Department of the Environment, T. (2021). Annual Fire Scars - Landsat, QLD DES Algorithm, QLD Coverage. Version 1.0. Terrestrial Ecosystem Research Network. Dataset. https://portal.tern.org.au/metadata/461074b3-5272-4e4e-886f-df26bd2426ad
Annual Fire Scars - Landsat, QLD DES Algorithm, QLD Coverage
Ver: 1.0
Status of Data: completed
Update Frequency: annually
Security Classification: unclassified
Record Last Modified: 2025-12-02
Viewed 2527 times
Accessed 456 times
Dataset Created: 2014-10-14
Dataset Published: 2021-09-23
How to cite this collection:
The data set is a statewide annual composite of fire scars (burnt area) derived from all available Landsat 5, 7 and 8 images acquired over the period January to December using time series change detection. Fire scars are automatically detected and mapped using dense time series of Landsat imagery acquired over the period 1987 - present. In addition, from 2013, products have undergone significant quality assessment and manual editing. The automated Landsat fire scar map products covering the period 1987-2012 were validated using a Landsat-derived data set of over 500,000 random points sampling the spatial and temporal variability. On average, over 80% of fire scars captured in Landsat imagery have been correctly mapped with less than 30% false fire rate. These error rates are significantly reduced in the edited 2013-2016 fire scar data sets, although this has not been quantified.
For the 2016 annual fire scar composite, the manual editing stage incorporated Landsat and Sentinel 2A imagery (resampled to match Landsat spatial resolution), allowing for increased cloud-free ground observations, and an associated reduction in the number of missed fires (not quantified). Sentinel 2A images were primarily used to map fire scars that were otherwise undetectable in the Landsat sequence due to cloud cover/Landsat revisit time. Additionally, Landsat-7 SLC-Off imagery (affected by striping) was excluded from the 2016 annual composite. It is expected that these modifications should result in improved mapping accuracy for the 2016 period.
A new fire scar detection algorithm has been developed, with a new edited product implemented in 2021.
For the 2016 annual fire scar composite, the manual editing stage incorporated Landsat and Sentinel 2A imagery (resampled to match Landsat spatial resolution), allowing for increased cloud-free ground observations, and an associated reduction in the number of missed fires (not quantified). Sentinel 2A images were primarily used to map fire scars that were otherwise undetectable in the Landsat sequence due to cloud cover/Landsat revisit time. Additionally, Landsat-7 SLC-Off imagery (affected by striping) was excluded from the 2016 annual composite. It is expected that these modifications should result in improved mapping accuracy for the 2016 period.
A new fire scar detection algorithm has been developed, with a new edited product implemented in 2021.
Credit
We at TERN acknowledge the Traditional Owners and Custodians throughout Australia, New Zealand and all nations. We honour their profound connections to land, water, biodiversity and culture and pay our respects to their Elders past, present and emerging. Landsat 5 TM, Landsat 7 ETM+ and Landsat 8 OLI images were acquired from United States Geologic Survey.
Copernicus Sentinel 2 data from the European Union and European Space Agency Copernicus Program.
Purpose
Characterising historic patterns of burning and changing fire regimes over time (spatial extent, timing, patchiness, frequency and intensity) is important for improving our understanding and management of fire, climate, land-use and vegetation interactions. These products may assist the development of appropriate fire management practices and benefit a range of conservation and resource management objectives, as well as ongoing scientific research.
Lineage
Applied outlier identification algorithm to timeseries of geometrically and radiometrically corrected Landsat and Sentinel 2 imagery, as described in https://doi.org/10.1016/j.rse.2014.03.021
Method DocumentationData not provided.
Procedure Steps
1.
Algorithm Summary:
Raster pixel values correspond to the month of detection (often different from the date of active fire). A pixel is mapped as burnt if there has been a significant change in reflectance relative to the time series due to the effects of fire e.g. presence of charcoal or ash, removal of foliage, scorch. Pixel values:0: no fire scar was detected during this period;1-12: month (of Landsat acquisition) when fire scar was first detected;254: crop/water masked - no fire scar detection conducted. For the 2014-2016 products, crop and water masking was applied during processing and these areas are not attributed as 254;255: no data value.Note: fire scars may persist and continue to be detected for several months in the image time sequence. The earliest month of detection within the compositing year is recorded.Data sets are 8 Bit GeoTiff with LZW compression and tiling (BigTIFF).
See https://doi.org/10.1016/j.rse.2014.03.021
See https://doi.org/10.1016/j.rse.2014.03.021
2.
Algorithm Accuracy:
Landsat does not provide a complete record of fire history for this period. This is mostly due to the sensor revisit time of 8-16 days which may be further limited by cloud and cloud shadow obstruction and striping in the imagery.
Annual composites from 2003 onwards are affected by data loss due to systematic striping in the Landsat-7 ETM+ imagery. This is due to the failure of the instrument's Scan Line Corrector (SLC). This is increasingly apparent in imagery acquired in the period 2010-2013, prior to the launch of Landsat-8, as image transmission from the Landsat-5 TM (unaffected by striping) was limited and finally ceased in November 2011 due to sensor malfunction. This has resulted in striping and fragmentation in the fire scar maps derived from SLC-Off imagery.
3.
Omissions:
In the 1987-2012 automated fire scar products, the average fire scar omission error for the State was measured at 15%. The omission error does not include fire scars missed due to Landsat data loss e.g. SLC-Off striping, or gaps in the Landsat record e.g. due to cloud or revisit time. This has not been quantified due to the lack of a validation data set which is independent from the sensor being used (Landsat). In addition, omission errors are likely to be higher for fire scar composites containing Landsat-7 SLC-Off striping and for wet season periods (Nov-February) in tropical and coastal regions where cloud cover may obscure the view of the surface for months at a time.
A fire scar signal may not be evident in the image sequence for long time periods, particularly in savanna regions in North Queensland. Ash/char can be blown or washed away over short periods of time (~weeks) and the fire scar is often rapidly masked by green-flush and vegetation resprouting in subsequent images.
4.
Omissions:
Additionally, fires may be captured in the Landsat imagery but missed or under-mapped by the classifier for the following reasons: the fire may be too small or patchy to detect; cool grass/understorey fires may be obscured by the unburnt tree canopy; or the fire may be misclassified as non-fire related change or cloud shadow.
An assumption that burnt areas decline in reflectance over time may not always be true and missed fire scars have been noted (e.g. spinifex grasses).
The additional step of manual editing applied to the 2013-2016 fire scar data sets should reduce the number of missed fires (due to misclassification) to well below the measured omission rate of 15%, although the edited mapping has not been validated. For 2016, the incorporation of Sentinel 2A into the Landsat image sequence at the manual editing stage has allowed for increased cloud-free ground observations, improved interpretation, and mapping of fire scars which might otherwise be missed in the Landsat sequence.
5.
False Fires:
In the 1987-2012 automated fire scar products, the average rate of false fires across Queensland was measured at 30%. This is likely to be less in some regions and more in others. False fires are far more common in image dates affected by SLC-Off striping as the data is fragmented and less reliable.
False fires or over-mapping of fire scars may result from the presence of cloud shadows, areas of high land-use change (e.g. cropping), black soils, and inundation e.g. tidal flats, wetlands, ephemeral lakes and channels. These features often spectrally and temporally resemble fire scars.
The additional step of manual editing applied to the 2013-2016 fire scar data sets should reduce the number of false fires to well below the measured rate of 30%, although the edited mapping has not been validated.
Queensland
Temporal Coverage
From 1987-01-01 to 2016-12-31
Spatial Resolution
Data not provided.
Vertical Extent
Data not provided.
Data Quality Assessment Scope
The input imagery was processed to level L1T by the USGS. Geodetic accuracy of the product depends on the image quality and the accuracy, number, and distribution of the ground control points.
Data Quality Report
Data not provided.
Data Quality Assessment Outcome
The USGS aims to provide image-to-image registration with an accuracy of 12m. Refer to the L8 Data Users Handbook for more detail.
For 1987-2012 automated fire scar products, the average fire scar omission error was measured at 15%.
Horizontal Resolution
Parameters
Temporal Resolution
Terrestrial Ecosystem Research Network
80 Meiers Road, Indooroopilly, Queensland, 4068, Australia.
Contact Us
TERN services are provided on an “as-is” and “as available” basis. Users use any TERN services at their discretion and risk. They will be solely responsible for any damage or loss whatsoever that results from such use including use of any data obtained through TERN and any analysis performed using the TERN infrastructure. Web links to and from external, third party websites should not be construed as implying any relationships with and/or endorsement of the external site or its content by TERN.