Illegal waste burning and field-fire detection via thermal anomaly

Mid-infrared brightness temperature exceedances betray unauthorised combustion events from orbit, but separating illegal waste fires from permitted agricultural burns demands land-cover context that spectral data alone cannot supply.

Sensors

VIIRS 375 m Active Fire (Suomi-NPP / NOAA-20): Primary workhorse. The I-band channels at 3.74 µm and 11.45 µm give 375 m pixel resolution at nadir, roughly twice the spatial detail of MODIS. Near-polar orbit yields one daytime and one night-time overpass per location per satellite per day; with both NPP and NOAA-20 operating, effective revisit is approximately 12 hours. Fire Radiative Power (FRP) is retrieved per pixel. Saturation occurs for large, intense fires when radiance exceeds the sensor's dynamic range, causing FRP to be underestimated.
Sentinel-3 SLSTR: The Sea and Land Surface Temperature Radiometer carries a dedicated fire channel (S7) at 3.74 µm with a 1 km pixel at nadir, plus a 500 m oblique view that aids smoke discrimination. ESA's Sentinel-3 Fire Radiative Power product provides global coverage with two satellites (Sentinel-3A and 3B) achieving roughly 1-day revisit at mid-latitudes. FRP retrievals are consistent with VIIRS but at coarser resolution, making it better suited to large industrial dump fires than small field burns.
MODIS MOD14 / MYD14: The original operational fire product, at 1 km resolution. Useful for historical comparisons back to 2000 and for cross-validation. Spatial resolution limits detection to fires substantially larger than VIIRS can see. Terra and Aqua together give approximately four overpasses per day globally, though at the cost of resolution.
Landsat 8 / 9 TIRS: Thermal Infrared Sensor bands at 10.9 µm and 12.0 µm, 100 m native resolution (resampled to 30 m in products). 16-day revisit per satellite, so not suited to near-real-time alerting. The value is retrospective: confirming fire extent, mapping burn scars, and providing high-resolution ground truth to calibrate coarser VIIRS detections. Landsat does not saturate on the same fires that overwhelm VIIRS, making it useful for post-event intensity mapping.

What a mid-infrared pixel is actually measuring

Every fire detection in the VIIRS Active Fire product rests on a straightforward physical principle. At mid-infrared wavelengths near 3.74 µm, the Planck function rises steeply with temperature, so a small area of burning material contributes a disproportionately large radiance signal to the mixed pixel. A fire occupying as little as 0.01 hectares under ideal conditions, clear sky, night-time, cool background, can produce a statistically significant brightness temperature exceedance above the surrounding pixel neighbourhood. That figure comes from published VIIRS algorithm documentation and represents a best-case floor, not a guaranteed operational threshold.

The detection algorithm is contextual rather than absolute. Each candidate pixel is compared against a dynamic background built from its neighbours. If the mid-infrared brightness temperature exceeds the background by a threshold that varies with local conditions, the pixel is flagged. This means a small fire on a hot, sun-baked surface in summer is harder to detect than the same fire on a cool surface at night. Enforcement programmes that rely on daytime passes alone will systematically miss events timed to exploit that blind spot.

The saturation problem nobody mentions in the brochure

VIIRS I-band channels saturate when fire radiance exceeds the sensor's upper dynamic range limit. For large, intensely burning waste dumps or extensive field fires, this means the retrieved Fire Radiative Power is a floor estimate, not a true value. Pixels flagged as saturated appear in the standard product with a quality flag, but downstream users who ingest only the FRP number without reading those flags will systematically underestimate the energy output of the worst events, which are often the ones regulators care most about.

Sentinel-3 SLSTR saturates at different thresholds and its 1 km pixel dilutes intensity further, so it can sometimes recover a usable FRP where VIIRS is saturated, and vice versa. Using both sensors together, cross-referenced against Landsat TIRS for post-event burn-scar extent, gives a more defensible picture of fire magnitude than any single source alone.

Why spectral data cannot tell you whether a fire is illegal

A burning tyre dump and a permitted crop-residue fire look nearly identical in mid-infrared. Both produce a brightness temperature exceedance. Both may generate similar FRP values. The sensor has no way to distinguish polyvinyl chloride combustion from dry wheat straw from the radiance signal alone.

Separating the two requires contextual layers: land-cover classification (is the pixel on agricultural land, a mapped landfill, or informal settlement?), calendar information (is this within a permitted burning window declared by the relevant authority?), and ideally a registered-permit database against which detections can be cross-referenced. None of that context comes from the satellite. It must be assembled from national GIS datasets, municipal records and regulatory schedules. Programmes that skip this step produce fire-detection maps, not enforcement intelligence. The distinction matters enormously when the output is intended to support prosecution or permit revocation.

Smoke plume composition, which would help discriminate waste burning from crop fires, is detectable via instruments such as TROPOMI for NO2 and CO columns, but attribution of a specific plume to a specific fire location at the scale of an individual dump site remains genuinely difficult at TROPOMI's 3.5 km pixel. That problem is covered separately in the sibling pages on NO2 and CO monitoring.

Revisit, latency and what enforcement agencies can realistically expect

With Suomi-NPP and NOAA-20 both operating, VIIRS delivers active fire detections with a latency of roughly three hours from overpass through the NASA FIRMS near-real-time system. That is fast enough to dispatch an inspection team to a site the same day, provided the fire is still burning or has left visible evidence. It is not fast enough to catch a fire that was lit and extinguished within a single two-hour window between overpasses.

Sentinel-3 SLSTR fire products from ESA's Copernicus services are typically available within a few hours of acquisition, though near-real-time latency varies by processing tier. MODIS data through FIRMS has an archive extending to 2000, which is its primary enforcement value: establishing whether a site has a documented history of repeated burning events, which is often the basis for escalated regulatory action.

Cloud cover is an unresolved constraint. Thermal infrared at 3.74 µm and 11 µm does not penetrate cloud. In tropical regions with persistent convective cloud, or during monsoon seasons, detection rates drop substantially. There is no satellite-based workaround for this at present; it is an honest limit of the method.

Building an enforcement-grade detection system

An operationally useful illegal-burning detection system has four layers. First, a near-real-time ingestion pipeline for VIIRS 375 m and Sentinel-3 SLSTR fire pixels, filtered to the jurisdiction of interest. Second, a spatial join against land-cover and facility databases to assign each detection a preliminary context label: agricultural, landfill, industrial, unclassified. Third, a permit calendar filter that flags detections falling outside declared burning windows or on non-agricultural land. Fourth, a human-review queue for flagged events, because false positives from industrial heat sources, volcanoes, or sensor artefacts do occur and should not reach an enforcement officer as confirmed violations.

Satellize applies this layered approach in analytics work for government clients. The Tonga crop-estimation programme demonstrated that combining open-constellation data with national land-registry layers produces outputs that national agencies can actually act on, rather than raw satellite products that require specialist interpretation. The same principle applies here: the satellite data is the starting point, not the finished product.

Repeat-offender analysis is where the archive depth of MODIS and Landsat becomes valuable. A site with ten documented fire events over five years, cross-referenced against inspection records, builds a legally defensible pattern of non-compliance in a way that a single detection cannot.

Typical figures

Primary spatial resolution	375 m (VIIRS I-band); 1 km (Sentinel-3 SLSTR, MODIS); 100 m native / 30 m resampled (Landsat TIRS)
Revisit frequency	~12 hours with NPP + NOAA-20 combined (VIIRS); ~1 day at mid-latitudes with Sentinel-3A + 3B; 16 days per satellite (Landsat 8/9)
Near-real-time latency	~3 hours from overpass via NASA FIRMS (VIIRS); a few hours via Copernicus services (Sentinel-3 SLSTR)
Key spectral bands	Mid-infrared ~3.74 µm (fire detection); thermal infrared ~11 µm (background temperature); both required for contextual thresholding
Minimum detectable fire size	~0.01 ha under ideal conditions (VIIRS, night, cool background, clear sky); larger under daytime or warm-surface conditions
Saturation behaviour	VIIRS I-band saturates for large intense fires; affected pixels flagged in quality layer; FRP becomes a floor estimate, not a true value
Cloud penetration	None. Mid-infrared and thermal infrared are blocked by cloud cover; no satellite workaround exists at these wavelengths
Archive depth	MODIS from 2000; VIIRS from 2012 (NPP) and 2018 (NOAA-20); Landsat from 1972 (TM/ETM+/TIRS)
Coverage	Global daily for VIIRS and Sentinel-3; global 16-day for Landsat
Delivery formats	CSV / GeoJSON fire point shapefiles (FIRMS); NetCDF (Sentinel-3 FRP product); GeoTIFF burn-scar layers (Landsat-derived)

Analytics Satellize can run

Near-real-time fire alert feed	VIIRS 375 m contextual brightness-temperature thresholding (NASA FIRMS algorithm class), filtered to jurisdiction boundary and land-cover type	Hourly GeoJSON alert feed with pixel coordinates, FRP value, quality flag and preliminary land-cover context label
Permit-calendar compliance filter	Spatial and temporal join of fire detections against regulatory burning-window schedules and facility permit registers	Daily flagged-event report distinguishing permitted from potentially unauthorised detections, formatted for inspection-team dispatch
Repeat-offender site history	Multi-year time-series aggregation across MODIS MOD14 and VIIRS archives, clustered by facility or land parcel	Site-level fire-frequency report with detection dates, FRP ranges and cloud-affected overpass counts, suitable for regulatory case files
Post-event burn-scar extent map	Landsat 8/9 TIRS and OLI differenced Normalised Burn Ratio (dNBR) applied to pre- and post-fire image pairs	GeoTIFF burn-scar polygon with area estimate in hectares and severity classification, delivered within 16 days of event
Fire radiative power cross-sensor reconciliation	Joint VIIRS and Sentinel-3 SLSTR FRP comparison to identify saturated pixels and constrain true energy release	Tabular FRP reconciliation with saturation flags highlighted, supporting defensible intensity estimates for large events
Land-cover contextual classification layer	Overlay of fire detections against national land-use GIS, municipal landfill registers and ESA WorldCover or equivalent open land-cover product	Static or annually updated GIS layer assigning each monitored site a land-cover category and regulatory risk tier

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.