News from Fire Reseach: The Global Fire Product: Fire Distribution from Satellite Data (IFFN No. 19 – September 1998)

NEWS FROM FIRE RESEARCH:

The Global Fire Product:
Fire Distribution from Satellite Data

(IFFN No. 19 – September 1998,p. 78-83)


Introduction

In 1991, following a workshop on the requirements for terrestrial biospheric data sets and in response to requirements from the International Geosphere Biosphere Programme (IGBP) core projects, IGBP-DIS set up the Fire Working Group (FWG) to develop a consensus algorithm for global fire mapping. From this was born the concept of a Global Fire Product (GFP). This would be based on the use of an active fire detection algorithm and the global daily Advanced Very High Resolution Radiometer (AVHRR) data being collected by the IGBP-DIS 1 km AVHRR Global Land Project (Eidenshink and Faundeen 1994). A consensus algorithm was developed (Flasse and Ceccato 1996) and approved by the FWG in 1996 (Malingreau and Justice 1996). Data processing was initiated at the Joint Research Centre (JRC) in 1996, and completed in November 1997 (Stroppiana et al. 1998).

Input data set

The input data set is composed of 5-channel, raw AVHRR scenes at 1.1 km (nadir) resolution for each daily afternoon orbital pass of NOAA-11 over all land and coastal zones. The data were collected over the 21 month period from April 1992 to December 1993. The data set was provided by the USGS-EROS Data Center and ESA-ESRIN; it is fully documented in Eidenshink and Faudeen (1994).

Algorithm

Each region of the globe has its own characteristic fire regime, biome, and seasonal pattern of surface temperature and consequently, a different response in each of the NOAA-AVHRR channels as a result of fire disturbance. In order to process a global data set automatically and without adjusting the algorithm for each geographic region a contextual algorithm was chosen since it gives better performance and global consistency compared to a conventional channel-threshold technique (Giglio et al. 1998). The chosen algorithm is essentially that of Flasse and Ceccato (1996), with very minor modifications. For each day processed, the system ingests 2 gigabytes of data from tape, which represents the 5 channels of the raw AVHRR data for the 14 orbits covering all land areas of the globe. Firstly the data is geolocated using an orbit model obtained from the Colorado Center for Astrodynamics Research (CCAR) (Rosborough et al. 1994). The orbit model is typically accurate to ±2 pixels. Then all ocean and large inland water bodies are masked out. A “no-burn” mask is applied to exclude regions where the surface is of a type which does not support any significant biomass burning. These masks significantly reduce the amount of data to be subsequently processed. A simple cloud detection algorithm based on that of Saunders and Kriebel (1988) is applied before finally testing the remaining pixels for the presence of hot sources using the algorithm mentioned above.

Product description

The GFP itself is composed of the following two kinds of data:

Daily fire position tables: These consist of daily lists of the latitude and longitude of each fire pixel detected by the system for the period April 1992 – December 1993.

10-day synthesis raster format data: These are 10-day composite rasters on latitude-longitude grids of 0.5º ´ 0.5º cells and contain the following bands:

  • Fire Density Map: The number of fire pixels detected in each grid cell (see Fig.1)

  • Cloud/No-Data Map: The percentage of cloud or “no-data” obscuring each cell, and

  • No-Burn Mask: The percentage of each grid cell masked out by the no-burn mask.

Global distribution of fire activity

Twelve months of the global fire product (April 1992 – March 1993) have been studied in detail and the spatial and temporal distribution of fires has been reported elsewhere (Dwyer et al. 1998a,b). A total of 6.5 million fire pixels were detected in the 12 months of 1 km resolution AVHRR data analyzed. However, these are not evenly distributed throughout the year (Fig.2). There is a peak in global fire activity in July and August. It then decreases slowly reaching a minimum in early November when the number detected is only 28% of those detected during the period of peak burning. From November fire activity increases again reaching another lower peak in late December and January after which activity reduces.

While over 70% of fire pixels are located within the tropics, 50% of all fire pixels detected were on the African continent. Most of the fires are set in the savanna regions. The reasons for burning are numerous and vary across the continent, but some of the more common ones are: burning to remove unpalatable stubble and to initiate off-season regrowth of fresh shoots, clearing ground for crops, establishment of fire-breaks around settlements, removal of parasites, to drive game out of hiding and to make pathways accessible. Other regions where very high concentrations of fire activity were seen are in mainland Southeast Asia, the Orissa province in Eastern India, parts of the Cerrado in Brazil and Arnhem land in the Northern Territories of Australia. Although the number of fires occurring in temperate and boreal biomes is much smaller than in the tropics, they can have a big impact on land cover and the global carbon cycle. Fires in boreal biomes can be of extremely large extent, consume very high fuel quantities and are often left to burn out naturally.

Uses of the product

The use of the Global Fire Product (GFP) was envisaged for two user communities which can be loosely collected under the subject areas of atmospheric chemistry and ecosystem studies. Biomass burning is responsible for large emissions of gaseous and particulate products into the atmosphere and has a significant role in ecosystem maintenance and change. Use of the data is foreseen in certain IGBP core projects such as the International Global Atmospheric Chemistry project (IGAC), the Land-Use and Land-Cover Change Project (LUCC) and the Global Change and Terrestrial Ecosystems project (GCTE) . Other international initiatives such as the Global Observation of Forest Cover (GOFC) project of CEOS and Forest Resources Assessment (FRA) –2000 (FAO) have expressed interest in utilizing the product. The product is unique in that a single algorithm was used for all the processing therefore guaranteeing an internally consistent data set. The full resolution of 1km is available to all users who may regrid the data for their own requirements. This flexibility allows the use of the product across a wide range of spatial scales.

Atmospheric Chemistry Studies

The highest resolution of the GFP is 1 km. In studies related to atmospheric chemistry, it is probable that a gridded product at a lower resolution is more appropriate. Figure 1 shows fire counts in 0.5° by 0.5° grid cells for a ten day period. Similar products of different grid sizes and over different time durations can easily be constructed from the basic product. The information provided by the GFP which can be of most use to the atmospheric chemistry community is:

  • Spatial localisation of fire events

  • Spatial variation in the number of fire events

  • Seasonality of fire.

With respect to the last point, although the day on which each fire event was detected is recorded, the seasonality i.e. the time period and duration of the burning season is likely to be of more interest. Figure 3 shows an example of such a derived product. The mid fire season month which is defined as the month in which 50% of all fire events were recorded for each grid cell is shown. Other parameters related to the seasonality of burning can be easily derived. Previous studies of emissions due to biomass burning have generally assigned an empirical time distribution of burning events throughout the fire season and across large areas. Hao and Liu (1994) and Kim and Newchurch (1998) used ozone measurements to identify the spatial location and burning period in their studies of gaseous emissions and transport from biomass burning. The GFP can give improved estimates of these parameters from direct observations.

 

https://gfmc.online/wp-content/uploads/res_1_1-1.gif (75167 Byte)

Fig.1. Summary of fire pixels per cell from 30 July to 8 August 1992.

 

3739 Byte

Fig.2. The number of fire pixels detected in each of the 12 months of AVHRR data.

 

It is not advised to use the GFP to estimate area burned as the product indicates only the presence or absence of fire in a pixel. Nor can it be used to count the absolute number of fire events in a given location as it is only a temporal sample. Although a research study using 1 km AVHRR data combined with high resolution LANDSAT data in Southern Africa has shown that it is feasible to estimate burned area from the 1 km data (Justice et al., 1996), its results cannot be universally applied. Extensive research for different vegetation types would need to be carried out if such a scheme were to be adopted. Current research is focused on retrieving burned area directly from low resolution satellite data.

Ecosystem Studies

Vegetation types affected: The GFP facilitates the study of fire in relation to landcover and ecosystem dynamics. The relative levels of fire occurrence in different vegetation types and regions can be estimated when the data is used with appropriate land cover maps. Using the 25 class United States Geological Survey (USGS) legend supplied with the IGBP-DIS 1km land cover map, fire distributions were determined for the different vegetation types. Almost 90% of all the fire pixels detected were found in 8 vegetation types. Table 1 shows the percentage of the global land surface covered by each of these vegetation types, the proportion of each type which was affected by fire and the percentage of the earth’s land surface this represents. Although over 6% of the earth’s surface was affected by fire in the course of the year, this does not mean that this much surface area was burned. As each fire pixel detected covers a 1 km2 surface area it can represent one or more fires of unknown dimensions within that area. The type of vegetation burned is also of interest in atmospheric chemistry studies and in research into carbon cycling.

Timing The timing of fire is a very important parameter in relation to the study of fire impact on ecosystems. In tropical regions, late dry season fires are generally more intense and difficult to control than those occurring in the early dry season, when the fuel is more moist. The GFP data combined with data on vegetation conditions or weather data for the year in question can be used to determine how the timing of burning varies spatially and in different vegetation types.

Fig.3. For each 0.5° by 0.5° cell the month of the mid fire season is shown. This is independent of the number of fire pixels detected in a cell (330 KB)

 

Tab.1. Vegetation types affected by fire. The eight vegetation types, as defined in the IGBP-DIS land cover map, which showed the most fire activity account for 66% of the earth’s land surface. Varying amounts of each vegetation type were affected by fire, however, savanna burning was the most widespread.

Vegetation Type

% of global land surface

% of vegetation type affected

% of global land surface affected by fire

Savanna

11

19

2.1

Evergreen Broadleaf forest

10

7

0.7

Deciduous Broadleaf forest

5

13

0.6

Dryland crops and pasture

9

6

0.5

Shrubland

12

4

0.5

Cropland/Woodland mosaic

7

7

0.5

Irrigated Crops and pasture

3

14

0.4

Grassland

9

4

0.3

 

Land use and Land Cover Change Fire is an indicator of land use and land cover conversion. Although the GFP is limited in time to 21 months, because of its global extent which covers all ecosystems it facilitates the study of spatial relationships between fire activity and land cover use and change.

Diurnal Cycle The GFP gives a snapshot of fire activity for each location at one instant – early afternoon- during the day. It is not a record of total fire activity. Until further information is available on the diurnal variation in burning in different regions and vegetation types, it is not possible to say what percentage of vegetation fires are captured in the GFP. Night time data from the Defence Meteorological Satellite Program (DMSP), and the Geostationary Operational Environmental Satellite (GOES) combined with the GFP can improve knowledge of the diurnal cycle in burning.

Limitations of the product

The GFP is the first map of global vegetation fire derived with a single algorithm directly from observations of the fires themselves, and it will undoubtedly prove to be of considerable value both in global and regional scale studies. The contextual algorithm gives better fire detection performance over that obtained with algorithms based on simple threshold tests and it provides the best consistency for global applications (Giglio et al. 1998). However there are a number of limitations to fire detection using the AVHRR sensor alone. The imagery only represents a snapshot of the total number of fires which burn in any 24 hour period, fire counts may be either overestimated or underestimated due to confusion with hot surfaces and sun glint from reflective surfaces such as water and clouds. Although flaming fires with fronts as short as 50 m can be detected, in general no information on the fire characteristics (e.g. size, temperature) is available. However, this single observation system approach will soon be qualitatively and quantitatively improved by combining global datasets of both active fires and burned areas from different Earth observing systems.

Product availability

In March 1998 the Fire Working Group (FWG) of the IGBP-DIS recommended an internal evaluation process to be completed by the end of the year before adoption of the GFP as an IGBP-DIS data set. The GFP has been distributed to the FWG and users involved in biomass burning research. During this time, the quality of GFP will be assessed in each of the major biomes. The results of the evaluation will be available with the product. In the meantime, the data set is available for use on application to the authors.

Acknowledgments

The Global Fire Project was conducted under the direction of Jean-Paul Malingreau, and was coordinated by the Fire Working Group of IGBP-DIS. This work was funded by the European Commission.

Edward Dwyer, Daniela Stroppiana, Simon Pinnock, and Jean-Marie Grégoire @
Global Vegetation Monitoring (GVM) Unit, Space Applications Institute, Joint Research Centre, European Commission, Ispra, Italy. Corresponding author:

Edward Dwyer
MTV/SAI, TP263
Joint Research Centre
I – 21020 Ispra, Varese, ITALY

Fax: ++39-0332-789073
Tel: ++39-0332-785608
e-mail: ned.dwyer@jrc.it

 

References

Dwyer, E., J.-M.Grégoire, and J.P.Malingreau. 1998a. A global analysis of vegetation fires using satellite images: spatial and temporal dynamics. Ambio 27 (3), 175-181.

Dwyer, E., S.Pinnock, J.-M.Grégoire, and J.M.C.Pereira. 1998b. Global spatial and temporal distribution of vegetation fire as determined from satellite observations International Journal of Remote Sensing (submitted)

Eidenshink, J.C., and J.L.Faudeen. 1994. The 1-km AVHRR fire detection. Int. J. Remote Sensing, 15, 3443- 3462.

Flasse, S., and P. Ceccato. 1996. A contextual Algorithm for AVHRR fire detection. Int. J. Remote Sensing, 17, 419-424.

Giglio, L., J.D.Kendall, and C.O.Justice. 1998. Evaluation of Global Fire Detection Algorithm Using Simulated AVHRR Infrared Data. Int. J. Remote Sensing (accepted).

Hao, W.M., and M.H.Liu. 1994. Spatial and temporal distribution of tropical biomass burning. Global Biogeochemical Cycles 8, 495-503.

Justice, C.O., J.D.Kendall, P.R.Dowty, and R.J.Scholes. 1996. Satellite remote sensing of fires during the SAFARI campaign using NOAA advanced very high resolution radiometer data. J. Geophys. Res. 101 (D19), 23851-23863.

Kim, J.H., and M.J.Newchurch. 1998 Biomass-burning influence on tropospheric ozone over New Guinea and South America. J. Geophys. Res. 103 (D1), 1455-1461.

Malingreau, J.P., and C.O.Justice. 1996. Definition and implementation of a global fire product derived from AVHRR data. IGBP-DIS Working Paper #17, 3rd IGBP-DIS Fire Working Group meeting report, Toulouse, France, on 13-15 November 1996.

Rosborough, G., Baldwin, D., and Emery, W.J., 1994, Precise AVHRR image navigation. IEEE Transactions on Geoscience and Remote Sensing 32, 644-657.

Saunders, R.W., and Kriebel, K.T., 1988, An improved method for detecting clear sky and cloud radiances from AVHRR data. Int. J. Remote Sensing 9, 123-150.Stroppiana, D., S.Pinnock, and J.M.Grégoire. 1998. The global fire product. Int. J. Remote Sensing (submitted)


IFFN No. 19