After a wildfire, mapping the affected area is only the first step. Forest managers also need to understand where vegetation changed most, where field inspections are needed, and which areas should be monitored during recovery. Comparing satellite imagery acquired before and after a fire can support these decisions, provided the analysis distinguishes measured spectral change from confirmed ecological damage.<\/p>\n
The calculations are relatively simple. The harder work is selecting comparable images, deciding what the index values mean in the local landscape, and checking the classification against independent evidence. A credible assessment preserves that distinction throughout the workflow.<\/p>\n
These terms describe different aspects of a wildfire:<\/p>\n
A pixel classified as high spectral burn severity therefore does not, by itself, establish a particular soil temperature, percentage of dead trees or loss of soil infiltration capacity. Those conclusions require measurements or locally validated relationships specific to the variable being reported.<\/p>\n
An image acquired soon after containment and an image acquired during the following growing season describe different stages of fire effects.<\/p>\n
An initial assessment<\/strong> captures relatively immediate changes, including vegetation loss, scorching and exposed burned surfaces. An extended assessment<\/strong> can capture delayed effects and vegetation response over a longer interval. Neither date is automatically better: the choice depends on whether the task is rapid inspection planning, ecological assessment or recovery monitoring.<\/p>\n A 2015 study in Fire Ecology<\/a> compared immediate and extended assessments against field-based severity measures in California conifer forests. Its findings demonstrate why assessment timing and the particular effect being estimated must be evaluated together, rather than assuming that one post-fire date suits every application.<\/p>\n For image selection, check the following:<\/p>\n If no suitable image is available, record the limitation. A later observation can fill a coverage gap, but it should retain its own acquisition date and assessment context.<\/p>\n For a new analysis, consistently processed surface-reflectance imagery provides a practical starting point. Keep the sensor, processing level, band selection and spatial grid consistent across dates wherever possible. Combining different sensors requires additional checks because their spectral responses and pixel footprints differ.<\/p>\n For Landsat 8 and 9, NBR commonly uses the near-infrared band B5 and shortwave-infrared band B7, both at 30 m resolution. For Sentinel-2, B8A and B12 provide a common pairing at 20 m. Workflows also use the broader B8 near-infrared band, but that choice must be documented and held consistent.<\/p>\n Resampling Sentinel-2 shortwave-infrared data onto a 10 m grid does not create an independently observed 10 m burn-severity measurement. Small burned patches and mixed forest edges remain subject to the information available in the original bands.<\/p>\n Apply the product\u2019s reflectance scaling and any additive offsets before calculating indices. For example, Landsat Collection 2 Level-2 surface-reflectance values require both a scale factor and an offset. Additive offsets do not cancel in a normalized ratio.<\/p>\n Mask clouds, cloud shadows, snow, invalid observations and other obscured areas in both dates. Retain these masks as a separate quality layer so that missing observations cannot be mistaken for low severity.<\/p>\n The Normalized Burn Ratio combines near-infrared and shortwave-infrared reflectance:<\/p>\n Healthy vegetation often has relatively high near-infrared reflectance and lower shortwave-infrared reflectance. Burning can change this relationship through vegetation loss, drying and exposure of different surface materials.<\/p>\n A low post-fire NBR can help identify a burn scar, but low values also occur over some unburned surfaces. One image alone provides limited evidence about how much a site changed.<\/p>\n The differenced Normalized Burn Ratio compares the pre-fire and post-fire indices:<\/p>\n Larger positive values generally indicate greater fire-related spectral change when the images are comparable. Negative values can reflect increased greenness or moisture, seasonal differences or other changes; they should not automatically be labelled post-fire recovery.<\/p>\n Keep track of numerical units. A floating-point dNBR of 0.60 is equivalent to 600 in a product scaled by 1,000. Thresholds must match the units of the raster being classified.<\/p>\n The relativized differenced Normalized Burn Ratio adjusts dNBR using pre-fire NBR. It was developed to address the influence of initial vegetation conditions: a sparse site has less vegetation available to lose than a dense forest, which affects the magnitude of absolute spectral change.<\/p>\n A 2009 study in Remote Sensing of Environment<\/a> calibrated and validated RdNBR against field measures in California\u2019s Sierra Nevada and Klamath Mountains. Such calibration provides evidence for defined ecosystems and response variables; it does not establish a universal conversion from RdNBR to damage.<\/p>\n Implementations must handle values near zero in pre-fire NBR carefully because the normalization can become unstable. Document the formula, scaling, any background-change adjustment and the treatment of unsuitable pixels. RdNBR still depends on sound image selection and validation.<\/p>\n NBR can be an input to a supervised classifier, while dNBR or RdNBR can also be classified directly using thresholds. These are different approaches. Neither produces reliable severity classes without an appropriate reference.<\/p>\n If a supervised model learns from labels generated by thresholding dNBR, agreement with those labels shows how well it reproduces the threshold scheme. It does not independently demonstrate ecological accuracy.<\/p>\n MTBS explicitly describes categorical severity mapping as an interpretive process. Analysts examine index values alongside pre-fire and post-fire imagery, ecological context and available reference information. The meaning of a class can vary across landscapes, so a published threshold table should be treated as a candidate scheme requiring evaluation.<\/p>\n A practical approach is to retain the continuous index, inspect its distribution within relevant vegetation types, and evaluate proposed class boundaries against independent observations. If evidence is insufficient to distinguish several classes, a broader classification may be more defensible than a detailed legend.<\/p>\n UN-SPIDER\u2019s published Sentinel-2 workflow uses Empedrado in Chile\u2019s Maule Region to demonstrate assessment following the 2017 wildfires. It specifies a pre-fire image from 20 December 2016<\/strong> and a post-fire image from 18 February 2017<\/strong>, using tile T18HYF<\/strong>.<\/p>\n The published sequence connects the original observations to the final classes:<\/p>\n\n
Prepare Comparable Reflectance Data<\/h2>\n
What NBR, dNBR and RdNBR Measure<\/h2>\n
NBR: The Surface Condition at One Date<\/h3>\n
NBR = (NIR \u2212 SWIR2) \/ (NIR + SWIR2)<\/code><\/p>\ndNBR: The Difference Between Two Dates<\/h3>\n
dNBR = NBR before fire \u2212 NBR after fire<\/code><\/p>\nRdNBR: Adjusting for Pre-Fire Conditions<\/h3>\n
From an Index to a Defensible Classification<\/h2>\n
A Public Example: Empedrado, Chile<\/h2>\n