Forest fire detection

june25_95 copy.jpg (29792 bytes)

 

Conventional ground-based or airborne fire detection techniques are severely limited because they can only cover small areas. For a country the size of Canada, a large-area monitoring capability would provide a definite advantage.  Researchers in Environmental Monitoring have developed algorithms to detect forest fires  [1].  The fire (hotspot) detection algorithm exploits the infrared channels (2-5) from AVHRR. The algorithm consists of two major steps: marking potential fire pixels using the strong thermal signal in the mid-infrared channel (3.7 :m), then removing false fires using a series of threshold tests (Fig. 1).  Threshold tests were optimized to detect the majority of real fires, while removing as many false fires as possible.  Thresholds were chosen following a trial-and-error approach based on a 1,200 by 1,200 km training data set.  Histogram analyses of reflectance and brightness temperature corresponding to burning and non-burning pixels provided an effective means of obtaining optimized values. 

The performance of the algorithms was evaluated against ground-based fire observation data obtained by provincial fire agencies in Saskatchewan and Manitoba. The test site, which is the study region of the Boreal Ecosystem-Atmosphere Study (BOREAS), spans an area of 1000 by 1000 km across the two provinces [3].

The hotspot algorithm was applied to archived single-date GEOCOMP composites for the years 1994-2000, producing a daily fire mask for all of Canada.  Since 1998 summer the algorithm has been applied operationally.  Images showing active fires and smoke were generated within 6-10 hours of satellite reception, integrated with fire behavior information into GIS system, and put on a web site for users of Canadian fire management agencies and media. 

Comparing the 1994 and 1995 locations to provincial fire surveys, the algorithm was found to detect the vast majority (> 90%) of fires larger than 300 ha.  Within individual burns, however, the distribution of hotspots is patchy due to limited satellite revisit frequency and obscuring by cloud (Fig. 2).  Consequently, the cumulative area of all hotspot pixels detected during each summer was, on average, about 30% less than estimates based on traditional reconnaissance.  To account  for these missed burned patches, a novel burned area mapping technique was developed.

Forest Fires, 1999

Given the size of Canada, the use of remote sensing data is considered to be the only cost effective way to achieve a comprehensive overview of forest fire activity in Canada in near-real time. These images were acquired as part of the Fire Monitoring, Mapping and Modeling System (Fire M3) an initiative of the Canada Centre for Remote Sensing and the Canadian Forest Service. 

A dry spring and human carelessness has resulted in more than 65 forest fires in Manitoba. About 400 residents of three small towns clustered on the northeast shore of Lake Winnipeg have been forced to flee their homes. 
 
 
 
 

In Northern Ontario, 54 fires broke out over the weekend of May 1, forcing for a time the closing of highways in the Temagami and Kenora areas. A huge forest fire has forced the evacuation of the town of Beardmore located on the southeast shore of Lake Nipigon. About 300 residents from the town have fled to Geraldton, 80 kilometres to the east. 


Sources: Winnipeg Free Press and CBC News online.
 
 

Canadian Forest Fires, 1998

These are time series of NOAA AVHRR satellite images acquired over areas of Canada that clearly show the extent of forest fires. The hot spot of a fire is marked as a red dot and the backgroud is a false-color composite using AVHRR channel 1 (visible) and channel 2 (near IR) reflectance measurements. 

AVHRR data are received at the Prince Albert satellite receiving station. Raw data are sent to CCRS, where they are first calibrated. A fire detection algorithm, developed by scientists in the Environmental Monitoring Section is then applied, and the images can now be used by forest fire management agencies for forest fire monitoring and supression.
 
 

Pink DotKluane Lake, Yukon
Purple DotGreat Bear Lake, NWT
Lt Blue DotGreat Slave Lake, NWT
Med Purple DotKasba Lake, NWT
Green DotSalmon Arm, B.C.
Yellow DotWilliston Lake, B.C.
Red DotSwan Hills, Alta.
Green DotLake Athabasca, Sask.
Yellow DotReindeer Lake, Sask.
Dk Purple DotS. Indian Lake, Man.
Pink DotJames Bay, Ont.
Purple DotLake Nipigon, Ont.
Blue DotJames Bay, Que

flowchart_s.jpg (23942 bytes)

Figure 1.

000807_id_mt_s.jpg (29521 bytes)

Figure 2.

hotspots94-99_s.jpg (22757 bytes)

Figure 3.

namerica_june25_s.jpg (36354 bytes)

Figure 4.

hotspots99_animation_s.gif (181611 bytes)

Figure 5. 1994

hotspots99_animation_s.gif (181611 bytes)

Figure 6. 1995

hotspots99_animation_s.gif (181611 bytes)

Figure 7. 1996

hotspots99_animation_s.gif (181611 bytes)

Figure 8. 1997

hotspots99_animation_s.gif (181611 bytes)

Figure 9. 1998

hotspots99_animation_s.gif (181611 bytes)

Figure 10. 1999

hotspots99_animation_s.gif (181611 bytes)

Figure 11. 2000
 
 

ont_199905_animation_s.gif (50180 bytes)

Figure 12.

Russia_fires_s.jpg (38095 bytes)

Figure 13.

wg_52_se_S.jpg (31854 bytes)

Figure 14.

ng_53_se_s.jpg (26410 bytes)

Figure 15.

ng_54_s_s.jpg (17727 bytes)

Figure 16.

bl_53_se_s.jpg (23890 bytes)

Figure 17.

firemap_s.gif (3631 bytes)

Figure 18.