Large fires have gone from being isolated episodes to becoming a recurring threat with ecological and social impacts increasingly visible. In this context, knowing and quantifying the severity of the fire is no longer just a technical matter: it is key to prioritizing restorations, planning the forest fire prevention and understand how much and how the landscape changes after the flames pass.
The assessment of burned areas and damage severity increasingly relies on remote sensing, with sensors such as Sentinel-2 or Landsat, which allow for capturing before and after events with remarkable precision. Thanks to spectral indices such as NBR and its temporal difference (dNBR), it is possible delineate scars, grade the intensity of the damage and monitor recovery, avoiding bias when applying appropriate time windows and masking clouds and artifacts.
What do we mean by fire severity?
Severity measures the degree of structural and functional alteration that fire causes in the ecosystem: from aboveground vegetation to the soil, the seed bank, and local hydrology. It's not limited to the extent of what's burned; it also describes how the living and the physical support that sustains it have changed.
Recent literature shows that severity directly conditions the time of forest recovery: the greater the severity, the slower and more difficult it is to return to previous coverageIn very intense fires, not only canopies and brush are burned; seeds, shoots, and much of the organic layer of the soil can also be destroyed, complicating natural regeneration.
A large global analysis of 3.281 fires between 2001 and 2021 notes that the frequency of large-scale fires and their severity have increased, with 2010 as the most turning point associated with the heating, droughts and extremesThe most exposed areas include western North America, southeastern Australia, northern South America, southern Asia, and central-eastern northern Siberia, where the impacts on forest structure are particularly pronounced.
On average, a forest can take about four years to recover its vegetative density after a fire, although the most sensitive ones need several more months, and there are cases where the recovery stagnates and declines. Therefore, measures are advocated for Ecological restoration and reforestation that accelerate resilience and preserve the climatic function of forests.
Remote sensing to map burned areas and severity
The Sentinel-2 and Landsat satellites have democratized access to quality multispectral imagery. The combination of near-infrared (NIR) and shortwave infrared (SWIR) bands allows for the detection of live and structural vegetation changes after fire, relying on the spectral signature of healthy vegetation, which reflects strongly in NIR and decreases towards SWIR, and in damaged vegetation, which shows the opposite pattern.
The Normalized Burn Ratio (NBR) is calculated as (NIR − SWIR) / (NIR + SWIR). Sentinel-2 typically uses band 8A for NIR and band 12 for SWIR; Landsat 8 typically uses band B5 (NIR) and band B7 (SWIR2). This normalization highlights the transition from healthy areas (more positive values) to burned or severely affected areas (negative or reduced values). NBR itself already helps to identify scarsHowever, its true potential appears when we compare two dates.
The dNBR is obtained by subtracting the post-fire NBR from the pre-fire NBR (dNBR = NBRpre − NBRpos). In doing so, We limit false positives on non-vegetated surfaces with low NBR values (roads, bare soils) and we highlight real changes due to fire. In practice, dNBR typically ranges from approximately −0,5 to +1,3, with high, positive values indicating greater severity and very negative values reflecting vigorous regeneration.
NIR and SWIR-based RGB compositions are a complementary visual aid. For example: 8‑4‑3 (Sentinel‑2) highlights vegetation in red; 4‑3‑2 (natural) allows intuitive reading; 12‑8A‑4 emphasizes thermal impacts and residual humidityIt is even possible to distinguish active fronts and plumes of smoke Playing with SWIR. These are approaches that, together with indices such as NBR or alternatives such as BAIS2, complete the diagnosis.
Workflow in QGIS with Sentinel‑2
To assess burned areas and severity, a typical sequence in QGIS encompasses data preparation, atmospheric correction, cloud masking, NBR/dNBR calculation, and classification of severity into categories, with statistics and final maps. A consolidated flowchart based on USGS and UCG procedures is provided below.
1) Image preparation and loading
Select two dates: one before the fire and one after. For Sentinel-2 Level 1C, work with bands 2, 3, 4, 8, 8A, and 12. You can open the bands from the browser panel or from Layer → Add Layer → Add Raster Layer, navigating to the product directory: PRODUCT → GRANULE → L1C… → IMG_DATAIn real-life studies, scenes close to the event are used (e.g., July vs. October in the same tile) to reduce unrelated changes.
The selected bands (especially 8A and 12) are the ones that support the NBR. Also loading 2, 3, 4 and 8 allows you to build display combinations that help interpret fire behavior between bands now better locate the area of interest before calculating indices.
2) TOA corrections and preprocessing
Before indexing, apply the TOA correction in the Semi-Automatic Classification Plugin (SCP). In SCP → Preprocess → Sentinel‑2, select the band folder and the MTD_MSIL1C metadata file, and select the atmospheric correction option. DOS1 and keep the NoData value for the black borders. Remove the bands you won't be using from the list (keep 2, 3, 4, 8, 8A, and 12) and run the process.
The progress is shown in the QGIS bar, and when finished, you'll see the new layers with an identifiable prefix (e.g., “RT”). Repeat the same procedure. pre-processed for the post-fire date so that both scenes are homogeneous and comparable.
3) Cloud masking
Level 1C clouds have a vector mask in QI_DATA → MSK_CLOUDS_B00.gml. Load it, save it as a shapefile, and set the CRS you are using (e.g., EPSG:32717 in some projects). In its attribute table, enable editing and create an integer field (e.g., "value") by assigning the number 1 to all geometries. Save and close edition.
Repeat the process for the second date and, if necessary, join both shapefiles on a single cloud layer (e.g., “Total_Clouds.shp”). Note that the mask included in L1C is useful, but not perfect: thin clouds or shadows may escape, so it’s a good idea to check it visually.
4) Band combinations for a first reading
Before calculating indices, generate color composites to explore scars and relevant elements. 8-4-3 emphasizes live vegetation in red and soils in brown; 4-3-2 offers a intuitive natural vision; and 12-8A-4 helps highlight moisture and residual heat. In urban areas, you'll see light cyans, recent burns can tend toward dark browns, and conifers often appear darker than hardwoods.
Calculation of NBR and dNBR
With the scenes corrected, calculate the NBR for each date. For Sentinel-2: NBR = (B8A − B12) / (B8A + B12). For Landsat 8: NBR = (B5 − B7) / (B5 + B7). This index ranges between −1 and 1 and captures the contrast between living vegetation (high NIR, lower SWIR) and damaged vegetation or burned soils (low NIR, high SWIR).
Once you have the pre and post NBRs, compute the dNBR as NBRpre − NBRpos. In practice, the dNBR can vary from approximately −0,5 to +1,3: the larger and more positive the value, the greater the severity; values close to zero usually indicate stability/no burn; very negative values indicate vigorous regrowth after fireIt is common to multiply by 1.000 to facilitate reading and classification by discrete ranges.
To visualize the NBR (and also the dNBR), define an appropriate palette. Tools like ColorBrewer can guide you in choosing perceptually ordered ramps with sufficient contrast. In the final mapping, it's a good idea to include legible legends and a descriptive titleFor example, some teaching exercises suggest a two-line title: “Fire Severity” and “Kalamos, Greece.”
Severity thresholds and reclassification
There are widely used classification schemes. One, with dNBR scaled ×1000, separates the categories as follows: vigorous regrowth (−500 to −251), moderate regrowth (−250 to −101), unburned (−100 to 99), low severity (100 to 269), low-moderate (270 to 439), moderate-high (440 to 659), and high severity (660 to 1300). These ranges allow mapping the spatial intensity of the damage. with a structured legend.
Another widely used reference (USGS) in unscaled values indicates: less than −0,25 (high postfire growth), between −0,25 and −0,1 (low postfire growth), between −0,1 and 0,1 (stable/unburned), 0,1 to 0,27 (low severity), 0,27 to 0,44 (low-moderate), 0,44 to 0,66 (moderate-high), and greater than 0,66 (high). When reclassifying, choose the scheme consistent with your scale (raw or ×1000) and documents the choice well.
In QGIS, you can build a reclassification matrix that translates dNBR values into class codes, symbolize them with a color ramp, and generate the legend. Placing the legend on the right side of the map is common practice for optimize reading in final composition. If needed, export the result as a raster for further use.
Statistical analysis and derivative products
Beyond the map, it's interesting to know the distribution of areas by severity category. You can calculate histograms or tables with the number of pixels per class and their percentage of the total. Add labels with counts and percentages facilitates communication with managers and planners.
In advanced workflows, it is common to build a rasterStack or rasterBrick with well-defined key bands and names, so that the calculation of indexes, clips, and masks is reproducible. If you work with Landsat, remember to convert DN (digital values) to reflectance using official calibration factors (OLI/TIRS) to ensure comparability between scenes.
Delimiting the study area is another good practice: create an AOI (fire-adjusted extent) and clip the rasters. This reduces noise outside the relevant area and speeds up rendering. Compare band combinations between dates helps identify smoke plumes or hot spots at the event and improves visual validation of the dNBR.
Limitations, common mistakes and good practices
Severity analysis is, in essence, a change detection exercise. Therefore, non-fire-related changes (deforestation, land clearing, phenological or humidity variations) can be confused with fire damage, especially at low severities. A poorly chosen time window (too long) increases the likelihood of conflating processes.
To minimize errors, use dates as close to the event as possible, check the cloud and shadow mask, and use RGB composite controls. Keep in mind that the Sentinel‑2 L1C cloud mask is useful but not perfect: may allow cirrus clouds or complex shadows to pass throughVisual inspection and support with auxiliary layers (e.g., slopes or land uses) improve the robustness of the result.
Low severity classes require special care: small variations in humidity or phenology can cause subtle changes in NIR/SWIR that result in falsely low dNBR. In large areas with temporal mosaics, radiometric consistency and homogeneity in the pre-processing (same correction, same projection, same bands) are key to an honest comparison.
Practical applications: from the Iberian Peninsula to the rest of the world
In the Valencian Community, where the Mediterranean climate, relief, and forest continuity increase the risk, quantifying the burned area and severity is considered a priority. Cases such as Bejís (Castellón, 2022) or the Luchente fire (Valencia, 2018) have prompted the use of Sentinel-2 with standardized methodologies (USGS and UCG) to delimit scars and grade damage with precision.
Applications have been developed that automate these calculations, faithfully following the USGS procedure or the UCG methodology, so that technicians, managers, and citizens can obtain comparable severity maps and statistics. These tools bring together Pre- and post-fire data, mask clouds, calculate NBR/dNBR, reclassify by ranges and generate cartographic outputs ready for dissemination.
Globally, trends point to an increase in the incidence of large-scale fires of 14% in the short term and potentially up to 50% by the end of the century. These figures, combined with the observation that Recovery slows as severity increases, reinforce the need for systematic monitoring and active restoration and prevention policies. increased incidence It is a warning sign for managers.
Spectrality, indices and visualization: pieces that fit together
The key to NBR is in the spectral response. Live vegetation: high NIR, lower SWIR; burned or stressed vegetation: NIR drops and SWIR rises. This contrast gives rise to the signal. Alternatives such as BAIS2 (specifically proposed for Sentinel-2) can complement or validate the diagnosis, especially if criteria are pursued. robustness against bright backgrounds or variable radiometric conditions.
Classic RGB compositions with NIR and SWIR, as used in world-famous cases (e.g., fires in the Amur region), work very well as an inspection tool: they help to locate active fronts, feathers and scars before even trading indices. But remember: for consistent severity grading, the comparative dNBR remains the gold standard.
Symbology and legend matter. Using ramps with a logical progression (from green/gray for unburned to intense reds for high severity, for example) makes it easier for the end user to understand the map at a glance. Add concise titles, scale, and references; and export the severity raster when you need post-analysis or integration into web viewers.
Quality operating tips
Before calculating indexes, ensure that both dates share the same reference system, resolution, and extent. Use AOI clipping to avoid edge artifacts and normalize layer nomenclature to don't get lost between pre and post versions. Document your flow: what atmospheric correction you applied, what thresholds you used, and what cloud mask you generated.
If you're working with Landsat, always convert from DN to reflectance using the official product constants; if you're working with Sentinel-2 L1C, preprocessing with SCP and DOS1 correction is an extended and effective starting point for separating atmospheric disturbances of surface reflectanceAnd, if possible, check several post-fire time windows (immediate, weeks later) to better understand the dynamics of ash and initial regeneration.
For reporting purposes, please include a statistics table (area by class and percentage) and a brief methodological note explaining the chosen dNBR, the band set, and how to mask clouds. This streamlines peer review, facilitates decision-making by managers and reduces misunderstandings about the interpretation of colors and classes.
Risks, recovery and management
The link between severity and recovery is close. While low severities can allow rapid regrowth, high severities often involve loss of seeds and organic horizons, which the forest takes longer to return to previous states or may remain anchored in scrub or grassland trajectories. In boreal regions, the tendency toward slow recovery is more pronounced due to their intrinsic conditions.
With scenarios of increasing severe fires, ecological restoration (from soil retention to assisted reforestation) will be increasingly necessary, as well as the use of operational products based on Sentinel-2 and Landsat that offer indicators comparable in time and spaceIn this way, critical areas are prioritized, results are monitored, and prevention feedback is provided.
For emergency managers, having two established methodologies (USGS and UCG) provides confidence. The first is standard in the use of dNBR; the second is applied in multiple case studies in Spain with specific guidelines. delimitation and severity assessment using Sentinel‑2. Both converge on the essentials: good temporal preselection, adequate correction, well-masked clouds, and transparent classification.
Measuring the severity of wildfires with NBR and dNBR, supported by reproducible flows in QGIS and clear classifications, allows the impact of fire to be translated into figures and maps that guide the recovery, planning and public communicationWith a relevant time window, carefully selected cloud masks, and consistent thresholds, the assessment gains reliability and becomes a practical land management tool.
