Canadian Wildland Fire Information System

Fire Monitoring, Mapping, and Modeling (Fire M3)

Frequently Asked Questions

How are satellite sensors able to detect forest fires?

Satellite sensors record the intensity of electromagnetic radiation from Earth in various spectral wavelengths or channels. These wavelengths can be classified into general ranges, including the visible (0.4–0.7 µm), near-infrared (0.7–1.3 µm), mid-infrared (1.3–3.0 µm), and thermal infrared (3.0-100 µm). Fires and other sources of intense heat can be detected if a sensor includes a channel near the 4-µm range, which is highly sensitive to thermal radiation emitted from objects hotter than about 200°C. For more background on satellite remote sensing see the Fundamentals of Remote Sensing tutorial on the Canada Centre for Remote Sensing (CCRS) web site.


Which sensors are currently being used to provide information about forest fires in Canada?

  • Thermal infrared scanners flown on board aircraft are often used to map fire hotspots and fire intensity over individual fires or small regions. This information allows fire management agencies to effectively target fire suppression efforts by water bombers and ground attack crews.

  • The US National Oceanic and Atmospheric Administrations's Advanced Very High Resolution Radiometer (AVHRR) is the most commonly used satellite sensor for detecting fires over large regions or entire countries. It can provide a daily snapshot of Canada at a 1-km nominal resolution in five spectral channels. AVHRR is used by Fire M3 for daily forest fire detection over Canada.

  • The Satellite Pour l'Observation de la Terre (SPOT) Vegetation (VGT) sensor, launched in 1998, has four channels that measure reflected energy from Earth. Like AVHRR, VGT provides daily coverage of Canada at a 1-km resolution. Because it lacks thermal channels, the sensor is not as well suited to detecting active fires as AVHRR. However, it does include near-infrared and short-wave infrared channels that are highly effective for mapping the extent of burned forest after a fire has stopped burning. VGT imagery is currently being used for annual mapping of burned forest across Canada.

  • Landsat's Thematic Mapper (TM) sensor allows areas to be observed at a 30-m resolution in seven channels. This high resolution comes at the expense of observing a given location only once every 16 days. TM is best suited to providing detailed maps of areas burned by individual fires or fire complexes. These maps can be used to plan salvage logging operations and to verify the extent of the burned areas mapped using coarser-resolution VGT imagery.

  • The National Aeronautics and Space Administration's Moderate Resolution Imaging Spectroradiometer (MODIS), launched at the end of 1999, has channels specifically designed for fire detection. These include 1-km resolution channels at 4 µm and 11 µm, with high saturation temperatures of about 450 K (177°C) and 400 K (127°C), respectively.


How are forest fires detected by the AVHRR satellite sensor?

The most important AVHRR channel for fire detection is the mid-infrared channel, which measures a combination of reflected and thermal energy at wavelengths around 3.7 µm. This channel is highly sensitive to objects that are emitting thermal energy at high temperatures (over 200°C), such as vegetation fires. For this reason, the AVHRR sensor can detect fires that cover only a fraction (less than 0.1% or 1200 m²) of a 1.2-km² AVHRR pixel. Since other types of objects, such as cloud edges and exposed soil, also produce a large response in the mid-infrared channel, information from other AVHRR channels is required to filter out extraneous data ("false alarms"). The AVHRR fire detection algorithm used in Fire M3 was developed and tested specifically for boreal forest fires. It is described in publications by Li et al. (2000b,2000c).


What are some of the benefits and limitations of using satellites for fire detection?

The main benefit of using satellites for detecting fires is that they can cover all of Canada on a daily basis at relatively low cost. This makes them effective for detecting fires in remote, unpopulated regions, where conventional fire monitoring is less intensive. Thick smoke plumes from forest fires, often extending several hundred kilometers, can also be identified by means of satellite imagery.

Satellite fire detection has some limitations that must be kept in mind when examining the daily fire images:

  • The tests used by the fire algorithm to remove "false alarms" sometimes fail, leading to false records of fires. Li et al. (2000c) found that about 14% of the fires detected are probably false alarms. Given the large number of satellite pixels examined each day across Canada (about 9.5 million), this represents a very low level of noise, on the order of 0.0001%. A fire hotspot can be confirmed if a conical smoke plume is observed emanating from it. However, it is sometimes impossible to see a plume from a small fire or a fire with clouds nearby.

  • The algorithm cannot detect fires through thick cloud or smoke. A given fire may therefore go undetected for several days and then later reappear. Because of this limitation, the total area of the satellite hotspots for an average year represents only about 65% of the actual area burned, which may vary according to amount and frequency of cloud cover.

  • The time lapse between retrieving the final AVHRR satellite orbit over Canada and distributing the daily fire images on the Fire M3 web site is about 6 hours (i.e., from 6 PM to midnight). This limits the utility of satellite detection for tactical fire operations.

  • The actual size of the actively burning area is difficult to determine from satellite imagery. Because of the sensitivity of AVHRR's mid-infrared channel, a 1-km² hotspot pixel may represent a fire as small as 100 m². In addition, an intense fire covering an area less than 1 km² may actually show up as a cluster of several 1-km² hotspot pixels. This is the result of the varying size and spatial overlap of the raw, unprojected AVHRR pixels.


What other information about forest fires (besides location) can be derived from satellite sensors?

In addition to detecting actively burning fires, AVHRR is used in Fire M3 to monitor the location of thick smoke plumes and cloud. Affected burned areas are also mapped after a fire using SPOT VGT, Landsat TM, and AVHRR. We are currently developing techniques to measure the severity of vegetation damage caused by fire.


How are smoke plumes detected by the AVHRR satellite sensor?

Smoke plumes from forest fires tend to be highly reflective and therefore are often difficult to distinguish from cloud. Yet AVHRR imagery does provide some information that can be used to effectively distinguish thick smoke from cloud cover. Thick smoke and the underlying land surface are generally warmer than cloud and therefore produce a larger signal in the thermal infrared channels. In addition, smoke plumes tend to have a smoother spatial texture than many cloud types at a 1-km² scale. A recent study by Li et al. (2000a) examined various techniques for identifying smoke using AVHRR, including artificial neural networks and multiple threshold tests. For smoke detection in Fire M3, a multiple threshold approach has been adopted. The thresholds are designed to be conservative, in that false smoke pixels are kept to a minimum, while only the more optically thick smoke is detected. Note that even this conservative approach to smoke detection can lead to false smoke pixels. Smoke plumes in the fire images can be confirmed by the presence of nearby hotspots, their smooth texture, and their typical conical shape.


How can satellite sensors be used to map burned areas?

Since forest fires exceeding 10 km² account for more than 95% of the annual burned area in Canada, 1-km resolution satellite imagery is effective for mapping the vast majority of burned areas. A technique has been developed at the Canada Centre for Remote Sensing (CCRS) that allows burned forest to be mapped at annual intervals across Canada (Fraser et al. 2000). The method works by combining an annual AVHRR hotspot map with observed annual changes in a vegetation index from SPOT VGT. The vegetation index is compared for each pixel from one year to the next (e.g., September 1998 and September 1999). Pixels with a significant drop in the index that are spatially coupled to an AVHRR hotspot are mapped as being burned. This technique is used to provide a coarse-resolution burned area product for Canadian forest at the end of each forest fire season.

A separate algorithm has been developed to map burned areas at 30-m resolution using single Landsat TM images. The procedure has been used to automatically map individual burns, which can aid in planning salvage logging operations after a forest fire. A representative sample of 1998 and 1999 forest fires was mapped with TM for use in validating and calibrating the 1-km burned area product from SPOT VGT. For more information contact Robert Landry (


What modeling activities are included as the third "M" in Fire M3?

For each AVHRR hotspot location, various fire attributes are modeled with the Canadian Forest Fire Weather Index System and Forest Fire Behavior Prediction System developed by the Canadian Forest Service. Fire weather indices are modeled on the basis of weather station data and include ratings for fine fuel moisture, duff moisture, drought, initial spread, buildup, and fire weather. The Forest Fire Behavior Prediction System uses information on fire weather, fuel type, topography, and foliar moisture content to predict the rate of spread, fire intensity, type of fire, and fuel consumption. This array of fire attributes can be queried for each hotspot using the i icon on the Internet map server toolbar.

The Canadian Forest Service and the CCRS have begun developing a spatially explicit model to predict the quantity of greenhouse gas emissions from forest fires across Canada. The model will combine:

  • daily burned area mapped using SPOT VGT and AVHRR

  • fuel consumption modeled according to the Canadian Forest Fire Behaviour Prediction System

  • measurements of emissions from forest fire experiments (as part of this project, the potential of Landsat TM and SPOT VGT for estimating fuel consumption will be investigated)




Fraser, R.H.; Cihlar, J. 2000. Hotspot and NDVI differencing synergy (HANDS): a new technique for burned area mapping over boreal forest. International Journal of Remote Sensing 74(3):362-376.

Li, Z.; Khananian, A.; Fraser, R.H. 2000a. Detecting smoke from boreal forest fires using neural network and threshold approaches applied to AVHRR imagery. IEEE Transactions on Geoscience and Remote Sensing 39:362-376.

Li, Z.; Nadon, S.; Cihlar, J. 2000b. Satellite detection of Canadian boreal forest fires: development and application of the algorithm. International Journal of Remote Sensing 21(16):3057-3069.

Li, Z.; Nadon, S.; Cihlar, J.; Stocks, B. 2000c. Satellite-based mapping of Canadian boreal forest fires: evaluation and comparison of algorithms. International Journal of Remote Sensing 21(16):3071-3082.

Fire M3 Summary

Fire M3 Data Sources and Methods


Date modified: