Savanna fire-regime mapping for biodiversity outcomes
Burn frequency, seasonal timing and spatial patchiness together determine savanna biodiversity more than any single fire event. Satellite data from MODIS, VIIRS and Sentinel-2 can reconstruct those patterns at landscape scale, with honest caveats about resolution and low-intensity burns.
Sensors
- MODIS MCD64A1 Burned Area Product: Monthly global burned-area product at 500 m resolution, derived from Terra and Aqua MODIS reflectance and active-fire detections. Archive runs from November 2000, making it the primary source for multi-decadal fire-frequency analysis. The 500 m pixel floor means burns smaller than roughly 25 hectares are systematically under-detected, and low-intensity surface fires in sparse grass are frequently missed.
- VIIRS 375 m Active Fire (VNP14IMGTDL_NRT): Near-real-time active-fire detections at 375 m from the Suomi-NPP and NOAA-20 VIIRS instruments, with twice-daily overpasses. Detection confidence flags distinguish high-confidence fire pixels from lower-confidence detections near the sensor's minimum detectable fire radiative power threshold of roughly 5–10 MW under clear-sky conditions. Used here to assign precise ignition dates and season of burn to complement the monthly MCD64A1 composites.
- Sentinel-2 MSI: 10 m (visible, NIR) and 20 m (red-edge, SWIR) multispectral imagery with a five-day revisit at the equator from the twin Sentinel-2A/B constellation. The SWIR bands (B11 at 1610 nm, B12 at 2190 nm) are particularly sensitive to char and ash immediately after fire, while the red-edge bands (B5, B6, B7) track post-fire green-up over weeks to months. Cloud cover in wet-season savannas can interrupt revisits substantially.
- Landsat OLI (Landsat 8 and 9): 30 m multispectral imagery with a 16-day revisit per satellite, or eight days when both Landsat 8 and 9 are used together. The OLI SWIR2 band (2110 nm) resolves burn-scar boundaries at finer spatial detail than MODIS, useful for mapping small patch burns and verifying MCD64A1 commission and omission errors. The archive back to 1972 (Landsat 1 MSS) supports very long-term fire-history reconstruction, though pre-OLI radiometric consistency requires careful cross-calibration.
Why pyrodiversity predicts biodiversity
The pyrodiversity-begets-biodiversity hypothesis, supported by field studies in African and Australian savannas, holds that a mosaic of burn ages and fire intensities creates a corresponding mosaic of vegetation structure. Early post-fire patches offer open foraging ground for some grazers and their predators. Older unburnt areas provide dense cover for nesting birds and small mammals. The spatial grain of that mosaic, how large each patch is and how frequently it burns, determines which species can persist at landscape scale.
Satellite data cannot measure species directly. What it can measure, with increasing precision, is the spatial and temporal structure of burning: which areas burned in which month, how many times in a decade, and at what apparent intensity. Those measurements are the inputs to habitat-suitability models and conservation planning decisions. The translation from fire pattern to biodiversity outcome still requires ground-truth ecological knowledge, but the fire-pattern layer is where satellite data earns its place.
Reading a burn scar: what the spectral signal actually says
A fresh burn scar in Sentinel-2 imagery is hard to miss. Charred grass and soil absorb strongly in the visible and reflect poorly in the NIR, producing a sharp drop in NDVI (Normalised Difference Vegetation Index) and a simultaneous rise in the Normalised Burn Ratio (NBR), calculated from NIR and SWIR2. The delta-NBR between a pre-fire and post-fire image is the standard index for burn severity. Values above roughly 0.1 are generally considered low-severity burns in savanna; values above 0.44 indicate high-severity combustion, though these thresholds vary with grass type and moisture state.
The difficulty is distinguishing a low-intensity surface burn from bare soil or heavily grazed ground, particularly at MODIS's 500 m resolution. A pixel that is 40 percent bare soil and 60 percent sparse grass can produce an NBR signal similar to a lightly burned mixed pixel. This is not a solvable problem at 500 m; it requires either Sentinel-2 or Landsat imagery to resolve the spatial structure within that pixel. For fire-regime mapping across continental areas, the practical approach is to use MCD64A1 for frequency statistics and Sentinel-2 for boundary precision and severity grading within the areas flagged by MODIS.
Building a fire-history layer: frequency, season and patch geometry
With the MCD64A1 archive running from 2000, it is now possible to compute per-pixel fire return intervals across 24 years for any savanna on Earth. A pixel that burned in 14 of those 24 years has a very different ecological character from one that burned twice. Annual fire frequency maps, stacked into decadal composites, reveal which parts of a landscape are locked into high-frequency burning (often associated with annual grass-dominated, species-poor areas) and which carry infrequent fire (often associated with woody encroachment or management intervention).
Season of burn is a separate and important variable. Early dry-season burns, when grass is still partially green, tend to be lower intensity and patchier. Late dry-season burns, when fuels are fully cured, run hotter and further. VIIRS active-fire detections, with their near-daily revisit and 375 m resolution, allow the month of ignition to be assigned to each MCD64A1 burn polygon with reasonable confidence. The resulting seasonality layer, expressed as the modal month of burning per pixel over the archive period, is a direct input to habitat models for species with season-specific resource needs.
Patch geometry adds a third dimension. Using Sentinel-2 to delineate burn boundaries at 10 to 20 m resolution, it is possible to calculate patch size distributions, perimeter-to-area ratios and the spatial arrangement of burn-age classes across a landscape. These metrics map directly onto concepts from landscape ecology: edge density, interior habitat area and connectivity of unburnt refugia.
Honest limits: what the sensors cannot tell you
The 500 m floor of MCD64A1 is not a minor technical footnote. In savannas managed with small, targeted patch burns for conservation purposes, many individual burns may cover fewer than 25 hectares and will be absent from the MODIS record entirely. Published validation studies in southern African savannas have found MCD64A1 omission errors of 20 to 50 percent for burns smaller than 100 hectares. Sentinel-2 closes much of that gap, but its five-day revisit means a fast-moving fire that ignites, burns and cools between overpasses may still go undetected if cloud cover is present on the relevant acquisition dates.
Fire intensity, in the ecologically meaningful sense of flame height and heat flux, cannot be read directly from post-fire reflectance. Delta-NBR correlates broadly with combustion completeness but conflates fuel load, moisture content and fire behaviour in ways that limit its use as a precise intensity metric. VIIRS fire radiative power (FRP) provides a real-time energy flux estimate in megawatts per pixel, but it captures only the instantaneous state of the fire at the moment of overpass, not the integrated energy release across the burn event.
Finally, cloud cover in the wet-season transition months, precisely when early dry-season burns are most ecologically significant, can reduce Sentinel-2 usable acquisitions to one or two per month across large parts of sub-Saharan Africa and northern Australia. SAR-based burn detection has been explored as a cloud-penetrating alternative, but Sentinel-1 C-band backscatter changes after fire are subtle in open savanna and remain an active research area rather than an operational standard.
From fire maps to conservation decisions
A fire-regime layer becomes operationally useful when it is connected to management questions. Protected-area managers in fire-dependent ecosystems often want to know whether their current burning programme is producing sufficient pyrodiversity across the landscape, or whether large areas are burning at the same time each year, homogenising the habitat. A multi-year fire-frequency and seasonality composite, updated annually, gives a quantitative basis for that assessment without requiring extensive field survey.
The same layer supports adaptive management at finer scales. If a particular zone of a reserve has not burned in eight years and woody encroachment is suspected, the fire-history archive provides the evidence base for a prescribed burn decision. Conversely, if a zone shows fire return intervals of less than two years, the data can support arguments for fire exclusion or altered ignition timing.
Satellize runs this class of analysis on open Sentinel, MODIS and VIIRS archives, with Landsat OLI added for historical depth. The Tonga crop-estimation programme demonstrated the same multi-sensor compositing logic in a different domain; the fire-regime workflow applies analogous time-series methods to a savanna context. Outputs are delivered as annual GIS layers with accompanying summary statistics, ready for integration into existing reserve management systems.
Typical figures
| Burn-area spatial resolution | 500 m (MODIS MCD64A1); 20–30 m (Sentinel-2 NBR / Landsat OLI); 375 m (VIIRS active fire) |
| Revisit frequency | Daily to twice-daily (MODIS, VIIRS); 5 days (Sentinel-2A+B combined); 8 days (Landsat 8+9 combined) |
| Archive depth | MODIS MCD64A1 from November 2000; Landsat from 1972 (MSS); Sentinel-2 from 2015 |
| Minimum detectable burn patch | Approximately 25 ha at MODIS 500 m; approximately 1–4 ha at Sentinel-2 20 m (cloud-free conditions) |
| Key spectral bands | NIR (842 nm), SWIR1 (1610 nm), SWIR2 (2190 nm) for NBR; red-edge (705–783 nm) for post-fire green-up (Sentinel-2) |
| Active-fire detection latency | VIIRS NRT product typically available within 3 hours of overpass |
| Fire radiative power range (VIIRS) | Minimum detectable FRP approximately 5–10 MW under clear-sky conditions |
| Typical cloud impact | Wet-season transitions in tropical savannas can reduce usable Sentinel-2 acquisitions to 1–2 per month |
| Delivery formats | GeoTIFF burn-frequency rasters, GeoPackage / Shapefile burn-polygon archives, CSV fire-history summary statistics |
Analytics Satellize can run
| Annual burned-area mosaic | MODIS MCD64A1 monthly compositing with VIIRS active-fire date assignment | GeoTIFF raster layer showing burn year and month per pixel, updated annually |
| Fire return interval map | Per-pixel frequency counting across MCD64A1 archive (2000 to present), expressed as fires per decade | GeoTIFF raster with decadal fire-frequency classes; summary statistics table by management zone |
| Season-of-burn layer | Modal month of burning per pixel derived from VIIRS NRT active-fire detection dates, aggregated over user-defined period | GeoTIFF raster; histogram of burn-month distribution per zone as PDF report |
| Burn-severity classification | Delta-NBR from Sentinel-2 pre/post image pairs, classified into low / moderate / high severity using published savanna thresholds | GeoTIFF severity raster per fire event; vector polygons with severity attributes |
| Pyrodiversity index layer | Shannon diversity index computed across fire-age classes within moving spatial windows, following published landscape-ecology methods | GeoTIFF pyrodiversity surface; ranked zone comparison table for reserve management reporting |
| Burn-patch geometry metrics | Object-based image analysis on Sentinel-2 / Landsat burn-scar polygons; patch size distribution, edge density and interior-area calculations | GIS polygon layer with per-patch attributes; landscape metrics summary in CSV |
| Multi-year fire-history change report | Comparison of fire-frequency and seasonality composites across user-defined periods (e.g. pre/post management intervention) | PDF report with mapped change layers and statistical significance assessment |
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.