Mangrove dieback and large-scale mortality event detection
Episodic mangrove die-off from cyclones, hypersalinity or extreme low tides leaves a spectral signature that collapses within days. Catching it requires near-real-time multispectral imagery and a clear-eyed understanding of when cloud cover makes that impossible.
Sensors
- Sentinel-2 MSI: 10 m resolution in red, NIR and red-edge bands; 5-day revisit at the equator with both satellites. Red-edge bands (B5, B6, B7 at 20 m) are particularly sensitive to early chlorophyll decline before visible browning appears. Free and open archive from 2015.
- Planet SuperDove: 3 m resolution, daily revisit over most tropical latitudes. Eight spectral bands including two red-edge channels. Commercial tasking; latency as low as same-day. The high spatial resolution resolves individual mangrove patches that Sentinel-2 conflates with adjacent mudflat or open water.
- Landsat 8/9 OLI: 30 m resolution, 8-day combined revisit. Useful for contextualising events within the 40-year Landsat archive and confirming that a reflectance anomaly is genuinely unprecedented for a given site. Less suited to rapid event tracking given its revisit cadence.
- MODIS MOD09GA: 500 m surface-reflectance product, daily global coverage. Too coarse to map patch-level mortality but valuable for flagging regional-scale events (tens of square kilometres or more) within 24 to 48 hours of cloud clearance, triggering tasking of finer sensors.
Why episodic mortality is a different problem from gradual decline
Most satellite-based mangrove monitoring is designed to track slow processes: annual extent loss, multi-year canopy thinning, incremental landward retreat under sea-level rise. Those workflows tolerate monthly or seasonal composites and can afford to wait out cloud cover. Episodic mortality events operate on a completely different timescale.
The 2015 to 2016 Gulf of Carpentaria die-off, which killed an estimated 7,400 hectares of mangrove across northern Australia, developed over a matter of weeks following an extreme low-tide anomaly coinciding with record sea-surface temperatures. By the time annual monitoring cycles would have flagged it, the canopy had been brown for months. The detection problem is not one of sensitivity; it is one of latency. A workflow built for annual extent mapping will always be fighting the last war.
What a dying canopy looks like from orbit
Healthy mangrove canopy has NDVI values typically in the range of 0.6 to 0.85, depending on species and season. Acute mortality collapses this within days to weeks as chlorophyll degrades: NDVI can fall below 0.2, and the canopy shifts from absorbing near-infrared radiation to reflecting it in a pattern more consistent with dry litter or bare sediment. The red-edge bands on Sentinel-2 (centred near 705 nm, 740 nm and 783 nm) are particularly diagnostic here. Chlorophyll decline suppresses red-edge reflectance before visible browning is obvious, giving a few extra days of early warning compared with a simple NDVI threshold.
Cyclone damage adds structural complexity. Wind-thrown canopy may show high NIR reflectance from exposed woody stems and litter rather than the low-NIR signature of hypersalinity kill. Distinguishing the two matters for management response: hypersalinity events often require drainage intervention, whereas cyclone-damaged stands may regenerate naturally if the root system survives. Spectral shape alone is rarely sufficient for this distinction; it needs to be combined with event metadata (cyclone track, tidal gauge records, salinity sensor data where available).
The cloud problem is not solvable, only managed
Cyclones and the heavy rainfall that accompanies them produce persistent cloud cover precisely when rapid post-event imagery is most needed. Sentinel-2 and Planet SuperDove are both optical sensors; neither sees through cloud. In practice, the window between storm passage and the next usable optical acquisition can be several days to two weeks in humid tropical environments. That gap is long enough for initial canopy collapse to be well advanced before it is first observed.
Synthetic aperture radar (SAR) from Sentinel-1 can acquire through cloud and provides structural information about canopy loss, but SAR backscatter from mangroves is complex: the double-bounce signal from flooded prop-root systems changes with inundation state, and interpreting mortality versus temporary flooding requires care. SAR is useful as a gap-filler and corroborating layer, but it does not replace multispectral reflectance for confident mortality attribution. Honest monitoring design acknowledges the cloud gap and builds in contingency: MODIS daily composites to flag regional anomalies, then commercial tasking of Planet for the first clear-sky opportunity.
Archive depth matters too. A single anomalous NDVI value means little without a baseline. Sentinel-2's archive from 2015 and Landsat's archive from 1982 allow analysts to establish site-specific seasonal envelopes, so that a post-event reading can be compared against the same calendar period across multiple years rather than against a fixed global threshold.
Detection thresholds and the limits of what is mappable
At Sentinel-2's 10 m resolution, a mortality patch needs to be at least two to three pixels across to be reliably distinguished from edge effects and mixed-pixel contamination with water or mudflat. In practice, patches smaller than roughly 0.05 hectares are below confident detection. Planet SuperDove at 3 m pushes that floor down to patches of a few hundred square metres, which matters when mortality begins as scattered individual tree death before coalescing into contiguous dieback.
MODIS at 500 m is only useful for events covering many hundreds of hectares. It remains valuable precisely because its daily revisit means it is often the first sensor to see through a break in cloud cover. The practical workflow is a two-stage trigger: MODIS flags a regional anomaly, Sentinel-2 or Planet provides spatial detail once cloud clears.
One persistent ambiguity is distinguishing genuine mortality from severe but recoverable defoliation. Mangroves defoliated by cyclone wind can look spectrally similar to dead stands for four to eight weeks before resprouting. A single post-event image cannot resolve this; a time series of at least two to three acquisitions over six to ten weeks is needed to separate recovery trajectories from confirmed dieback.
From detection to a number a manager can use
The output that conservation managers and government agencies actually need is not a false-colour image; it is an affected-area estimate in hectares, a severity classification, and ideally a probable cause attribution. Producing that requires classifying each flagged pixel into one of at least three states: confirmed dead canopy, severely stressed but potentially recovering, and unaffected. Severity is typically derived from the magnitude of NDVI or red-edge departure from the seasonal baseline, binned into two or three classes with documented thresholds.
Satellize runs this workflow on Sentinel-2 and Landsat open data, with Planet tasking added on client licence when sub-weekly revisit is required. The analytic approach is directly analogous to the vegetation-index monitoring underlying the Tonga crop-estimation programme, adapted for the spectral characteristics of mangrove canopy rather than agricultural crops. Outputs are delivered as GIS-ready polygon layers with per-polygon severity scores, alongside a written event summary that situates the affected area in the context of the site's historical baseline.
Typical figures
| Finest spatial resolution (operational) | 3 m (Planet SuperDove); 10 m (Sentinel-2 red, NIR); 20 m (Sentinel-2 red-edge) |
| Minimum detectable mortality patch | ~0.05 ha at Sentinel-2 10 m; ~few hundred m² at Planet 3 m; ~tens of ha at MODIS 500 m |
| Revisit interval | Daily (MODIS, Planet over most tropical sites); 5 days (Sentinel-2 two-satellite); 8 days (Landsat 8+9 combined) |
| Key spectral bands for mortality detection | Red (~665 nm), NIR (~842 nm), red-edge (~705, 740, 783 nm on Sentinel-2 MSI) |
| Primary data gap | Cloud cover during and immediately after cyclone events; optical sensors only |
| Archive depth | Sentinel-2 from 2015; Landsat from 1982; MODIS from 2000 |
| Latency (open data) | Sentinel-2: typically 1 to 3 hours after acquisition via Copernicus Data Space; MODIS: same day via FIRMS/EARTHDATA |
| Latency (commercial Planet tasking) | Same-day to next-day delivery on client licence |
| Severity classification time series needed | Minimum 2 to 3 acquisitions over 6 to 10 weeks to distinguish mortality from recoverable defoliation |
| Delivery formats | GeoTIFF, GeoPackage, Shapefile polygon layers; PDF event summary report |
Analytics Satellize can run
| Rapid mortality alert | MODIS daily surface-reflectance anomaly detection against site-specific seasonal baseline; threshold trigger on NDVI departure | Automated alert with affected-region bounding box and estimated area, issued within 48 hours of cloud clearance |
| Confirmed mortality extent map | Sentinel-2 or Planet red-edge and NIR change detection against multi-year seasonal envelope; per-pixel NDVI and red-edge index departure scoring | GIS polygon layer (GeoPackage or Shapefile) with per-polygon severity class and area in hectares |
| Severity classification | Three-class binning (confirmed dead, severely stressed, unaffected) based on magnitude of NDVI and red-edge departure from baseline percentile thresholds | Attributed polygon layer with severity score; included in event summary report |
| Recovery trajectory monitoring | Multi-date NDVI and red-edge time series over 3 to 6 months post-event to separate confirmed dieback from recoverable defoliation | Time-series chart per affected zone; updated GIS layer at each new clear-sky acquisition |
| Historical baseline and anomaly context | Landsat and Sentinel-2 archive compositing to establish site-specific seasonal NDVI envelopes; event reading expressed as percentile departure | Site baseline report with anomaly magnitude in context of full archive; included in initial event summary |
| Probable cause attribution layer | Spectral shape analysis (NIR vs. SWIR ratio, red-edge slope) combined with event metadata (cyclone track, tidal anomaly records) to distinguish wind damage from hypersalinity or thermal kill | Written attribution assessment in event summary; flagged as provisional where spectral evidence is ambiguous |
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.