Nitrogen deposition in tropical forests from savanna and deforestation fires

We used satellite‐derived estimates of global fire emissions and a chemical transport model to estimate atmospheric nitrogen (N) fluxes from savanna and deforestation fires in tropical ecosystems. N emissions and reactive N deposition led to a net transport of N equatorward, from savannas and areas undergoing deforestation to tropical forests. Deposition of fire‐emitted N in savannas was only 26% of emissions – indicating a net export from this biome. On average, net N loss from fires (the sum of emissions and deposition) was equivalent to approximately 22% of biological N fixation (BNF) in savannas (4.0 kg N ha−1 yr−1) and 38% of BNF in ecosystems at the deforestation frontier (9.3 kg N ha−1 yr−1). Net N gains from fires occurred in interior tropical forests at a rate equivalent to 3% of their BNF (0.8 kg N ha−1 yr−1). This percentage was highest for African tropical forests in the Congo Basin (15%; 3.4 kg N ha−1 yr−1) owing to equatorward transport from frequently burning savannas north and south of the basin. These results provide evidence for cross‐biome atmospheric fluxes of N that may help to sustain productivity in some tropical forest ecosystems on millennial timescales. Anthropogenic fires associated with slash and burn agriculture and deforestation in the southern part of the Amazon Basin and across Southeast Asia have substantially increased N deposition in these regions in recent decades and may contribute to increased rates of carbon accumulation in secondary forests and other N‐limited ecosystems.


Introduction
Nitrogen (N) is an essential element in Earth's atmosphere, biosphere, and hydrosphere (Galloway et al., 2003). The vast majority of N in the atmosphere is N 2 , which is biologically unavailable to most organisms. N must be converted from N 2 to reactive N (N r ), which includes inorganic oxidized forms (e.g., NO, NO 2 , and HNO 3 ), inorganic reduced forms (NH 3 and NH 4 1 ), and organic forms (e.g., amino acids and urea), before its use by microbes or plants. N 2 to N r conversion (N fixation) occurs during electrical discharges in lightning and by some bacteria and algae via a process known as biological N fixation (BNF) (Cleveland et al., 1999). In terrestrial ecosystems, humans have more than doubled the flux of N 2 to N r through fossil fuel use, industrial N fertilizer production, and cultivation of N-fixing crops Smil, 1999;Galloway et al., 2004Galloway et al., , 2008. Much of our knowledge of the N cycle and the consequences of N enrichment comes from studies of tempe-rate ecosystems, where industrial and agricultural modification of N availability has been substantial. Tropical ecosystems respond differently to perturbations in N cycling (Matson et al., 1999;Davidson et al., 2004;Bustamante et al., 2006) and important uncertainties remain with respect to our understanding of N deposition, fixation, and loss pathways in these ecosystems. With implementation of stricter air quality standards, N emissions and deposition are likely to decrease in temperate regions, particularly in developed countries. In contrast, N emissions and deposition in tropical and subtropical regions are likely to increase (Galloway et al., 1994;Lamarque et al., 2005;Dentener et al., 2006). Since tropical forests and savannas account for more than half of global terrestrial net primary production (NPP) (Field et al., 1998) and may contribute substantially to contemporary (Stephens et al., 2007) and future land C sinks (Friedlingstein et al., 2006), it is important to quantify the processes contributing to changing levels of N deposition in these ecosystems and subsequent impacts on N cycling and carbon storage.
Fires, including wildfires and prescribed fires, influence the N cycle within tropical ecosystems by changing both the availability and the mobility of N. Fire-induced losses of N by means of volatilization are substantial (Raison et al., 1985;Cook, 1994;Bustamante et al., 2006) and contribute to wet and dry deposition of N in downwind ecosystems. Nitric oxide (NO) and ammonia (NH 3 ) are the primary N r gases emitted from these fires, accounting typically for over 90% of N r emissions (Andreae & Merlet, 2001). These gases are converted to other N r gases and to particulate species (NO 3 À , NH 4 1 , and organic aerosols) by means of multiple reaction pathways (Crutzen & Andreae, 1990;Atkinson, 2000). The ultimate fate of N r in the atmosphere is removal by wet and dry deposition -processes regulated by the amount and intensity of precipitation, surface roughness, wind speed, and regional patterns of atmospheric transport (Dentener et al., 2006). Emissions of N r probably account for only 40%-50% of the N present in fuels before combustion, as some is retained within the ecosystem in combusted residues (approximately 20%) and the remainder (another 30%-40%) is transformed directly into N 2 during combustion (Lobert et al., 1990;Kuhlbusch et al., 1991). N 2 emissions from fires represent a net loss of N from the biospherecontrasting with N r fluxes that instead lead to redistribution of N r across different ecosystems. Fires influence nutrient cycling in tropical ecosystems in multiple ways, apart from the emissions during the initial event and subsequent deposition of the reactive component described above. Immediately after fire, soil NO and N 2 O losses often increase for relatively short periods (Verchot et al., 2006;Davidson et al., 2008), probably from N additions in ash to the soil surface and mortality of fine roots that increase levels of mineralization and nitrification (Dunn et al., 1979;Neff et al., 1995). Yienger & Levy (1995) incorporated these responses into a global model, and estimated that postfire pulses of NO account for approximately 10% (0.6 Tg N yr À1 ) of tropical soil emissions annually. Over a period of years to decades, fire-induced changes in species composition (e.g., Scholes & Archer, 1997) can alter N cycling by several different pathways, including by changing the abundance of N-fixing trees and shrubs (Okello et al., 2008), by changing the abundance of asymbiotic N fixers (Mack et al., 2001), or by triggering a series of feedbacks that change the long-term fire regime (Mack & D'Antonio, 1998;Cochrane et al., 1999;Hoffmann et al., 2002) and thus the volatilization losses described above. In many grassland and savanna ecosystems, grazing intensity plays an important role in regulating N losses during fire (Hobbs et al., 1991) and N limitation in postfire ecosystems (Buis et al., 2009). Deposition of N from tropical fires is likely to have spatially heterogeneous effects on NPP, depending in part on ecosystem type, soil properties, past disturbance history, and N inputs from other sources. Multi-ple studies and meta-analyses document widespread N limitation in savanna and grassland ecosystems (Elser et al., 2007;Lebauer & Treseder, 2008;Xia & Wan, 2008), with the magnitude of sensitivity often modulated by grazing intensity and fire frequency (Cech et al., 2008). Patterns in tropical forests are more complex, but may allow for several generalizations. First, in many but not all fertilization studies in secondary forests, N limits multiple components of NPP (Campo & Vázquez-Yanes, 2004;Davidson et al., 2004). Postdisturbance trajectories of litter and soil chemical composition and N 2 O fluxes provide additional indirect evidence that N limitation is strong during early stages of secondary succession (Davidson et al., 2007). N is also limiting in forests on relatively young and unweathered soils (Vitousek & Farrington, 1997;Harrington et al., 2001). In contrast, in primary tropical forests on highly weathered soils, phosphorus (P) limits NPP and in many instances fertilization by N alone does not stimulate significant increases in production (Harrington et al., 2001). However, stimulation of some NPP components by N has been observed in a lowland forest, including increases in the reproductive fraction of litterfall (Kaspari et al., 2008). Further, other studies in lowland forests document N by P interactions, in which the combined response to the two nutrients exceeds the response to P alone (Tanner et al., 1992;Mirmanto et al., 1999;Harrington et al., 2001). In the broader context of atmospheric deposition from many sources, including P from fires, dust, and biogenic sources (Mahowald et al., 2008), this suggests that N deposition from fire has the potential to synergistically interact with other deposition elements to sustain productivity.
Here we assessed the spatial and temporal patterns of N emissions and N r deposition caused by fires in tropical ecosystems. We estimated fire emissions using the Global Fire Emission Database version 2 (GFEDv2; van der Werf et al., 2006), which combines satellite observations of burned area (BA)  with estimates of fuel loads (FL) obtained from a biogeochemical model constrained by other satellite data. We estimated pyrodenitrification losses to N 2 for different ecosystems by scaling up measurements obtained from chamber combustion experiments (Kuhlbusch et al., 1991). Transport and deposition of N r were simulated using the GEOS-Chem global chemical transport model (CTM) . We performed 10-year (1997-2006) global-scale simulations, focusing our analysis and discussion on tropical ecosystems. We partly validated our modeled fluxes by comparing our results with published observations of N deposition. By quantifying the net N balance associated with both emissions and deposition in each model grid cell, we assessed the spatial distribution of donor and recipient regions. Emissions and deposition fluxes were compared with terrestrial and marine BNF fluxes (Cleveland et al., 1999;Deutsch et al., 2001;Lee et al., 2002) as a means to gauge the relative impact of fires on ecosystem N cycling.

N r emissions from biomass burning
The GFEDv2 product consists of 11 Â 11 gridded monthly BA, FL, combustion completeness (CC), emission factors (EMFs), and emissions (E) for different gas and aerosol species . BA was derived using Moderate Resolution Imaging Spectroradiometer (MODIS) active fire and BA datasets for the 2000-2006 period as described by Giglio et al. (2006). BA estimates were extrapolated back in time using Tropical Rain Monitoring Mission and Along Track Scanning Radiometer active fire observations. FL and CC values were estimated using the Carnegie-Ames-Stanford Approach (CASA) biogeochemical model constrained by additional satellite observations of fractional tree cover and the fraction of absorbed photosynthetically active radiation by plant canopies . Emission rates for each gaseous N species were calculated by applying EMFs to fire-emitted carbon fluxes from CASA Organic dry matter (DM) in the fuel was assumed to be comprised of 45% carbon. EMFs, defined as the emission of species (g) per 1 kg burned DM, were obtained from Andreae & Merlet (2001) with updates from M. Andreae (personal communication).
Comparison with atmospheric carbon monoxide (CO) observations provides some validation of our carbon emissions and CTM. GFEDv2 CO emissions in equatorial Asia, for example, required a small negative adjustment (from 0% to 22% depending on atmospheric inversion approach) to match Measurements of Pollution in the Troposphere (MOPITT) satellite observations during 2000.
Atmospheric CO anomalies, obtained after removing an annual mean cycle and long-term trend, have large variations from year-to-year that are primarily driven by fires (e.g., . This is because, at remote flask sampling stations in both hemispheres, other important CO sources such as methane and volatile organic compound oxidation and fossil fuel emissions vary by only a small amount from yearto-year. In contrast, fires on different continents may vary by a factor of 30 or more (e.g., van der Werf et al., 2008) as a consequence of year-to-year changes in climate and patterns of land use. CO anomalies from GFEDv2 in the GEOS-CHEM model had a mean correlation (r) of 0.78 with observations from NOAA Global Monitoring Division (GMD) stations in the northern hemisphere (NH) and 0.67 with GMD stations in the southern hemisphere (SH) during 1997-2008 (Table 1; Fig.  S1 in supporting information). Similarly, the ratio of modeled to observed standard deviations of CO anomalies from GMD was 0.92 and 1.01, respectively, for NH and SHs. These results suggest that while important uncertainties still remain in our emissions and transport estimates, these two model components are probably adequate for preliminary assessments of the impact of fires on atmospheric chemistry. N r can be emitted into the atmosphere in different chemical forms and physical phases. NO x and NH 3 account for the majority (490%) of N r emissions from biomass burning (Andreae & Merlet, 2001). In our model simulations and calculations, we used the sum of gaseous NO x and NH 3 emissions to represent total N r emissions from biomass burning. The low bias we expected from neglecting the remaining N r emissions (e.g., N 2 O, HCN, and acetonitrile) was small compared with uncertainties associated with BA (e.g., Giglio et al., 2006), FL, CC, and NO and NH 3 EMFs. The model results were sampled at the time and place of the measurements. s is the standard deviation of modeled or measured CO anomalies during this period. To estimate CO anomalies, both an annual cycle and a linear trend were removed from the observations and model estimates at each station using the same procedure. NOAA GMD CO observations were analyzed following the approach described by Novelli et al., (1998).

Direct emissions of N 2 during biomass burning
A large gap has been observed in the N balance between its content in the fuel and the sum of measured N in gaseous emissions and what remains after combustion in the ash (Lobert et al., 1990). By burning different biomass materials in a stainless steel chamber, Kuhlbusch et al. (1991) showed that the missing N is mostly molecular N (N 2 ), which accounts for approximately one third of fuel N. This experiment also showed that N 2 is formed primarily during the flaming stage and thus the total amount of N 2 production (as a fraction of fuel N) depends on fire intensity. The CO to CO 2 emission ratio serves as an indicator for the relative extent of flaming combustion and may serve as a means to scale fire intensity estimates over larger regions. Analysis of the Kuhlbusch et al. (1991) measurements showed that the N 2 -N to N r -N ratio of fire emissions was inversely related to its CO/CO 2 ratio (Fig. S2). The N 2 -N to N r -N ratio derived from this relationship was used to estimate global N 2 emissions from biomass burning. In this study, we estimated the CO/CO 2 ratio for three different fire types: savanna and grassland fires, tropical forest fires, and extratropical forest fires using the EMFs of CO and CO 2 listed in Table 1 of Andreae & Merlet (2001). A 21(latitude) Â 2.51 (longitude) vegetation map of these three types (Andreae & Merlet, 2001) was used to derive the global gridded N 2 -N to N r -N ratio. For each grid cell, we scaled GFED N r emissions by this ratio to obtain the fire emissions rate of N 2 .

Model simulations of N transport and deposition
We used the GEOS-Chem model to calculate the transport, transformation, and deposition of biomass burning emitted N species. GEOS-Chem is a global 3-D CTM  driven by assimilated meteorological observations from the Goddard Earth Observing System (GEOS) of the NASA Global Modeling and Assimilation Office (GMAO). In this study, we used version 08-01-01 of the model with 21 Â 2.51 horizontal resolution and 30 vertical layers between the surface and 0.01 hPa.
The GEOS-Chem model includes a detailed description of tropospheric O 3 -NO x -hydrocarbon chemistry (Wang et al., 1998). Emission sources of N include NO x from lightning, soil and fertilizer, aircraft, biofuel, fossil fuel, biomass burning, and NH 3 from biofuel, fossil fuel, biomass burning, and natural sources (e.g., soils, vegetation, and wild animals). N tracers are present in gas phase [e.g., NO x , peroxyacyl nitrate (PAN), HNO 3 , N 2 O 5 , and NH 3 ] or aerosol phase (NH 4 1 , N on sulfate, and N on sea salt). These tracers are transported by meteorology-driven advection and convection processes, and are removed by the action of precipitation (wet deposition) or by interaction with vegetation or soils in the lowest atmospheric layer (dry deposition) or vegetation (dry deposition). We performed GEOS-Chem full chemistry simulations over a 10-year period (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) using the GFEDv2 inventory, which resolves the interannual variability of biomass burning emissions. N emissions from other sources were obtained from climatological inventories which represent contemporary (circa 2000) emissions. After a 3-month spin-up period with the same initial conditions, two sets of simulations were conducted: one with biomass burning emissions prescribed from GFEDv2, and another without these emissions. The difference between these two simulations represents the effect of biomass burning emissions on N transport and deposition. We recorded monthly mean deposition rates for each N species through dry and wet deposition. Wet deposition included convective and large-scale components (Liu et al., 2001). Total deposition rate of N from biomass burning was then calculated for each 21 Â 2.51 model grid cell. Cleveland et al. (1999) reviewed published studies and generated three linear regressions (conservative, central, and upperbound) between BNF and evapotranspiration (ET) in terrestrial ecosystems. Galloway et al. (2004) argued that contemporary N fixation rates are at the lower end of this range, due to sampling biases in many plot-scale studies. In this study, as a simple measure to gauge the relative importance of fire impacts on ecosystem N fluxes, we constructed maps of BNF using the mean of central and conservative regressions between BNF and ET from Cleveland et al. (1999). The global ET estimates we used for this spatial extrapolation were developed using satellite data and a bio-meteorological approach by Fisher et al. (2008). The spatial pattern of BNF, as well as probability distributions of ET and BNF across tropical terrestrial ecosystems, are shown in Fig.  1. Based on the distribution of ET in savanna and tropical forest areas, we estimated the mean BNF rates in these two biomes to be 18.6 and 26.2 kg N ha À1 yr À1 (Fig. 1), respectively. As expected, these globally extrapolated BNF values were similar to mean estimates from Cleveland et al. (1999).

BNF
Previous estimates of global ocean BNF range from 60 to 200 Tg N yr À1 (Duce et al., 2008). Large uncertainty remains in the estimation of regional marine N fixation (Mahaffey et al., 2005). In this study, we used reported estimates of BNF rates for the tropical Atlantic (Lee et al., 2002) and Pacific (Deutsch et al., 2001). Owing to a dearth of data for other regions, we assumed that the mean of these two rates represented the N fixation rate in other tropical oceans.

Validation of wet and dry deposition
Relative to NH industrial regions, few N deposition measurements exist for tropical land and ocean regions. Here we compiled wet and dry N deposition measurements in tropical regions published in recent years. We only considered observations from 1996 to 2006 to ensure overlap with the period of our satellite-derived fire emissions time series. We differentiated urban and nonurban observations according to the description of the site in the original literature or the proximity of the measurement station to an urban center. Based on the original data, we calculated wet and dry deposition rates for total oxidized N (NO 3 -N) and reduced N (NH 4 -N) with units of kg N ha À1 yr À1 . Altogether, we reported wet and dry deposition rates from 26 nonurban stations. The location and time period of these measurements, as well as observed and modeled deposition rates, are summarized in Table S1. We compared these observations with modeled deposition fluxes, which were sampled in grid cells corresponding to observations during the month when observations were available. For locations that had measurements only during 1996, we compared the deposition rates with our 10-year (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) mean model results.
Overall, the model simulation corresponded reasonably well to wet and dry deposition observations at nonurban sites (Fig. 2). About 75% of the wet deposition rates and 80% of the dry deposition rates agreed within AE 50% of the observations. This agreement was similar to previous comparisons of N deposition between models and observations (Lamarque et al., 2005;Dentener et al., 2006). The model somewhat underestimated NO 3 -N wet deposition and overestimated NH 4 -N dry deposition. Simulations were more strongly correlated with measurements for dry deposition than for wet deposition, as indicated by the r values shown in Fig.  2. For sites substantially affected by fire emissions (filled points in Fig. 2), the agreement between our model simulation and measurements was similar.
In contrast to nonurban sites, the model was generally biased low for urban data, which were likely affected by nearby fossil fuel emissions (Fig. 2). Urban data were generally not representative of the coarse 21 Â 2.51 grid cells in the model, so they were not included in our linear regressions. Most of these urban measurement sites were located in Southeast Asia.

Gross N emissions from fires
Emissions of N r and N 2 by fires were 11.4 Tg N yr À1 (5.4 kg N ha À1 yr À1 ) in savannas and 5.3 Tg N yr À1 (3.8 kg N ha À1 yr À1 ) in tropical forests, with these two biomes accounting for 75% of global fire N emissions ( Fig. 3a and Table 2). High emission rates occurred in African savannas, as well as in deforestation regions in South America and Southeast Asia (see Fig. S3). Losses directly to N 2 accounted for about half of the total N emissions flux. This fraction was slightly higher for savannas than for tropical forests because of more complete combustion (Fig. S2).

N deposition rates from biomass burning
Fires contributed substantially to both N emissions and deposition in the tropics compared with other natural and anthropogenic components (Fig. 4). Deposition rates from fires were highest near source regions -with maxima on the perimeter of the Congo basin in Africa (up to 9.9 kg N ha À1 yr À1 ), in southern Borneo in equatorial Asia (up to 9.7 kg N ha À1 yr À1 ), and across the Brazilian state of Mato Grosso, eastern Bolivia, and Paraguay in South America (up to 5.3 kg N ha À1 yr À1 ) (Fig. 3b). In these areas, fires accounted for between 50% and 80% of the total N deposition flux -considering all other natural and anthropogenic sources. These results implied that unburned forest and savanna ecosystems in these regions (including parks and other protected areas) experienced relatively high levels of N deposition from nearby savanna and deforestation fires. Seasonal variations in N deposition were closely linked with the timing of fire emissions (Fig. 5). Total wet deposition was comparable to total dry deposition across the tropics, although regionally the relative importance of these two pathways varied. Wet deposition accounted for a larger fraction of total deposition in Southeast Asia (57%) and NH Africa (55%) than in South America (46%) or Southern Hemisphere (SH) Africa (39%). About 2/3 of the total N wet deposition occurred in the gas phase (NH 3 and HNO 3 ). The relative importance of wet deposition as an atmospheric loss pathway increased toward the end of the fire season on all three continents (Fig. 5), as a consequence of increasing precipitation levels. Model estimates of the total deposition rate for oxidized N and reduced N were similar. Separation of N deposition by its oxidation form is important for assessing N deposition effects on ecosystem function. Reduced N, for example, can reduce base cation uptake by plants (de Graaf et al., 1998), causing changes in species composition in ecosystems such as heathlands (van den Berg et al., 2008) and peatlands (Paulissen et al., 2004).

Changes in N surface budgets caused by fire emissions
The emissions and deposition of N from fires led to a net loss of N from the terrestrial biosphere and a redistribution of N among ecosystems. Savanna ecosystems in Africa and South America and tropical deforestation frontiers in all three regions experienced a net loss of N (Figs. 3c, S2). Net gains at a regional scale occurred in interior tropical forests and over tropical oceans.
The net loss of N (N r and N 2 ) from savanna ecosystems was largest in Africa (Table 2), where fires burned more frequently than on other continents . Net fire losses were equal to approximately 34% of BNF in savanna ecosystems in Africa, corresponding to 6.7 Tg N yr À1 or 5.7 kg N ha À1 yr À1 in savanna ecosystems (Tables 2, S2). Deposition from fires in African savannas offset only 26% of emissions to N 2 and N r ( Table 2). Considering N r , deposition from fires was only 53% of emissions, indicating that African savannas were a net exporter of N r to other biomes, including tropical forests in the Congo basin and the tropical Atlantic Ocean (Fig. 3b). For savannas across all three continents, net fire N losses were equal to 22% of BNF (4.0 kg N ha À1 yr À1 ), with only 55% of N r emissions offset by deposition inputs.
Tropical forests received large N deposition inputs from savanna fires but also lost N through deforestation fires. Overall, tropical forest fires had a net N loss equal to approximately 8% of BNF (2.8 Tg N yr À1 or Observed wet deposition (kg N ha -1 yr -1 ) Modeled wet deposition (kg N ha -1 yr -1 ) Observed dry deposition (kg N ha -1 yr -1 ) Modeled dry deposition (kg N ha -1 yr -1 ) (a) (b) Fig. 2 Comparison of observed and modeled (a) wet and (b) dry N deposition rates (kg N ha À1 yr À1 ) in tropical regions, with NO 3 -N points in red and NH 4 -N points in blue. Gray points are observation stations near urban areas. Circles, diamonds, and triangles represent data from South America, Africa, and Southeast Asia, respectively. Dashed lines indicate 1 : 1, 1 : 2, and 2 : 1 relationships. Filled points represent a subset of nonurban stations where more than 30% of the annual N deposition originated from fires. Wet and dry deposition comparisons had similar regression statistics (wet deposition had a slope of 1.01, an intercept of À0.34, 42 points, r 5 0.89, and Po0.01 whereas dry deposition had a slope of 0.84, an intercept of À0.08, 10 points, r 5 0.84, and Po0.01). For stations where fire deposition was significant, the linear fits of wet deposition had a slope of 0.64, an intercept of À0.41, eight points, r 5 0.74, and P % 0.02 whereas dry deposition had a slope of 1.36, an intercept of À0.02, six points, r 5 0.94, and Po0.01.
2.0 kg N ha À1 yr À1 ). In Southeast Asia, net losses were higher (20%), because here deforestation fires were more widely distributed within the ecosystem. In 'interior' tropical forests that were removed from the deforestation frontier (areas with fire emissions below a 500 kg C ha À1 yr À1 threshold), remote tropical fires led to a net gain of N by means of deposition. In interior forests in Africa, for example, the net atmospheric flux from fires was 0.50 Tg N yr À1 (3.4 kg N ha À1 yr À1 ), corresponding to approximately 15% of BNF. In contrast, forests at the deforestation frontier in South America and Southeast Asia had a large net N loss at a regional scale from fire emissions. Expressed as a percentage of BNF, net N losses at the deforestation frontier (38%) were even higher than in savanna ecosystems.
Tropical oceans were a large recipient of N from fire emissions. Approximately 37% (2.85 Tg N yr À1 ) of fireemitted N r was deposited over oceans. The tropical Atlantic, in particular, received high N deposition fluxes -up to 3.8 kg N ha À1 yr À1 (Fig. 3), owing to high levels of fire emissions in Africa and the prevalent easterly wind during the fire season (see Fig. S4). A small amount of the N from African fire emissions was transported across the Atlantic and deposited over South America. This pattern is consistent with analysis of lidar mea-surements from Manaus that show long-range transport of smoke and dust from Africa (Ansmann et al., 2009). The southern part of South China Sea also received substantial N deposition (up to 2.7 kg N ha À1 yr À1 ) from fires in Indonesia and Peninsular Malaysia.

Redistribution of N between ecosystems in Africa
Fires caused a net transport of N from savannas to interior forests in tropical Africa ( Fig. 3 and Table 2). Evergreen broadleaf forests were distributed mostly between 51N and 51S and were bordered to the north and south by savannas (Fig. 6a). BNF was highest near the equator and decreased to the north and south with decreasing ET (Fig. 6b). The highest levels of N r and N 2 fire emissions occurred at $ 71N in the NH and $ 101S in the southern hemisphere. The N deposition pattern, particularly for dry deposition, was similar to that of fire emissions but with a small equatorward shift in location of the peak deposition in the NH. A portion of N r emitted from savanna fires was deposited back in savannas. However, a substantial amount of fireemitted N r was transported further equatorward and deposited in tropical forests. In NH winter, which corresponded to the NH fire season, prevailing winds were to the south toward the Intertropical Convergence Zone (ITCZ) in the SH (Fig. 6c). Convective lifting of air masses near the equator increased precipitation and, subsequently, rates of wet deposition in tropical forests. The transport was reversed in NH summer, when winds to the north carried emissions from SH savannas across interior forests. The Sahara and Kalahari deserts, which were located upwind of the African savanna regions during the fire seasons in the northern and southern hemispheres, respectively, received relatively small N deposition fluxes from fire emissions. The combined effects of biomass burning emissions, atmospheric transport, and deposition resulted in a net transport of N from savannas to forests in Africa (Figs 6f  and 3c). This equatorward transport pattern was not unique to Africa. Prevailing winds in South America and Southeast Asia also caused transport of fire emissions toward interior (and relatively undisturbed) areas of the Amazon and the Indonesian archipelago (Fig. S4). This may be explained more generally by a covariance between fire emissions and relatively strong surface winds (toward the ITCZ) associated with the winter hemisphere of a Hadley cell (e.g., Plumb & Mahlman, 1987).

Consequences for terrestrial ecosystem processes
In terms of assessing the impact of the N fluxes described above for tropical ecosystem functioning, it may be useful The domains of tropical regions in Africa, South America, and Southeast Asia are shown in Fig. 3. MODIS-IGBP vegetation classes from the MODIS 500 m land cover product (MOD12) were used to determine biome types. Biological nitrogen fixation (BNF) rates were estimated using the mean of central and conservative regressions between BNF and evapotranspiration (ET) as given in Cleveland et al. (1999). *'Tropical Forest' refers to 'evergreen broad leaf forest' in the MODIS-IGBP classification. w'Interior' refers to tropical forest regions that are rarely affected by human activities. In this study, they are defined as 21 Â 2.51 grid cells where the fraction of evergreen broad leaf forest (from MODIS 500 m land cover product) is over 30% and the 10-year average fire emission rate (from GFEDv2) is smaller than 500 kg C ha À1 yr À1 . z'Frontier' refers to tropical forest regions undergoing substantial anthropogenic deforestation. Here they are defined as regions with forest cover 430%  to consider two timescales -one over the last millennium (but before the Industrial Revolution) and another covering the period after World War II when rates of tropical deforestation have been high (Houghton, 2003;Hurtt et al., 2006). Before the Industrial Revolution, humans had important roles in shaping the fire regime across South America, Africa, Australia, and Southeast Asia, particularly in savannas (e.g., Marlon et al., 2009). More recently, fire has been deployed at an unprecedented scale as a tool in clearing tropical forests and peatlands for pastures and croplands, often in conjunction with large-scale agricultural efforts (Page et al., 2002;DeFries et al., 2008;Langner & Siegert, 2009) and more extensively in South America and Southeast Asia than in Africa (e.g., Hansen et al., 2008). The fire emissions time series we used here contained components associated with human activities spanning both timescales. Fires in savannas and at the deforestation frontier account for a substantial amount of global fire emissions. Fires in savannas were 1.2 Pg C yr À1 during 1997-2006, the most of any biome globally and accounting for 50% of global fire carbon emissions . In Africa, where most savannas are located, the impact of recent increases in population and concurrent changes in land use have not necessarily increased fire emissions (Archibald et al., 2009). Fire emissions at the deforestation frontier, defined here as fires in areas where humid tropical forest cover loss was measured using independent Landsat and MODIS observations (Hansen et al., 2008), were also substantial during 1997-2006, accounting for 30% of global fire emissions (0.7 Pg C yr À1 ) (Le Quere et al., 2009). This latter component of global fire emissions is a relatively new phenomena and one that is directly linked with rates of forest and peatland clearing for pastures and crops van der Werf et al., 2008). In South America, this component of fire emissions was distributed across the 'arc of deforestation' across Bolivia and Brazil (Schroeder et al., 2005). In Southeast Asia, these emissions were interspersed among forests across the Indonesian archipelago and other countries including Malaysia, Mynamar, Vietnam, and Thailand. Deforestation rates in Africa during 2000-2005 were much lower than in Asia or South America (Hansen et al., 2008), implying that the fire-emitted component of carbon losses associated with this flux were lower as well.
On millennial timescales, the cross-biome transport of N documented here may help to sustain productivity levels within tropical forests in central Africa, particularly given that this flux is accompanied in many areas by finemode black carbon aerosols containing phosphorus and potassium (Mahowald et al., 2005) and phosphorus from other natural sources (Mahowald et al., 2008). Although many soils in central Africa are oxisols and utisols (USDA-NRCS, 2005) where P would be expected to be the primary element limiting NPP, some studies suggest Fig. 4 Latitudinal distribution of reactive N (N r ) emission and deposition rates (kg N ha À1 yr À1 ) from fires and other sources. Inserted plots show the mean values and fire contribution fractions for three tropical regions as defined in Fig. 3. The values are 10-year (1997-2006) averages. that even on highly weathered soils the effect of combined N and P additions can exceed that of P alone (Tanner et al., 1992;Mirmanto et al. 1999;Harrington et al., 2001;Campo & Vázquez-Yanes, 2004). Long-term N deposition fluxes from fires are likely to influence other ecosystem processes, including rates of decomposition in actively cycling soil carbon components and carbon storage in the bulk soil (e.g., Cusack et al., 2009). For Africa, until better long-term records of fire activity become available, it remains difficult to assess unidirectional shifts in the fire regime and thus consequences for trends in NPP and carbon storage within tropical forests (e.g., Lewis et al., 2009) by means of the atmospheric transport pathways assessed here.
In contrast with Africa, deforestation fires in South America and Southeast Asia have probably substantially modified N and P deposition rates in recent decades, particularly in areas near the deforestation frontier. On Borneo, for example, airport visibility records show marked increases in atmospheric aerosol concentration during El Nino events only after large-scale government efforts to increase settlements on the island occurred during the 1980s and 1990s (Field et al., 2009). In the southern Amazon, long-term active fire records show increases in fire activity near transportation corridors (Schroeder et al. 2005) that have increased in spatial extent in recent decades. Secondary forest areas on these two continents are widespread and interspersed within areas of active deforestation (Hirsch et al., 2004;Hurtt et al., 2006). Multiple types of observations provide evidence for N limitation in these secondary forests (Campo & Vázquez-Yanes, 2004;Davidson et al., 2004;Davidson et al., 2007), suggesting they have the potential to respond to changes in deposition caused by fires.
The redistribution of N and P from deforestation fires may increase carbon storage in nearby primary and secondary forests in South America and Southeast Asia by stimulating NPP, and thus offsetting some of the net carbon loss associated with changes in carbon stocks in deforestation areas. These remote deposition-driven increases would likely be small in magnitude but widely distributed, thus serving potentially as one of the drivers for the observed increases in aboveground biomass measured over the last several decades in the Amazon (Philips et al., 2009). In the context of these atmospheric linkages, the amount of deforested biomass emitted as fire sets the relative importance of this negative feedback and highlights the need to improve our understanding of this highly uncertain component of the deforestation flux. Other fire effects on remote forests need to be considered in parallel, including increases in photosynthesis caused by aerosol effects on diffuse light  Cleveland et al. (1999), (c) meridional wind speed (m s À1 ) at 850 hPa, with northward wind denoted as positive, (d) N emissions from fires, (e) N deposition from fires, and (f) net fire N and reactive N (N r ) fluxes (kg N ha À1 yr À1 ) for a north-south swath across Africa between 101E and 301E. The net fire N r flux includes losses from N r emissions and gains from N r deposition. This domain covers, from north to south, the Sahara desert, northern hemisphere (NH) savanna, tropical forest within the Congo Basin, southern hemisphere (SH) savanna, and the Karahari desert in Southwest Africa. The values are 10-year (1997-2006) averages. The meridional wind speed, derived from GEOS-4 reanalysis data, was averaged over February (during the time of peak fire activity in the NH) and August (during the time of peak fire activity in the NH). and canopy temperature (Oliveira et al., 2007;Mercado et al., 2009) as well as the deleterious effects of O 3 created by fire-emitted NO.
Deposition patterns associated with a changing fire regime will also likely interact with other drivers of global change. Modeling studies suggest tropical forests have great potential to increase NPP in response to elevated levels of CO 2 (Friedlingstein et al., 2006;Ciais et al., 2008), although a key uncertainty is whether nutrient availability can keep pace with photosynthesis to allow for long-term increases in biomass (e.g., Hungate et al., 2003;Luo et al., 2004;Thornton et al., 2009). Indeed, free air CO 2 enrichment studies from mid-latitude forests show that rates of carbon accumulation in response to elevated levels of CO 2 increase when nutrient availability is enhanced (Oren et al., 2001). If similar relationships exist for tropical secondary or primary forests, increasing deposition from deforestation fires may be enhancing the sensitivity of intact and recovering forests to increasing levels of atmospheric CO 2 .
In other regions where N deposition rates are large from other sources, additional inputs from deforestation fires may have a negative effect on carbon storage by influencing nitrification and nitrate mobility in soil (e.g., Aber et al., 1998). We estimate that at present day, only 9.3% of the tropical forests have N deposition rate 410 kg N ha À1 yr À1 , a critical load over which the negative effect is evident (Dentener et al., 2006;Phoenix et al., 2006). However, Lamarque et al. (2005) suggest that a three-fold increase in N deposition over forested areas is possible from increasing fossil fuel combustion by 2100 compared with present-day conditions. Therefore, the combined influences of fire and energy production could increase the area of tropical forests receiving over 10 kg N ha À1 yr À1 in coming decades with potentially negative effects of N deposition occurring at these higher levels.

Consequences for ocean ecosystems
Detailed analysis of the effect of atmospheric N deposition on marine ecosystems is complex for several reasons. First, currently there are not enough observations to derive global maps of N fixation . Second, increases in atmospheric N deposition by anthropogenic activity may be partially offset by lower rates of N fixation (Krishnamurthy et al., 2007). Nevertheless, N is generally considered a primary limiting nutrient for phytoplankton biomass accumulation in marine ecosystems, particularly in coastal systems (Rabalais, 2002). A large portion ($37%) of N r from savanna and forest fires is carried to and deposited over tropical oceans (Table 2). Averaged over the whole tropical ocean, this deposition is relatively small (2%) compared with BNF and also with respect to N transport to surface waters by upwelling . However, the ratio is much higher in coastal regions in Africa and Southeast Asia (Table 2) and the majority of the flux distributed in time over a period of only a few months. Further work is needed to identify how nutrient loading from fires affects marine ecosystem function in these highly impacted areas, including phytoplankton primary production and the health of coral reefs. In equatorial Asia after the 1997/1998 El Nino, for example, coral reef death was attributed to extraordinary red tide caused by deposition of iron from fire emissions (Abram et al., 2003).

Conclusions
We estimated the spatial patterns of emissions and deposition of N from tropical fires. Our results indicated a redistribution of N between different types of tropical ecosystems, with equatorward transport of N from fires in savannas and active deforestation frontiers to interior, tropical forests. On average, net fire losses were equivalent to 22% of BNF in savannas and 38% of BNF for ecosystems at the deforestation frontier. Interior tropical forests gained N equivalent to 3% of their BNF. This atmospheric transport of N was closely related to the seasonality of fire emissions and meteorological patterns in tropics. The surface branch of the Hadley cell carried emissions from these fires toward the Intertropical Convergence Zone (ITCZ). In Africa, the prevailing winds carried fire emissions from savannas toward the Congo Basin during both NH and SH fire seasons. As a consequence of widespread savanna burning in Africa in both hemispheres, the net transfer of N from savannas to forests was larger than that observed in South America or Asia.
Improving our understanding of atmospheric N fluxes in tropical regions will require new measurement programs to collect key observations, including N deposition measurements near deforestation regions and improved estimates of N emissions from fires associated with forest clearing. More detailed measurements of N losses to N 2 during fires also are needed to reduce uncertainties associated with pyrodenitrification. Although considerable effort has gone into improving our understanding of the deposition of chemical species in tropics [e.g., the Deposition of Biogeochemically Important Trace Species (DEBITS) network, http://www.igac.noaa.gov/DEBITS.php], measurements of tropical N deposition are still sparse compared with the density of observations in available in temperate countries. Satellite measurements of NO 2 have the potential to provide important constraints in this regard, particularly with respect to emissions and some aspects of chemistry and transport (e.g., Chai et al., 2009). In terms of assessing the impacts of a changing fire regime on carbon storage and species diversity, key uncertainties remain with respect to understanding interactions between deposition of N and other species such as P and potassium associated with fires and other natural and anthropogenic sources.

Supporting Information
Additional Supporting Information may be found in the online version of this article: Figure S1. Time series of monthly mean CO anomalies from NOAA GMD stations and the GEOS-Chem model. Figure S2. The fraction of N lost to N 2 during fire emissions as a function of the CO to CO 2 ratio. Figure S3. Forest cover loss in tropical regions. Figure S4. Biomass burning emissions and 850 kPa wind vectors for South America, Africa, and Southeast Asia. Table S1. Summary of N deposition observations. Table S2. N fluxes from fires in tropical regions -same as Table 2 but expressed in units of N emissions or deposition per unit of land area.
Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.