Skip booklet index and go to page content

Technical Thematic Report No. 6. - Trends in large fires in Canada, 1959 to 2007

Methodology

An analysis of long-term forest fire characteristics was performed using a combination of data from the large fire database and data derived from remote sensing methods. The primary national source of historic fire occurrence is the large fire database (Stocks et al., 2003). The database spans from 1959 to 1999 and includes all documented fires larger than 2 km². Although these large fires only make up approximately 3% of the total number of fires that occur, they represent 97% of the total area burned. Documentation of fires in the database is from aerial mapping and/or remote sensing gathered from the provinces, territories, and Parks Canada. A limitation of the large fire database is missing data in northern parts of the country prior to the 1970s (Stocks et al., 2003). Caution should therefore be exercised in interpretation of the results for these early years in the northern ecozones+. These omissions were corrected with the advent of satellites in the 1970s. Also, complete mapping of large fire polygons did not start to occur until the 1980s, preventing spatial analysis of large fires prior to this date.

More recently remote sensing methods have become more widely used to detect and map burned areas. For the analysis for this report, remote sensing of burned areas was completed for 1995 to 2007 using the HANDS (Hotspot and NDVI – Normalized Difference Vegetation Index – Differencing Synergy (Fraser et al., 2000)) processing method that uses AVHRR (Advanced Very High Resolution Radiometer) or SPOT-VGT (Système Probatoire d'Observation Terrestre – VEGETATION) satellite images obtained from the Canadian Centre for Remote Sensing. Fires detected from these methods have been shown to be as good as conventional large fire mapping methods (Fraser et al., 2000; Fraser et al., 2004). The caveat to coarse remote sensing methods is that areas burned are often over-estimated because they include large unburned islands within the fire polygon. Also, remote sensing data do not include metadata from the fire protection agencies (provinces, territories, and Parks Canada) that identifies the cause of fires. Lastly, because burned area maps are calculated at the end of the fire season, the start and end dates of the fires are not provided through this data source.

To make the best use of available data a combination of the large fire database and remote sensing data were used. For the analysis of area burned, data from the large fire database was used from 1959 to 1994, and remote sensing data was used to complete the record from 1995 to 2007. Based on a comparison of fire polygons during the period of overlap (1995 to 1999) there was a 1.7% difference in burned area between the large fire database and remote sensing methods.

Average annual area burned was calculated as the total area burned from 1959 to 2007 divided by the number of years (49). Percent annual forested area burned was calculated as the annual area burned divided by the forested area of the region of concern, multiplied by 100. Forested areas were defined as those areas that contained forest fuels, that is conifer and deciduous forests, shrubland, grasslands, previous burned areas, and woodlands interspersed within croplands. These forested areas were calculated by subtracting water bodies and other non-fuel areas (croplands, urban and built-up areas, snow and ice, and barren land) from the total ecozone+ area. Land cover areas were determined by averaging data from land cover matrices developed by Ahern (2011) for 1985 to 2005. Land cover maps were not available for this study prior to 1985, so an assumption was made that there were minimal changes in forest cover from 1959 to 1985.

Long-term trends in annual area burned were calculated by summing the area burned by decade. Totals for the 2000s decade were pro-rated over 10 years based on the average from 2000 to 2007. The seasonality and cause of large fires (those > 2 km²) was derived from data in the large fire database for 1959 to 1999 inclusive.