Global AVHRR fire probability map 1982-1999

Introduction

On a global scale, biomass burning may have significant impacts on atmospheric chemistry and global climate.

The fluctuation of the global biomass burning (excluding savannas contribution) could be at the origin of the disturbance of global carbon cycle on the scale of several years. On a long time scale base (several hundreds of years), the contribution of the biomass burning would be neglected only whether the burned surfaces (use or cover) would not be anthropogenically changed during this temporal cycle.

It has been estimated that approximately 85% of biomass burning takes place in the tropical countries, and the tropics play a peculiar role in troposphere chemistry. At this level, presently there are no reliable assessments of the global emission of CO2 into the atmosphere as a result of biomass burning computed on the base of a long-term period. A long-term data set on burned surfaces will provide an insight into the mechanisms of such phenomenon. The assessment of the spatial-temporal dynamics of these phenomena at global scale could be provided based on remote sensing data.

In this perspective, a new weekly satellite Earth Observation product called Global Burnt Surfaces 1982-1999 (GBS 1982-1999) time series appears as an opportunity to increase our understandings on the global fire activity phenomenon. This product, derived from the daily NOAA-AVHRR GAC 8km data set (1982-1999), allows to characterize fire activity in both northern and southern hemispheres on the basis of average seasonal cycle and inter-annual variability. Details can be found in Carmona-Moreno et al. (2005a).

Processing of the NOAA-AVHRR GAC 8km dataset

The burned surface detection algorithm implemented for obtaining burned surface maps uses weekly composites of the daily NOAA-AVHRR GAC 8km data. The algorithm is a global extension of the multi-temporal multi-threshold algorithm developed for the Africa continent by Barbosa et al. (1999). A detailed description of the processing system can be found in Moreno-Ruiz et al. (1999).

AVHRR GAC images are representative samplings of the original AVHRR-LAC images and as such only qualitative interpretations, comparisons and trends can be deduced from them. The main spatial and temporal patterns of fire activity at global scale are clearly depicted, but caution should be exercised when interpreting the GBS time series on a quantitative basis because of its sampling character.

From a point of view of the error detection, fire and burnt surfaces presenting high spatial variability like fires in gallery forests, “small” extensions of fires (in relation with the resolution of the image) (< 1500 ha), dark soils (some over-estimations has been detected in this product even if these errors have been minimized by the temporal and automatic detection analysis), under-story forest fires and low temperature peat fires (these important sources of CO cannot be detected by this algorithm) are a source of error in this data-set. Ever-cloudy areas (in boreal and some tropical regions, mainly) represents also another source of detection problem

Spatial-temporal distribution of the GBS time series (1982-1999). Data unavailable for 1994 (trimesters: 48-52)

The Fire Season Probability Maps 1982-1999

An important aspect in defining the global fire regime is the probability of fire occurring in a particular season for any given area. This can be described as the probability for a given area at (lat. i, long. j) to burn in a given unit of time. Here we consider that the unit of time is a trimester and a unit area is a 0.5° x 0.5° cell.

where P(BSp(i,j)) is the probability of a 0.5° x 0.5° cell to burn in a trimester p of the year (p = {1, 2, 3, 4}); å BSp(i,j) is the number of burned pixels detected in the cell (i, j) for the trimester p of the year; where m is the number of pixels (8 km resolution) in the cell; and, åå BSp(i,j) is the total number of burned pixels detected in the cell (i, j) along the 17 years considered here.

The outputs of this computation are four fire season probability maps (one per season of the year) and are downloadable from this web site (GEOTiff format). By definition, let be the sum of the probabilities of the 4 periods equal to 0% for those regions where the total number of burned pixels is equal 0 (no fire detected during the seventeen years time series) in order to have a homogenous system.

December-January-February (DJF)
March-April-May (MAM)
June-July-August (JJA)
September-October-November (SON)

As expected, these results show the low probability of fire occurring in the first trimester of the year (winter) in the Mediterranean area and also in high latitudes of the boreal region. The probability strongly increases during the second and third periods of the year.

This time series clearly depicts a temporal shift (~2 trimesters) between the fire activity in the northern and the southern hemispheres, and this temporal shift between the tropical and medium latitudes in the northern hemisphere is around 1 trimester.

Fire season in some regions are very stable over the entire 17 year record. These areas have either very high or very low probability during any particular trimester. This is the case for the high and medium latitudes in the Northern Hemisphere (JJA), Far East of China-Russia (MAM), Central Africa (JJA) and Africa Savannahs (DJF and SON). We can hypothesize that in these areas inter-annual variations in climate have little or no effect on fire seasonality – though of course may still influence the size, intensity and efficiency of fire. However, in other regions, even if there is always a maximum of probability in a given season, the probability peak is less evident, i.e. the timing of the fire activity across the 17 years varies quite widely. This temporal homogeneity probability can be interpreted as particularly fire prone areas/ ecosystems being more sensitive to inter-annual climate variability introducing temporal shifts in the fire activity. This is the case in Indonesia, Mediterranean Basin, Southern and East Africa, California, Australia, Southern East Asia (Indonesia), Central and Latin America.

For more information on this topic, please contact Cesar Carmona-Moreno

KEY PUBLICATIONS:

An Algorithm for Extracting Burned Areas from Time Series of AVHRR GAC Data Applied at a Continental Scale, Barbosa P.M., Stroppiana D., Pereira J.M.C. 1999. Remote Sensing of Environment. 69:38-49.

Characterising interannual variation in global fire calendar using data from Earth Observing satellites, Carmona-Moreno, C., Belward, A., Malingreau, J.P., Hartley, A., Garcia-Alegre, M., Antonovskiy, M., Buchshtaber, V., Pivovarov, V. 2005a. Global Change Biology. 11( 9), 1537-1555, doi: 10.1111/j.1365-2486.2005.001003.x, September 2005.

GLINTS-BS – Global Burn Scar Detection System, Moreno-Ruiz, J.A., Barbosa, P.M., Carmona-Moreno, C., Gregoire, J.M., Belward, A.S. 1999. European Union Technical Note I.99.167. May 1999.