Preindustrial nitrous oxide emissions from the land biosphere estimated by using a global biogeochemistry model

To accurately assess how increased global nitrous oxide (N2O) emission has affected the climate system requires a robust estimation of the preindustrial N2O emissions since only the difference between current and preindustrial emissions represents net drivers of anthropogenic climate change. However, large uncertainty exists in previous estimates of preindustrial N2O emissions from the land biosphere, while preindustrial N2O emissions on the finer scales, such as regional, biome, or sector scales, have not been well quantified yet. In this study, we applied a processbased Dynamic Land Ecosystem Model (DLEM) to estimate the magnitude and spatial patterns of preindustrial N2O fluxes at the biome, continental, and global level as driven by multiple environmental factors. Uncertainties associated with key parameters were also evaluated. Our study indicates that the mean of the preindustrial N2O emission was approximately 6.20 Tg N yr−1, with an uncertainty range of 4.76 to 8.13 Tg N yr−1. The estimated N2O emission varied significantly at spatial and biome levels. South America, Africa, and Southern Asia accounted for 34.12, 23.85, and 18.93 %, respectively, together contributing 76.90 % of global total emission. The tropics were identified as the major source of N2O released into the atmosphere, accounting for 64.66 % of the total emission. Our multi-scale estimates provide a robust reference for assessing the climate forcing of anthropogenic N2O emission from the land biosphere


Introduction
Nitrous oxide (N 2 O) acts as the third most important greenhouse gas (GHG) after carbon dioxide (CO 2 ) and methane (CH 4 ), contributing greatly to the current radiative forcing (Myhre et al., 2013). Nitrous oxide is also the most longlived reactant, resulting in the destruction of stratospheric ozone (Prather et al., 2015;Ravishankara et al., 2009). The atmospheric concentration of N 2 O has increased from 275 to 329 parts per billion (ppb) since the preindustrial era until 2015 at a rate of approximately 0.26 % per year, as a result of human activities (Davidson, 2009;Forster et al., 2007;NOAA2006A, 2016. The human-induced N 2 O emissions from the terrestrial biosphere have offset about 50 % of terrestrial CO 2 sink and contributed a net warming effect on the climate system (Tian et al., 2016). In the contemporary period, anthropogenic N 2 O emissions are mainly caused by the expansion of agricultural land area and an increase in nitrogen (N) fertilizer application, as well as industrial activities, biomass burning, and indirect emissions from reactive N (Galloway et al., 2004;Reay et al., 2012). Human-induced biogenic N 2 O emissions were calculated by subtracting the preindustrial emissions (Tian et al., 2016), even though a small amount of anthropogenic N 2 O emissions was present before 1860, which was estimated as being 1.1 Tg N yr −1 for 1850 by Syakila and Kroeze (2011) and 0.7 (0.6-0.8) Tg N yr −1 (including anthropogenic biogenic emissions from soils and biomass burning) for 1860 by David-son (2009). Therefore, it is necessary to provide a robust reference of preindustrial N 2 O emission for assessing the climate forcing of anthropogenic N 2 O emission from the land biosphere.
Numerous studies have reported the sources and estimates of N 2 O emission since the preindustrial era (Davidson and Kanter, 2014;Galloway et al., 2004;Kroeze et al., 1999;Prather et al., 2012Prather et al., , 2015Syakila and Kroeze, 2011). According to the Intergovernmental Panel on Climate Change Guidelines (IPCC, 1997), the global N 2 O emission evaluated by Kroeze et al. (1999) is 11 (8-13) Tg N yr −1 (natural soils: 5.6-6.6 Tg N yr −1 ; anthropogenic: 1.4 Tg N yr −1 ), which is consistent with the estimation from global preagricultural N 2 O emissions in soils (6-7 Tg N yr −1 ) (Bouwman et al., 1993). While taking into account the new emission factor from the IPCC 2006 Guidelines (Denman et al., 2007), Syakila and Kroeze (2011) conducted an updated estimate based on the study of Kroeze et al. (1999) and reported that the global preindustrial N 2 O emission is 11.6 Tg N yr −1 (anthropogenic: 1.1 Tg N yr −1 ; natural soils: 7 Tg N yr −1 ). Based on the IPCC AR5, Davidson and Kanter (2014) indicated that the central estimates of both top-down and bottom-up approaches for preindustrial natural emissions were in agreement at 11 (10-12) Tg N yr −1 , including natural emission from soils at 6.6 (3.3-9.0) Tg N yr −1 (Syakila and Kroeze, 2011). Prather et al. (2015) provided an estimate of the preindustrial emissions (total natural emission: 10.5 Tg N yr −1 ) based on the most recent study with a corrected lifetime of 116 ± 9 years. Although these previous estimates intend to provide a baseline of preindustrial N 2 O emission at a global level, information on preindustrial N 2 O emissions at fine resolutions such as biome, sector or country, and regional levels remains unknown but is needed for climate change mitigation.
Large uncertainties in the estimates of preindustrial N 2 O emission could derive from different approaches (i.e., topdown and bottom-up), as mentioned above. Nitrous oxide, as an important component of the N cycle, is produced by biological processes such as denitrification and nitrification in terrestrial and aquatic systems (Schmidt et al., 2004;Smith and Arah, 1990;Wrage et al., 2001). In order to accurately estimate preindustrial N 2 O emissions using the processbased Dynamic Land Ecosystem Model (DLEM; Tian et al., 2010), uncertainties associated with key parameters, such as maximum nitrification and denitrification rates, biological N fixation (BNF) rates, and the adsorption coefficient for soil ammonium (NH + 4 ) and nitrate (NO − 3 ), were required to be considered in model simulation. Upper and lower limits of these parameters were used to derive a range of preindustrial N 2 O emissions from terrestrial ecosystems.
In this study, the DLEM was used to simulate global N 2 O emission in the preindustrial era at a resolution of 0.5 • × 0.5 • latitude × longitude. Since there are no observational data of N 2 O emission in the preindustrial period, the estimates of natural emission from Prather et al. (2012Prather et al. ( , 2015 were used to validate the simulation results. In addition, site-level N 2 O emissions from different natural vegetation were used to test model performance in the contemporary period. The objectives in this study include (1) providing a global estimation of N 2 O emission from terrestrial soils in 1860, (2) offering the continental-, biome-, and country-scale N 2 O emission amounts and flux rates, and (3) discussing uncertainties in estimating N 2 O budget in the preindustrial era. Finally, our estimates on global and biome scales were compared with previous estimates.

Model description
The DLEM is a highly integrated process-based ecosystem model, which combines biophysical characteristics, plant physiological processes, biogeochemical cycles, vegetation dynamics, and land use to make daily, spatially explicit estimates of carbon, nitrogen, and water fluxes and pool sizes in terrestrial ecosystems from site and regional to global scales (Lu and Tian, 2013;Tian et al., 2012. The DLEM is characterized of cohort structure, multiple soil layer processes, coupled carbon, water, and nitrogen cycles, multiple GHG emissions simulation, enhanced land surface processes, and dynamic linkages between terrestrial and riverine ecosystems (Liu et al., 2013;Tian et al., 2010. The previous results of GHG emissions from DLEM simulations have been validated against field observations and measurements at various sites (Lu and Tian, 2013;Ren et al., 2011;Tian et al., 2010Tian et al., , 2011. The estimates of water, carbon, and nutrient fluxes and storage were also compared with the estimates from different approaches on regional, continental, and global scales (Pan et al., 2014;Yang et al., 2015). Different soil organic pools and calculations of decomposition rates have been described in . The decomposition and nitrogen mineralization processes in the DLEM have been described in previous publications (Lu and Tian, 2013;Yang et al., 2015).

The N 2 O module
Previous work provided a detailed description of trace gas modules in the DLEM (Tian et al., 2010). However, both denitrification and nitrification processes have been modified based on first-order kinetics (Chatskikh et al., 2005;Heinen, 2006).
In the DLEM, the N 2 O production and fluxes are determined by soil inorganic N content (NH + 4 and NO − 3 ) and environmental factors, such as soil texture, temperature, and moisture: where F N 2 O is the N 2 O flux from soils to the atmosphere (g N m 2 d −1 ), R nit is the daily nitrification rate (g N m 2 d −1 ), R den is the daily denitrification rate (g N m 2 d −1 ), F (T soil ) is the function of daily soil temperature on nitrification process (unitless), and F (Q wfp ) is the function of water-filled porosity (unitless). Nitrification, a process converting NH + 4 into NO − 3 , is simulated as a function of soil temperature, moisture, and soil NH + 4 concentration: where k nit is the daily maximum fraction of NH + 4 that is converted into NO − 3 or gases (d −1 ), F (ψ) is the soil moisture effect (unitless), and C NH 4 is the soil NH + 4 content (g N m −2 ). Unlike in Chatskikh et al. (2005), who set k nit to 0.10 d −1 , k nit varies with different plant functional types (PFTs) in the DLEM with a range of 0.04 to 0.15 d −1 . The detailed calculations of F (T soil ) and F (ψ) have been described in  and Yang et al. (2015).
Denitrification is the process that converts NO − 3 into three types of gases, namely, nitric oxide, N 2 O, and dinitrogen. The denitrification rate is simulated as a function of soil temperature, water-filled porosity, and NO − 3 concentration C NO 3 (g N g −1 soil): where F N (C NO 3 ) is the dependency of the denitrification rate on NO − 3 concentration (unitless) and α is the maximum denitrification rate (g N m −2 d −1 ). The detailed calculations of F (Q wfp ), F N (C NO 3 ), and α have been described in .
In each grid cell, there are four natural vegetation types and one crop type. The sum of N 2 O emission in each grid d −1 is calculated by the following formula: where E is the daily sum of N 2 O emission from all PFTs in total grids (Tg N yr −1 d −1 ); N ij (g N m −2 ) is the N 2 O emission in the grid cell ifor PFTj ; f ij is the fraction of cell used for PFT j in grid cell i; and A i (km 2 ) is the area of the ith grid cell. The factor 10 6 converts square kilometers to square meters, and 10 12 converts grams to teragrams.

Input datasets
Input data to drive the DLEM simulation include static and transient data (Tian et al., 2010). Several additional datasets were generated to better represent the terrestrial environment in the preindustrial period as described below. The natural vegetation map was developed based on LUH (Land Use Harmonization; Hurtt et al., 2011) and a new joint 1 km global land cover product (SYNMAP) (Jung et al., 2006), which rendered the fractions of 47 vegetation types in each 0.5 • grid. These 47 vegetation types were converted to 15 PFTs used in the DLEM through a cross-walk table (Fig. 1). Cropland distribution in 1860 was developed by aggregating the 5 arcmin resolution HYDE (History Database of the Global Environment) v3.1 global cropland distribution data (Fig. 2). Half-degree daily climate data (including average, maximum, minimum air temperature, precipitation, relative humidity, and shortwave radiation) were derived from CRU-NCEP (Climate Research Unit -National Centers for Environmental Prediction) climate forcing data (Wei et al., 2014). As a global climate dataset was not available prior to the year 1900, long-term average climate datasets from 1901 to 1930 were used to represent the initial climate state in 1860. The nitrogen deposition dataset was developed based on the atmospheric chemistry transport model (Dentener, 2006)

Model simulation
The implementation of the DLEM simulation includes three steps: (1) equilibrium run, (2) spin-up run, and (3) transient run. In this study, we first used a land use and land cover (LULC) map for 1860, long-term mean climate during 1901-1930, N input datasets for 1860 (the concentration levels of N deposition and manure application rate), and atmospheric CO 2 in 1860 to run the model to an equilibrium state. In each grid, the equilibrium state was assumed to be reached when the interannual variations in carbon, nitrogen, and water storage were less than 0.1 g C m −2 , 0.1 g N m −2 and 0.1 mm, respectively, during two consecutive periods of 50 years. After the model reached equilibrium state, the model was spun up by the de-trended climate data from 1901 to 1930 to eliminate system fluctuation caused by the model mode shift from the equilibrium to transient run (i.e., three spins with 10year climate data each time). Finally, the model was run in the transient mode with daily climate data, annual CO 2 concentration, manure application, and N deposition inputs for 1860 to simulate preindustrial N 2 O emissions. An additional description of model initialization and the simulation procedure can be found in previous publications (Tian et al., 2010.

Model validation
Observations of annual N 2 O emission accumulations (g N m −2 yr −1 ) were selected to compare with the simulated emissions at different sites. As there were no field measurements in the preindustrial era, observations during 1970-2009 were collected to test the model performance in the contemporary period. All environmental factors (climate, CO 2 concentration, soil property, N deposition, LULC) in the exact year were used as input datasets for N 2 O simulations. The selected sites include temperate forest, tropical forest, boreal forest, savanna, and grassland globally. As shown in Fig. 3, the simulated N 2 O emissions have a good correlation with field observations (R 2 = 0.79, p < 0.001). This indicates that the DLEM has the capacity to simulate N 2 O emissions in the preindustrial era driven by environmental factors back then. Detailed information for each site can be found in Table S1 in the Supplement.

Estimate of uncertainty
In this study, uncertainties in the simulated N 2 O emission were evaluated through a global sensitivity and uncertainty analysis as described in Tian et al. (2011). Based on sensitivity analyses of key parameters that affect terrestrial N 2 O fluxes, the most sensitive parameters were identified to conduct uncertainty simulations with the DLEM. These parameters include potential denitrification and nitrification rates, BNF rates, and the adsorption coefficient for soil NH + 4 and NO − 3 (Gerber et al., 2010;Yang et al., 2015). The ranges of five parameters were obtained from previous studies. Chatskikh et al. (2005) set k nit to 0.10 d −1 ; however, it was set to a range of 0.04 to 0.15 d −1 and varied with different PFTs in the DLEM simulations. The uncertainty ranges of potential nitrification rates were based on previous studies (Hansen, 2002;Heinen, 2006); the global preindustrial N fixation was estimated to be 58 Tg N yr −1 , ranging from 50 to 100 Tg N yr −1 (Vitousek et al., 2013). The spatial distribution of BNF was based on the estimates by Cleveland et al. (1999). The potential denitrification rate was set to an uncertainty range of 0.025-0.74 d −1 and varied with different PFTs in the DLEM. The uncertainty ranges of the adsorption coefficient were based on the sensitivity analysis conducted in Yang et al. (2015). Parameters used in the DLEM simulations for uncertainty analysis were assumed to follow a normal distribution. The improved Latin hypercube sampling (LHS) approach was used to randomly select an ensemble of 100 sets of parameters (R version 3.2.1) .
In the DLEM, after the model reached equilibrium state, a spin-up run was implemented using de-trended climate data from 1901 to 1930 for each set of parameter values. Then, each set of the model was run in transient mode for 1860 to produce the result of the preindustrial N 2 O emissions. All results from 100 groups of simulations are shown in Table S2. The Shapiro-Wilk test was used on 100 sets of results to check the normality of DLEM simulations. It turned out that the distribution is not normal (P value < 0.05, R version 3.2.1), as shown in Fig. S1 in the Supplement. Thus, the uncertainty range was represented as the minimum and maximum value of 100 sets of DLEM simulations.

Magnitude and spatial distribution of N 2 O emission
The global mean preindustrial soil N 2 O emission was 6.20 Tg N yr −1 . We define the parameter-induced uncertainty of our global estimates as a range between the minimum (4.76 Tg N yr −1 ) and the maximum (8.13 Tg N yr −1 ) of 100  sets of DLEM simulations. The terrestrial ecosystem in the preindustrial period acted as a source of N 2 O, and its spatial pattern mostly depends on the biome distribution across the global land surface. The spatial distribution of annual N 2 O emission in a 0.5 • × 0.5 • grid (Fig. 4) shows that strong sources were found near the Equator, such as Southeast Asia, central Africa, and Central America, where N 2 O emission reached as high as 0.45 g N m −2 yr −1 . Weak N 2 O sources were observed in the northern areas of North America and Asia, where the estimated N 2 O emission was less than 0.001 g N m −2 yr −1 . Microbial activity in soils determined the rate of nitrification and denitrification processes, which accounts for approximately 70 % of global N 2 O emissions (Smith and Arah, 1990;Syakila and Kroeze, 2011). The tropical regions near the Equator could provide optimum temperatures and soil moistures for microbes to decompose soil organic matter and release more NO x and CO 2 into the atmosphere (Butterbach-Bahl et al., 2013). Referring to the observational data from field experiments and model simulations in the tropics, it has been argued that the tropics are the main sources within the total N 2 O emissions from natural vegetation (Bouwman et al., 1995;Werner et al., 2007;Zhuang et al., 2012).
In this study, Asia is divided into two parts: Southern Asia and Northern Asia, where the PFTs and climate conditions are significantly contrasting. As shown in Fig. 1, tropical forest and cropland were dominant PFTs in Southern Asia. In contrast, temperate and boreal forests were the main PFTs in Northern Asia. The estimates of N 2 O emissions from seven land regions are shown in Fig. 5. On continental scales, the N 2 O emission was 2.09 (1.63-2.73) Tg N yr −1 in South America, 1.46 (1.13-1.91) Tg N yr −1 in Africa, and 1.16 (0.90-1.52) Tg N yr −1 in Southern Asia. South Amer-ica, Africa, and Southern Asia accounted for 33.77, 23.60, and 18.73 %, respectively, together, which was 76.10 % of global total emission. Europe and Northern Asia contributed 0.45 (0.32-0.66) Tg N yr −1 , which was less than 10 % of the total emission.
Nitrous oxide emissions varied remarkably among different ecosystems. Forest, grassland, shrub, tundra, and cropland contributed 76.90, 3.11, 13.14, 0.18, and 6.67 %, respectively, to the total emission globally (Fig. 6). In different biomes, the tropics accounted for more than half of the total N 2 O emission, which is comparable to the conclusion drawn by Bouwman et al. (1993). In the preindustrial era, the major inputs of reactive N to terrestrial ecosystems were from BNF, which relies on the activity of a phylogenetically diverse list of bacteria, archaea, and symbioses (Cleveland et al., 1999;Vitousek et al., 2013). Tropical savannas have been considered "hot spots" of BNF by legume nodules that provide the major input of available N (Bate and Gunton, 1982). The substantial inputs of N into tropical forests could contribute to higher amount of gaseous N losses as N 2 O or nitrogen gas (Cleveland et al., 2010;Hall and Matson, 1999). In contrast, as the largest terrestrial biome, boreal forests lack available N because the rate of BNF is restricted by cold temperatures and low precipitation during the growing season (Alexander and Billington, 1986). Morse et al. (2015) conducted field experiments in northeastern North American forests. They found that denitrification does vary coherently with patterns of N availability in forests, and there are no significant correlations between atmospheric N deposition, potential net N mineralization, and nitrification rates. Thus, it is reasonable that boreal forests contributed the least amount of N 2 O emission among different forests. As shown in Fig. 2, cropland areas varied spatially. The regions with large areas of cropland were all of Europe, India, eastern China, and the central-eastern United States. The global N 2 O emission from croplands was estimated as being 0.41 (0.32-0.55) Tg N yr −1 , which is about 10 times less than the estimate reported in the IPCC AR5 (Ciais et al., 2014). As no synthetic N fertilizer was applied to the cropland in 1860, leguminous crops were the major source of N 2 O emission from croplands, most of which were planted in the centraleastern United States (Fig. 4). Rochette et al. (2004) conducted the experiments on the N 2 O emission from soybean without the application of N fertilizer. Their work was in agreement with the suggestion that legumes may increase N 2 O emissions compared with non-BNF crops (Duxbury et al., 1982) The background emission from ground-based experiments was as high as 0.31-0.42 kg N ha −1 in Canada (Duxbury et al., 1982;Rochette et al., 2004).
Preindustrial N 2 O emission at country level could serve as a reference for calculating human-induced N 2 O emission in today's nations. We estimated preindustrial N 2 O emissions from 17 countries that are hot spots of N 2 O sources in the contemporary period (  era. Previous studies indicated that agriculture produces the majority of anthropogenic N 2 O emissions (Ciais et al., 2014;Davidson and Kanter, 2014). Our estimate on country scales could be used as a reference to quantify the net increase in N 2 O emissions from agriculture activities in countries of hot spots.
There is a debate about whether the natural wetlands and peatlands act as sinks or sources of N 2 O. Previous studies showed that N 2 O emissions from natural peatlands are usually negligible; however, the drained peatlands with lower water tables might act as sources of N 2 O (Augustin et al., 1998;Martikainen et al., 1993). High water tables in wetlands might block the activity of nitrifiers and limit the denitrification (Bouwman et al., 1993). The fluxes of N 2 O were negligible in the pelagic regions of boreal ponds and lakes due to the limitation of nitrification and/or nitrate inputs (Huttunen et al., 2003). Couwenberg et al. (2011) mentioned that N 2 O emissions always decreased after rewetting when conducting field experiments, which had been excluded from their future analysis of GHG emissions in peatlands. Hadi et al. (2005) pointed out that tropical peatlands ranged from sources to sinks of N 2 O, highly affected by land use and hy- drological zone. We were unable to examine N 2 O fluxes from wetlands and peatlands in 1860 as human-induced land use in those ecosystems was unknown. Thus, we excluded the N 2 O emissions from wetlands and peatlands in this study.

Revisiting preindustrial global N 2 O emission by incorporating top-down estimates
The "top-down" methodology used to estimate N 2 O emissions is based on atmospheric measurements and inversion modeling (Thompson et al., 2014). Prather et al. (2012) provided an estimate of 9.1 ± 1.0 Tg N yr −1 of natural emissions in the preindustrial era using observed preindustrial abundances of 270 ppb and model estimates of lifetime decreases from 142 years in the preindustrial era to 131 ± 10 years in the present day. Later, Prather et al. (2015) reevaluated N 2 O lifetime based on Microwave Limb Sounder satellite measurements of the stratosphere, which were consistent with modeled values in the present day. The lifetime in the preindustrial era and the present day was estimated to be 123 and 116 ± 9 years, respectively. The current lifetime increases the preindustrial natural emission from 9.1 ± 1.0 to 10.5 Tg N yr −1 . Natural sources for N 2 O include soil under natural vegetation, oceans, and atmospheric chemistry (Ciais et al., 2014). The emission from atmospheric chemistry was estimated as being 0.6 with an uncertainty range of 0.3-1.2 Tg N yr −1 . Syakila and Kroeze (2011) estimated global natural emissions from oceans as 3.5 Tg N yr −1 . Oceanic emission was estimated as being 3.8 with an uncertainty range of 1. 8-5.8 Tg N yr −1 in the IPCC AR4. However, the uncertainty range became larger (1. 8-9.4 Tg N yr −1 ) in the IPCC AR5. In our study, the simulated N 2 O emission came from agri-cultural and natural soils. The natural emission was estimated as being 5.78 (4.4-7.72) Tg N yr −1 . Combining the atmospheric chemistry and the ocean emissions in the IPCC AR5 with the natural emissions from our study, the global total natural N 2 O emissions were 10.18 (6.5-18.32) Tg N yr −1 . The large uncertainty range was attributed to the uncertainty from oceanic emission, atmospheric chemistry emission, and our estimation. The estimated global total amount (10.18 Tg N yr −1 ) in this study was comparable to the estimate (10.5 Tg N yr −1 ) by Prather et al. (2015) using the topdown approach.

Comparison with estimates using the bottom-up methodology
The "bottom-up" approach includes the estimations based on inventory, statistical extrapolation of local flux measurements, and process-based modeling (Tian et al., 2016). The global preagricultural N 2 O emission was estimated as being 6.8 Tg N yr −1 based on the regression relationship between measured N 2 O fluxes and modeled N 2 O production indices (Bouwman et al., 1993). This estimate was adopted to retrieve the trends of atmospheric N 2 O concentration in Syakila and Kroeze (2011). In our study, the preindustrial N 2 O emission from natural vegetation was estimated as being 5.78 (4.4-7.72) Tg N yr −1 , which is about 1 Tg N yr −1 lower than the estimate from Bouwman et al. (1993). An estimate from the tropics (±30 • of the Equator) was about 4.57 Tg N yr −1 , which is 0.83 Tg N yr −1 lower than the estimate from Bouwman et al. (1993). For the rest of natural vegetation, our estimate was 1.21 Tg N yr −1 , which is close to 1.4 Tg N yr −1 estimated in Bouwman et al. (1993).
Although Bouwman et al. (1993) has studied the potential N 2 O emission from natural soils, our study provided a first estimate of spatially distributed N 2 O emission in 1860 using the biogeochemical process-based model. Bouwman et al. (1993) provided 1 • × 1 • monthly N 2 O emission using the monthly controlling factors without considering the impact of N deposition. In their study, the soil fertility and carbon content were constant for every month, which cannot reflect the monthly dynamic changes of carbon and N pools in natural soils. Moreover, although their study represented a spatial distribution of potential N 2 O emission from natural soils, they did not provide that estimate on biome, continent, and country scales. Thus, their result was hardly to be used as a regional reference for the net human-induced N 2 O emissions from some hot spots, such as Southern Asia. In contrast, in our study, using a daily climate and N deposition dataset better reflects the real variation in N 2 O emission through the growing season in natural ecosystems. The comparison with field observations during 1997-2001 indicated that the DLEM can catch the daily peak N 2 O emissions in Hubbard Brook Forest (Tian et al., 2010) and Inner Mongolia .
Regarding the N 2 O emission from croplands, our estimate is comparable to the estimate of 0.3 (0.29-0.35) Tg N yr −1 extracted from Syakila and Kroeze (2011) by digitizing graphs using the Getdata Graph Digitizer. In their study, the estimation was based on the relationship between the crop production and human population during 1500-1970. In contrast, the result in our study was estimated based on the cropland area of a specific crop type (mainly soybean, rice, corn, and wheat) in 1860.
Thus, the DLEM is capable of providing an estimate of N 2 O emission from natural ecosystems on regional and biome scales with a higher spatial resolution. This could be a useful reference for quantifying the effects of human activities -such as LULC change, N fertilizer and manure application, and, increasingly, atmospheric N deposition -on N 2 O emissions in different terrestrial ecosystems or sectors in the contemporary period.

The N 2 O budget in the preindustrial era
The observed N 2 O concentration is the result of dynamic production and consumption processes in soils as soils act as sources or sinks of N 2 O through denitrification and nitrification (Chapuis-Lardy et al., 2007). There was a slight increase in atmospheric N 2 O concentration during 1750-1860 according to the ice core records, but these showed a rapid increase from 1860 to present (Ciais et al., 2014). Natural sources of N 2 O emissions have been discussed in Sect. 3.2 and 3.3. Previous studies found that there were some anthropogenic N 2 O emissions along with the natural sources in the preindustrial era (Davidson, 2009;Syakila and Kroeze, 2011). Syakila and Kroeze (2011) found that anthropogenic N 2 O emission began in 1500 because of biomass burning and agriculture. The total anthropogenic N 2 O emission in their study was estimated as being 1.1 Tg N in 1850. In addition, Davidson (2009) derived a time-course analysis of sources and sinks of atmospheric N 2 O since 1860. The preindustrial anthropogenic N 2 O sources in his study included biomass burning, agriculture (e.g., manure and fertilizer application and the cultivation of legumes) and human sewage, the sum of which was 0.7 (0.6-0.8) Tg N yr −1 (Davidson, 2009). Thus, anthropogenic N 2 O emission already existed in 1860, but it was smaller than the contemporary amount. Davidson (2009) mentioned that there was possibly a certain amount of N 2 O loss in the preindustrial period through atmospheric sink and the reduced emission from tropical deforestation. He estimated the anthropogenic sink as being 0.26 Tg N in 1860. In addition, the deforestation of tropical forest may have caused a loss of N 2 O emissions in 1860, which was estimated as being 0.03 Tg N (Davidson, 2009). However, studies have shown that the conversion of forest to pasture and cropland could increase or have no effect on N 2 O emissions because the effects depended on the disturbance intensity of human activities on soil conditions (van Lent et al., 2015). For instance, N 2 O emissions tended to increase during the first 5-10 years after conversion and thereafter might decrease to average upland forest or low canopy forest levels in the non-fertilized croplands and pastures. In contrast, emissions were at a high level during and after fertilization in fertilized croplands (van Lent et al., 2015). Thus, more work is needed to study how forest degradation affects N 2 O fluxes (Mertz et al., 2012).

Future research needs
Large uncertainty still exists in the DLEM simulation associated with the quality of input datasets and parameters applied in simulations. Although input datasets could play a significant role in the variety of the model output, it is difficult to obtain accurate datasets going back to the year 1860. Average climate data from 1901 to 1930 were used to run the model simulation, which could raise the uncertainty in estimating N 2 O emission in 1860. The datasets of LULC, N deposition, and manure application in 1860 could introduce uncertainties into this estimate. The average oceanic and atmospheric chemistry emissions cited from the IPCC AR5 could introduce uncertainty into the calculation of the total natural emissions in 1860 when compared with the estimate done by Prather et al. (2015). Thus, a more accurate estimate of oceanic N 2 O emissions is significant for narrowing the confidence estimate of the preindustrial terrestrial sources. The N 2 O fluxes from wetlands and peat need to be included in any future study.

Conclusions
Using the process-based land ecosystem model DLEM, this study provides a spatially explicit estimate of preindustrial N 2 O emissions for major PFTs across global land surface. Improved LHS was performed to analyze uncertainty ranges of the estimates. We estimated that preindustrial N 2 O emission is 6.20 Tg N yr −1 . The modeled results showed a large spatial variability due to variations in climate conditions and PFTs. The tropical ecosystem was the dominant contributor of global N 2 O emissions. In contrast, boreal regions contributed less than 5 % to the total emission. China, India, and the United States are the top countries accounting for emissions from croplands in 1860. While uncertainties still exist in the N 2 O emission estimation for the preindustrial era, this study offered a relatively reasonable estimate of the preindustrial N 2 O emission from land soils. Moreover, this study provided a spatial estimate for N 2 O emission from global hot spots, which could be used as a reference to estimate net human-induced emissions in the contemporary period.
Data availability. All input data and output files used in this paper are available from the corresponding author upon request (tian-han@auburn.edu).
Author contributions. HT, SP, and RX initiated this research and designed model simulations. RX performed DLEM simulations, analyses, and calculations. CL contributed to the model calibration and data analysis. JC contributed to the data processing and statistical analysis. JY took charge of input datasets preparation (environmental factors), data description, and model verification. BZ provided manure N input data. All coauthors contributed to the development of the paper.