The work presented here was part of the Ph.D. dissertation delivered by Dr. Paulo Marinho Barbosa in April 2000 to the Instituto Superior de Agronomia at the Universidade Técnica de Lisboa to obtain the degree of Doutor em Engenharia Florestal.
Original title: Utilização de Imagens de Satélite para a Estimação de Áreas Ardidas, Biomassa Queimada e Emissões Atmosféricas na África.
The present product makes available the first published time series of burned area maps of Africa, covering an eight year period, 1981-1983 and 1985-1991. These maps were derived from the analysis of the Advanced Very High Resolution Radiometer (AVHRR) Global Area Coverage (GAC) images at 5 km resolution. The period 1983 to 1985 was discarded due to poor data quality and lack of images.
Several studies have tried to evaluate the global amounts of burned biomass and the resulting emissions, in particular in the tropical belt.
The estimates of burned biomass, reported for tropical areas vary from 3260 Tg/year to 10450 Tg/year are usually based on the classification method which relies on the following equation:
This approach is based on statistical data retrieved from different sources, namely vegetation maps, biomass density maps, and burning efficiencies derived from experiments in different types of vegetation. The burned area extent is normally based on the assumption that a given vegetation type will have a certain percentage burned every year, or on empirical knowledge of the return intervals between fires, and is by far the factor with the higher level of uncertainty.
There is a need to accurately assess the spatial and temporal distribution of burned area in order to reduce the uncertainty of the estimates of burned biomass and consequently of the amount of gaseous and particulate emissions to the atmosphere
The daily time series of AVHRR-GAC data was composited into weekly images using the Minimum Value Composite of Albedo. A multitemporal multithreshold technique was then applied to the weekly images in order to detect the burned pixels. This technique is based on the changes recorded in the visible and infrared spectral domain of the AVHRR sensor as a result of vegetation burning. Validation of the burned area maps by comparison with a number of Landsat TM images, which were classified in terms of burned and unburned surfaces, showed an overall accuracy of 71%. More detailed information on the burned area algorithm and validation can be found in Barbosa et al., 1998,1999.
Burned area maps
Each burned area map covers the period from November to October of the following year. These two months correspond to the period where the northern hemisphere fire season has already started and the southern hemisphere fire season is reaching its end. The results are available weekly although some of the years lacked a few weeks of data.
Since we cannot tell exactly the percentage burned, inside the AVHRR GAC pixel, we set two different scenarios, that should give the range between the minimum and maximum burned area. Given the large pixel size (25 Km2) of the AVHRR-GAC images a considerable percentage of pixels was affected by burning in more than one week and so they were flagged as burned more than once during a one-year period.
The following scenarios were set:
- Scenario 1: all the pixels were considered as totally burned independently of the number of times they were flagged as burned during a one-year period
- Scenario 2: pixels that were flagged as burned once were considered as 33% burned, twice 66% burned, and three or more times 100% burned.
The first scenario probably over estimates the total area burned and we believe that for this scenario we are near the upper limit of the burned area estimations.
Under the second scenario, if the minimum area detectable is bigger than 33 %, we are presently underestimating the area burned for the pixels flagged as burned. Conversely, if the minimum burned area detectable is lower than 33%, which is unlikely, we are overestimating the area burned. However, for this second scenario we believe that this is near the minimum value for the burned area estimations.
For more information on this topic, please contact Jean-Marie Grégoire
An Algorithm for Extracting Burned Areas from Time Series of AVHRR GAC Data Applied at a Continental Scale, Barbosa P.M., J-M. Grégoire, and J.M.C. Pereira, 1999. Remote Sensing of Environment, 69(3), 253-263.
An assessment of vegetation fire in Africa (1981-1991): burned areas, burned biomass and atmospheric emissions, Barbosa P.M., D. Stroppiana, J-M. Grégoire, and J.M.C. Pereira, 1999. Global Biogeochemical Cycles, 13(4) , p. 933-950.
Compositing criteria for burned area assessment using multi-temporal low resolution satellite data, Barbosa P.M., J.M.C. Pereira, and J-M. Grégoire, 1998. Remote Sensing of Environment, 65:38-49.