Satellite Fire Monitoring: A Status Report
Contributed by Chris Justice, IGBP-DIS, Global Environmental Change Program, Department of Environmental Sciences, University of Virginia*
* This article includes contributions from Luke Flynn, University of Hawaii; Ned Dwyer, Joint Research Center, Ispra; Olivier Arino, European Space Agency; Chris Elvidge, NOAA; Yoram Kaufman, NASA/GSFC; Jose Pereira, Tapada da Ajuda, Lisbon.
|
A Note from the Chair Science Features 2 BIBEX 3 STARE, TRACE-A, and SAFARI 7 Satellite Fire Monitoring 9 EXPRESSO 11 Domestic vs. Wild Fires in Africa 13 Boreal Forest Fire Research 15 GFMC and BIBEX 16 SAFARI-2000 19 BIBEX in the Future |
The risk to human livelihood posed by fires and the large economic cost of extreme fire events is leading disaster-management agencies to evaluate their monitoring and management response systems. The satellite component of these early-warning systems is focussed on securing near real-time information on the susceptibility of vegetated landscapes to fire, the early identification of fires and the tracking of fire spread. The satellite component is just one part of the information systems needed by fire managers, which include weather data, field and aircraft reports, and fire prediction and management response models.
The global change research community has rather different requirements for satellite data. Their focus is to quantify the location, areal extent and intensity of fire events, their return frequency, and associated aerosol and trace gas emissions. Global change researchers are also concerned with the impacts of fire on ecosystem disturbance, on nutrient cycling, and on the surface radiation budget and their associated feedbacks. Global change research questions which are getting increasing attention involve quantifying and predicting changes in fire frequency associated with variability and trends in climate conditions and human population dynamics, and the likely impacts on trace gas and particulate release, ecosystem function, land management, and biological conservation.
At the Dahlem Conference on Fire in the Environment in 1992 some members of the BIBEX community articulated a goal for a Vegetation Fire Information System that would provide the underpinning data for BIBEX research. Through the activities of the IGBP-DIS Fire Working Group there has been a concerted effort to move towards this goal by making available samples of regional data from existing satellite sensing systems and developing a Global Fire Product using daily data from the NOAA AVHRR sensing system.
The Fire Working Group has also promoted the global change research requirements with the space agencies. There is an increasing availability of data from existing satellite systems and new sensing systems are being designed with fire monitoring specifications.
Current sensing systems can be used to generate products of fire susceptibility using time-series of vegetation state, the occurrence and location of active fires using middle and thermal infrared sensors, and smoke and area burned using visible and near and middle infrared sensors. Microwave sensors can also be used to identify burned areas. The current sensors that provide these have different spatial resolutions and different overpass times and repeat frequencies, resulting in very different accuracies. Some of these products are currently being generated for specific regions and are being made available to the public. However at this time there is no ongoing global fire data set production activity.
Research is being undertaken to enhance and compare algorithms for the current suite of sensing systems, develop product validation protocols, test algorithms with data from new sensors, prototype new algorithms for upcoming sensors, and evaluate algorithms using data from computer, laboratory and airborne simulations. Efforts are also underway to increase the availability of data through the World Wide Web and develop packages suitable for local to regional application by ground receiving stations. Examples of some of the recent development activities by researchers associated with the IGBP-DIS Fire Working Group are described below.
Researchers from the Hawaii Institute of Geophysics at the University of Hawaii have developed a hot spot monitoring web site using 4 km spatial resolution GOES-8 and GOES-10 data. Image data collected at the Naval Research Laboratory in Monterey, CA are forwarded to Hawaii where they are processed through a hot spot location algorithm using the difference of GOES Band 2 (3.8-4.0 microns) and 4 (10.2-11.2 microns). A set of six images including a map key showing hot spot locations, a 1-km GOES Band 1 (0.55-0.70 microns) cloud location image, and a 4-km resolution color image highlighting hot spots are displayed to the web site within 15 minutes of when the data are first received from the GOES satellite. Current study areas include a 500 km ¥ 500 km section of Amazonia north of Cuiaba, Brazil, where fire duration and location for the current fire season are being tabulated, and a 500 km ¥ 500 km area centered on Anaheim, CA, where Santa Ana wind-stoked fires in the Santa Barbara, CA area have been observed as late as October 19, 1998. The monitoring system has the capability to shift data collection to other study areas in the Americas within 24 hours of the receiving a report of particularly large and potentially persistent. While GOES coverage does not extend to the European and African continents, a similar real-time system could be envisioned to ingest geostationary satellite data from Meteosat Second Generation-1 (MSG-1) planned for launch in the year 2000.
The Global Vegetation Monitoring unit of SAI/JRC and their collaborators are continuing research using low resolution satellite data for monitoring of active fires, burned surfaces, and vegetation status. Following the production of 21 months of the AVHRR Global Fire Product, the World Fire Web project that is currently in its pilot phase, is attempting to build an extensive vegetation fire monitoring network. The pilot network consists of satellite receiving stations in Brazil, Italy, Vietnam and Australia. During 1999, following commissioning tests on the software, the network will be expanded to include more stations. Plans have been prepared for Venezuela, Central African Republic, and Mongolia.
A multi-temporal burn-scar mapping algorithm currently under test will also be added to active fire detection and the fire maps will be made accessible to members of the network via the World Wide Web. Active fires and burned areas have been mapped as part of the EXPRESSO Campaign for the Central African Republic and are now being produced for the northern part of South America as part of the Amazon (LBA/CLAIRE) experiment. Work has also been completed on determining burned areas for the African continent for an 8 year period in the 1980s using AVHRR-GAC-4 km data. The accuarcy of applying the methods to determine global burned area from GAC data is being evaluated. Research has begun to determine vegetation moisture status to infer fire risk and burning efficiency from time series of AVHRR 1 km and SPOT VEGETATION data. The methods will initially be tested for West Africa and, if successful, they will be applied to other areas.
The NOAA National Geophysical Data Center (NGDC) in Boulder, Colorado archives data from the U.S. Air Force Defense Meteorological Satellite Program (DMSP). The suite of DMSP sensors includes the Operational Linescan System, which features a visible band which is intensified at night, permitting detection of city lights, fires, and gas flares. Most OLS data have a ground sample distance of 2.7 km. NGDC has developed algorithms to generate nightly fire products from the OLS data. This involves screening incoming data against a set of known light sources from cities, towns, and gas flares to identify ephemeral fire events. The thermal band on OLS is used for cloud detection and is at too long a wavelength for general use in fire detection. NGDC has completed a six-month global fire product for October 1, 1994 through March 31, 1995. Other project areas have included Madagascar (1992-97), Indonesia (1997), the Mexico (1998), and the Amazon region (1998). Graphical products for many of these areas are available online at NOAA Defense Meteorological Satellite Program at NGDC. Compressed versions of the georeferenced fire and cloud images are made available through the NGDC ftp site. During the current fire monitoring of the Amazon region graphical products are being prepared for the individual Amazon states of Brazil. In this case the DMSP observed fire and cloud data are "stamped" onto forest/non-forest land cover maps derived primarily from Landsat TM data by the NASA Landsat Pathfinder project.
Researchers at ESA ESRIN have been developing a World Fire Atlas using AVHRR daytime data and Along Track Scanning Radiometer (ATSR) day and night time data. Three continents have been monitored for several years with AVHRR. The first continental atlas generated with ATSR data was for 1996. These data are readily available using the Fire Atlas Web Server. An ATSR World Fire Atlas for 1997 is currently being generated and ATSR data will eventually be processed for 1995 to the present. Planning is underway to generate a global burned area product in mid-1999. ESA/ESRIN has also been examining the use of combined ATSR and Synthetic Aperture Radar (SAR) for the study of forest fires in Borneo. As part of their development program ESRIN has developed the capacity for an ATSR Rush Fire Product. This rush fire product could process 10 out of 14 orbits in near real-time permitting active fire maps to be on the World Wide Web in less than 3 hours from sensing for most part of the world. This product could be used in supporting in situ validation campaign.
The MODerate Resolution Imaging Spectroradiometer (MODIS) fire team is developing the scientific basis for remote sensing of fires and fire products (CO, CO2, aerosols) using MODIS. MODIS is to be launched into a 10:30 am orbit on Earth Observing System-AM (EOS-AM1) and into a 1:30 pm orbit on EOS PM. Each MODIS will have 3.9 µm and 11 µm channels with high saturation (450 K and 400 K respectively) specially designed for fire monitoring. MODIS will have almost twice daily coverage. Its data will be used to detect fires, to estimate the rate of emission of radiative energy from the fire, and to estimate the fraction of biomass burned in the smoldering phase. The rate of emission of radiative energy is a measure of the rate of combustion of biomass in the fires. As a step towards inferring emitted gases from remote sensing of fire radiance, simultaneous measurements of fire infrared radiance and gas emissions for several different fuel types were made in a laboratory. These measurements correlate gaseous emission with radiant emission from biomass fires. If these correlations hold in the field, then satellite-observed infrared radiance can be used to estimate quantities such as total carbon burned, total biomass burned, and total carbon dioxide, carbon monoxide, and energy emitted from biomass fires on a global, daily basis.
In the Smoke, Clouds, and Radiation in Brazil (SCAR-B) experiment, the MODIS Airborne MAS was flown with a 50-m spatial resolution. These multispectral data were used to observe the thermal properties and sizes of fires in the cerrado grassland and Amazon forests of Brazil and to simulate and foresee the performance of the MODIS 1-km resolution fire observations. Although some fires saturated the MAS 3.9 µm channel, all the fires were well within the MODIS instrument saturation levels. Analysis of MAS data over four sites, representing different ecosystems, showed that the fire size varied from single MAS pixels (50 ¥ 50 m) to over 1 km2. The 1 ¥ 1 km resolution MODIS instrument will observe only 30-40% of these fires, but the observed fires are mostly larger fires that are responsible for 80 to nearly 100% of the emitted radiative energy and therefore for 80 to 100% of the rate of biomass burning in the region. The rate of emission of radiative energy from the fires was found to correlate very well with the formation of the burn scar from the fire (correlation coefficient = 0.97). Therefore this new remotely sensed quantity should be useful in regional estimates of biomass consumption when the two MODIS instruments are launched and in orbit.
The MODIS fire team is working with collaborators from the Southern Africa region and other NASA researchers to develop the SAFARI 2000 field experiment. In addition to a number of regional science objectives, this field experiment will be used to validate satellite-based fire and emissions data products. High spatial resolution satellite data, airborne imagery and field measurements will be used to validate MODIS data products. Field validation will be undertaken in conjunction with the IGBP Miombo Network. A number of test sites have been selected and will be used for MODIS product validation.
Given the wide range of demands from the various user communities, the relatively small community of satellite fire researchers needs to adopt strategies that will satisfy critical and sometimes conflicting user demands. There is a need to balance the development of near real-time monitoring capability with the processing of archival data back to the beginning of the satellite record. There is a need to balance the development of new experimental algorithms with developing community consensus algorithms and processing systems for consistent distributed processing. There is a need to balance the generation and provision of new data sets, with documentation of the limitations and a clear statement of data set accuracy. Users are often unclear as to the utility of different data sets and their associated algorithms; algorithm and product inter-comparison studies can help clarify the pros and cons of different products. It is often easier to generate remotely sensed data products than it is to validate them quantitatively.
Although there has been a noticeable increase in fire product data availability, users remain frustrated at the lack of long term operational fire products. It often appears that the remote sensing community is proceeding to the latest sensing system without fully exploiting the utility of existing data archives and making time series data more widely available. This is in part due to the costs of data and large-volume processing, an awareness of the limitations of the existing systems, and the desire to provide improved products. In most cases the user community has relatively little insight into the state of various fire product development activities or where to obtain data. To this end NASA recently developed a satellite monitoring Web Site providing a summary of products that are available. This fire Web Site will transition into the Earth Observatory planned for the Earth Observing System.
Mining and processing the existing long-term archives will pay off in terms of our understanding of trends in fire frequency. A series of regional demonstration studies need to be developed to provide a convincing case for different ecosystems with an explicit estimate of product accuracy. There also needs to be a conscious effort to provide data product continuity with the evolving technology of new sensing systems to secure a well-characterized long-term product record. There is an immediate need to secure fire monitoring as an integral part of the operational satellite monitoring systems being planned for the next decade.
Over the last five years there has been a small number of initiatives in Europe and in the United States to develop satellite systems specifically for fire monitoring. Of the various plans and designs that are being discussed for a single system, an operational global network of high temporal (30 minutes), mixed spatial resolution (250-500m-1km) geostationary satellites has considerable appeal to the fire monitoring community. Such a network would provide a good sampling of the diurnal cycle of fire events, an increased possibility of cloud free surface observation, an ability to detect and track smoke, and the possibility to develop rapid response systems.
Recognizing that no one sensing system can currently fulfill all the user data requirements, efforts are being initiated to develop multi-source fire data sets combining data with different spatial, spectral and temporal resolutions. Given the logistical and technical complexity of combining data from multiple sensing systems these initiatives are being developed at the regional scale and require enhanced data archive, management and query structures.
Considerable progress has been made in the last five years in terms of the ability to monitor fires by satellite both in near real-time for fire management and retrospectively for environmental research. In the next five years with the upcoming satellite missions for example from the US, Europe, and Japan, which include a number of high and moderate spatial resolution sensors, there will be a continued growth in the availability of fire related data products. Optimizing data products to address global change research questions and to improve fire management remains a continuing challenge to the research and operational communities.
