Peat fire combustion depth and ground-level loss measurement

Subsurface peat combustion during drought can lower the ground surface by tens of centimetres in weeks. Combining InSAR displacement time-series with VIIRS and MODIS fire radiative power separates combustion-driven loss from seasonal peatland oscillation, though attribution between burning, drainage and post-fire compaction requires field validation.

Sensors

Sentinel-1 A/B (C-band SAR, ESA): 20 m ground range resolution in Interferometric Wide swath mode, 6-day repeat at mid-latitudes (12-day with single satellite). Provides surface displacement via InSAR and coherence maps that drop sharply over actively burning or freshly burned peat, recovering as char consolidates. C-band penetration is limited to a few centimetres into vegetation canopy, so bare or low-biomass burned surfaces are best suited.
ALOS-2 PALSAR-2 (L-band SAR, JAXA): L-band (1.27 GHz) penetrates residual standing dead vegetation and sparse scrub better than C-band, reducing decorrelation from canopy remnants over partially burned areas. Stripmap mode offers 3 m resolution; ScanSAR offers 100 m at wider swath. Repeat cycle is 14 days, which limits rapid-response interferogram pairs but improves coherence over vegetated peat.
VIIRS 375 m active fire product (Suomi-NPP / NOAA-20): 375 m pixel resolution active fire detections, twice-daily overpass. Fire radiative power (FRP) in megawatts per pixel provides a proxy for combustion intensity. Used here to timestamp fire onset and intensity episodes that are then matched against InSAR displacement epochs. Minimum detectable FRP is roughly 5 MW at 375 m under clear sky; smoke and cloud obscure detection.
MODIS MCD64A1 burned area product (Terra/Aqua): Monthly burned area at 500 m resolution, globally consistent archive back to 2000. Defines the spatial extent of each burn event as a boundary layer against which InSAR deformation fields are clipped and compared. Temporal precision is limited to monthly composites, so it complements rather than replaces VIIRS for event timing.
MODIS MOD14/MYD14 fire radiative power (Terra/Aqua): 1 km active fire product with FRP, four overpasses daily when Terra and Aqua are combined. Longer archive than VIIRS (back to 2000) makes it useful for retrospective analysis of past fire seasons against historical InSAR stacks.

What a lowering surface actually means for peat

Peat is mostly water and organic matter. When it burns subsurface during drought, the solid matrix oxidises and the ground collapses into the void left behind. Published field measurements from Kalimantan and Sumatra document losses of 20 to 100 centimetres of surface elevation per fire event, depending on peat moisture and burn duration. That is not erosion or compaction in the conventional sense. It is the physical removal of material from below.

The complication is that peatlands breathe. Seasonal wetting and drying causes surface oscillations of several centimetres that are real, repeatable and entirely unrelated to fire. Any InSAR time-series over peatland must establish a pre-fire baseline long enough to characterise that seasonal cycle before attributing anomalous subsidence to combustion. Without that baseline, a drought-year drawdown signal and a fire-driven collapse look uncomfortably similar.

How InSAR reads the collapse, and where it goes wrong

Sentinel-1 interferometric SAR measures line-of-sight displacement to millimetre precision under ideal conditions. Over actively burning peat, coherence collapses almost entirely: the surface is changing faster than the 6-day repeat allows, and smoke and heat distort the tropospheric phase. This coherence dropout is itself diagnostic. A spatially coherent zone of near-zero coherence that appears suddenly and coincides with VIIRS fire detections is a strong indicator of active combustion rather than flooding or land-cover change.

Once the fire front passes and the surface stabilises as char, coherence recovers over days to weeks. The displacement measured between a pre-fire reference scene and the first coherent post-fire scene captures the net surface loss, but it is a single snapshot rather than a continuous record of how the collapse progressed. Multi-temporal small-baseline subset (SBAS) processing can reconstruct a time-series of displacement from overlapping interferogram networks, giving a better picture of subsidence rate versus fire intensity. The honest limit: phase unwrapping fails where displacement between consecutive epochs exceeds half a wavelength, roughly 2.8 cm for C-band. Rapid collapse during peak combustion may exceed this, creating unwrapping errors that underestimate total loss.

L-band versus C-band over burned peat: a practical trade-off

After a fire, burned peatland is rarely bare. Standing dead trees, partially consumed shrubs and wind-blown debris create a heterogeneous surface that decorrelates C-band returns quickly. ALOS-2 PALSAR-2 at L-band (23 cm wavelength versus Sentinel-1's 5.6 cm) penetrates this residual structure and scatters from the soil and char surface below. Published comparisons over Indonesian peatlands show substantially higher post-fire coherence in L-band pairs than in contemporaneous C-band pairs over the same scenes.

The trade-off is revisit. ALOS-2's 14-day repeat means that a post-fire interferogram pair spans at minimum 14 days, during which secondary compaction and drainage-driven subsidence are already accumulating. Separating those signals from the primary combustion loss becomes harder. For operational monitoring, the practical approach is to use Sentinel-1 for rapid detection and temporal density, and ALOS-2 to validate displacement magnitudes in vegetated zones where C-band coherence is unreliable.

Fire radiative power as a combustion depth proxy

VIIRS 375 m FRP measures the radiant energy flux from the fire surface in megawatts. For surface fires, FRP correlates with fuel consumption rate. For subsurface peat fires, the relationship is less direct because much of the combustion occurs below the surface and the thermal signal is attenuated by overlying char and soil. That said, cumulative FRP integrated over the duration of a fire event has been used in published research as a rough proxy for total carbon consumed, and by extension for volumetric peat loss.

The practical use here is temporal alignment. VIIRS provides near-real-time detection of when and where fire intensity peaked. Those timestamps anchor the InSAR analysis: which interferogram pairs straddle the peak, which post-date it. Without this alignment, it is easy to attribute post-fire compaction or drainage rebound to the combustion event itself. MODIS MCD64A1 adds spatial boundary confirmation at monthly resolution, useful for clipping the InSAR deformation field to the confirmed burned perimeter rather than relying on the coherence dropout alone.

The attribution problem: combustion, drainage or compaction?

Surface lowering over burned peat has at least three concurrent causes. First, direct combustion removes peat mass. Second, fire events on tropical peatlands typically occur during severe drought, when water table drawdown has already caused elastic and inelastic compaction. Third, post-fire, the loss of surface vegetation removes evapotranspiration, which can paradoxically raise water tables and cause some elastic rebound, partially masking the combustion signal. Drainage infrastructure in degraded peatlands adds a fourth variable.

InSAR cannot disaggregate these processes on its own. The displacement field is a sum. Field measurements of bulk density, water table depth and peat stratigraphy are required to partition the signal. This is not a caveat to be buried in a footnote. It is the central analytical challenge, and any programme that reports combustion depth from satellite data alone, without ground truth, is overstating what the physics allows. Satellize's approach to analogous attribution problems, developed through work such as the Tonga crop-estimation programme, is to flag uncertainty explicitly in deliverables rather than smooth it away.

What a monitoring programme actually needs to set up

A credible peat fire ground-loss programme requires a pre-fire InSAR baseline of at least one full seasonal cycle, preferably two, to characterise the natural oscillation envelope. Sentinel-1 archive data back to late 2014 makes this feasible for most tropical peatland regions. The MODIS active fire archive extends to 2000, so retrospective analysis of past events is possible even where InSAR coverage is patchy.

Operationally, VIIRS active fire alerts can trigger automatic interferogram processing within 24 to 48 hours of satellite overpass, flagging coherence dropout zones for analyst review. The output is a displacement map with an explicit uncertainty layer, not a single number. For regulatory or insurance purposes, that uncertainty layer is as important as the central estimate. Ground validation, even sparse GPS benchmarks or repeat levelling surveys at a handful of points, transforms the satellite product from indicative to defensible.

Typical figures

InSAR spatial resolution (Sentinel-1 IW)	20 m ground range x 5 m azimuth (multi-looked to ~20 x 20 m for displacement maps)
InSAR spatial resolution (ALOS-2 Stripmap)	3 m (Stripmap); 100 m (ScanSAR Wide)
Sentinel-1 repeat cycle	6 days (two-satellite constellation); 12 days (single satellite)
ALOS-2 repeat cycle	14 days
VIIRS active fire detection resolution	375 m; minimum detectable FRP approximately 5 MW under clear sky
MODIS burned area resolution (MCD64A1)	500 m monthly composite
Minimum detectable displacement (InSAR, C-band)	~5 mm line-of-sight under high coherence; phase unwrapping fails above ~2.8 cm per epoch
Archive depth	Sentinel-1: late 2014 to present; ALOS-2: 2014 to present; MODIS fire: 2000 to present; VIIRS: 2012 to present
Cloud and smoke limitation	SAR is cloud-independent; VIIRS and MODIS fire detection degraded under thick smoke or cloud
Typical displacement time-series latency	24-48 hours after SAR overpass for rapid-response coherence alert; 3-5 days for full SBAS time-series update

Analytics Satellize can run

Pre-fire peatland oscillation baseline	SBAS InSAR time-series over 1-2 seasonal cycles to characterise natural surface displacement envelope	GIS raster stack of monthly displacement with seasonal amplitude and phase maps
Coherence dropout fire alert	Automated coherence difference between consecutive Sentinel-1 pairs, spatially intersected with VIIRS active fire detections	Near-real-time polygon alert layer (GeoJSON or shapefile) flagging active combustion zones within 48 hours of overpass
Post-fire surface displacement map	Differential InSAR between pre-fire reference and first coherent post-fire scene; phase unwrapping with SNAPHU or equivalent	Line-of-sight displacement raster with uncertainty layer; vertical component estimated using incidence angle correction
Cumulative ground-level loss time-series	SBAS multi-temporal InSAR network anchored to fire onset timestamps from VIIRS FRP	Per-pixel subsidence time-series chart and GIS layer, updated at each new SAR acquisition
L-band versus C-band coherence comparison	Co-registered ALOS-2 and Sentinel-1 coherence maps over burned perimeter from MCD64A1	Side-by-side coherence comparison report identifying zones where C-band underestimates displacement due to residual vegetation
Combustion depth uncertainty report	Displacement partitioning using published peat bulk density ranges and water table proxy from SAR backscatter change; explicit uncertainty quantification	PDF technical report with confidence intervals on combustion depth estimates and list of required field validation points
Multi-year fire recurrence and cumulative loss assessment	Stack of annual InSAR displacement epochs cross-referenced with MODIS MCD64A1 burn perimeters from 2014 onwards	Cumulative subsidence map per concession or administrative unit, suitable for carbon accounting or regulatory reporting

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.