Atmospheric Environment

Transcription

1 Atmospheric Environment 44 (2010) 1469e1477 Contents lists available at ScienceDirect Atmospheric Environment journal homepage: Emissions of gases and particles from biomass burning during the 20th century using satellite data and an historical reconstruction A. Mieville a,b, *, C. Granier a,b,c,d, C. Liousse e, B. Guillaume e, F. Mouillot f, J.-F. Lamarque c,d,g, J.-M. Grégoire h, G. Pétron d,i a Université Pierre et Marie Curie-Paris 6, UMR8190, Paris, France b LATMOS CNRS, UMR8190, Paris, France c NOAA Earth System Research Laboratory, Chemical Sciences Division, Boulder, CO, USA d Cooperative Institute for Research in Environmental Sciences (CIRES), University of Colorado at Boulder, CO, USA e Laboratoire d'aérologie, UMR 5560, Toulouse, France f IRD, CEFE/CNRS, Montpellier, France g National Center for Atmospheric Research, Boulder, CO, USA h Joint Research Center, European Commission, Ispra, Italy i NOAA Earth System Research Laboratory, Global Monitoring Division, Boulder, CO, USA article info abstract Article history: Received 23 April 2009 Received in revised form 11 January 2010 Accepted 13 January 2010 Keywords: Emissions Climate change Gases Particles Biomass burning Burnt areas Historical Satellite A new dataset of emissions of trace gases and particles resulting from biomass burning has been developed for the historical and the recent period (1900e2005). The purpose of this work is to provide a consistent gridded emissions dataset of atmospheric chemical species from 1900 to 2005 for chemistry-climate simulations. The inventory is built in two steps. First, fire emissions are estimated for the recent period (1997e2005) using satellite products (GBA2000 burnt areas and ATSR fire hotspots); the temporal and spatial distribution of the CO 2 emissions for the 1997e2005 period is estimated through a calibration of ATSR fire hotspots. The historical inventory, covering the 1900e2000 period on a decadal basis, is derived from the historical reconstruction of burned areas from Mouillot and Field (2005). The historical emissions estimates are forced, for each main ecosystem, to agree with the recent inventory estimates, ensuring consistency between past and recent emissions. The methodology used for estimating the fire emissions is discussed, together with the time evolution of biomass burning emissions during the 20th century, first at the global scale and then for specific regions. The results are compared with the distributions provided by other inventories and results of inverse modeling studies. Ó 2010 Elsevier Ltd. All rights reserved. 1. Introduction During the 20th century, human activities and biomass burning have produced substantial emissions of trace gases and particles into the atmosphere. These emissions have resulted in significant perturbations in the radiative balance of the atmosphere and in air quality at regional and global scales. The definition of emission regulation policies and the evaluation of the effects of these policies require an accurate estimate of atmospheric emissions and of their temporal evolution. Since the late seventies, biomass burning has been known to be a major source of aerosols and gases in the atmosphere * Corresponding author at: LATMOS/Université Pierre et Marie Curie-Paris 6, UMR8190, Paris, France. address: aude.mieville@latmos.ipsl.fr (A. Mieville). (Seiler and Crutzen, 1980; Andreae et al., 1988). Globally, biomass burning contributes today to about 50% of the total direct CO emissions (Pétron et al., 2004) and about 15% of surface NO x emissions (IPCC, 2001). Pollutants emitted as a result of biomass burning can be injected into the atmosphere at relatively high altitudes, and be rapidly transported in the free troposphere. For example, it has been shown that polluted air masses resulting from fires in Brazil can be transported over the tropical Atlantic towards Africa and the Indian Ocean (Singh et al., 1996); plumes originating from Alaskan fires in 2004 have also been detected in Europe, leading to an increase in the ozone background concentration, and even to high ozone episodes (Real et al., 2007). Wildfires are known to have a large interannual variability (Duncan et al., 2003), and the resulting emissions are therefore very variable in time and space. Satellite sensors such as the Along-Track Scanning Radiometer (ATSR, Arino and Plummer, 2001) and Moderate Resolution Imaging Spectroradiometer /$ e see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi: /j.atmosenv

2 1470 A. Mieville et al. / Atmospheric Environment 44 (2010) 1469e1477 (MODIS, Justice et al., 2002), provide routinely the location of fires observed under the satellite overpass at a spatial resolution of 1 km; the quasi-global coverage from these observations has allowed a better identification of fires interannual variability at the global scale over the past few years (Duncan et al., 2003; van der Werf et al., 2006). These studies have shown that biomass burning accounts for most of the interannual variability of the emissions of several chemical compounds. In the tropics, this variability is in part linked to the El Niño-Southern Oscillation (ENSO), which produces episodes of droughts in equatorial and tropical regions, and therefore increases fire activity in these regions. During the 1997e1998 EL Niño episode, intense fires were observed in Indonesia, more particularly in peatland areas, leading to strong emissions of chemical species (van der Werf et al., 2006). In boreal regions, a very high interannual variability is observed as well, which has been proven to be correlated with the Arctic Oscillation and associated temperature anomalies (Baltzer et al., 2005). These large changes in the spatial and temporal distribution of emissions have to be taken into account when simulating the evolution of the atmospheric composition, and its impact on climate. A quantitative description of this variability has been made possible by the analysis of satellite data, but the satellite record is limited to the recent years (Liousse et al., 2004). The evolution of fire emissions during the last century is not well established and only a few studies have evaluated these emissions over several decades (Mouillot and Field, 2005; Mouillot et al., 2006; Schultz et al., 2008a); the quantification of the evolution of historical emissions is however essential for understanding the distribution of tropospheric species during the past century, and for the understanding of chemistry-climate interactions. The purpose of the work described here is to provide a consistent gridded emissions dataset of atmospheric chemical species from 1900 to 2005 for long-term chemistry-climate simulations. This dataset covers the period 1900e2000 on a decadal basis, as well as the recent period (1997e2005) on a annual basis. The historical and the recent emission inventories are built with different methods and types of data: for the recent period, observations from space-borne instruments are used, while for the historical period we have used an historical reconstruction of burnt areas from Mouillot and Field (2005). The recent period 1997e2005 is used as the reference point for the 1990s decade, which is also provided in the historical inventory: the biomass burning emissions estimate provided for the 1990s decade from the historical reconstruction is scaled, for each ecosystem, so that it matches the average 1997e2005 estimate, ensuring consistency between past and recent emissions. For the recent period, different types of products have been derived from space-borne instruments for the observation of fires. Distributions of fire counts obtained from instruments such as ATSR (since 1995) or MODIS (since November 2000) provide the location of fires but are not sufficient for estimating the real area and amount of biomass which burned, and consequently the emissions. Satellite imagery products providing directly the burnt areas are just becoming available for longer time periods, and are currently under evaluation (Brivio et al., 2009). In this study, the GBA2000 (Grégoire et al., 2003) satellite product providing burnt areas for the year 2000 is used, together with ATSR fire hotspots time series over the 1997e2005 period. Similar methods, where different satellite products are combined, have been described by van der Werf et al. (2006) and Giglio et al. (2006). This inventory is referred to as GICC (Global Inventory for Chemistry-Climate studies). After a description of the methodology used, the emission estimates obtained for the entire period are discussed and compared with other available inventories. Although the inventory provides the emissions of a long list of species (greenhouse gases, ozone precursors, particles and their precursors), the paper focuses mainly on the emissions of carbon dioxide (CO 2 ), carbon monoxide (CO), nitrogen oxides (NO x ) and black carbon (BC). 2. Methodology: estimation of biomass burning emissions for the 1997e2005 period The emission of a species X resulting from fires, E(X), is expressed (Seiler and Crutzen, 1980) as the product, for each vegetation class i, of the burnt area BA i (in m 2 ), the biomass density BD i (in kg m 2 ), the burning efficiency BE i and the emission factor of species X, EF i (X): EðXÞ ¼ XN i ¼ 1 BA i $BD i $BE i $EF i ðxþ (1) where N is the number of vegetation classes taken into account. The biomass density (BD) or fuel load provides the available biomass per surface unit; the burning efficiency (BE) corresponds to the percentage of the biomass which effectively burns. The emission factor gives the amount of chemical species emitted for a given amount of biomass burned. In this section, the quantification of each component of equation (1) is discussed Satellite fire observations and land cover map In this study, burned areas over the past decade are obtained or derived from the following space-borne observations: - The GBA2000 (Tansey et al., 2004) product provides global burnt areas for the year 2000 on a monthly basis. It is derived from visible and infrared measurements from the SPOT4/ VEGETATION space-borne instrument, launched in March An analysis of the images provides a cartography of the burnt areas at a 1 km resolution. - The ATSR WFA (World Fire Atlas; Piccolini and Arino, 2000) product provides daily maps of fire hotspots at a 1 km resolution since It is derived from the ATSR radiometer (Arino and Plummer, 2001) on-board ERS-2 (1995e2003) and ENVI- SAT (since 2003), which measures the radiation originating at the earth surface in the visible, mid-infrared and thermal infrared. The algorithm applied to detect fire spots is identical throughout the 1997e2005 period considered in this study, and provides therefore a consistent dataset for the full period under consideration. Only nighttime fires are provided, which means that not all fire events are detected, compared to the total fire activity. Other hotspots such as volcanoes or gas flares can be removed since these are observed continuously (Mota et al., 2006). In this study we used the dataset from Mota et al. (2006), regridded at a 0.5 resolution. The global land cover distribution is needed to determine the type of vegetation which is burning, when a fire is detected. In this study we used the GLC2000 map (Bartholomé and Belward, 2005), derived from the SPOT/VEGETATION space-borne system, at a 1 km resolution. The biomass densities (BD), burning efficiencies (BE), and emission factors (EF) of GLC2000 vegetation types were derived from values published in Michel et al. (2005) for the UMD land cover map (Defries et al., 1998; Hansen et al., 2000). Correspondences between the UMD and GLC2000 vegetation cover classes have been established (Liousse et al., 2005, P. Mayaux, pers. com.); the related BD, BE and CO 2 emission factors are indicated in Table 1.

3 A. Mieville et al. / Atmospheric Environment 44 (2010) 1469e Table 1 Values of BD (biomass density), BE (burning efficiency), EF (Emission Factor) for each GLC class, and factor BA/FC (Burnt Area per Fire Count). The unit kg dm corresponds to kg of dry matter. GLC classes 2.2. CO 2 emissions from GBA2000 BD (kg m 2 ) As a first step, CO 2 emissions related to each GLC2000 vegetation type were estimated from GBA2000 burnt areas using equation (1). The use of two products derived from the same instrument ensures a good co-localisation between burned areas and the associated vegetation type at a 1 km resolution; therefore the allocation of burned areas to vegetation types is achieved accurately at a high resolution. The product of the burnt area, biomass density, burning efficiency and CO 2 emission factor provides the amount of CO 2 emitted in each corresponding vegetation class. Table 2 gives the total biomass burnt and the associated CO 2 emissions obtained for year 2000 in each vegetation class Emissions for the 1997e2005 period BE EF(CO 2 ) (g(co 2 ) kg dm) BA/FC (km 2 FC 1 ) Broadleaf evergreen GLC Closed broadleaf deciduous GLC Open Broadleaf deciduous GLC Evergreen needleleaf forest GLC Deciduous needleleaf GLC Mixed leaf type GLC Mosaic: tree cover/other natural vegetation GLC9 Shrub, closed-open, evergreen GLC11 Shrub, closed-open, deciduous GLC12 Herbaceous cover, closed-open GLC13 Sparse herbaceous or sparse shrub cover GLC14 Cultivated and managed areas GLC16 Mosaic: cropl./tree/other natural veget. GLC17 Mosaic: cropland/shrub or grass GLC In the second step, emissions for the 1997e2005 period are estimated using the time series from ATSR fire hotspots. The ATSR dataset provides a sampled image of the fire activity and is used here to get the spatial and temporal patterns of fire activity. The emissions obtained from GBA2000 for year 2000 are used to calibrate ATSR fire hotspots; the correspondence between burnt areas and ATSR fire hotspots is made possible since ATSR fire data are available for year In our approach, we determine relationships between fire hotspots and burned areas in 2000, and the corresponding quantity of species emitted. As stated by van der Werf et al. (2006), Giglio et al. (2006), Tansey et al. (2008), Smith et al. (2007), we found out that the correlation between fires hotspots and burned areas is highly driven by the vegetation cover type, especially by the tree and herbaceous cover fraction. Therefore the relationship (burnt area per fire count) was first determined separately for each vegetation class, as indicated in Table 1. Within each vegetation class, the ratio of the burnt area to the number of fire hotspots is spatially very heterogeneous, as well as highly variable in time; in the case of exceptional fire events, this ratio is unusually higher, because of an «undersampling» of the fire counts. In the tropical band, where fire activity is predominant, the ratio is higher as well. Moreover, for some GLC2000 classes, Table 2 Total burnt biomass et CO 2 emission by vegetation type in 2000, as determined from GBA2000. Allocation of each vegetation class to either forest (F), savanna (S) or crops (C) ecosystems is indicated. Vegetation type GLC class Global burned biomass (Tg) Tree Cover, broadleaved, evergreen Tree Cover, broadleaved, deciduous, closed Tree Cover, broadleaved, deciduous, open Tree Cover, needle-leaved, evergreen Tree Cover, needle-leaved, deciduous 1 (F) (F) (S) (F) (F) Tree Cover, mixed leaf type 6 (F) Mosaic: Tree Cover/Other 9 (F) natural vegetation Shrub Cover, closed-open, 11 (S) evergreen Shrub Cover, closed-open, 12 (S) deciduous Herbaceous Cover, 13 (S) closed-open Sparse herbaceous or 14 (S) sparse shrub cover Cultivated and managed areas 16 (C) Mosaic: Cropland/Tree/Other 17 (F þ C) natural vegetation Mosaic: Cropland/Shrub and/or grass cover 18 (S þ C) Total CO 2 emissions (Tg-CO 2 year 1 ) the type of vegetation is somewhat specific for tropical regions: for example, the closed broadleaf deciduous class (GLC2) corresponds to woodland in tropical areas (intermediate between forest and savanna), whereas in temperate region it corresponds to deciduous tree forests. Vegetation patterns have also different characteristics (soil and air humidity) in temperate and tropical latitudes, which affect fire propagation; this also explains that the ratio is higher in tropical regions. Therefore the calibration was determined using three latitude zones ( 90 Sto 15 S; 15 Sto15 N; 15 N to 90 N). The relationship within one vegetation class and latitude zone is heterogeneous; splitting the globe into more zones would have led to a critical statistical number of fire counts per region and vegetation class. This approach is an approximation since the correlation between active fires and burnt surfaces is not straightforward, as we observed on the scatterplots of CO2 emissions (from GBA) versus the number of fire hotspots per vegetation class and latitude zone (not shown). The calibrations of the fire hotspots, obtained for each vegetation class and latitude zone from these relationships, are used to derive the monthly CO 2 emissions distribution for the 14 GLC2000 vegetation classes considered. Emissions for other gaseous and particulate species are calculated using the emissions ratios compiled by Andreae and Merlet (2001) for the following four main ecosystems: tropical forest, temperate/boreal forests, savannas and agriculture fires (Table 3). For black carbon (BC) and organic carbon (OC) however, Table 3 CO, NO x, BC and OC emissions ratios to CO 2 (unit: mol mol 1 ). CO NO x BC OC Forest fires: boreal and temperate forests Tropical forest fires Savanna fires Cultivated areas

4 1472 A. Mieville et al. / Atmospheric Environment 44 (2010) 1469e1477 the emission ratios were taken from a compilation of observations reported in Liousse et al. (2004). Each GLC2000 class is allocated to these four ecosystems as indicated in Table 2. For the mosaic classes with crops (GLC17 and GLC18), arbitrarily 50% of crops cover and 50% of the other vegetation type cover is used. 3. Methodology: estimation of historical biomass burning emissions 3.1. Historical burned areas For the historical period, the reconstruction of burnt areas compiled by Mouillot and Field (2005) is used. The burned areas were reconstructed on the basis of published data, data on land-use practices, qualitative reports, as well as from local studies such as tree ring analysis. The dataset provides, for each decade since 1900, the percentage of annual burnt surfaces on a 1 1 spatial grid (Fig. 1) Emissions estimation The historical biomass burning emissions are determined using the GLC2000 vegetation map which is derived from satellite imagery taken in the year Since the beginning of the century, land cover has very likely changed, especially in the amazonian basin. The uncertainties linked to the use of the same vegetation land cover throughout the period are assessed and discussed in Section 4.3. The GLC2000 map is first derived at , which provides the percentage of each vegetation class within each cell. The area burnt associated to each vegetation class is calculated according to the vegetation percentages within each cell. As indicated in Section 2.3 for the crops mosaic classes (GLC17 and GLC18), 50% of the surface is allocated to crops and 50% to the other vegetation type. The CO 2 emissions for the 14 vegetation classes are calculated using equation (1), and added in order to obtain gridded annual CO 2 emissions, for each decade from 1900 to A discrepancy was found between the historical burnt areas estimate for the 1990s decade and recent satellite-based estimate, with an annual burnt area of 5.65 Millions km 2 for the 1990s in the historical dataset, compared to 3.6 Millions km 2 in the GBA2000 product. Fig. 1 shows the distributions of burnt areas in both datasets. The ultimate goal being to provide a consistent set of emissions throughout the 1900e2005 period, the recent period 1997e2005 is used as the reference point for the 1990s decade, and the average 1997e2005 recent estimate serves as an initialization for the 1900e2000 inventory: the burnt area and thus the related emissions provided for the 1990s decade from the historical reconstruction are scaled, for each main ecosystem (forest and savanna), so that they match the average 1997e2005 estimate, ensuring consistency between the past and the recent emissions. On Fig. 2, the distribution of the CO 2 emissions obtained for the last decade is shown and is compared to the average 1997e2005 estimate and the average 1997e2005 estimate from GFED-v2, showing a good consistency with recent inventories. The scaling factors obtained (1/3.1 for forest ecosystem and 1/1.51 for savanna ecosystems) are then applied on burnt areas for every decade, for forest and savanna fires, respectively. The large seasonality of fire activity needs to be taken into account for the analysis of observations and for modeling studies. Here, we considered that the seasonal variation over the century is similar to the one observed during the recent 1997e2005 period. Fig. 1. Distribution of burnt areas (in km 2 ) from Mouillot and Field (2005), for the 1900e1910 (a) and 1990e2000 (b) decades, and comparison with GBA2000 burnt areas (c).

5 A. Mieville et al. / Atmospheric Environment 44 (2010) 1469e Fig. 2. Distribution of CO 2 emissions obtained for the 1990e2000 historical inventory (a), the average 1997e2005 estimate (b) and the average 1997e2005 estimate from GFED-v2 (c). The relative average seasonal variation is determined from the recent 1997e2005 inventory and is applied to the historical annual emissions, in order to derive historical monthly emissions. Such an approach will not allow the determination of short-term interannual variability of the distributions of chemical species during the past decades, but will help better determine the impact of fire emissions on trends in the distribution of chemical species during the past century. Emissions for the other chemical compounds are then calculated from the CO 2 emissions, using the emissions ratios indicated in Table Results: evolution of biomass burning emissions during the 1900e2005 period 4.1. Recent inventory (1997e2005) Average total annual emissions of carbon dioxide as a result of biomass burning are estimated to be 9950 Tg-CO 2 year 1 over the 1997e2005 period. Fires in cultivated areas contribute to a total of 325 Tg-CO 2 year 1, which represents approximately 3% of the total biomass burning emissions. Forest fire emissions are estimated at about 4470 Tg-CO 2 year 1 (45%), while savanna fire emissions represent about 5155 Tg-CO 2 year 1 (52%). Emissions display a very high interannual variability, as seen in Fig. 3, which displays yearly average CO 2 emissions in 1997 and This variability is mainly linked to the very high interannual variability of forest fire emissions: Fig. 3 shows large emissions in Indonesia in 1997 linked with an intense ENSO episode (van der Werf et al., 2006), and large emissions in the high latitude boreal forests of Canada and Siberia in Savanna fire and agriculture waste burning emissions display a much lower interannual variability. Monthly global CO 2 emissions are represented in Fig. 4, from 1997 to Biomass burning emissions display two seasonal maxima: a first is observed during the DecembereJanuaryeFebruary period and is due to fires in the northern tropics, especially in West Africa and in the savanna areas of Central Africa; a second maximum is seen in AugusteSeptember as a result of fires in the southern tropics (South Africa, South America, Australia), as well as in boreal regions (Siberia and Canada). In 1998, unusually high emissions were observed in East Siberia, and in Canada (e.g. Spichtinger et al., 2004). These emissions followed an intense drought from May to September During summer 2003, intensive fires were again observed in Siberia (from May to July), emitting high quantities of CO 2 (Jaffe et al., 2004). During summer 2004 and 2005, large fires occurred in the western part of Alaska and also led to very high emissions (Pfister et al., 2005). The amounts of CO, NO x, and BC emitted as a result of forest and savanna fires for the 1997e2005 period are given in Table 4. An evaluation of the emissions resulting from forest and savanna fires provided by the recent GICC inventory is performed through a comparison with the GFED-v2 fire emissions inventory (van der Werf et al., 2006), and the fire emissions estimated by inverse modeling of MOPITT CO observations. MOPITT is a satellite instrument on-board TERRA, operational since 2000, which measures CO in the troposphere (Edwards et al., 2006). Global estimates of CO emissions were optimized for the March 2000eDecember 2003 period, using a Kalman filter technique together with the MOZART chemistry-transport model and the MOPITT data at 850 hpa (Pétron et al., 2004 and unpublished

6 1474 A. Mieville et al. / Atmospheric Environment 44 (2010) 1469e1477 Table 4 Total emissions of CO (Tg CO year 1 ), NO x (Tg NO 2 year 1 ) and BC (Tg C year 1 ) from 1997 to CO NO x BC Fig. 3. CO 2 annual emissions in 1997 (a) and 1998 (b). updates). Emission ratios as shown in Table 3 were used together with this inverse modeling exercise to derive the emissions of the other compounds, including CO 2, based on the optimized CO emissions. Fig. 5 gives total annual CO 2 emissions from the three inventories: in this comparison, only forest and savanna fires from the GICC inventory are included, as in the other inventories. The GICC inventory exhibits higher CO 2 emissions than GFED-v2, for every year except for 1997, when GFED emissions are higher ( Tg- CO 2 in GFED versus 8945 Tg-CO 2 in this study). Over the whole 1997e2005 period, average annual emissions due to open fires from both inventories are rather close, with values of 9620 Tg- CO 2 year 1 and 8850 Tg-CO 2 year 1 in the GICC and GFED-v2 inventory, respectively, i.e. a 9% difference. On the other hand, both GFED-v2 and GICC inventories provide smaller emissions than those estimated from inverse modeling. The regional distribution of CO 2 emissions estimates provided by the three datasets is compared in Fig. 6 for the 2000e2003 period. For Eurasia, a region dominated by Siberian fire emissions, a good agreement is obtained for the three inventories. In North America, GICC emissions are systematically higher than GFED-v2, and closer to the inverse modeling results. In Africa, which provides about half of total CO 2 emissions, GICC and GFED agree within 10%, but display the largest difference with the inverse modeling results. In South America, GICC emissions are close to inverse modeling results, but higher than GFED emissions. A similar pattern is obtained for Oceania. These differences reflect the large uncertainties that remain in the determination of biomass burning emissions, from both bottomeup and topedown techniques. The largest uncertainties in the bottomeup approach are related to uncertainties in the calculation of emissions using calibrated fire hotspots, as well as to uncertainties on GBA burnt areas, biomass densities, burning efficiencies and emission factors values. Both GICC and GFED-v2 use a combination of burned areas and active fires products, and different land cover distributions. All of these products are significantly different and lead to the differences highlighted in Figs. 5 and Historical inventory (1900e2000) The evolution of total burnt surface areas from 1900 to 2000 is shown in Fig. 7a. In 1900, the total burnt area is about km 2 year 1. It decreases until about 1920, mainly in boreal regions (Siberia, Canada and North America). In this area, according to Mouillot and Field (2005), forest fires representing high carbon stocks have burned at a rate of km 2 year 1 at the beginning Fig. 4. Time series of monthly CO 2 emissions resulting from biomass burning for the 1997e2005 period.

7 A. Mieville et al. / Atmospheric Environment 44 (2010) 1469e Fig. 5. Comparison of annual CO 2 emissions as given by the GICC and GFED-v2 inventories, and as derived from CO emissions estimated by inverse modeling. of the century. The forest burning rate decreased to km 2 year 1 in 1960e1970, and to km 2 year 1 during the last decade of the 20th century. This decrease in boreal and temperate forest fires is the consequence of fire suppression policies, the improvement of fire fighting systems, and the end of intense deforestation in these regions. The total area burnt remained rather stable for the 1920s to 1970s decades, when it started to increase significantly, mainly as a result of deforestation in tropical regions, particularly in South America. The location of burnt surfaces has not changed significantly in savanna regions, whereas large changes have occurred in tropical forested areas: a clear progression of burnt areas from the southern border of Amazonia, towards the Amazonian forest is observed since the 1970s. It should be mentioned, however, that the evolution and variability of fires in these regions are very uncertain at the beginning of the century, since only a few observations are available for the tropical regions during this period. The evolution of CO 2 emissions for 8 world regions from 1900 to 2000 is given in Fig. 8. In North America and Siberia, emissions have decreased significantly during the 20th century. A steady decrease is also observed in Australia, which is likely due to less pronounced intense deforestation practices in this region. A large increase in emissions is shown in Africa and in South America during the second half of the century. In the southern part of South America, emissions were rather high at the beginning of the 20th century. This is probably linked to intense fires in savannas and shrub ecosystems around the BrazilianeArgentinian border, as a result of the expansion of agriculture in these areas. Large increases in the emissions in tropical and sub-tropical regions are also observed at the end of the century, in Central Asia, South-East Asia and especially Indonesia. Total CO 2 emissions from open fires, shown in Fig. 7b, follow the temporal evolution of burnt areas, with emissions of about 8150 Tg-CO 2 year 1 in 1900, and an increase since the 1970s, up to 9620 Tg-CO 2 year 1 in the 1990s. The GICC CO 2 emissions are compared in Fig. 7b with the RETRO emission inventory (Schultz et al., 2008a), derived on a decadal basis; GFED-v2 emissions averaged over 1997e2005, and estimates from the inverse modeling of MOPITT CO observations averaged over 2000e2003 are also shown. GICC emissions in the 1990s (1960s) are higher by 15% (25%) than the RETRO emissions, respectively, for the same decades. Such differences can however be considered as satisfactory, considering the high uncertainties on all the different datasets used, as discussed in Section 2. The differences between GICC and RETRO for the past decades are also due to the very large uncertainties in the distribution of the burned areas before 1990, when no satellite observations were available. Fig. 7c and d show the evolution of the global GICC emissions estimates of CO and BC from wildfires during the 20th century, together with values of the RETRO inventory. The relative differences obtained in the trace species emissions between the different datasets are very similar to the ones obtained for CO 2. BC emissions are also compared with the historical emissions provided by Ito and Penner (Ito and Penner, 2005), which were determined from a scaling of CH 4 emissions based on land-use changes. In 2000, Ito and Penner (2005) give emissions that are about 10% higher than in the GICC inventory. Furthermore, GICC emissions in 2000 are higher by 25% and 20% than RETRO and GFED-v2 emissions, respectively. The BC emissions from Ito and Penner (2005) are approximately 50e60% lower than the GICC emissions during the first decades of the 20th century. Considering the differences in the adopted methodologies and in the input data, this comparison shows a reasonable agreement, and leads furthermore to a quite robust evidence that the emissions of black carbon and of other species have not increased by more than a factor of 2 during the past 100 years Uncertainties on burnt areas and emissions Fig. 6. Comparison over different regions of the average CO 2 emissions for the 2000e2003 period, as given by the GICC and GFED-v2 inventories, and as derived from an estimation of CO emissions by inverse modeling. The study of Mouillot et al. (Geophys. Res. Lett., 2006) provides an estimation of the uncertainties on the burnt areas estimates. In 1960 the total burnt surface is estimated to be a higher best guess value with an uncertainty of about 1% per year, going back in time. This leads to an uncertainty of about 60% for For the recent period, based on the van der Werf approach and data from Grégoire et al. the uncertainties on burnt areas were evaluated to be around 30e40%. The GBA2000 product is known to underestimate burnt surfaces, especially over tropical humid forests (Amazonia, Indonesia), where

8 1476 A. Mieville et al. / Atmospheric Environment 44 (2010) 1469e1477 Fig. 7. Total burned area for the 1990e2000 period (a: top left); Emissions of CO 2 (b: top right), CO (c: bottom left) and BC (d: bottom right). Fig. 8. CO 2 emissions in different world areas from the 1900s to the 1990s. burnt scars are too small to be detected by the SPOT-Vegetation instrument at 1 km resolution. This problem is not seen in tropical humid forests of Africa, where fires are almost inexistent. It should be noted however that the burnt area estimates in GBA2000 are very similar to those of L3JRC and GFED-v2 estimates, at the global scale (Tansey et al., 2008). For the historical period, the use of the same land cover map also leads to uncertainties in the emissions estimates, especially in Amazonia where land cover has changed rapidly; the related error results from the adopted type of vegetation that is burned. Since the 1960s, due to deforestation in Amazonia, the percentage of forest coverage has decreased, to be replaced by savanna-type cover, although the southern forest boundary has not moved noticeably. Therefore, before the 1960s, the use of the same vegetation map for the different decades is not likely to lead to significant errors since most fires were occurring essentially in areas dominated by savanna/crops/shrub covers. After 1960, more fires have occurred in the evergreen dense forest with a proportion of approximately 20% of forest and 80% of savanna burning (Mouillot and Field, 2005). In the historical inventory the proportion of forest burnt areas between the 1960s and the 1990s ranges between 20% (1960s) and 38% (1990s) for forest burning (after scaling historical burnt areas on recent values). The emissions are estimated to be actually overestimated by a factor of 1 for 1960 to 1.25 for the 1990s. The resulting error is likely to be within the uncertainties on burnt areas values. Errors in the allocation of burnt surface to land cover type could as well lead locally to errors in the historical emission estimates of about 100%; these errors are likely to prevail on errors due to the uncertainties on the factors BD (biomass density), BE (burning efficiency), and EF (Emission Factor). 5. Conclusions This paper describes the methodology used to estimate historical and recent global emissions of gases and particles from biomass burning. Such an inventory can be used in modeling studies of the evolution of the composition of the atmosphere over the past century, and for assessing the impact of biomass burning on atmospheric chemistry during the past decades. The inventory provides, for each decade from 1900 to 2000, gridded emission data at a 1 1 resolution, with an average

9 A. Mieville et al. / Atmospheric Environment 44 (2010) 1469e seasonal variation. The recent inventory provides for each year from 1997 to 2005, gridded monthly emissions at a resolution. Newly released annual or pluri-annual burnt area datasets, such as L3JRC (Tansey et al., 2008) and MODIS burnt area product (Roy et al., 2008), are now available; some of these datasets, such as the L3JRC product are still under evaluation for use at the global scale; we intend to use these new products in the future for a new version of the GICC inventory. The historical inventory shows a decrease in biomass burning emissions from the beginning of the century to the 1920s. The emissions estimates have remained stable then until the 1970s, with an average of 7400 Tg-CO 2 year 1. Since the 1980s, emissions have increased rapidly and are currently about 9950 Tg-CO 2 year 1. This global evolution of emissions is the result of two different features: in boreal regions, burnt surfaces and emissions have decreased during the past century, as a result of fire suppression policy. On the contrary, in tropical regions, fire emissions have been increasing rapidly since the 1980s, especially in South America and Indonesia. This trend is related to human pressure in these areas, and to the use of fires for deforestation as part of agriculture expansion. Large uncertainties remain, however, in the determination of the trends in biomass burning emissions. Comparisons with available inventories show a good agreement for the recent years, and agreements within 15e50%, depending on the species, for the past four decades. The main uncertainties are related to the limited knowledge of the evolution of the distribution of burned areas, to differences between available vegetation maps, and to uncertainties in biomass densities, burning efficiencies in the different ecosystems, and on emissions factors. Recent developments based on the use of Fire Radiative Energy (Ickoku et al., 2008; Schultz et al., 2008b) could also help reduce these uncertainties as the radiative energy from fires can be directly converted into burnt biomass. The full GICC dataset (1900e2005) will be made available to the research community through the GEIA/ACCENT emissions portal, Acknowledgments This work has been partly supported by the French program Gestion et Impacts du Changement Climatique (GICC), by the LEFE/INSU program and by the CITYZEN European FP7 project. The distribution of the emissions is supported by the ACCENT European network. The authors would like to thank Martin Schultz, Guido van der Werf and Guy Brasseur for very valuable comments on the manuscript. G. Pétron acknowledges the support of the NCAR Advanced Study Program for the inverse modeling emissions estimates. References Andreae, M.O., et al., Biomass-burning emissions and associated haze layers over Amazonia. J. Geophys. Res. 93, 1509e1527. Andreae, M.O., Merlet, P., Emissions of trace gases and aerosols from biomass burning. Global Biogeochem. Cycles 15, 955e966. Arino, O., Plummer, S. (Eds.), Along Track Scanning Radiometer World Fire Atlas: Validation of the 1997e98 Active Fire Product Validation Report, IGBP-DIS Working Pap. 23, Int. Geosph. Biosph. Progr., Stockholm. Baltzer, H., et al., Impact of the Arctic Oscillation pattern on inter-annual forest fire variability in Central Siberia. Geophys. Res. Lett. 32, L doi: /2005gl Bartholomé, E., Belward, A.S., GLC2000: a new approach to global land cover mapping from Earth observation data. Int. J. Remote Sens. 26 (9), 1959e1977. Brivio, P.A., et al., 2009, Intercomparison of global estimates of CO emissions derived from satellite-based burnt biomass products. In: Proceedings of the 33rd International Symposium on Remote Sensing of Environment (ISRSE), Stresa, Italy. Defries, et al., Global land cover classification at 8 km spatial resolution: the use of training data derived from Landsat imagery in decision tree classifiers. Int. J. Remote Sens. 19, 3141e3168. Duncan, et al., Interannual and seasonal variability of biomass burning emissions constrained by satellite observations. J. Geophys. Res. 108 (D2). doi: /2002jd Edwards, D.P., et al., Southern Hemisphere carbon monoxide interannual variability observed by Terra/Measurement of Pollution in the Troposphere (MOPITT). J. Geophys. Res. 111, D doi: /2006jd Giglio, L., et al., Global estimation of burned area using MODIS active fire observations. Atmos. Chem. Phys. 6, 957e974. Grégoire, J.-M., Tansey, K., Silva, J.M.N., The GBA2000 initiative: developing a global burned area database from SPOT-VEGETATION imagery. Int. J. Remote Sens. 24 (6), 1369e1376. Hansen, M.C., et al., Global land cover classification at 1 km spatial resolution using a classification tree approach. Int. J. Remote Sens. 21, 1331e1364. Ickoku, C., et al., Global characterization of biomass-burning patterns using satellite measurements of fire radiative energy. Remote Sens. Environ. 112, 2950e2962. IPCC, Climate Change, Synthesis report. In: Watson, R.T., the Core Writing Team (Eds.), A Contribution of Working Groups I, II, and III to the Third Assessment Report of the Integovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom/New York, NY, USA, p Ito, A., Penner, J.E., Historical emissions of carbonaceous aerosols from biomass and fossil fuel burning for the period 1870e2000. Global Biogeochem. Cycles 19, GB2028. doi: /2004gb Jaffe, D., et al., Long-range transport of Siberian biomass burning emissions and impact on surface ozone in western North America. Geophys. Res. Lett. 31, L doi: /2004gl Justice, C., et al., The MODIS fire products. Remote Sens. Environ. 83, 244e262. Liousse, C., et al., Deriving global quantitative estimates for spatial and temporal distributions of biomass burning emissions. In: Granier, C., Artaxo, P., Reeves, C. (Eds.), Emissions of Atmospheric Trace Compounds. Kluwer Academic Publishers, Dordrecht, The Netherlands, p Liousse, C., et al., D-modelling of carbonaceous aerosols over greater Paris focusing on the organic particle formation. J. Atmos. Chem. 51 (2), 207e221. Michel, C., et al., Biomass burning emissions inventory from burnt area data given by SPOT-Vegetation satellite for TRACE-P and ACE-Asia case studies. J. Geophys. Res. 110, D doi: /2004jd Mota, B.W., et al., Screening the ESA ATSR-2 world fire Atlas (1997e2002). Atmos. Chem. Phys. 6, 1409e1424. Mouillot, F., Field, C.B., Fire history and the global carbon budget: a 1 1 fire history reconstruction for the 20th century. Glob. Chang. Biol. 11, 398e420. Mouillot, F., et al., Global carbon emissions from biomass burning in the 20th century. Geophys. Res. Lett. 33 doi: /2005gl Pétron, G., et al., Monthly CO surface sources inventory based on the 2000e2001 MOPITT satellite data. Geophys. Res. Lett. 31, Pfister, G., et al., Quantifying CO emissions from the 2004 Alaskan wildfires using MOPITT CO data. Geophys. Res. Lett. 32, L11809, doi: / 2005GL Piccolini, I., Arino, O., March Towards a global burned surface World atlas. Earth Observation Quarterly, ATSR Special Issue, ESA publication. Real, E., et al., Processes influencing ozone levels in Alaskan forest fire plumes during long-range transport over the North Atlantic. J. Geophys. Res. 112, D10S41. doi: /2006jd Roy, D.P., et al., The collection 5 MODIS burned area product e global evaluation by comparison with the MODIS active fire product. Remote Sens. Environ. 112, 3690e3707. Schultz, M.G., et al., 2008a. Global emissions from Wildland fires from 1960 to Global Biogeochem. Cycles 22, GB2002. doi: /2007gb Schultz, M.G., et al., 2008b. Evaluation of a Fire Radiative Power product derived from Meteosat-8/9 and identification of operational user needs, Final report to EUMETSAT contract. Seiler, W., Crutzen, P.J., Estimates of gross and net fluxes of carbon between the biosphere and the atmosphere from biomass burning. Clim. Change 2, 207e247. Singh, H.B., et al., Impact of biomass burning emissions on the composition of the South Atlantic troposphere: reactive nitrogen and ozone. J. Geophys. Res. 101 (D19), 24203e Smith, R., et al., Estimating the area of stubble burning from the number of active fires detected by satellite. Remote Sens. Environ. 109, 95e106. Spichtinger, N., et al., Boreal forest fires in 1997 and 1998: a seasonal comparison using transport model simulations and measurement data. Atmos. Chem. Phys. 4, 1857e1868. Tansey, K., et al., A global inventory of burned areas at 1 km resolution for the year 2000 derived from SPOT VEGETATION data. Clim. Change 67, 345e377. doi: /s Tansey, K., et al., A new, global, multi-annual (2000e2007) burnt area product at 1 km resolution. Geophys. Res. Lett. 35, L doi: / 2007GL van der Werf, G.R., et al., Interannual variability in global biomass burning emissions from 1997 to Atmos. Chem. Phys. 6, 3423e3441.