Fire radiative energy for quantitative study of biomass burning: derivation from the BIRD experimental satellite and comparison to MODIS fire products

Abstract

A major focus in global change research is to quantify the amount of gaseous and particulate pollutants emitted from terrestrial vegetation fires. Determination of the emitted radiant energy released during biomass combustion episodes (the so-called fire radiative energy or FRE) has been suggested as a new tool for determining variations in biomass combustion rates and the rate of production of atmospheric pollutants. We review the physical principals behind the remote determination of FRE and present an alternative method for its derivation via analysis of ‘fire pixel’ radiances in the middle infrared spectral region. We compare our method to the existing FRE retrieval approach used in the EOS Moderate Resolution Imaging Spectro-radiometer (MODIS) fire products, and to retrievals of FRE based on derived fire temperature and area made via the so-called Bi-spectral method. We test each FRE retrieval method using both simulated data and imagery from a new experimental space mission, the Bi-spectral InfraRed Detection (BIRD) small satellite, which has sensors specifically designed for the study of active fires. We analyse near simultaneous MODIS and BIRD data of the fires that burned around Sydney, Australia in January 2002. Despite the markedly different pixel size and spectral coverage of these sensors, where the spatial extent of the fire pixel groups detected by MODIS and BIRD are similar, the derived values of FRE for these fires agree to within ±15 %. However, in certain fires, the lower spatial resolution of MODIS appears to prevent many of the less intensely radiating fire pixels being detected as such, meaning MODIS underestimates FRE for these fires by up to 46% in comparison to BIRD. Though the FRE release of each of these low intensity fire pixels is relatively low, their comparatively large number makes their overall FRE significant. Thus, total FRE release of the Sydney fires on 5 January 2002 is estimated to be 6.5×10⁹ J s⁻¹ via BIRD but 4.0×10⁹ J s⁻¹ via MODIS. The ability of BIRD to resolve individual fire fronts further allows the first accurate calculation of ‘radiative’ fireline intensity from spaceborne measurements, providing values of 15–75 kJ s⁻¹ m⁻¹ for fire fronts that are up to 9 km in length. Finally, we analyse the effectiveness of the satellite-based FRE retrieval methods in estimating the FRE from the active flaming and smouldering components only (FRE_Active, believed to be proportional to the rate of biomass combustion), despite the sensor receiving additional radiance from the ‘cooling ground’. The MIR radiance method appears particularly strong in this regard, allowing FRE_Active to be estimated to within ±30% in the range 100–100,000 J s⁻¹ m⁻². These results provide further confidence in the ability of spaceborne missions to derive physically meaningful values of FRE that could be used to support biomass burning emissions inventories. Future comparisons between FRE derived via MODIS and those from higher spatial resolution BIRD or airborne imagery may allow the MODIS-derived FRE values to be ‘calibrated’ for any systematic underestimation. We therefore expect FRE to become an important tool for enhancing global studies of terrestrial vegetation fires with infrared remote sensing, particularly as the majority of large fires are now imaged four times per day via the MODIS instruments on the Terra and Aqua spacecraft.

Introduction

The worldwide combustion of forest and grassland vegetation releases large volumes of radiatively active gases, pyrogenic aerosols, and other chemically active species that significantly influence Earth's radiative budget and atmospheric chemistry (Andreae & Merlet, 2001). Scholes and Andreae (2000) have recently estimated that around 9200 Tg±50% (dry weight) of terrestrial vegetation is combusted per year, but improved data on regional and interannual variations is required to fully investigate the specific emissions sources and their subsequent transportation and effect (Wittenberg, Heimann, Esser, McGuire, & Sauf, 1998). This requirement has led to pyrogenic emissions estimation becoming a major research focus in the global change community, particularly so because of the need to better understand the effects of these emissions on the global climate Crutzen & Andreae, 1990, Scholes & Andreae, 2000. A few studies such as Kaufman, Tucker, and Fung (1990) and Spichtinger et al. (2001) have directly analysed emissions of smoke or particular chemical species via satellite sensor observations. However, the majority of pyrogenic emissions inventories are derived via relationships linking the concentrations of pollutant products to the amount of biomass combusted, the latter being commonly estimated from historical information on fire-frequency or from fire counts or maps of ‘burn-scars’ derived from Earth observation imagery (Andreae & Merlet, 2001). These data sources provide important information relating to burned area, but converting such measurements into estimates of burnt biomass involves additional knowledge of the pre-burn biomass density and the combustion factor (essentially the fraction of available biomass actually burnt). For the large areas affected by biomass burning, these factors are difficult or impossible to measure on the ground, but are also problematic to measure remotely (Pereira et al., 1999). Hence, biomass combustion and pyrogenic emissions estimates are often subject to potentially large errors. Andreae and Merlet (2001) demonstrate the current level of uncertainty by comparing emissions estimates for the savanna regions of southern African, based on both the traditional fire-frequency and current Earth observation approaches. Results from the two techniques differ by an order of magnitude, leading Andreae and Merlet to conclude that new and independent routes for providing pyrogenic emissions estimates are urgently required to resolve these differences and to improve regional and global assessments.

To address the requirement for new remote sensing tools useful for the estimation of biomass combustion rates and the associated emissions, Kaufman et al. (1996) introduced the concept of remotely measured fire radiative energy (FRE). Terrestrial, airborne, and spaceborne remote sensing has long been used to provide data on ‘fire counts’, essentially by detecting the strong infrared thermal emission signal associated with active fires (e.g., Hirsch et al., 1971, Matson et al., 1984, Waggoner, 1991). However, the enhancement proposed by Kaufman et al. was that quantification of the amount of energy radiated during the combustion process could provide a remote measurement related directly to the fire intensity and the amount of vegetation consumed per unit time. Remotely sensed FRE is therefore a candidate for the independent emissions-estimation route called for by Andreae and Merlet (2001). Kaufman et al. introduced the FRE concept for use with the Moderate Resolution Imaging Spectro-radiometer (MODIS), launched on the Terra and Aqua satellites in December 1999 and May 2002, respectively; this being possible due to the inclusion of low-gain infrared channels on MODIS that, unlike those on previous satellite imagers, provides unsaturated infrared measurements over even the largest fires. Kaufman, Justice, et al. (1998) and Kaufman, Kleidman, and King (1998) first used the method with data from the MODIS Airborne Simulator and reported the encouraging result that time-integrated FRE was better related to the observed growth of burnt areas than was a simple count of active fire pixels. Recently, Wooster (2002) quantitatively tested the relationship between FRE and the amount of biomass combusted using field spectro-radiometer observations of small experimental burns. Over almost two orders of magnitude the results show a linear agreement (r²=0.76, n=12) between the total radiative energy emitted during the burn (so-called time-integrated FRE) and the mass of vegetation combusted. We believe these results strongly support the idea of FRE as a valuable addition to the array of techniques used to assess biomass combustion in natural wildfires.

There are some limitations, however, not least the issue of cloud-cover that will obstruct satellite-derived FRE observations. There is also the issue that during combustion energy is lost by a variety of processes in addition to radiation, such as convection of the airmass above the fire and conduction into the ground (Asensio & Ferragut, 2002). However, the high temperatures involved in vegetation fires means that radiant energy loss is intense and the early studies cited above suggest that FRE is a valuable addition to the tools used to inform studies of biomass burning and the production and transport of atmospheric pollutants. Furthermore, its determination via spaceborne remote sensing means it can be derived on a subdaily basis for all regions subject to large-scale vegetation fire activity, cloud-cover permitting. In the current paper, we explain further the physical principals behind the remote measurement of FRE and present a new algorithm for its derivation, which we compare to existing approaches. We apply the differing FRE-derivation techniques to simulated observations and to real data from a newly launched spaceborne mission, the Bi-spectral InfraRed Detection (BIRD) small satellite, analysing the benefits and limitations of each approach. BIRD has been specifically designed to target high temperature events such as active fires, and the BIRD Hot Spot Recognition System (HSRS) operates with a 370-m pixel size compared to the 1–4-km pixel size of existing sensors most commonly used to study fire Fuller, 2000, Robinson, 1991. As such, in addition to providing important data in its own right, BIRD is expected to play an important role in validating lower spatial resolution fire products such as those from the AVHRR, ATSR, AATSR, GOES, and MODIS space missions. We provide the first comparison of near-simultaneous FRE observations of natural fires made from spaceborne sensors having markedly different characteristics, namely the BIRD-HSRS and Terra-MODIS instruments. The targets are the intense fires that burned around Sydney, Australia in January 2002. Finally, we analyse the effectiveness of the satellite-based FRE retrieval methods in estimating FRE from the active fire (flaming and smouldering) components only, believed to be the quantity most likely to be proportional to the rate of biomass combustion, despite the sensor recording an additional radiance contribution from the cooling ground that has recently been heated by the fires passage.