The vector dataset is based on the filtered version (3x3 pixel filter) of the class fire severity classification raster product as derived from a classification of a SPOT and Landsat TM woodlands fire severity indexes to map the fire severity of the February 2009 Victorian bush fires using ground control and air photograph analysis provided by DSE to Ecowise for validation. The noise filtered classification is the final product derived using SPOT 4/5 satellite imagery with Landsat 5 imagery used on small areas where cloud or where SPOT imagery coverage was unavailable. . The five of the eight classes have been labeled based on severity classes provided with the ground survey data. The remaining classes account for vegetation structures with no specific burn severity assigned. The area covered by the classification is approximately 460 by 300 km and is clipped to the extents of the fire effected lands (approximately 400 000 hectares).
Search Words:
Environment
Publication Date:
02 February 2018
Dataset Status:
Complete
Completeness Verification: Incomplete
Progress:
Completed
Access Constraint:
Creative Commons Attribution 4.0 (CC-BY)
(
License Text
)
Data is not to be distributed without authorisation from the Department of Sustainability and Environment (DSE), Victoria
Data Existence:
Metadata Name
Descriptions
Resource Name:
FIRE_SEV09_POLY
Title:
Victorian Bushfires Severity Map 2009 (Polygons)
Anzlic ID:
ANZVI0803003677
Custodian:
Owner:
Jurisdiction:
Victoria
Abstract:
The vector dataset is based on the filtered version (3x3 pixel filter) of the class fire severity classification raster product as derived from a classification of a SPOT and Landsat TM woodlands fire severity indexes to map the fire severity of the February 2009 Victorian bush fires using ground control and air photograph analysis provided by DSE to Ecowise for validation. The noise filtered classification is the final product derived using SPOT 4/5 satellite imagery with Landsat 5 imagery used on small areas where cloud or where SPOT imagery coverage was unavailable. . The five of the eight classes have been labeled based on severity classes provided with the ground survey data. The remaining classes account for vegetation structures with no specific burn severity assigned. The area covered by the classification is approximately 460 by 300 km and is clipped to the extents of the fire effected lands (approximately 400 000 hectares).
Search Words:
Environment
Purpose:
Geographic Extent Polygon:
Geographic Bounding Box:
Beginning to Ending Date:
2009-03-27 - 2009-07-17
Maintainence and Update Frequency:
Not planned
Stored Data Format:
Available Format(s) Types:
DIGITAL - All major formats available
Positional Accuracy:
There are three different data sources that are reliant on spatial position used to derive the product, however, the primary data source (satellite imagery) is an orthorectified product as supplied by DSE Victoria. This product has been found to be accurate to within one hundred metres against all other reference information available. Largest issues are seen in scenes taken at shallow angles and in steep terrain.
Attribute Accuracy:
Error Assessment Landsat
Class | % within class | %of one category in error | %beyond one category in error
1 81.3 18.13 0.58
2 74.3 24.64 1.07
3 58.8 34.83 6.37
4/5a 44.16 53.03 2.8
5b 66.6 23.8 9.6
Error Assessment SPOT
Class | % within class | % of one category in error | % beyond one category in error
1 55.1 42.0 2.8
2 64.7 32.8 2.5
3 50.6 43.8 5.6
4/5a 45.1 50.8 4.1
5b 52.9 41.2 5.9
Error Assessment SPOT (3x3 filter)
Class | % within class | % of one category in error | % beyond one category in error
1 56.8 41.5 1.7
2 70.7 27.8 1.6
3 52.8 42.3 4.9
4/5a 47.6 50.1 2.3
5b 76.5 17.6 5.9
Logical Consistency:
Not specified
Data Source:
Dataset Source: The dataset is a derived product from raw satellite data as acquired by the SPOT 4 and 5 Satellite. The imagery is a product orthorectified by DSE.
Calibration of the imagery has been limited to the application of the gain values as supplied with the SPOT images for the post fire scenes while the pre fire scenes were spectrally matched to the post fire scenes through the use of "invarient" targets including water bodies, highly reflective features and unchanged forests. Where multiple post fire images are mosaiced, the central scene or scene of best quality is used as the reference to which all other scenes are matched.
A fire severty index was prepared through the digital processing of the pre fire and post fire imagery. The analysis is largely based on a ratio of the SWIR and NIR bands though a Transform Vegetation Index (TVI) ratio was used on Wilsons Promitory.
The index for each region was then compared against ground information and split into five classes that can be statistically justified from the six descrete classes supplied by DSE.
The Churchill, Dargo, Redesdale and Horsham areas do not have any ground survey points available for verification. The Churchill fire severity classes were extrapolated across from the Kinglake and Bunyip SPOT scenes while all other areas are classified through visual interpretation of the imagery. In wooded areas within Dargo, Redesdale and Horsham, only severity classes one to three have been interpreted with all other wooded areas left as unclassified burnt woodlands. The Redesdale and Horsham areas are dominated by grassland burns with grasslands either within classes seven or eight dependent on their likelyhood of actually being burnt.
Noise filtering has been performed using a 3X3 majority filter. This filter simply applies the majority value within a 3X3 window to the central pixel thus removing the majority of speckle within the classification.
A basic conversion to vector has been undertaken with no additional simplifications performed. The constructed attributes table contains a measure of area (in hectares) and a class description.
Dataset Originality: Derived