Showing results for 
Search instead for 
Do you mean 

Generating Burn Indices from Satellite Imagery

by Technical Evangelist ‎12-19-2018 12:00 PM - edited ‎01-14-2019 06:21 AM (1,019 Views)

Download models

 Description:

As has been seen extensively in 2018 in the western United States, wildfires can destroy structures, property and lives. Being able to map the exact extent of the fire, both during and after such events, is crucial to efforts to fight the fire, to assist with evacuation orders and to help remediation efforts to prevent further damage due to mudslides. Remote sensing data can provide a vital resource for mapping fire extent whether it be real-time drone imagery or evaluation of large swath satellite imagery such as the freely available Landsat and Sentinel-2 platforms. 

 

Landsat-8 imagery dated (left to right) July 26, August 11 and August 27, 2018, Mendocino Complex fires, California
Fire_Images.jpg

 

The large size, inaccessibility, and spatial variability of wildland fires have caused multispectral satellite data to become a common tool for mapping fire locations and effects. Remotely sensed indices mapping burn severity of wildfires are commonly used by land managers to assess fire effects.

 

The Normalized Burn Ratio (NBR) is a common such technique and has been available via ERDAS IMAGINE's Indices dialog for several years. The NBR works by comparing the NIR band  (which is sensitive to the cell structure of plants and is related to plant abundance and health) and the Shortwave Infrared (SWIR) band (which is sensitive to the cellulose content and water content of the plants, and increases with greater cover of soil, ash, or carbon). In forested systems recent burns have negative NBR values (NIR < SWIR) whereas unburned vegetated has strongly positive NBR values (NIR > SWIR).

 

NBR = (NIR – SWIR) / (NIR + SWIR)

 

This article looks at three other indices which are now in common use.

 

 

Differenced Normalized Burn Ratio (dNBR)

 

Detail of the River Fire, August 11 Landsat-8 (left), dNBR result (middle), Burn Severity categories (right) 
dNBR_Results.jpg

 

Many fire severity mapping applications to date have subtracted a post-fire NBR image from a pre-fire NBR image in an absolute change detection methodology to derive the “differenced NBR” (dNBR) as follows:

 

dNBR = prefireNBR−postfireNBR

 

One advantage of the dNBR technique is that higher values indicate higher severity of burn.

 

 dNBR_thresholds_reflectance_v_16_5_0.gmdx
dNBR_Model.PNG

 

 

Relative differenced Normalized Burn Ratio (RdNBR)

 

 Detail of the River Fire, Landsat-8 (left), RdNBR result (middle), Burn Severity categories (right) 
RdNBR_Results.jpg

Since chlorophyll contents vary due to vegetation type and density, each absolute differenced image should ideally be stratified by pre-fire vegetation type and independently calibrated. Miller and Thode (2007) therefore proposed the creation of a Relative differenced NBR (RdNBR) to remove the biasing of the pre-fire vegetation by dividing dNBR by the square-root of the pre-fire NBR as follows:

 

RdNBR = dNBR / Sqrt (ABS (prefireNBR /1000))

 

Positive RdNBR values represent a decrease in vegetation cover, just like dNBR, while negative values represent an increase in vegetation cover.

 

RdNBR_thresh_reflectance_v_16_5_1.gmdx
RdNBR_Model.PNG

 

 

Normalized Burn Ratio Thermal (NBRT)

 

Detail of the River Fire, Landsat-8 (left), NBRT result (right)
NBRT_Results.jpg

This index uses a thermal band to enhance the NBR. It theoretically results in a better separability between burned and unburned land. 

 

NBRT = (NIR - SWIR * (Thermal/1000)) / (NIR + SWIR * (Thermal/1000))

 

The Thermal band in this equation must be calibrated to brightness temperatures (in Kelvins). The NIR and SWIR bands should be calibrated to top-of-atmosphere reflectance (as they should be for all these techniques). Consequently the model converts to Reflectance and temperature in Kelvins if a Metadata file (*_mtl.txt) is provided as input.

 

NBRT_v16_5_0.gmdx
NBRT_Model.PNG

 

Burn Severity Categorization

 

NBR, dNBR, RdNBR and NBRT are all Indices and therefore result in a continuous, athematic output (usually stored as floating point values, or sometimes multiplied by 1000 and truncated to integer), which can sometimes be difficult to interpret. End-users often prefer a thematic map of burn extent or of categorized burn severity (also known as a Burned Area Reflectance Classification or BARC). Consequently a couple of the provided Spatial Models (dNBR and RdNBR) include an optional output which thresholds the index result into burn severity classes. The Preview ability of the Spatial Modeler can be used to manipulate the default threshold values to produce the desired results prior to clicking Run and producing an output file. 

 

If you do not wish to output a Burn Severity Raster leave that input field blank and that section of the Spatial Model will not be executed when Run.

 

Conversely if you only want a Burn Severity Raster (BARC) output file and do not want the continuous dNBR/RdNBR Raster file you can leave that field blank. Obviously if you leave both output raster fields blank the Spatial Model will produce no output files!

 

Assumptions

 

The provided Spatial Models were built with Landsat-8 MSI & TIR imagery in mind. If using other sensors care should be taken to identify the appropriate input band numbers corresponding to NIR, uSWIR and possibly TIR bands). For example, in Landsat-7 imagery, the NIR band is often band 4, whereas it is 5 for Landsat-8.

 

If the input data is already corrected to Reflectance the optional Metadata File input fields can be left blank. If blank the spatial model bypasses the step which applies the Landsat-8 TOA reflectance conversions (i.e. the input is assumed to already represent Reflectance values)

 

If the Reflectance conversion is to be applied, the appropriate Metadata file containing adjustment parameters must be provided for each input image. For Landsat-8 images this is usually a file with the naming pattern *_mtl.txt. 

 

If the input data requires Reflectance correction, but is not Landsat-8, the Sensor port on the Read sensor Metadata operators will need to be altered appropriately and the subsequent Dictionary Item operators will need to be altered to pull the correct field values required by the reflectance calculation. The Convert to TOA Reflectance sub-models may also require modification depending  on the reflectance calculation required for the particular sensor.

 

The default Burn Severity thresholds are based on the paper by Key and Benson (for the dNBR technique).

 

References: 

 

USE OF MULTIPLE SPECTRAL INDICES TO ESTIMATE BURN SEVERITY IN THE BLACK HILLS OF SOUTH DAKOTA: http://www.asprs.org/a/publications/proceedings/pecora17/0012.pdf

Key and Benson : http://www.fs.fed.us/rm/pubs/rmrs_gtr164/rmrs_gtr164_13_land_assess.pdf

http://www.fs.fed.us/postfirevegcondition/documents/publications/miller_etal_rse_2009.pdf

Common terms - National Fire Danger Rating System (NFDRS): http://gacc.nifc.gov/rmcc/predictive/fuels_fire-danger/drgloss.htm

USFS Wildland Fire Assessment System (WFAS) http://www.wfas.net/

CBI: remotesensing-04-00456.pdf

https://projects.sfchronicle.com/2018/fire-tracker/

  

 

Courses
Contributors