapefile) with canopy-closure class and area statistics per administrative unit
Annual mangrove loss layer	Bitemporal log-ratio change detection on Sentinel-1 composites, confirmed by NDVI differencing on Landsat or Sentinel-2 annual composites	Change polygon layer with loss-driver attribution (aquaculture, reclamation, erosion) and per-patch area in hectares
Canopy density stratification	ALOS-2 PALSAR-2 L-band backscatter intensity classification correlated with published biomass-backscatter relationships	Raster density map (3–10 m) classified into intact, degraded and sparse strata; updated on PALSAR-2 acquisition cycle
Tidal-phase-corrected SAR composite	Acquisition-weighting using FES2014 tidal model predictions to stratify scenes by inundation depth before compositing	Processed GeoTIFF composite with provenance metadata listing scene dates, tidal heights and weighting applied
Loss-alert notification	Threshold-based monitoring on rolling 30-day Sentinel-1 backscatter anomaly relative to baseline composite	Automated alert (email or API) with bounding coordinates, estimated area and acquisition date when patch loss exceeds configurable threshold
Multi-decade trend report	Landsat archive time-series analysis (1990–present) using annual NDVI composites and supervised classification	PDF and tabular report showing decadal extent, net loss/gain by coastline segment, and trend attribution

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.