Super typhoon Haiyan makes landfall
Super typhoon Haiyan
Subscribe here to receive special images

VEGETATION - 2000

Lake Maggiore - Italy, 3-6 April 2000


Detecting active fires with the VEGETATION instrument

V. Gond*, M. Maggi*, P. Henry┬░, J.-M. Grégoire*, E. Bartholomé*
*Joint Research Ispra, Italy
┬░ CNES, Toulouse, France

Active fires may, in certain conditions, be observed on VEGETATION images. Indeed fires generate a surplus of energy that can be detected with the Middle Infrared channel if they are sufficiently large and warm.

To assess instrument sensitivity, VEGETATION S1 data have been analysed over 3 different windows, one over Northern Australia (Darwin), and the 2 others over West and Central Africa (Burkina Faso and Central African Republic). The area covered in Australia is typified by open forest, woodlands and some swamps in the coastal region. The combination of both African windows includes a wide range of ecological conditions, stretching from the sub-arid conditions of Northern Sahel, to the moist evergreen forest of the Congo basin. In addition to the S1 day images, night images were acquired over N. Australia by CNES on an ad-hoc basis for calibration purposes.

The World Fire Web network provides external validation material by giving access to fire geo-location retrieved from AVHRR data processed locally: by CSIR-EOC for Australia and PRGIE-OFB for Central Africa. Some ATSR images were also used over the N. Australia site for comparison. VEGETATION data are also used for consistency assessment: smoke plumes of active fires show up clearly in the Blue channel (B0), day-to-day image difference of the Middle Infrared channel (MIR) allow newly burnt surfaces to be delineated. This is easily achieved thanks to the excellent image co-location.

For day images the conclusions of this study may be summarised as follows: 1) suspected features are confirmed to be fires, 2) fires influence exclusively the MIR channel 3) these features cannot be extracted by simple threshold, but rather by taking advantage of their local contrast (contextual analysis), 4) they can also be detected using day-to-day image comparison thanks to excellent image co-registration, 5) detectable fires occur only in specific ecological conditions, typically in humid and sub-humid rather than in arid regions 6) fires detected with VEGETATION are only a small fraction of those detected with AVHRR.

For night images the conclusions are the following: 1) features observed are confirmed to be fires, 2) they can be easily retrieved by simple threshold, 3) given the small size of area it is difficult to draw conclusions on the efficiency in terms of fire count, 4) there is further uncertainty due to lack of cloud coverage information, 5) fire activity (including gas flares of off-shore oil platforms) should be taken into account when MIR calibration campaigns are performed 6) more observations would be necessary to carry out an accurate benchmark of VEGETATION night image efficiency for active fire detection.