Smouldering peat fire trace gas mapping from thermal and atmospheric sensors

Smouldering peat emits CO and CH4 at ratios that betray its origin, yet MODIS and VIIRS active-fire counts routinely miss it. TROPOMI and Sentinel-3 SLSTR close that gap, with caveats.

Sensors

Sentinel-5P TROPOMI: Provides daily global CO and CH4 total-column retrievals at 5.5 × 3.5 km (post-2019 upgrade from 7 × 3.5 km). CO sensitivity in the lower troposphere makes it the primary tool for detecting peat-burn plumes; CH4 columns support mass-balance attribution when cloud fraction is below roughly 0.3.
Sentinel-3 SLSTR: Dual-view thermal imager with a 1 km pixel at nadir in the 3.7 µm (S7) and 10.85 µm (S8) bands, revisit approximately 1.7 days at the equator. The 3.7 µm channel is sensitive to sub-pixel thermal anomalies down to roughly 500 K, but smouldering peat at 400–600 °C produces weaker signatures than flaming fire, raising the detection floor considerably.
MODIS (Terra/Aqua): Active-fire product (MOD14/MYD14) at 1 km, twice-daily combined overpass. Designed for flaming combustion above approximately 800 °C; systematically undercounts smouldering peat, which burns at lower temperatures and often beneath a thin ash or soil cap. Useful as a baseline comparison to show what conventional fire products miss.
VIIRS (Suomi-NPP and NOAA-20): 375 m active-fire product improves spatial resolution over MODIS and detects smaller hotspots, but the same physics applies: the 3.74 µm band threshold favours flaming combustion. VIIRS Nightfire data from Colorado School of Mines adds temperature-class discrimination, which helps flag anomalously cool thermal events consistent with smouldering.
Sentinel-2 MSI (contextual only): 10 m optical imagery used to map peat drainage canals, land-cover change and burn-scar extent after cloud clears. Not a trace-gas or thermal sensor, but essential for ground-truthing the spatial footprint that TROPOMI columns cannot resolve at 5.5 km.

Why peat smoke is chemically distinct from forest fire smoke

Flaming biomass combustion is relatively efficient. It runs hot, above 800 °C, and converts most carbon to CO2 with a modified combustion efficiency (MCE) typically above 0.95. Smouldering peat is the opposite: a slow, oxygen-limited oxidation of deep organic material at 400–600 °C that produces a far higher ratio of CO and CH4 relative to CO2. Published MCE values for tropical peat smouldering cluster around 0.80–0.85, meaning roughly 15–20 percent of emitted carbon leaves as incompletely oxidised gases rather than CO2.

That chemistry is detectable from orbit. TROPOMI's CO retrieval in the 2.3 µm shortwave infrared band responds to the elevated CO burden in the troposphere above a peat burn, and the CO-to-CH4 ratio in the column can, under favourable wind and cloud conditions, be compared against published emission-factor ratios to distinguish peat combustion from open savanna or agricultural burning. This is not a clean separation: overlapping plumes, transport mixing and retrieval noise all degrade the signal. But the chemical fingerprint is real and has been used in published atmospheric-inversion studies of Indonesian fire seasons.

What the 3.7 µm band gives away, and what it conceals

The 3.7 µm mid-infrared band sits at the crossover between reflected solar radiation and thermally emitted radiation. At night, it is purely thermal; during the day, the solar component must be subtracted. This band is exceptionally sensitive to sub-pixel hot spots because thermal emission scales with the fourth power of temperature: a small area at 700 °C dominates the pixel radiance even when it covers a tiny fraction of the 1 km footprint.

Smouldering peat, however, burns cooler than flaming fire and often underground. Surface temperatures over a smouldering front may be only 50–150 °C above ambient, which is barely above the detection threshold of SLSTR and VIIRS in a 1 km pixel. The practical consequence is that large areas of active peat combustion produce no active-fire flag at all in standard products. Studies of the 2015 Indonesian fires, one of the largest peat-burn events on record, found that MODIS active-fire counts explained only a fraction of the total burned area later mapped from Landsat and Sentinel-2 imagery. TROPOMI CO columns, by contrast, showed persistent elevated anomalies over Kalimantan and Sumatra throughout that season.

Cloud cover: the honest problem over tropical peat

Tropical peatlands in Indonesia sit in one of the cloudiest regions on Earth. TROPOMI requires a scene cloud fraction below approximately 0.3 for reliable CO and CH4 retrievals; in the wet season, cloud fractions routinely exceed 0.8 for days at a stretch. Even in the dry season, convective cloud builds by mid-morning, which limits useful TROPOMI overpasses to the early-morning window.

Sentinel-3 SLSTR has a dual-view geometry that marginally improves cloud detection, but it cannot see through cloud in the thermal bands any more than MODIS can. The practical result is that satellite-based peat-fire monitoring is intermittent rather than continuous. A week of persistent cloud over southern Borneo can produce a complete gap in the thermal and trace-gas record. Analysts must be explicit about this: a clean TROPOMI CO map does not mean no fire; it may mean no retrievals. Gap-filling via atmospheric transport modelling (e.g., HYSPLIT back-trajectories) can reconstruct plume origin during clouded periods but introduces its own uncertainty.

Russian and Canadian boreal peat fires face different cloud regimes. Summer cloud cover over western Siberia and the Hudson Bay Lowlands is lower than over equatorial Indonesia, and solar illumination is longer. TROPOMI coverage in these regions is also improved by the convergence of orbital tracks at higher latitudes, giving revisit intervals below one day above 60° N. Detection conditions are meaningfully better, though permafrost peat fires can still smoulder beneath a surface that looks intact from above.

Combining TROPOMI columns with SLSTR anomalies: the method and its limits

The most informative approach pairs TROPOMI CO anomaly maps with SLSTR thermal anomaly detections on the same day, then cross-references both against Sentinel-2 burn-scar mapping when cloud permits. Where SLSTR flags a thermal anomaly and TROPOMI shows elevated CO in the same grid cell, confidence in an active smouldering event is high. Where TROPOMI shows elevated CO but SLSTR shows nothing, the fire is likely either cloud-obscured or burning at temperatures below the thermal detection floor. That second scenario is common in peat.

Quantifying emission rates from TROPOMI columns requires assumptions about boundary-layer height, wind speed and plume age. Operational estimates typically carry uncertainties of 30–50 percent for individual events, and larger uncertainties when the plume has aged beyond 24 hours and mixed vertically. Published studies using TROPOMI data over Indonesian peat fires have estimated seasonal CO emissions in the range of tens of teragrams of carbon, but these figures depend heavily on the emission factors assumed and the fire-area estimates used to constrain the inversion. Users should treat column-derived emission estimates as order-of-magnitude indicators rather than precise measurements.

Applying this to enforcement and policy: what the data actually supports

Governments and concession monitors have used MODIS fire alerts for years as a basis for enforcement action against illegal land-clearing fires on peatland. The systematic undercount of smouldering events in those products means that enforcement based solely on active-fire flags will miss a substantial fraction of actual burning. Adding TROPOMI CO as a secondary signal extends detection to events that are thermally invisible.

What the combined data cannot do is attribute a specific plume to a specific land parcel at the 5.5 km TROPOMI pixel scale. That attribution step requires Sentinel-2 burn-scar mapping, concession boundary overlays and, often, field verification. Satellize runs this kind of multi-layer analysis for clients who need spatially resolved outputs rather than raw plume maps; the Tonga crop-estimation programme used similar multi-sensor fusion logic, though in a very different context. The analytic chain from column anomaly to parcel-level evidence is well-established in the literature, but it is not automatic and it is not instantaneous.

For Canadian and Russian boreal peat, the enforcement context is different: the primary users are fire-management agencies tracking carbon budget commitments rather than concession regulators. TROPOMI seasonal CO anomaly time series over the West Siberian Plain and the Hudson Bay Lowlands have been used in published carbon-accounting studies to constrain national inventory estimates. The data is open; the analysis is not trivial.

Typical figures

TROPOMI CO pixel size	5.5 × 3.5 km (post-August 2019); 7 × 3.5 km before
TROPOMI revisit	Daily global coverage; sub-daily at latitudes above ~55° N/S due to orbital overlap
TROPOMI CO detection limit	~1 ppbv precision per pixel; plume attribution practical above ~10 ppbv column anomaly
Sentinel-3 SLSTR thermal pixel size	1 km at nadir (3.7 µm and 10.85 µm bands)
Sentinel-3 SLSTR revisit	~1.7 days at equator; better at higher latitudes
VIIRS active-fire pixel size	375 m (I-band); 750 m (M-band)
Smouldering detection floor (thermal)	Surface temperature anomaly of ~50–150 °C above ambient is often sub-threshold for 1 km sensors; flaming detection floor ~800 °C
Cloud-fraction retrieval limit (TROPOMI)	Scene cloud fraction must be below ~0.3 for reliable CO/CH4 columns
TROPOMI archive depth	From May 2018 (Sentinel-5P launch) to present; data available via Copernicus Dataspace
Emission rate uncertainty	30–50% for individual event estimates from column inversion; higher for aged or mixed plumes

Analytics Satellize can run

Daily TROPOMI CO anomaly map over target peatland region	Background-subtracted CO column retrieval using TROPOMI L2 product; anomaly defined against rolling 30-day baseline for the same calendar period	GeoTIFF layer and PDF report flagging grid cells exceeding threshold anomaly, with cloud-fraction mask applied
Smouldering versus flaming fire classification	Cross-comparison of SLSTR 3.7 µm thermal anomaly temperature class against VIIRS Nightfire temperature estimates; low-temperature events flagged as probable smouldering	Classified point shapefile with confidence tier (high/medium/low) and associated CO column value
Burn-scar area mapping for TROPOMI plume attribution	Sentinel-2 dNBR (differenced Normalised Burn Ratio) computed from pre- and post-event imagery when cloud-free; overlaid with concession boundary polygons	Burn-scar polygon layer in GeoPackage format with area statistics per land-tenure unit
Seasonal CO emission time series for carbon accounting	Integration of daily TROPOMI CO column anomalies over fire season using published peat emission factors (van der Werf et al. GFED framework) to estimate carbon flux	Monthly tabular emission estimate (Tg C) with uncertainty range, delivered as CSV and summary report
Cloud-gap analysis and data-availability calendar	Scene cloud-fraction statistics extracted from TROPOMI L2 QA flags and SLSTR cloud mask; gap periods identified and flagged	Calendar heatmap showing retrieval availability by day; explicit notation of periods where absence of anomaly reflects cloud, not absence of fire
Multi-year peat-fire trend analysis	TROPOMI archive (2018 to present) combined with MODIS Collection 6.1 active-fire and burned-area products for pre-TROPOMI years; anomaly trend fitted per administrative region	Annual trend report with inter-year comparison charts and a ranked list of highest-emission seasons by region

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.