Post-fire and post-disturbance forest recovery monitoring
Dense Landsat and Sentinel-2 time-series, combined with spectral indices and the LandTrendr segmentation algorithm, can track canopy regrowth trajectories after fire, storm or clearance. The critical caveat: spectral recovery consistently outpaces structural and biodiversity recovery, a gap that matters enormously for carbon accounting.
Sensors
- Landsat 8/9 OLI: 30 m multispectral resolution, 16-day repeat at the equator (8-day with both satellites combined). The archive back to 1984 is the backbone of any long-term recovery trajectory analysis. Shortwave infrared bands (SWIR1 ~1.6 µm, SWIR2 ~2.2 µm) are essential for computing burn severity indices such as dNBR and for separating char, bare soil and recovering vegetation.
- Sentinel-2 MSI: 10 m (visible/NIR) and 20 m (red-edge, SWIR) resolution, 5-day revisit at mid-latitudes with both Sentinel-2A and 2B. Red-edge bands (B5, B6, B7 at ~705, 740, 783 nm) add sensitivity to early-stage canopy stress and thin regrowth that broadband NDVI can miss. Free archive from 2015.
- MODIS Terra/Aqua: 250–500 m resolution, near-daily global coverage. Too coarse for patch-level recovery mapping but useful for regional phenology composites and for cross-calibrating Landsat time-series at landscape scale. MODIS NDVI and EVI products (MOD13) are well-validated and freely available.
- GEDI (Global Ecosystem Dynamics Investigation): Spaceborne lidar operating from the ISS, delivering waveform returns that characterise canopy height and vertical structure at 25 m footprints. Critical for distinguishing spectral recovery from genuine structural recovery: a plot can appear green in multispectral imagery while remaining structurally immature. Coverage limited to approximately 51.6° N/S latitude.
Why a green pixel is not a recovered forest
This is the central problem in post-disturbance monitoring. Within two to five years of a moderate-severity fire, shrubs and pioneer grasses can drive NDVI values back toward pre-fire levels. The satellite sees green. The ecologist sees a structurally simplified stand with no snag habitat, no understorey stratification and a carbon stock that may be 20 to 40 per cent of the original. Spectral recovery is a necessary condition for biological recovery; it is not sufficient.
The distinction has direct financial consequences. Carbon offset methodologies that rely on spectral proxies alone risk overstating sequestration rates. Governments purchasing forest carbon credits or reporting under national REDD+ frameworks need to know whether their monitoring system is measuring greenness or biomass. Honest remote sensing practice keeps those two questions separate and uses different sensor combinations to answer each.
The spectral toolkit: dNBR, RdNBR and what they actually measure
The Normalised Burn Ratio (NBR) uses the near-infrared and SWIR2 bands to exploit a well-understood contrast: healthy vegetation reflects strongly in NIR and absorbs in SWIR, while burned surfaces do the opposite. The differenced NBR (dNBR, pre-fire minus post-fire) maps burn severity. The Relativised dNBR (RdNBR) corrects for pre-fire vegetation density, which matters because a sparse dry woodland and a dense boreal stand can produce the same raw dNBR from very different fire intensities. Published work using Landsat has validated RdNBR against field-measured Composite Burn Index scores across a range of North American and Mediterranean fire regimes.
For recovery specifically, the delta NBR trajectory over successive annual composites tells you how fast the ratio is returning to baseline. NDVI adds a complementary signal, particularly in the early regrowth phase when photosynthetically active biomass is accumulating faster than canopy structural complexity. Neither index tells you anything about understorey composition or stem density. That is where GEDI waveform data becomes indispensable, even with its sampling limitations.
LandTrendr: reading the archive as a story
LandTrendr (Landsat-based Detection of Trends in Disturbance and Recovery) is a temporal segmentation algorithm developed at Oregon State University and published in the peer-reviewed literature. It fits piecewise linear segments to annual spectral composites, identifying the year of abrupt disturbance, the magnitude of spectral loss and the slope and duration of subsequent recovery. The algorithm operates on the full Landsat archive, giving trajectories going back to 1984 for any point on Earth where cloud-free imagery exists.
Practically, LandTrendr outputs a disturbance year, a recovery year (defined as the point at which the spectral index returns to a specified fraction of pre-disturbance value), and a recovery rate expressed as spectral units per year. These can be stratified by forest type, elevation, aspect and climate zone to build recovery typologies. A boreal black spruce stand recovers spectrally at a very different rate from a tropical dry forest or a Mediterranean cork oak woodland. The algorithm does not distinguish cause of disturbance without ancillary data, so fire perimeters, storm damage records or harvest permits are needed to attribute segments correctly.
One honest limitation: LandTrendr's annual compositing step means it can miss sub-annual disturbance dynamics and may attribute recovery to the wrong year when cloud contamination corrupts a single annual composite. Sentinel-2's denser time-series can plug some of those gaps, though its archive only extends to 2015.
What cloud cover actually costs you
Tropical and temperate maritime forests, which include some of the most ecologically significant post-disturbance landscapes, can have persistent cloud cover for months at a time. A single cloudy season can introduce a gap in the spectral trajectory that LandTrendr misinterprets as slow recovery or no change. Dense time-series compositing, using all available cloud-free observations within a year rather than a single scene, reduces but does not eliminate this problem.
SAR (synthetic aperture radar) penetrates cloud and is sensitive to canopy structure changes, but it is handled on a separate use-case page covering SAR and lidar fusion for biomass. For optical-only workflows, the practical advice is to report confidence intervals on recovery dates and rates rather than point estimates, and to flag pixels where fewer than a threshold number of cloud-free observations contributed to any given annual composite.
Structuring a recovery monitoring programme
A credible programme needs three things: a pre-disturbance baseline (ideally multi-year to capture phenological variability), a disturbance characterisation layer (burn severity, storm damage extent, clearance boundary), and a time-series of post-disturbance spectral indices at consistent compositing intervals. Landsat provides the long baseline; Sentinel-2 adds spatial detail and red-edge sensitivity for recent events. GEDI spot-checks provide structural ground truth at the footprint level, though GEDI's mission lifetime and data continuity beyond its current operational phase remain uncertain.
Forest type and climate zone stratification is not optional. Recovery rates vary by roughly an order of magnitude between fast-growing tropical secondary forests and slow-recovering high-elevation conifers. Reporting a single landscape-average recovery metric obscures the variation that matters most to forest managers and carbon verifiers. Satellize applies this kind of stratified trajectory analysis within its analytics workflows, drawing on the same open constellations used in programmes such as the Tonga crop-estimation engagement, where multi-temporal compositing under variable cloud conditions is a routine operational challenge.
Validation against field plots remains the only way to confirm that spectral recovery corresponds to meaningful ecological or carbon recovery. The monitoring design should specify plot revisit intervals and the structural metrics (basal area, canopy height, stem density) that will be compared against satellite-derived indices. Without that ground link, the spectral trajectory is informative but not certifiable.
Honest limits of the method
Spectral indices cannot resolve species composition within a recovering stand. A monoculture of invasive grasses and a diverse native shrub layer can produce identical NDVI values. Red-edge bands on Sentinel-2 help at the margins but do not solve the problem. Hyperspectral sensors (not yet routinely available at global scale) would improve species discrimination substantially.
GEDI's 25 m footprint sampling is not wall-to-wall coverage; it provides transects, not continuous canopy height maps. Interpolating between footprints introduces uncertainty that grows with the heterogeneity of the recovering stand. At 30 m Landsat resolution, patches smaller than roughly one to two pixels (roughly 900 to 1800 m²) are effectively invisible to the recovery signal. For fragmented landscapes with small clearings or strip disturbances, this is a genuine detection floor, not a footnote.
Typical figures
| Spatial resolution (optical) | 10–20 m (Sentinel-2 MSI), 30 m (Landsat 8/9 OLI) |
| Spatial resolution (lidar footprint) | ~25 m footprint, sampled transects (GEDI) |
| Revisit interval | ~5 days at mid-latitudes (Sentinel-2A+2B combined); ~8 days (Landsat 8+9 combined) |
| Temporal archive depth | Landsat: 1984–present; Sentinel-2: 2015–present; MODIS: 2000–present |
| Key spectral bands | NIR (~865 nm), SWIR1 (~1.6 µm), SWIR2 (~2.2 µm) for NBR; red-edge (~705–783 nm) for early regrowth sensitivity |
| Minimum detectable patch (optical) | ~900 m² at Landsat resolution; ~100–400 m² at Sentinel-2 10 m bands |
| Recovery detection latency | Annual composites: 12-month lag; dense time-series: seasonal (3–4 month) composites feasible |
| Cloud sensitivity | Optical only; persistent cloud cover can reduce usable observations to fewer than 4 per year in humid tropics |
| Delivery formats | GeoTIFF rasters, vector polygons (recovery zones), tabular trajectory summaries (CSV/JSON), GIS-ready layers |
| Latitude coverage (GEDI) | Approximately 51.6° N to 51.6° S |
Analytics Satellize can run
| Burn severity classification | dNBR and RdNBR computed from pre/post-fire Landsat or Sentinel-2 image pairs; validated against published CBI thresholds | Raster layer with low/moderate/high/very-high severity classes; polygon summary by severity class and area |
| Spectral recovery trajectory per pixel | LandTrendr temporal segmentation on annual Landsat NBR or NDVI composites; disturbance year, recovery year and recovery rate extracted per segment | Raster stack of disturbance year, recovery duration and recovery rate; tabular export by forest type stratum |
| Recovery rate comparison by forest type and climate zone | Stratified zonal statistics on LandTrendr outputs, cross-referenced with global forest type and Köppen climate zone layers | Summary report with recovery rate distributions per stratum; suitable for inclusion in national forest monitoring reports |
| Spectral vs. structural recovery gap assessment | Co-registration of optical spectral recovery index with GEDI canopy height and RH metrics at coincident footprints; divergence flagged where NDVI recovery exceeds structural recovery | Divergence map and tabular gap statistics; flagged polygons for field verification prioritisation |
| Multi-year recovery monitoring dashboard | Dense Sentinel-2 time-series compositing (seasonal or annual) with NDVI and red-edge index tracking; change flagged against pre-disturbance baseline | Annually updated GIS layer with recovery status per mapped disturbance patch; optional API feed for integration into client forest management systems |
| Cloud-gap confidence reporting | Per-pixel count of cloud-free observations contributing to each annual composite; confidence tier assigned to recovery date estimates | Confidence raster layer accompanying all spectral recovery outputs; methodology note for carbon accounting documentation |
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.