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


Lake Maggiore - Italy, 3-6 April 2000

Detection and mapping of burnt areas and active fires in tropical woodland ecosystems with the VEGETATION sensor: the SMOKO-FRACTAL case study over Northern Australia

D. Stroppiana1, M. Maggi1, J-M. Pereira2, D. Graetz3, J-M. Grégoire1, J. Silva2, A. Sà2, P. Henry3, V. Gond1 and E. Bartholomé1
JRC-Space Applications Institute, Ispra, Italy
2 Instituto Superior de Agronomia, Lisbon, Portugal
3 CSIRO-Earth Observation Centre, Canberra, Australia
4 CNES, Toulouse, France

Paper (pdf file, 1.15 M)

The SMOKO-FRACTAL field campaign was conducted during the dry season [June 1999] in Kakadu National Park, Northern Territory, Australia, by four partners: the CSIRO Earth Observation Center [EOC], the Technical University of Lisbon [TUL], the Centre National d’Etudes Spatiales [CNES] and the Space Applications Institute (SAI) of the European Commission Joint Research Center. The central scientific objective of the experiment is to develop and test methodologies for burnt areas assessment, from satellite imagery, in order to quantify the contribution of vegetation fires to gas and particulate emissions, as well as to improve knowledge of the regional carbon budget.

A complete data set of SPOT-VEGETATION imagery, S1 products, was acquired from May 15th to July 15th. The imagery is composed of daily four channel images of ground reflectances over the study area of one million square kilometers [10-17 S, 125-135 E]. In addition, some VEGETATION night images were acquired to assess MIR channel sensitivity to active fires.

The vegetation cover is mainly characterized by woodland formations with tall grassy understoreys. Other important vegetation types include mangroves, monsoon rainforest, freshwater wetlands and heathlands.

Ground and helicopter observations were collected on the extent of the main burns, the characteristics of the vegetation cover [specific composition and vegetation structure] and fire behaviour [cold and hot fires].

Using a classification trees approach, an algorithm has been developed to detect burnt areas from temporal composites of daily SPOT-VGT images. Burnt area maps have been produced for the 2 months period of the time series.

High resolution images [Landsat TM] have been used to validate the results: to assess the performance of the algorithm and to estimate the total and per each vegetation type commission and omission errors.

In order to evaluate the specificities of the VEGETATION sensor for burnt area mapping, three other satellite coverages were acquired for the experimental period [AVHRR-HRPT, ATSR, and ERS2-SAR] and processed to burnt area maps, and the results compared with those obtained from SPOT4-VGT.

The main conclusions of the study are:


  • SPOT-VEGETATION images performed well for burnt area detection and mapping, when a temporal change detection approach is used, due mainly to the presence of the NIR and SWIR channels and to the excellent geometry of the system.

  • The detection of burnt areas must be done on composited images, using a modified MinNIR criteria, rather than on single date images.

  • The analysis and interpretation of the results obtained with the other types of imagery collected for the experiment show that each sensor provides a specific contribution to what could be a multi-sensor approach to burnt area mapping. The main elements of such an approach are proposed.

  • The results obtained over the SMOKO-FRACTAL experimental site will be further developed in the framework of a global scale initiative for burnt area mapping from VEGETATION data: a network of partners is currently being built to develop a set of methodologies adapted to a range of ecological conditions over the globe. The acquisition of a daily global coverage, of VEGETATION imagery, has been initiated on October 1999 and will continue until December 2000. The resulting time series will be the basic experimental data set for the development of a prototype system for mapping burnt areas at medium resolution globally, and will contribute to the Millenium Assessment intitiative.

Features observed on night images are confirmed to be fires that can be easily retrieved by simple threshold. Given the small area size and possible cloud coverage it is difficult to draw conclusions on fire counts. More observations would be necessary to carry out an accurate benchmark of VEGETATION night image efficiency for active fire detection.