Peatland burn-scar mapping and combustion-depth estimation
Peat fires consume organic soil metres below the surface, leaving scars that optical imagery underestimates and conventional fire-detection algorithms miss entirely. Sentinel-1 InSAR coherence loss and surface subsidence, combined with Sentinel-2 burn-severity indices, offer the most tractable remote-sensing path to combustion-depth estimation, though field calibration remains non-negotiable.
Sensors
- Sentinel-1 SAR (C-band, ESA): 20 m resolution in Interferometric Wide Swath mode; 6-day repeat at mid-latitudes with both satellites active. Coherence decorrelation between pre- and post-fire image pairs reveals surface disruption caused by subsidence; differential InSAR (DInSAR) quantifies vertical displacement at centimetre-to-millimetre sensitivity under favourable conditions. Cloud-penetrating C-band is essential over humid tropical peatlands where optical access can be blocked for weeks.
- Sentinel-2 MSI (ESA): 10 m (visible/NIR) and 20 m (SWIR) resolution; 5-day revisit at the equator. The Normalised Burn Ratio (NBR = (NIR minus SWIR2) / (NIR plus SWIR2)) and its differenced form dNBR map surface burn severity. SWIR bands at 1610 nm and 2190 nm are sensitive to char and moisture loss in surface vegetation but do not directly detect subsurface combustion depth.
- ALOS-2 PALSAR-2 (JAXA): L-band SAR at 3–10 m resolution in spotlight and stripmap modes; longer wavelength than C-band penetrates residual canopy and dry peat surface more effectively, improving coherence retention in vegetated margins of burn scars. Repeat cycle is 14 days, which limits dense time-series analysis but provides a complementary coherence baseline to Sentinel-1.
- MODIS MCD64A1 (NASA Terra/Aqua): 500 m burned-area product derived from surface reflectance and active-fire detections; monthly global composites available from 2000. Coarse resolution makes it unsuitable for mapping individual peat fire scars in fragmented landscapes, but the 20-plus-year archive is valuable for establishing regional fire-regime context and inter-annual recurrence patterns in major peatland basins.
Why peat fire is a different detection problem
A surface grass or forest fire consumes above-ground biomass and leaves a spectral signature that standard burned-area algorithms handle reasonably well. Peat fires are different. Combustion propagates laterally and downward through waterlogged organic soil, sometimes at depths of one to several metres, while the surface may appear only lightly charred or even intact. The fire can smoulder for weeks or months without a visible flame front. MODIS active-fire detections, which rely on mid-infrared brightness temperature anomalies, frequently miss low-intensity subsurface smouldering altogether.
The practical consequence is systematic underestimation of carbon emissions. Published studies of Indonesian peat fires, particularly the 2015 El Niño season, found that combustion-depth estimates derived from field sampling produced emission figures substantially higher than those inferred from satellite burned-area products alone. Accurate depth estimation is not an academic refinement; it is the dominant source of uncertainty in peat-fire carbon accounting.
What a subsidence bowl gives away
When peat burns underground, the overlying surface collapses. The resulting subsidence bowl, sometimes only a few centimetres deep over a small scar but accumulating to tens of centimetres across a large fire, is detectable by interferometric SAR. The technique compares the phase of two SAR acquisitions separated in time. Where the surface has moved away from the satellite between passes, the phase shifts in proportion to the displacement. At C-band with Sentinel-1, the theoretical sensitivity is around half a wavelength, roughly 2.8 cm per phase cycle, and sub-centimetre displacement can be resolved by phase-unwrapping algorithms under low-noise conditions.
The practical limit is coherence. Peat surfaces, particularly wet tropical peatlands, lose interferometric coherence rapidly due to soil moisture change, vegetation regrowth and the physical disruption of burning itself. A coherence value near zero means the phase comparison is noise-dominated and no displacement signal can be extracted. This is why analysts typically use coherence loss itself as a first-pass burn-scar indicator, then apply DInSAR only where coherence is sufficient. In practice, the best results come from short temporal baselines, 6 or 12 days with Sentinel-1, acquired soon after fire cessation before regrowth destroys the signal.
Translating subsidence into combustion depth: the calibration problem
Surface subsidence and combustion depth are related but not identical. The relationship depends on peat bulk density, the degree of compaction after burning, and whether the surface has been additionally disturbed by drainage or vehicle access. Published field studies from Borneo and Sumatra suggest a rough proportionality, with subsidence accounting for perhaps 50 to 80 percent of the depth of burned peat, but the ratio varies considerably with peat type and moisture regime. No satellite sensor can resolve this ambiguity from orbit alone.
Field calibration is therefore not optional. A useful operational workflow pairs the InSAR-derived subsidence map with ground-truth depth-of-burn measurements, taken by probing the char layer at georeferenced points, to fit a site-specific conversion factor. Where field access is impossible during or immediately after a fire, the uncertainty in depth estimation can span a factor of two or more. Analysts should report subsidence in physical units, centimetres of vertical displacement, and state the assumed conversion explicitly rather than presenting a single depth figure as if it were directly measured.
Combining SAR and optical: what each layer adds
Sentinel-2 dNBR maps the extent and relative severity of surface char with 10 to 20 m spatial detail. It is fast to compute, cloud-permitting, and provides a spatially continuous picture of the burn perimeter. The limitation is that dNBR responds to surface vegetation and moisture changes; it does not distinguish a shallow surface scorch from deep subsurface combustion. High dNBR values can coincide with minimal peat loss if only the surface vegetation burned.
SAR coherence loss, by contrast, is sensitive to physical surface disruption and is cloud-independent. The two layers are complementary. Optical burn severity delineates the outer boundary and grades severity across the scar; InSAR subsidence identifies the zones of greatest volumetric peat loss within that boundary. Where the two signals diverge, for instance high optical severity but low subsidence, the interpretation is typically a surface fire with limited peat involvement. Where subsidence is large but optical severity moderate, deep smouldering is the likely explanation. Neither layer alone tells the full story.
Honest limits of the current state of the art
Several constraints are worth stating plainly. First, Sentinel-1's 6-day repeat is adequate for post-fire monitoring but may miss the subsidence accumulation if fire cessation and the next acquisition are poorly timed. ALOS-2's 14-day repeat is worse in this respect, though its L-band coherence can be better preserved in partially vegetated areas. Second, atmospheric phase delay, particularly water vapour over humid tropical peatlands, introduces artefacts that can be misinterpreted as ground deformation. Atmospheric correction using ERA5 reanalysis or GACOS tropospheric delay models reduces but does not eliminate this error source. Third, the MCD64A1 product at 500 m resolution is too coarse for small or fragmented peat scars; it is most useful for multi-year trend analysis across large basins rather than incident-level mapping.
Quantifying combustion depth from space remains an active research problem. The InSAR subsidence approach is the most physically grounded method currently available at operational scale, but published validation datasets are sparse and largely confined to Indonesian and Malaysian peatlands. Transferability to boreal peat systems, where permafrost, different peat types and distinct fire behaviour apply, has not been thoroughly demonstrated in the open literature.
From detection to carbon accounting
The practical output of a combined SAR-optical analysis is a georeferenced map of burn-scar extent, a subsidence raster in centimetres, and, where field calibration data exist, a combustion-depth estimate with stated uncertainty bounds. These layers feed directly into emission calculations using published emission factors for tropical peat, such as those in the IPCC Wetlands Supplement, multiplied by burned area, estimated depth and bulk density. The result is a carbon-loss estimate that is substantially more constrained than one derived from burned area alone, though still carrying meaningful uncertainty.
Satellize runs this combined SAR-optical workflow on open Sentinel archives and can incorporate commercial SAR tasking for higher-frequency post-fire monitoring where the situation demands it. The analytical approach is the same one we apply in structured programmes, including the Kingdom of Tonga crop-estimation work, where multi-sensor fusion and honest uncertainty quantification are standard practice rather than afterthoughts. If you are managing peatland concessions or national greenhouse-gas inventories, the first concrete step is a coherence-feasibility assessment over your area of interest using the existing Sentinel-1 archive, which will tell you whether the temporal baseline and coherence conditions are adequate before any new tasking is commissioned.
Typical figures
| SAR spatial resolution (Sentinel-1 IW mode) | 20 m range × 22 m azimuth (ground range) |
| InSAR vertical displacement sensitivity (Sentinel-1 C-band) | Theoretically sub-centimetre under low-noise conditions; practically 1–5 cm depending on coherence and atmospheric correction |
| Sentinel-1 revisit (both satellites active) | 6 days at mid-latitudes; 12 days with single satellite |
| Optical burn-severity resolution (Sentinel-2 dNBR) | 10 m (NIR band); 20 m (SWIR bands used in NBR) |
| MODIS MCD64A1 burned-area resolution | 500 m; monthly composites |
| ALOS-2 PALSAR-2 resolution (stripmap) | 3–10 m depending on mode; 14-day repeat |
| Sentinel-1 archive depth | From 2014 (Sentinel-1A launch); global coverage |
| Minimum detectable subsidence bowl (InSAR) | Spatially, features larger than ~1–2 pixels (20–40 m) are tractable; smaller features are below detection floor |
| Combustion-depth estimation uncertainty (without field calibration) | Factor of 2 or greater; stated explicitly in all Satellize deliverables |
| Delivery formats | GeoTIFF (subsidence raster, dNBR), GeoPackage / Shapefile (burn perimeter), PDF technical report with uncertainty statement |
Analytics Satellize can run
| Burn-scar extent map | Sentinel-2 dNBR thresholding with pre/post compositing to reduce cloud contamination; cross-validated against Sentinel-1 coherence-loss extent | GeoTIFF and vector polygon layer of burn perimeter with severity classification (low / moderate / high dNBR) |
| SAR coherence-loss map | Sentinel-1 IW SLC pair processing: co-registration, interferogram formation, coherence estimation over 5×20 pixel window, pre/post comparison | GeoTIFF of coherence difference layer, flagging pixels with decorrelation consistent with surface disruption |
| Surface-subsidence raster | Differential InSAR (DInSAR) using Sentinel-1 or ALOS-2 PALSAR-2 pairs; topographic phase removal using Copernicus DEM; atmospheric correction via ERA5 or GACOS tropospheric delay models | GeoTIFF of line-of-sight displacement in centimetres, with coherence mask applied and uncertainty band stated |
| Combustion-depth estimate (calibrated) | Site-specific linear regression of InSAR subsidence against field-measured depth-of-burn points; bulk-density assumption documented and sourced from published peat-type literature | Raster of estimated combustion depth in centimetres with explicit confidence interval; requires client-supplied or third-party field calibration data |
| Carbon-loss estimate | Burned area × combustion depth × peat bulk density × emission factor, following IPCC 2013 Wetlands Supplement methodology; uncertainty propagated from depth estimation error | Tabular carbon-loss report (tonnes CO₂-equivalent) with low/central/high scenario columns, suitable for national GHG inventory submission |
| Multi-year fire-recurrence layer | MODIS MCD64A1 annual burned-area stack (2000 to present) aggregated to peatland land-cover mask; recurrence frequency mapped per pixel | GeoTIFF of fire-recurrence count and time-since-last-fire, suitable for concession risk screening |
| Coherence-feasibility assessment | Retrospective analysis of existing Sentinel-1 archive over the area of interest: coherence statistics, temporal baseline distribution, atmospheric noise level | Technical memo stating whether InSAR-based subsidence mapping is viable for the site, with recommended acquisition strategy if additional tasking is needed |
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.