Estimate of changes in agricultural terrestrial nitrogen pathways and ammonia emissions from 1850 to present in the Community Earth System Model

Nitrogen applied to the surface of the land for agricultural purposes represents a significant source of reactive nitrogen (Nr) that can be emitted as a gaseous Nr species, be denitrified to atmospheric nitrogen (N2), run off during rain events or form plant-useable nitrogen in the soil. To investigate the magnitude, temporal variability and spatial heterogeneity of nitrogen pathways on a global scale from sources of animal manure and synthetic fertilizer, we developed a mechanistic parameterization of these pathways within a global terrestrial land model, the Community Land Model (CLM). In this first model version the parameterization emphasizes an explicit climate-dependent approach while using highly simplified representations of agricultural practices, including manure management and fertilizer application. The climate-dependent approach explicitly simulates the relationship between meteorological variables and biogeochemical processes to calculate the volatilization of ammonia (NH3), nitrification and runoff of Nr following manure or synthetic fertilizer application. For the year 2000, approximately 125 Tg N yr−1 is applied as manure and 62 Tg N yr−1 is applied as synthetic fertilizer. We estimate the resulting global NH3 emissions are 21 Tg N yr−1 from manure (17 % of manure production) and 12 Tg N yr−1 from fertilizer (19 % of fertilizer application); reactive nitrogen runoff during rain events is calculated as 11 Tg N yr−1 from manure and 5 Tg N yr−1 from fertilizer. The remaining nitrogen from manure (93 Tg N yr−1) and synthetic fertilizer (45 Tg N yr−1) is captured by the canopy or transferred to the soil nitrogen pools. The parameterization was implemented in the CLM from 1850 to 2000 using a transient simulation which predicted that, even though absolute values of all nitrogen pathways are increasing with increased manure and synthetic fertilizer application, partitioning of nitrogen to NH3 emissions from manure is increasing on a percentage basis, from 14 % of nitrogen applied in 1850 (3 Tg NH3 yr−1) to 17 % of nitrogen applied in 2000 (21 Tg NH3 yr−1). Under current manure and synthetic fertilizer application rates we find a global sensitivity of an additional 1 Tg NH3 (approximately 3 % of manure and fertilizer) emitted per year per C of warming. While the model confirms earlier estimates of nitrogen fluxes made in a range of studies, its key purpose is to provide a theoretical framework that can be employed within a biogeochemical model, that can explicitly respond to climate and that can evolve and improve with further observation. Published by Copernicus Publications on behalf of the European Geosciences Union. 3398 S. Riddick et al.: Estimate of changes in agricultural terrestrial nitrogen pathways

Abstract.Nitrogen applied to the surface of the land for agricultural purposes represents a significant source of reactive nitrogen (N r ) that can be emitted as a gaseous N r species, be denitrified to atmospheric nitrogen (N 2 ), run off during rain events or form plant-useable nitrogen in the soil.To investigate the magnitude, temporal variability and spatial heterogeneity of nitrogen pathways on a global scale from sources of animal manure and synthetic fertilizer, we developed a mechanistic parameterization of these pathways within a global terrestrial land model, the Community Land Model (CLM).In this first model version the parameterization emphasizes an explicit climate-dependent approach while using highly simplified representations of agricultural practices, including manure management and fertilizer application.The climate-dependent approach explicitly simulates the relationship between meteorological variables and biogeochemical processes to calculate the volatilization of ammonia (NH 3 ), nitrification and runoff of N r following manure or synthetic fertilizer application.For the year 2000, approximately 125 Tg N yr −1 is applied as manure and 62 Tg N yr −1 is applied as synthetic fertilizer.We estimate the resulting global NH 3 emissions are 21 Tg N yr −1 from manure (17 % of manure production) and 12 Tg N yr −1 from fertilizer (19 % of fertilizer application); reactive nitrogen runoff during rain events is calculated as 11 Tg N yr −1 from manure and 5 Tg N yr −1 from fertilizer.The remaining nitrogen from manure (93 Tg N yr −1 ) and synthetic fertilizer (45 Tg N yr −1 ) is captured by the canopy or transferred to the soil nitrogen pools.The parameterization was implemented in the CLM from 1850 to 2000 using a transient simulation which predicted that, even though absolute values of all nitrogen pathways are increasing with increased manure and synthetic fertilizer application, partitioning of nitrogen to NH 3 emissions from manure is increasing on a percentage basis, from 14 % of nitrogen applied in 1850 (3 Tg NH 3 yr −1 ) to 17 % of nitrogen applied in 2000 (21 Tg NH 3 yr −1 ).Under current manure and synthetic fertilizer application rates we find a global sensitivity of an additional 1 Tg NH 3 (approximately 3 % of manure and fertilizer) emitted per year per • C of warming.While the model confirms earlier estimates of nitrogen fluxes made in a range of studies, its key purpose is to provide a theoretical framework that can be employed within a biogeochemical model, that can explicitly respond to climate and that can evolve and improve with further observation.

Introduction
Nitrogen is needed by all living things for growth.However, it is relatively inert in its most abundant form, diatomic nitrogen (N 2 ), and needs to be converted to a form of reactive nitrogen (N r ) before it can be used by most plants for growth (Visek, 1984).Supplying sufficient N r for maximum crop yield is a major concern in agriculture.In preindustrial times N r demand was partly solved with the use of animal manure and seabird guano as well as crop rotation and the use of nitrogen-fixing crops (Smil, 2000).However, by the early 20th century the supply of these N r sources could not match the demands of an increasing population and the Haber-Bosch process of creating synthetic N r was developed (Galloway et al., 2004).
The use of N r to improve crop yield has become an environmental concern as N r in synthetic fertilizer and manure cascades through the soil, water and the atmospheric nitrogen cycles.Plants can readily use applied N r for plant growth; however, N r washed off fields or volatilized as gas can reduce ecosystem biodiversity through acidification and eutrophication (Sutton et al., 2013).Increased N r in the hydrosphere can lead to the subsequent degradation of riverine and nearshore water quality as the water becomes more acidic and the growth of primary producers blooms (Turner and Rabalais, 1991;Howarth et al., 2002), which can alter the local interspecies competition and biodiversity (Sutton et al., 2012).Reactive nitrogen emissions into the atmosphere impact air quality through the ozone generation associated with the emissions of nitrogen oxide (e.g., Hudman et al., 2010) and the contribution of ammonia (NH 3 ) to aerosol formation (e.g., Gu et al., 2014).Nitrogen cycling also impacts climate through the stimulation of plant growth and the associated increase in carbon storage; through the associated emissions of N 2 O, a strong greenhouse gas; through emissions of nitrogen oxides and the associated ozone production; and through the emissions of NH 3 , with its potential to cool the climate through aerosol formation (e.g., Adams et al., 2001).
As a result of their dependence on environmental conditions, N r pathways following manure or synthetic fertilizer application are likely to change as the climate changes.This study describes a biogeochemically consistent processdriven parameterization suitable for use in Earth system models for simulating N r flows following the addition of N r as manure or synthetic fertilizer.The parameterization is evaluated against local measurements of NH 3 fluxes and against global NH 3 flux estimates.This parameterization allows simulations of the coupling between the nitrogen cycle and climate to explicitly include NH 3 emissions and the N r runoff from manure and synthetic fertilizer application.To our knowledge, no Earth system model has yet to explicitly predict changing nitrogen pathways from manure and synthetic fertilizer in response to climate.We note at the outset that the representation of agricultural processes is highly simplified in the initial model version described here.Sources of N r largely fall into two categories: "new" sources, created by chemical and biological processes, and those that are "recycled", such as manure excretion by animals.The largest natural new N r creation is by biological nitrogen fixers found in the ocean, biological nitrogen fixers on land, and as the by-product of lightning estimated at 140 Tg N yr −1 ± 50 %, 58 Tg N yr −1 ± 50 % and 5 Tg N yr −1 ± 50 %, respectively (Fowler et al., 2013).The dominant anthropogenic sources of new N r are Haber-Bosch-derived fertilizer (estimated at 120 Tg N yr −1 ± 10 % in 2005), the burning of fossil fuels (30 Tg N yr −1 ± 10 % in 2000), and a further 60 Tg N yr −1 ± 30 % (ca.2005) from biological nitrogen fixers grown for human consumption, such as legumes (Fowler et al., 2013).Since preindustrial times, anthropogenic N r creation has increased from 15 Tg N yr −1 to the present estimate of 210 Tg N yr −1 (Galloway et al., 2004;Fowler et al., 2013).
Animal manure is used to stimulate plant growth in agriculture.It contains N r recycled from the soil produced when animals eat plants.A comprehensive increase in livestock population is estimated to have increased global manure production from 21 Tg N yr −1 in 1850 to the present estimate of 141 Tg N yr −1 (Holland et al., 2005).It is suggested that this increase in recycled N r production speeds up the decay and processing of plant biomass, releasing different N r products to the atmosphere when compared to natural decay processes (Davidson, 2009).
Projections of agricultural activity (Bodirsky et al., 2012) suggest continued increases in the application of synthetic fertilizers until the mid-21st century (and possibly beyond) concurrent with increases in manure production (Tilman et al., 2001).In addition to the increased use of organic and synthetic fertilizers in the future, NH 3 emissions are expected to increase because of the impact of changing climate on nitrogen biogeochemistry (Tilman et al., 2001;Skjøth and Geels, 2013;Sutton et al., 2013).Skjøth and Geels (2013) predict increases in future NH 3 emissions of up to 60 % over Europe by 2100, largely due to increased NH 3 emissions with temperature.Sutton et al. (2013) predict future temperature increases may enhance global NH 3 emissions by up to approximately 40 % assuming a 5 • C warming.In addition to future changes in climate-induced NH 3 volatilization from manure and synthetic fertilizer application, future changes in agro-management practices, soil microbiological processes and nitrogen runoff may also be expected.
Current estimates of the direct forcing of nitrate aerosols present as ammonium nitrate encompass the range from −0.03 to −0.41Wm −2 in the ACCMIP (Atmospheric Chemistry and Climate Model Intercomparison Project) (Shindell et al., 2013) and AeroCom Phase II (Myhre et al., 2013) simulations.With a future reduction in sulfate emissions the relative importance of nitrate aerosols is expected to dominate the direct aerosol forcing by 2100, with a resulting increase in radiative forcing of up to a factor of 8.6 over what it would have been without accounting for nitrate aerosols (Hauglus-taine et al., 2014).These estimates do not consider the impact of climate change on NH 3 emissions.
Modeling studies calculating NH 3 emission from manure and synthetic fertilizer have broadly fallen into two categories: models that use empirically derived agriculturally based emission factors and more complex processbased models.Global emissions have almost been universally estimated using the former approach.Emission factors were used by Bouwman et al. (1997) to estimate global NH 3 emissions in 1990 of 54 Tg N yr −1 , with an emission of 21.6 Tg N yr −1 from domestic animals (Bouwman et al., 1997).Beusen et al. (2008) also used emission factors to estimate global NH 3 emission from agricultural livestock (21 Tg N yr −1 ) and synthetic fertilizers (11 Tg N yr −1 ) in 2000; Bouwman et al. (2013) estimated emissions of 34 Tg NH 3 yr −1 on agricultural land, with 10 Tg NH 3 yr −1 from animal housing.A number of more recent global models have included emission factors explicitly as a function of temperature (e.g., Huang et al., 2012;Paulot et al., 2014).Paulot et al. (2014) estimate current global NH 3 emissions of 9.4 Tg yr −1 for synthetic fertilizer and 24 Tg yr −1 for manure.
Alternatively, process-based or mechanistic models have been developed that estimate N r flows, equilibria and transformations between different nitrogen species as well as N r emissions from synthetic fertilizer and manure.Process models have been used on the field to regional scale, but not on the global scale.These models generally do not simulate the runoff of N r .For example, Génermont and Cellier (1997) simulate the emissions of NH 3 (g) to the atmosphere after considering the physical and chemical equilibria and transfer of Nr species (NH 3 (g), NH 3 (aq), and NH + 4 (aq)) in the soil.The resulting model is used to calculate the NH 3 emissions from synthetic fertilizer over France within the air quality model, Chimere (Hamaoui-Laguel et al., 2014).Other examples include Pinder et al. (2004), who describe a process model of NH 3 emissions from a dairy farm, while Li et al. (2012) describe a farm-scale process model of the decomposition and emission of NH 3 from manure.
The overall goal of this paper is to describe and analyze a global model capable of simulating nitrogen pathways from manure and synthetic fertilizer added to the surface of the land under changing climatic conditions.The model will allow for a better global quantification of the climate, health and environmental impacts of a changing nitrogen cycle under climate change.The resulting model is designed out of necessity for use within an Earth system model so as to simulate the interactions between the climate and the carbon and nitrogen cycles.Section 2 presents the overall methodology used here including a detailed description of the process model used to calculate climate-dependent nitrogen pathways.Section 3 analyzes the model and includes a comparison of simulated vs. site level measurements of NH 3 fluxes, an analysis of the globally heterogeneous nitrogen pathways from applied manure and synthetic fertilizer over a range of climatic regimes, model predictions for changes in nitrogen pathways from 1850 to present and the sensitivity of the results to model parameters.Section 4 gives our conclusions.

Methods
In this section we describe a process model for the flow of agricultural nitrogen (FAN) that simulates NH 3 emissions and other N r flows from applied manure and synthetic fertilizer applications, including their spatial and temporal variations, within an Earth system model, the Community Earth System Model 1.1 (CESM1.1).The FAN process model developed here simulates the incorporation of manure and fertilizer N r into soil organic matter and soil nitrogen pools (Chambers et al., 1999), its volatilization as NH 3 to the atmosphere and the direct runoff of N r from the surface (Fig. 1).The model is global in nature, is designed to conserve carbon and nitrogen, and responds to changes in climate.It is designed to provide an interface between the application of manure and synthetic fertilizer and the nitrogen cycling developed within the Community Land Model 4.5 (CLM4.5), the land component of the CESM.
Nitrogen pathways subsequent to the application of manure or synthetic fertilizer depend on the complex interaction between both human and natural processes.In particular they depend on the biology and physics of the applied substrate, agricultural practices and climate.Bottom-up emission inventories with specified emission factors that take into account the animal feed, the type of animal housing, if any, and the field application of the synthetic fertilizer or manure (e.g., Bouwman et al., 1997) are generally used in global chemistry and chemistry-climate applications.For example, this type of emission inventory (e.g., Lamarque et al., 2010) was used in the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) (Lamarque et al., 2013a) for assessing historical and future chemistry-climate scenarios and in assessing nitrogen deposition (Lamarque et al., 2013b) with implications for the carbon cycle.However, these inventories include very simplified representations of the effect of climate on emissions, for example, by grouping countries into industrial or developing categories (Bouwman et al., 1997).A seasonal emission dependence is not implicit in these bottom-up inventories although sometimes an empirical relationship is applied (e.g., Adams et al., 2001; also see Skjøth et al., 2011).
In the first application of the model described here we take the opposite tact.We have minimized the description of agricultural practices, and instead emphasize the physically based climate-dependent biogeochemistry of manure and synthetic fertilizer decomposition and the resultant nitrogen pathways.While traditional bottom-up NH 3 emission inventories reflect regional differences in agro-management practices and the resulting regional differences in NH 3 emissions, they do not account for physically based geographical ) is directly input to the TAN pool, while fecal matter is split into three parts that decompose into the TAN pool at a rate determined by their C : N ratio (N available , N resistant , N unavailable ).Manure nitrogen that does not mineralize (N unavailable ) is added to the soil organic nitrogen pool.Nitrogen applied as synthetic fertilizer is added to the N fertilizer pool where it decomposes into the TAN pool.Losses from the TAN pool include ammonia (NH 3 ) emission (into CAM-chem), nitrogen runoff (into the RTM), nitrate (NO − 3 ) formation and diffusion to the soil nitrogen pools.and meteorological influences, including temperature, atmospheric mixing and rainfall.However, these influences are accounted for in the parameterization described below.As with regional differences in agro-management practices, meteorological impacts may also induce large regional and interannual variations in NH 3 emissions.For example, increasing the ground temperature from 290 to 300 K (at a pH of 7) increases the NH 3 emissions by a factor of 3 (see Sect. 2.2.6).
We recognize in this first application that we are simplifying many important agro-management processes.First, we assume that all synthetic fertilizer is urea and the pH of the soil is given.Different applied synthetic fertilizers have a strong impact on the pH of the soil-fertilizer mixture with the overall emission factor very dependent on the pH (Whitehead and Raistrick, 1990).Urea is the most commonly used synthetic fertilizer accounting for over 50 % of the global nitrogenous synthetic fertilizer usage (Gilbert et al., 2006) and has one of the highest emission factors for commonly used synthetic fertilizers (Bouwman et al., 1997).Emission factors for other types of fertilizers can be significantly smaller.Thus the estimates here for fertilizer NH 3 emissions may be considered as an upper estimate.Second, we do not account for manure management practices.Instead we assume all manure is continuously spread onto fields.In a global study Beusen et al. (2008) considered four primary pathways for manure nitrogen: (i) manure nitrogen lost from the system (14 % of the manure nitrogen, range 5-26 %), (ii) manure nitrogen excreted in animal houses followed by storage and subsequent spreading onto croplands (35 % of manure nitrogen; range 24-51 %), (iii) manure nitrogen excreted in animal houses followed by storage and subsequent spreading onto pasture lands (7% of manure nitrogen; range 3-11 %), and (iv) manure nitrogen excreted by grazing animals onto pastures (44 % of manure nitrogen; range 29-59 %).Of the 42 % of manure nitrogen excreted in housing, 20 % (range: 12-28%) is emitted as NH 3 from housing and storage facilities (Beusen et al., 2008).An additional 15-23 % of the remaining manure nitrogen is emitted as NH 3 (range: 11-30 %) after it is spread onto crop or pasture land.Of the 44 % of manure nitrogen excreted by grading animals on pasture land 11-12 % (range 6-17 %) is emitted as ammonia.Considering these various pathways the overall emission factor for manure nitrogen is estimated as 19 % in Beusen et al. (2008) (compare with 17 % in this study).Third, we do not account for specific fertilizer application techniques.For example, the soil incorporation of manure leads to a 50 % reduction in NH 3 emissions compared to soil broadcasting (Bouwman et al., 2002).We recognize that there are large spreads in all these ranges and that regional practices may alter these numbers, although large errors may be unavoidable due to insufficient characterization of regional agro-management practices.
In the present application we also simplify the representation of NH 3 fluxes to the atmosphere.The aerodynamic resistances used to compute the flux of NH 3 to the atmosphere are calculated with CLM4.5, but due to the configuration of the CLM they are not calculated at the plant function type (PFT) level.Instead, the canopy capture of the NH 3 flux is calculated as a global number.The high spatial heterogeneity of NH 3 emissions may preclude an accurate local representation of exchange processes on the approximately 2 × 2 • grid cell used here, although even on similar coarse resolutions Zhu et al. (2015) show the implementation of a bidirectional scheme has significant global and pronounced regional impacts.Nevertheless, the simulation of dynamic NH 3 emissions, as described below, with NH 3 emissions responding to temperature on the model time step is a first step towards more accurately coupling terrestrial NH 3 fluxes with the atmosphere.
A number of requirements are necessary to model NH 3 emissions following synthetic fertilizer or manure application within an Earth system model, specifications that are not necessary in more traditional formulations.(1) The model must be global in nature to characterize global interactions between applied N r and climate.(2) The model must conserve nitrogen.In particular the nitrogen associated with ma-nure does not add new nitrogen to the system, but merely represents a recycling of available nitrogen.Artificial sources or sinks of nitrogen may have serious repercussions, especially when simulating the global nitrogen cycle on the timescale of centuries.
(3) The model must be able to simulate the changing impact of climate on the fate of manure and synthetic fertilizer N r .In particular, NH 3 emissions are sensitive to both temperature and the water content of the soil.In addition, the runoff of N r is likely to change under climate change.The process model developed here is capable of simulating the physics of changing nitrogen pathways under a changing climate.
An ideal model would incorporate a globally explicit representation of agro-management practices, including manure treatment (housing, storage and spreading) and fertilizer application (e.g., see Sutton et al., 2013).It would also include an explicit representation of the bidirectional exchange of NH 3 between the land and atmosphere, including the incorporation of PFT dependent canopy deposition and aerodynamic resistances.While the model developed here captures many of the regional and global features seen in models based on emission factors, here we emphasize the importance of regional and climatic changes in meteorology.

Relation between the FAN process model and the CESM1.1
The parameterization developed here acts as the interface between specified manure and synthetic fertilizer application and the CESM1.1.The CESM1.1 simulates atmospheric, ocean, land and sea ice processes, linked together using a coupler, and includes a land and ocean carbon cycle (Hurrell et al., 2013;Lindsay et al., 2014).The CESM participates in the Climate Model Intercomparison Project (CMIP5) and has been extensively evaluated in the literature (see Hurrell et al., 2013).The land model within the CESM1.1,CLM 4.5, includes representation of surface energy and water fluxes, hydrology, phenology, and the carbon cycle (Lawrence et al., 2007;Oleson et al., 2008).CLM4.5 retains the basic properties of the previous model version (CLM4), which has been extensively tested and evaluated by many studies at the global (Lawrence et al., 2007;Oleson et al., 2008;Randerson et al., 2009) and the site (Stoeckli et al., 2008;Randerson et al., 2009) scale.CLM4.5 includes improvements to better simulate: (1) water and momentum fluxes at the Earth's surface; (2) carbon and nitrogen dynamics within soils and (3) precipitation runoff rates (Koven et al., 2013).CLM4.5 simulations can be forced by meteorology (as done here), or as a part of a coupled carbon-climate model (Lawrence et al., 2007;Oleson et al., 2008).The current version of the carbon model is an improved version of the coupled carbonclimate model used in Keppel-Aleks et al. (2013), Lindsay et al. (2014) and Thornton et al. (2009).The carbon model includes a nitrogen limitation on land carbon uptake, described in Thornton et al. (2007Thornton et al. ( , 2009)).Further improvements have been made to the below ground carbon cycle, as well as other elements of the land model in order to improve its performance (e.g., Koven et al., 2013;Lawrence et al., 2012).The impact of increases in nitrogen deposition (from fossil fuels, fires and agriculture; Lamarque et al., 2010) have been evaluated (Thornton et al., 2007(Thornton et al., , 2009) ) and extensively compared to observations (e.g., Thomas et al., 2013).
As described in Koven et al. (2013), CLM4.5 simulates the basic flows of N r within soils following the Century N model (Parton et al., 1996(Parton et al., , 2001;;Del Grosso et al., 2000), including the processes of nitrification, denitrification, and emissions of N r and N 2 and the loss of N r from leaching and runoff.CLM4.5 also simulates the transfer of N r between soils and vegetation, and the loss of N r from fire.Sources of N r within CLM4.5 are from biological nitrogen fixation and from surface deposition.The process model developed here adds an additional source of N r to CLM4.5, the addition of synthetic fertilizer.It also adds an additional pathway whereby N r is recycled: the creation and application of manure (Fig. 1).
The relation between nitrogen cycling within the FAN process model developed here and that within the atmospheric, land and river components of the Community Earth System Model (CESM1.1) is given in Fig. 1.In this first study the subsequent fate of N r from synthetic fertilizer or manure application after it is incorporated into the soil organic matter or the soil nitrogen pools of CLM4.5 is not considered (see Fig. 1).As described in more detail below, synthetic fertilizer and manure are not applied to particular PFTs (e.g., pasture or grassland) within CLM4.5.This is because soil related properties including soil nitrogen are not specified at the PFT level within CLM4.5, but instead are specified at the column level that includes many PFTs.In practice we expect that the impact of this contamination across PFTs will be small since the major N-application regions (central US, northern India, eastern China) are not PFT-diverse but contain almost exclusively crop and grass PFTs.
Note that as a first approximation the model described here does not simulate the direct emission loss of species other than NH 3 .Atmospheric emission losses of N 2 O or N 2 (and potentially NO x ) are simulated in CLM4.5 (Koven et al., 2013), the land component model of the CESM1.1,"downstream" from the pathways explicitly considered here.In addition, the fate of N r emitted into the atmosphere as NH 3 directly from synthetic fertilizer or manure is handled by the atmospheric chemistry component of the CESM (CAM-chem) and is not considered here (Fig. 1).The runoff of N r from manure or synthetic fertilizer nitrogen pools has been coupled to the river transport model (RTM) in Nevison et al. (2016) (Fig. 1) but is also not considered here.

FAN process model
A schematic of the overall model is given in Fig. 1.All the equations and variables used in the model are presented in the Appendix.The assumptions used in constructing this model www.biogeosciences.net/13/3397/2016/Biogeosciences, 13, 3397-3426, 2016 are detailed below where appropriate.Sensitivity to model parameters is given in Sect.3.4.The N r pathways are calculated separately for manure and synthetic fertilizer.While this model assumes that synthetic fertilizer application and manure application can take place in the same approximately 2 × 2 • grid cell, we assume that manure and synthetic fertilizer are not applied in the exactly the same place.Therefore the NH 3 emissions, the nitrogen incorporation into soil pools, and the nitrogen runoff are separately calculated for manure and synthetic fertilizer in each column.This means that the total ammoniacal nitrogen (TAN) pools (consisting of NH 3 (g), NH 3 (aq), and NH + 4 ) for manure and synthetic fertilizer are discrete and hence the nitrogen pathways are not combined.
The application rate and geographical distribution used for manure and synthetic fertilizer application is taken from the synthetic fertilizer application and manure production data sets developed in Potter et al. (2010).These data sets are valid for ca.2000 for synthetic fertilizer and 2007 for manure (Potter et al., 2010).As discussed above we assume that manure is continuously spread onto fields, by-passing the use of animal houses and storage.Manure and synthetic fertilizer is assumed to be spread across all PFTs in CLM4.5, as these share a common nitrogen pool.Future model versions will refine these initial assumptions.
To adequately model the conversion from manure to TAN it is necessary to separate the manure into four different pools depending on the decomposition timescales (Sects.2.2.1 and 2.2.2 and Fig. 1).A similar strategy was adopted by Li et al. (2012) for manure and is commonly used in simulating litter decomposition.Synthetic fertilizer N r is added to one pool, whereafter it decomposes into the TAN pool (Fig. 1).Once in the TAN pool N r (1) washes off during rain events (Brouder et al., 2005); (2) volatilizes as NH 3 (Sutton et al., 1994;Nemitz et al., 2000), whereafter it is redeposited onto the canopy (not shown) or enters the atmospheric flow; (3) nitrifies to form nitrate (NO − 3 ) (Stange and Neue, 2009); or (4) is incorporated into the soil nitrogen pools.Nitrate, in turn, becomes incorporated into the soil (Fig. 1).A number of other smaller loss processes are not explicitly simulated.
Manure must be added to the model in such a manner as to conserve nitrogen (Fig. 1).Here, we assume animals consume carbon and nitrogen from plants and then subsequently excrete this as manure.Within the CLM, carbon and nitrogen in the plant-leaf pool is thus converted to carbon and nitrogen in manure, conserving overall carbon and nitrogen.The conversion rate from carbon and nitrogen in plants to that in manure is set to equal the rate of manure production.The external data set of Potter (2010) gives the rate of N r production from animals, and thus allows us to specify the nitrogen flows.The specified C to N ratio in the plant-leaf pool determines the associated carbon flows due to ruminant consumption of plant material.The input manure production rate from animals implicitly includes that produced from transported feed.Thus the subsequent NH 3 emission rate includes the nitrogen contained in transported feed grown elsewhere.
Here we make the simplification that the consumption rate of plant matter to balance the manure production is local.That is, we do not explicitly consider the import of animal feed to match the carbon and nitrogen flows associated with manure production.While this is not entirely consistent, the development of the requisite data set for feedstock flows from 1850-2000 is outside the scope of this study.We do not know of an Earth system model that does consider the anthropogenic import of nitrogen or carbon.This inconsistency could produce cases where there is insufficient local plant material to balance the overall manure and urine production, but this is generally not the case.The parameterization also ignores export of N r in ruminant products such as milk and protein, and other N r losses, which could create an additional source of uncertainty.

Manure and urine
Prescribed manure (including urine) is input at a constant annual rate (α applied (m)) (g m −2 s −1 ) into the manure nitrogen pools depending on latitude and longitude.It is assumed that a fraction (f u = 0.5) of nitrogen excreted is urine, with the remaining 50 % excreted as fecal matter (Gusman and Mariño, 1999).In practice the fraction of nitrogen excreted as urine is highly variable and depends on the type of animal feed and other parameters (Jarvis et al., 1989).The excreted urine is directly added to the TAN pool (g N m −2 ).This is consistent with urea forming the dominant component of urine N and the subsequent rapid conversion of urea to ammoniacal form (Bristow et al., 1992).Feces are composed of matter with varying carbon to nitrogen ratios that take different times to decompose depending on how easily they can be digested by microbes.Excreted feces are assumed to form three different pools (g m −2 ) depending on their rate of mineralization (e.g., Gusman and Mariño, 1999): (1) we assume a fraction f un = 5 % is excreted as unavailable nitrogen (N unavailable ), the lignin component of manure where the nitrogen remains immobilized by bacteria (C : N ratio > 25 : 1); (2) a fraction f r = 45 % goes to the resistant pool (N resistant ), which forms the cellulose component of manure (C : N ratio ca. 15 : 1), which forms TAN relatively slowly; (3) and a fraction f a = 50 % goes to the available pool (N available ) that is readily available to form TAN (N available ).In reality the fractions within each of these broadly defined pools will be dependent on the type of animal and the type of feed.
The equations governing the three manure pools (see Fig. 1) are where α applied (m) is the amount of manure applied (g m −2 s −1 ); f a , f r and f un are the fractions of manure applied to each pool; and K a and K r (s −1 ) are temperaturedependent mineralization rates of these manure pools and into soil nitrogen pools.The decay constants, K a and K r , are measured as the fast and slow decomposition rates for biosolids added to various soils and incubated at 25 • C (Gilmour et al., 2003), where a two-component decay model accurately fits approximately 73 % of the samples incubated.
The decay timescales for manure in the available and resistant pools at 25 • C are 48 and 667 days, respectively.The temperature dependence of the decay constants is derived from a fit of temperature-dependent mineralization rates (see Appendix A) (Vigil and Kissel, 1995) corresponding to a Q 10 value of 3.66.To prevent the manure pools from building up over long timescales we assume that manure is incorporated into soils with a time constant of 365 days with a mechanical rate constant k m .This timescale is consistent with the base bioturbation rate of 1 cm 2 yr −1 assumed in Koven et al. (2013) and a typical length scale of 1 cm.The sensitivity of the subsequent nitrogen pathways to this timescale is small (Sect.3.4).Note that nitrogen in the N unavailable pool does not mineralize and is thus only incorporated into soil organic matter on the timescale determined by k m .We assume that nitrogen prior to conversion to TAN comprises a range of insoluble organic compounds that do not wash away or otherwise volatilize.

Synthetic fertilizer
Synthetic fertilizer nitrogen is added to the N fertilizer pool (g N m −2 ) (Fig. 1) at a rate (α applied (t)(f )) (g N m −2 s −1 ) that depends on geography and time.The amount of nitrogen within the synthetic fertilizer pool is subsequently released into the TAN pool at the rate k f (s −1 ): Here we assume all synthetic fertilizer is urea.We set the decay timescale of urea fertilizer to be 2.4 days, consistent with the decay rate measured in Agehara and Warncke (2005) for temperatures from 15 to 20 • C. In a series of experiments Agehara and Warncke (2005) show that 75 % of the urea hydrolyzes in a week at temperatures from 10 to 25 • C without a significant dependence on temperature especially for temperatures above 15 to 20 • C.
The timing for synthetic fertilizer application is determined internally within the CLM4.5 crop model (Levis et al., 2012) as the spring planting date for corn.Note, however, that except for the determining the timing of fertilizer application, the crop model is not used.We use the planting date for corn since the CLM4.5 crop model only specifically includes corn, soybean and temperate cereals and the planting date for corn lies between the earlier planting date for temperate cereal crops and the later planting of soy.The date for fertilizer application is determined for each grid point location using the surface temperature-based criteria developed by Levis et al. (2012) for simulating the planting date of corn: the 10-day running mean temperature, 10-day running mean daily minimum temperature and growing degree days must all surpass fixed threshold values (283.15 and 279.15K and 50 days, respectively) before planting can take place.Fertilizer application dates can have a large influence on the seasonality of the emissions (e.g., see Paulot et al., 2014) and the subsequent flows of N r (Sect.3.4).Future applications will assume more complete algorithms for fertilizing the spectrum of crops, as well as multiple fertilizer applications and double cropping.A global accounting of fertilization practices and application techniques (e.g., fertilizer injection) nevertheless remains a considerable source of uncertainty in a global simulation of the NH 3 emissions from agriculture.

Total ammoniacal nitrogen (TAN)
We consider two TAN pools (g N m −2 ), one for the nitrogen produced from synthetic fertilizer N TAN (f ) the other for nitrogen from manure N TAN (m).The budget for the manure and synthetic fertilizer TAN pools respectively is given by Here is the loss of nitrogen by runoff from the manure or synthetic fertilizer TAN pool; is the loss rate of nitrogen to the soil nitrogen pools; F NH 3 (m) and F NH 3 (f ) (g N m −2 s −1 ) are the NH 3 emissions from the manure and fertilizer TAN pools, respectively, to the atmosphere; and F NO 3 (m) and F NO 3 (f ) (g N m −2 s −1 ) are the loss of nitrogen through nitrification from the manure and synthetic fertilizer TAN pools.The formulation of each of these terms is given below.Inputs into N TAN (m) pool are from the fraction (f u ) of applied manure as urine (α applied (m)), and from the decomposition of the nitrogen within the available and resistant manure pools.Input into the N TAN (f ) pool is from the synthetic fertilizer pool.

Runoff of nitrogen to rivers
The immediate runoff of fertilizer and manure nitrogen to rivers from the manure and fertilizer TAN pools is derived from the runoff rate of water (R) (m s −1 ) in the CLM multiplied by concentration of nitrogen in the TAN water pool: The value of R is calculated within the CLM and is a function of precipitation, evaporation, drainage and soil saturation.The amount of water within the TAN pool ).An expression for N water (m/f ) is given in Sect.2.2.9.It should be emphasized that this is the immediate runoff of manure and synthetic fertilizer nitrogen from the TAN pools.Subsequent loss of nitrogen derived from manure and synthetic fertilizer application occurs following the nitrogen transfer to the soil pools, but is not tracked in these simulations.Additional hydrological losses will also occur following the deposition of NH 3 volatilized to the atmosphere.These losses are also not tracked in the current simulation.Initially, we attempted to use the runoff parameterization based on the global Nutrient Export from Watersheds 2 (NEWS 2) model (Mayorga et al., 2010), where runoff is also parameterized in terms of R.However, the amount of nitrogen that runs off in NEWS 2 is represented in terms of the annual nitrogen initially applied to the land and thus is not directly related to the amount of nitrogen in the TAN pool.

Diffusion through soil
Nitrogen is assumed to diffuse from the TAN pool to the soil pools.Génermont and Cellier (1997) represent the diffusion coefficient of ammonium through soils as dependent on soil water content, soil porosity, temperature and an empirical diffusion coefficient of ammonium in free water (see Appendix A).For example, assuming a temperature of 21 • C, a soil porosity of 0.5 and a soil water content of 0.2 the resulting diffusion coefficient is approximately 0.03 cm 2 day −1 , in reasonable agreement with measurements in Canter et al. (1996).Here we assume a typical length scale of 1.0 cm to convert the diffusion rate to a timescale.The resulting diffusion of ammoniacal nitrogen is added to preexisting nitrogen pools in CLM4.5 and is not further tracked.

Flux of ammonia to the atmosphere
The flux of NH 3 (F NH 3 , g m −2 s −1 ) to the atmosphere is calculated from the difference between the NH 3 concentration at the surface (NH 3 (g), g m −3 ) of the TAN pool and the freeatmosphere NH 3 concentration (χ a , g m −3 ) divided by the aerodynamic (R a ) and boundary layer (R b ) resistances (Ne-mitz et al., 2000;Loubet et al., 2009, Sutton et al., 2013).
The calculation of NH 3 (g) is given below.For compatibility with the NH 3 emission parameterization we compute average values of R a and R b over each CLM soil column, which may contain several PFTs.We specify χ a to be 0.3 µg m −3 , representative of concentrations over low-activity agricultural sites (Zbieranowski and Aherne, 2012).Continental NH 3 concentrations between 0.1 and 10 µg m −3 have been reported by Zbieranowski and Aherne (2012) and Heald et al. (2012).A concentration of 0.3 µg m −3 is intermediate between concentrations at low to moderate pollution sites as diagnosed in GEOS-chem (Warner et al., 2016).The sensitivity to this parameter is small as NH 3 (g) is usually very large (Sect.3.4).While Eq. ( 8) allows for negative emissions (NH 3 (g) < χ a ) or deposition of atmospheric NH 3 onto the soil we disallow negative emissions in the current simulations.In future studies the atmospheric concentration of NH 3 will be calculated interactively by coupling the FAN model to the atmospheric chemistry component of the CESM (CAMchem), thus allowing the dynamics of the NH 3 exchange between the soil, the atmosphere and vegetation to be captured (e.g., Sutton et al., 2013).
A large fraction of the NH 3 emitted to the atmosphere is assumed captured by vegetation.The amount emitted to the atmosphere is given by where f capture is set to 0.6.Plant recapture of emitted NH 3 is often reported to be as high as 75 % (Harper et al., 2000;Nemitz et al., 2000;Walker et al., 2006;Denmead et al., 2008;Bash et al., 2010).Using seabird nitrogen on different substrates (rock, sand, soil and vegetation) inside a chamber, Riddick (2012) found NH 3 recapture to be 0 % on rock, 32 % on sand, 59 % on soil and 73 % on vegetation.We set f capture to 0.6 as a mid-point between the value for soil (when the crops are planted) and when crops are fully grown (Wilson et al., 2004).Bouwman et al. (1997) also used canopy capture to estimate emissions with the captured fraction ranging from 0.8 in tropical rainforests to 0.5 in other forests to 0.2 for all other vegetation types, including grasslands and shrubs.Bouwman et al. (1997) omitted canopy capture over arable lands and intensively used grasslands.Overall, the deposition of NH 3 onto the canopy (or even the soil surface) is poorly constrained (e.g., see Erisman and Draaijers, 1995) and often ignored in model simulations.In reality canopy capture is not constant but depends on surface characteristics and boundary layer meteorology.Variations in canopy capture will induce temporal and regional variations in NH 3 emissions.Explicitly including canopy capture fraction allows a differentiation between biogeochemical pathways in future studies.In the future, when the model is fully coupled with the atmospheric NH 3 cycle a compensation point approach would be desirable for calculating the net NH 3 emissions, but this is outside the scope of the present study.
It is assumed that the nitrogen in the TAN pool is in equilibrium between NH 3 (g) (g m −3 ), NH 3 (aq) (g N m −3 ) and NH + 4 (aq) (g N m −3 ).The equilibrium that governs the speciation of these species is determined by the Henry's law coefficient (K H ), where K H is a measure of the solubility of NH 3 in water, and the disassociation constant of NH + 4 in water (K NH 4 ) (mol L −1 ) (e.g., Sutton et al., 1994): NH + 4 (aq) Combining these two expressions NH 3 (g) can be expressed as a function of the total TAN (e.g., Pinder et al., 2004, although note their different units for K H and K NH 4 ): where [H + ] is the hydrogen ion concentration in mol L −1 .
Both K H and K NH 4 are temperature-dependent.As temperature and pH increase the concentration of NH 3 (g) increases.The pH of the solution depends on the type of soil, the exposure of the manure to air, and the type of fertilizer used and may change with the aging of the manure or synthetic fertilizer.In Eghball et al. (2000) the majority of the reported measurements of pH for beef cattle feedlot manure are between 7 and 8, although in one case a pH of 8.8 was measured.The recommended pH for various crops ranges from approximately 5.8 to 7.0 depending on the crop (e.g., http://onondaga.cce.cornell.edu/resources/soil-ph-for-field-crops).For now we simply set the pH of the solution to 7 for both the synthetic fertilizer and manure TAN pools.Sensitivity to pH is explored in Sect.3.4.

Conversion of TAN to NO − 3
The flux of N r from the TAN pool to NO − 3 by nitrification (F NO − 3 , g m −2 s −1 ) was adapted from Stange and Neue (2009) for the gross nitrification rates in response to fertilization of a surface with manure or synthetic fertilizer.In particular, Stange and Neue (2009) fit measured gross nitrification rates to an expression using a maximal nitrification rate, r max (µg N kg −1 h −1 ), modified by a soil temperature response function (f (T )) and a soil moisture response function (f (M)) (Stange and Neue, 2009).However, since r max is fit from their experimental data, the dependence of the nitrification rate on the ammonium concentration is not explicitly included in the formulation of Stange and Neue (2009).We have remedied this by setting the maximum nitrification rate (r max ) in the formulation of (Stange and Neue, 2009) to be consistent with the formulation in Parton et al. (2001): where (T g ) and (M) are functions of soil temperature and moisture and the ammonium concentration is assumed to be in equilibrium with the other forms of ammoniacal nitrogen and is thus expressed in terms of pH, K H and K NH 4 and NH 3 (g) (m/f ) and r max is equal to 1.16 × 10 − 6 s −1 .

Nitrate
The rate of change of the nitrate pool is given by The source of nitrate ions is nitrification from the TAN pool.
Nitrate is lost to the soil nitrate pool through diffusion.Nitrate leaching is not explicitly taken into account in the current model as the diffusion of nitrate into the soil pools occurs very rapidly.The loss of nitrate through runoff and leaching can, however, occur within the CLM, but is not tracked in the current simulations.Nitrate ions diffuse significantly faster than the NH + 4 ions because they are not subject to immobilization by negatively charged soil particles (Mitsch and Gosselink, 2007).Diffusion rates used in this study are derived from the same formulation as assumed for the diffusion of ammonium (e.g., see Jury et al., 1983) with a different base diffusion rate.The summary of measurements given in Canter et al. (1996), where both the diffusion of ammonium and nitrate were measured in the same soil types and wetness, suggest the base diffusion rate of NO − 3 is approximately 13 times faster than that of ammonium.

TAN and manure water pools
The evolution of the TAN manure and synthetic fertilizer water pools depends on the water added during manure or synthetic fertilizer application and the subsequent evolution of the water in the pools.The equations for the manure and synthetic fertilizer water are These equations include a source of water (s w (m) or S w (f )) added as a fraction of the synthetic fertilizer or manure applied and a relaxation term (k relax , s −1 ) to the soil water (M water (m)) calculated in the CLM for the top 5 cm of soil.
M water explicitly takes into account the modification of the water pool due to rainfall, evaporation and the diffusion of water in the soil layers.We assume the TAN pool equilibrates with water within the top 5 cm of the soil with a rate of 3 days −1 .The solution is insensitive to this parameter within the ranges examined of 1 to 10 days −1 (Sect.3.4).The water content of manure applied to fields depends on the animal, its feedstock and on agricultural practices.74.2 g kg −1 and a nitrogen content of 1.6 g kg −1 , resulting in 5.7 × 10 −4 m water applied per gram of manure nitrogen applied (Sommer and Hutchings, 2001).In the case of synthetic fertilizer we assume urea is added as a liquid spread, where water added is calculated from the temperature-dependent solubility of urea in water (UNIDO and IFDC, 1998).

Present-day control simulation
This simulation uses the manure and synthetic fertilizer input as given in Potter et al. (2010).Forcing at the atmospheric boundary is set to the Qian et al. (2006) reanalysis for solar input, precipitation, temperature, wind and specific humidity.The simulation is run for 15 model years (1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004), with the last 10 years of the simulation used for analysis.The spin-up period allows for the more decomposition resistant N pools to approach a steady state with respect to the loss from mechanical incorporation into the soil.

Historical simulation
The historical simulation uses transient forcing conditions (accounting for changes in atmospheric CO 2 , nitrogen deposition, aerosol deposition and land use change forcings) and the Qian et al. (2006) atmospheric forcing data set.Quality 6-hourly meteorological data sets for the period prior to 1948 do not exist.Therefore from 1850 to 1973 CLM4.5 is driven by recycled meteorological data, using meteorological data from the 1948-1973 time period.During this time long-term temperature changes were small: the statistically significant long-term changes in temperature (outside of natural variability) occur after 1973.After 1973 the meteorological data are not recycled but are valid for the year applied.
The temporal distribution of manure and synthetic fertilizer application from 1850-2000 is specified by applying the temporal distribution of Holland et al. (2005) to the base values as calculated in Potter et al. (2010).Due to the lack of detailed information on the geography of historical manure and synthetic fertilizer we use the scaled spatial distribution from Potter et al. (2010).We assume manure production has changed from 26.3 Tg N yr −1 in 1860 to 125 Tg N yr −1 in 2000 (Holland et al., 2005;Potter et al., 2010), but we acknowledge that these temporal changes are uncertain.Synthetic fertilizer was first used in the 1920s, with use increasing to 62 Tg N yr −1 in 2000.

Model evaluation
To evaluate model output, measurements of the percentage of applied nitrogen that was emitted as NH 3 (P v ) from the literature were compared against corresponding model predictions.The model predictions are obtained from the presentday control simulation.The percent-volatilized NH 3 was used as a metric because it can be compared across time, irrespective of the absolute amount of nitrogen applied to the surface.To be able to compare emissions to published measurements, we require field studies with published data on nitrogen excretion rates, NH 3 emissions, ground temperature, location, and date of measurement.Given all of these requirements we found that only a small selection of publications had sufficient data.
For the manure emissions, 44 measurements in a range of climates (temperatures from 1.4 to 28 • C) and a range of livestock management methods (commercial beef cattle feed yard, dairy cows grazing on ryegrass, beef cattle grazing on ryegrass and dairy cattle grazing on pasture land) were used (Supplement Table S1).Each P v reported by the measurement campaign was compared against the P v at the corresponding grid cell in the model.For the synthetic fertilizer scenario, 10 measurements in a range of latitudes (43 • S to 50 • N) over a range of land use surfaces (pasture, sown crops, turf and forest) were used (Table S2).Each total annual P v reported by the measurement campaign was compared against the annual P v of the corresponding grid cell.

Nitrogen volatilized as NH 3 from manure
There is a general increase in P v with temperature, in both the model and measurements (Fig. 2).However, temperature is not the only factor in determining NH 3 emissions, where wind speed, water availability and below ground soil properties can also effect NH 3 emission.This is particularly demonstrated by the measurements of Todd et al. (2007) at temperatures less than 5 • C where the measured emissions are higher than those predicted at higher temperatures (e.g., Bussink, 1992).It is also worth noting that the model predicts the emissions of Todd et al. (2007) at lower temperatures with relative success.
The agreement between measured and modeled P v from manure appears reasonable, with an R 2 of 0.76 that is significant at the 99.9 % confidence level.On closer inspection, the model appears to agree best with measurements made on grassland and differs considerably with measurements made by both campaigns for beef cattle feedlots in Texas, where beef cattle feedlots are commercial operations to prepare livestock for slaughter and comprise thousands of animals contained in a pen.This is perhaps not surprising, as the parameterization developed here explicitly represents  S1).Symbol color indicates temperature at which the emissions occurred; shape gives the study.
emissions from manure spreading as opposed to the more managed conditions in feedlots.

Nitrogen volatilized as NH 3 from synthetic fertilizer
The comparison between measured and modeled annual average P v from synthetic fertilizer applied to a range of land use types is weak, with an R 2 of 0.2 that is significant at the 90 % confidence level (Fig. 3).The lowest emissions in the model and measurements tend to be associated with the higher latitudes of both hemispheres.There does not appear to be any noticeable bias with land use type where the model estimates are both higher and lower than measured values of P v for surfaces covered in turf, pasture land and crops.The fact that the R 2 for the synthetic fertilizer measurements is lower than the R 2 of the manure measurements is potentially caused by differences between actual farming practices and model assumptions.

Nitrogen runoff
In this study we simulate the direct nitrogen runoff from the manure and synthetic fertilizer TAN pools, but do not track the resulting nitrogen flows.These flows are tracked, however, in Nevison et al. (2016), where the nitrogen runoff from manure and synthetic fertilizer pools is routed into the river transport model (RTM) (Dai and Trenberth, 2002;Branstetter and Erickson III, 2003) within the CESM.Nevison et al. ( 2016) assume denitrification occurs within the simulated rivers at a rate inversely proportional to the river depth (amounting to approximately 30 % of the nitrogen in- gives confidence that the runoff is reasonably simulated, the complications in simulating river runoff preclude tight model constraints.

Geography of nitrogen inputs
Global maps of nitrogen input from synthetic fertilizer and manure application during the present-day simulation are given in Potter et al. (2010)

Geography of nitrogen losses
There are strong geographical differences in the loss pathways of nitrogen following manure or synthetic fertilizer application.The importance of the various loss pathways from the TAN pool (the amount nitrogen volatilized as NH 3 , runoff, nitrified or diffused directly into the soil, Figs.4-8) is dependent on temperature, precipitation and soil moisture.In hot, arid climates, the percentage volatilized is high (Figs. 4 and 5).For example, regions of high NH 3 volatilization of applied manure N r approach 50 % across the southwest US and Mexico, eastern South America, central and southern Africa, parts of Australia, and across southern Asia from India to Turkey (Fig. 5).The absolute highest emissions of NH 3 from applied synthetic fertilizer and from applied ma-nure approach 20 kg N ha −1 yr −1 over hot regions with high applications, e.g., the Indian subcontinent and parts of China (Figs. 4 and 5).Ammonia emissions from manure are more broadly distributed globally than those of synthetic fertilizer with high NH 3 emissions not only over the synthetic fertilizer hotspots, characterized by heavy application of both synthetic fertilizer and manure but also over southeastern South America and central Africa.For the most part, the largest synthetic fertilizer NH 3 emissions occur during April-June, reflecting the single fertilization date used in this study as calculated in the CLM for corn.While Paulot et al. ( 2014) also show the maximum synthetic fertilizer emissions generally occur from April-June, they obtain relatively higher emissions than simulated here during the other seasons.This is likely due to differences in the assumed timing of applied Runoff of N r from applied synthetic fertilizer and manure TAN pools, as well as nitrification and diffusion into the soil, depends on precipitation and soil moisture.High manure and synthetic fertilizer N r runoff from the TAN pools (see Fig. 6) occurs particularly across parts of China, Europe (particularly the northern parts) and the east central US.The global hotspot for simulated N r runoff from the TAN pools is China, where runoff approaches 20 kg N ha −1 yr −1 for nitrogen applied as either manure or synthetic fertilizer.In general the importance of runoff as a nitrogen loss pathway becomes more important in the wetter and cooler regions.In contrast, over India and Spain the agricultural nitrogen input is high, but the runoff is relatively low.In these regions with high temperatures (and dry conditions) NH 3 volatilization is the preferred pathway for nitrogen losses from the TAN pool.
The percent of the TAN pool nitrified or diffused directly into the soil (see Figs. 7 and 8) also tends to be largest in wetter and cooler regions.The amount of nitrogen nitrified has an optimal temperature of 28 • C and tends to occur more rapidly under moist conditions; the diffusion of nitrogen into the soil is also promoted under wet conditions.

Regional and global accounting of nitrogen losses
As nitrogen cascades through the environment it can be emitted as NH 3 or runoff or leached at many different stages.
Here we only examine the losses directly from manure or fertilizer TAN pools.Globally, the direct loss of applied niwww.biogeosciences.net/13/3397/2016/Biogeosciences, 13, 3397-3426, 2016 trogen to the atmosphere as NH 3 is similar for manure and synthetic fertilizer (17 % for manure, 19 % for synthetic fertilizer; see Fig. 9).Our global estimates of the percent of manure and synthetic fertilizer volatilized as NH 3 are similar to Bouwman et al. (2002) and Beusen et al. (2008), although our estimate for synthetic fertilizer volatilization as NH 3 is somewhat high.Bouwman et al. (2002) estimate 19-29 % of applied manure and 10-19 % of applied synthetic fertilizer volatilizes as NH 3 ; Beusen et al. (2008) conclude 15-23 % of applied manure is lost as NH 3 (including losses from housing and storage, grazing and spreading) and 10-18 % of applied synthetic fertilizer is lost.We calculate the global direct runoff from manure or fertilizer TAN pools as 8 % for manure N r and 9 % for synthetic fertilizer (Fig. 9).Bouwman et al. (2013) find that 23 % of deposited N r (comprised of synthetic fertilizer, manure and atmospheric nitrogen deposition) runs off, higher than our estimate.However, our estimate only includes the direct runoff from the TAN pool.Additional nitrogen is lost downstream from the TAN pool due to runoff and leaching from the soil nitrogen pools or following NH 3 emission and re-deposition.
Our simulations assume a large fraction of emitted nitrogen is captured by the canopy, where canopy capture accounts for 25.5 % of manure losses and 30 % of synthetic fertilizer losses (Fig. 9).The nitrogen captured by the canopy may have a number of fates.First, Sparks (2009) posits that foliar nitrogen uptake could more readily influence plant growth than uptake from soils since it is a direct addition of N to plant metabolism.As such it would decrease plant demand on soil uptake and thus conserve the soil nitrogen reservoirs.Secondly, nitrogen uptake by the plants, even if not directly used in plant metabolism, may redeposit onto the surface with litterfall.Finally, it may be emitted back to the atmosphere from plants.The latter process can be represented through a compensation point model between the atmosphere, the ground and stomata (e.g., Massad et al., 2010).A full accounting of this requires the simulation to be fully coupled with the atmosphere and soil chemistry and biogeochemistry, which is beyond the scope of the present study.
In the case of synthetic fertilizer the direct diffusion of TAN N r into the soil pool (22 %) is larger than nitrification (17 %); for manure it is just the opposite: the nitrification (29 %) is larger than the direct diffusion (14 %) (Fig. 9).In practice, as simulated here, this makes little difference as the diffusion of nitrate into the soil pool occurs very rapidly, an order of magnitude faster than the diffusion of nitrogen from the TAN pool.Thus, NO − 3 is directly incorporated into the soil nitrate pool without any subsequent loss.It should also be noted that a small percentage of manure is mechanically stirred into the soil organic nitrogen pools.Accounting for the N r diffused from the TAN pool into the soil pools, and assuming the NH 3 emissions captured by the canopy, as well as the ammonium nitrified to NO − 3 that also ends up in the soil pools, we find that globally 75 % of manure nitrogen and 71 % of synthetic fertilizer nitrogen ends up in the soil nitrogen or soil organic nitrogen pools.Emitted NH 3 may also end up in the soil nitrogen pools downstream of the emissions.Of course, once in these soil pools, there may be subsequent losses of nitrogen due to runoff and leaching or emissions, but these are not calculated in this initial study.
The global percentages given above change appreciably when examined over subsets of countries (Fig. 10).For example, over all developed countries the percentage of emissions of manure and synthetic fertilizer TAN as NH 3 (13 %) is substantially smaller than for developing countries (21 %).These differences can be largely explained by the fact that developing countries tend to be located in warmer climates than developed countries (e.g., see Bouwman et al., 2002).Bouwman (2002) calculated NH 3 emission factors for manure of 21 and 26 % for developed and industrialized countries, respectively and for synthetic fertilizer of 7 and 18 %, respectively.
In our simulations 16 and 9 % of applied agricultural nitrogen is emitted as NH 3 in the US and the European Union, respectively.The direct runoff of nitrogen accounts for 9 and 14 % of the losses of agricultural nitrogen in the US and the European Union, respectively.Nitrogen runoff is favored in the cooler moister climate of Europe.However, note the large contrast between India and China, where for India NH 3 emissions are 27 % of the applied N r with very little runoff, whereas for China the runoff and emissions are approximately equal (13 and 10 %, respectively).

Comparison to other emissions inventories
Figure 11 gives a comparison of manure and synthetic fertilizer NH 3 emissions from the FAN process model for the year 2000 to various emission inventories.These inventories rely on emission factors depending on animal husbandry, types of synthetic fertilizer usage and other details of agricultural practices.Only the NH 3 emission inventory of Huang et al. (2012) for China and that of Paulot et al. (2014) explicitly account for temperature to modify their emission factors; the inventory of Paulot et al. (2014) also uses wind speed to modify the emission factors.The inventories of Paulot et al. (2014) for 2005-2008, Beusen et al. (2008) for 2000, and EDGAR v4.2 (European Commission JRC and PBL, 2011) for 2005-2008 are global inventories.The EDGAR inventory does not strictly separate the NH 3 emissions into those from manure and synthetic fertilizer, so we simply show the overall NH 3 emissions from both manure and synthetic fertilizer.Over the US we also give an estimate for synthetic fertilizer NH 3 from 1995 (Goebes et al., 2003) and for NH 3 emissions from animal agricultural operations (US EPA, 2004).Over the China region, emission estimates are from Huang et al. (2012) for 2006and Streets et al. (2003) for 2000.Over Europe, results using the Greenhouse Gas and Air Pol-lution Interactions and Synergies (GAINS) model are given (Klimont and Brink, 2004) as reported in Paulot et al. (2014).
Globally all inventories give approximately the same overall NH 3 emissions of 30-35 Tg N yr −1 .The global apportionment of emissions between manure and synthetic fertilizer in this study is in approximately the ratio of 2 : 1, roughly consistent with that of Paulot et al. (2014) and Beusen et al. (2008).In this study the overall NH 3 emissions are high over the US compared to the other inventories: the manure NH 3 emissions are close to the estimate of other inventories, while the synthetic fertilizer emissions are higher.The European and Chinese NH 3 emissions estimated in this study are on the low side of the other inventories.Over Europe the current parameterization underestimates the manure emissions compared to the other studies while the synthetic fertilizer emissions are somewhat high.The EDGAR emission estimate is somewhat higher than the other estimates over Europe, although this may depend on exactly what is assumed for the boundary of the European region.Over China our synthetic fertilizer emissions are similar to those of Huang et al. ( 2012), but we underestimate the manure NH 3 emissions compared to the other inventories.Of the three regions examined all inventories suggest the Chinese emissions are highest.

Site-specific simulated pathways
The hourly time series of the fate of applied nitrogen from manure and synthetic fertilizer at a single site better illustrates the relationship between the different nitrogen flow pathways and the local meteorology (Fig. 12).The large fluctuations in the NH 3 emissions and the resultant implications for atmospheric chemistry also demonstrate the desirability of emissions that respond on hourly timescales to meteorological conditions.The site shown in Fig. 12 is near the Texas panhandle.It experiences several large rain events and surface temperatures ranging from 0 to 18 • C over a period of about 2 months during the spring season examined.The response of the NH 3 emissions to the diurnal temperature range is clearly evident.Near the beginning of the examined period the nitrogen losses of manure TAN due to NH 3 volatilization are initially small, comparable with the diffusive loss to the soil pools and somewhat less than the loss due to nitrification.The loss by nitrification and the diffusion from the TAN manure pool remain roughly constant throughout the period examined, although both processes show some response to precipitation.Note in particular that the diffusive loss reaches a maximum near 21 May, presumably caused by the increased water content in the soil due to the prior rain event.With the rise in temperatures towards the end of the period, the NH 3 emission loss of manure N r becomes the dominant loss pathway and the TAN manure pool decreases.Closer inspection suggests, however, that the large increase in the NH 3 emissions towards the end of the period cannot solely be attributed to temperature but must also  be attributed to decreased water in the TAN pool as the soil dries.The latter process increases the concentration of nitrogen species within the TAN pool.The TAN manure pool is punctuated by sharp declines, associated with precipitation and increased runoff (Fig. 12c).Synthetic fertilizer TAN responds similarly during this period but is clearly impacted by the temporal application of synthetic fertilizer.The decrease in the synthetic fertilizer TAN pool occurs on a timescale of approximately a week, consistent with the timescale used in the MASAGE_NH3 model (Paulot et al., 2014).

Global nitrogen pathways: historical
Historical nitrogen pathways are simulated since 1850 in a simulation with changing climate and changing application amounts.These simulations do not include changing agricultural practices, for example changes in animal housing and storage, changes in animal diet and explicit changes in land use.All of these changes may substantially alter the nitrogen loss pathways.Thus the results must be treated with caution.
The nitrogen produced as manure increases in the historical simulation from 21 Tg N yr −1 in 1850 to 125 Tg N yr −1 in 2000 (Fig. 13).In 1900 we estimate that 37 Tg N yr 1 of manure is produced, similar to the Bouwman et al. (2013) estimate of 35 Tg N yr −1 .Emissions of NH 3 from applied manure increase from approximately 3 Tg N yr −1 in 1850 (14.3 % of the manure produced) to 22 Tg N yr −1 in 2000 (17.6 % of the applied manure).On the other hand, the percentage of manure nitrogen that is nitrified decreases from 33 to 27 % since the preindustrial period.Note that the year 2000 emissions in the historical simulation differ slightly from the year 2000 emissions reported in the present-day control simulation, as the latter consists of the average emissions from 1995 to 2004.
Synthetic fertilizer nitrogen application has increased dramatically since the 1960s, with an estimated 62 Tg N yr −1 applied as synthetic fertilizer in 2000.We estimate that the volatilization of synthetic fertilizer as NH 3 is 12 Tg N yr −1 in 2000 (19 % of that applied).The percent of synthetic fertilizer nitrogen volatized to the atmosphere as NH 3 in 1920 was 8 %.On the other hand, the percentage of synthetic fertilizer that is lost through runoff decreased since the preindustrial period by 8 %.These changes can be explained by the fact the runoff of synthetic fertilizer can act to completely drain the TAN synthetic fertilizer pool when the application rate is small.
In part the historical emission increases in NH 3 can also be explained by changes in climate.The globally averaged temperature has warmed by approximately 1 • C since the preindustrial period.In a sensitivity experiment the temperature was artificially increased by 1 • C in the rate equations governing the nitrogen pathways following manure and synthetic fertilizer application.Under current manure and synthetic fertilizer application rates we find that globally an additional 1 Tg NH 3 is emitted from the manure and synthetic fertilizer pools per degree of warming.The resulting manure emissions increase by 4 % and the synthetic fertilizer emissions by 3 % per degree of warming.

Sensitivity tests
We have conducted a large number of sensitivity tests to evaluate the effect of changes in individual model parameters on NH 3 emissions.The various parameters may covary, of course, with nonlinear impacts on the NH 3 emissions; however, we have not attempted to evaluate these effects.The sensitivity tests for manure are given in Table 1, and those for synthetic fertilizer in Table 2.The sensitivity tests are labeled with a number denoting the sensitivity parameter perturbed and a letter denoting whether the test is with respect to manure emissions (m) or synthetic fertilizer emissions (f ).
In each case we give the percent change in NH 3 emissions due to the parameter change and the relative emission change with respect to the relative parameter change (the sensitivity).Rationale for the assumed parameter bounds is given in the Supplement.
Except for changes in the canopy capture parameter (EX8m/f, EX9m/f) and changes in the timing or composition of manure or synthetic fertilizer inputs (EX18m, EX18f, EX19f, EX20f), changes in the sensitivity parameters directly change the nitrogen cycling within the TAN pool (as described below).For the most part the synthetic fertilizer and manure TAN pools respond similarly to the parameter changes.Note also that, except for EX18m, where the amount of nitrogen input into the TAN pools is reduced, the total input and loss of nitrogen from the TAN pools remain the same for all sensitivity experiments.In general, the senwww.biogeosciences.net/13/3397/2016/Biogeosciences, 13, 3397-3426, 2016  4 NH 3 emissions (Tg N yr −1 ).5 Runoff (Tg N yr −1 ).6 Diffusion to soil (Tg N yr −1 ).7 Nitrification (Tg N yr −1 ).8 Canopy capture (Tg N yr −1 ).9 Percent change in NH 3 emissions due to parameter change (%).10 Percent change in NH 3 emissions per % change in parameter value11 Control simulation.12Soil pH from the ISRIC-WISE data set (Batjes, 2005). 13Change in manure composition to urine 41 %, available 21 %, unavailable 25 %, and resistant 13 %.Emissions from this study are labeled as "Riddick".
sitivity of NH 3 emissions to the imposed parameter changes are within the range of ±20 %, with many processes within the range of ±10 %.The sensitivity to the mechanical mixing of manure into the soil pools (EX1m, EX2m), the adjustment timescale for the water pool (EX3, EX4), the diffusion rate into the soil pools (EX14, EX15), the assumed depth of the water pool (EX12, EX13) and the maximum nitrification rate (EX16, EX17) all impact NH 3 emissions by less than 20 %.The sensitivity to the assumed background NH 3 concentration is also low (EX10, EX11).
The NH 3 emissions are most sensitive to changes in pH (EX5, EX6, EX7).The NH 3 emissions decreased by approximately 60 % when the pH is decreased from 7 to 6 and increased by 50 to 70 % (for manure and synthetic fertilizer, respectively) when the pH is increased from 7 to 8. We also test the sensitivity of the emissions to the spatially explicit pH input from the ISRIC-WISE data set (Batjes, 2005), with a global pH average of 6.55 (EX7).The spatially explicit pH reduced the manure NH 3 emissions from the control simulation by 23 % and the synthetic fertilizer NH 3 emissions by 14 %.Changes in pH also have a large impact on nitrification.Increased pH reduces NH + 4 (aq) and thus the rate of conversion of NH + 4 (aq) to NO − 3 .The effect of pH on the rate constant for nitrification is not included in the current parameterization.Parton et al. (2001) suggest this effect is small, between a pH of 6 and 8, varying only on the order of 15 %.Changes in pH also result in marked changes in the runoff and soil diffusion due to the large changes in emissions and nitrification: a low pH acts to increase the flux of nitrogen through these loss pathways, while a high pH acts to decrease them.
Emissions are also highly sensitive to changes in canopy capture (i.e., the parameter f capture ) as shown in EX8 and EX9.Decreasing the fraction captured by the canopy by a factor of 2 increases the emissions by approximately a factor of 3. Changes in this fraction modify the fixed ratio between the amount of nitrogen captured by the canopy and that emitted to the atmosphere.Of course, the nitrogen captured in the canopy can further impact the nitrogen budget, but this impact is not simulated here.
The NH 3 emissions are somewhat sensitive to the depth of the assumed water pool (EX12, EX13).Smaller depths (less water) give higher concentrations of all the constituents within the TAN pool resulting in higher NH 3 emissions (Eqs.7 and 12) and larger nitrogen runoff (see Sect. 2.2.4).Larger depths (more water) have the opposite effect.The diffusion of nitrogen into the soil is somewhat sensitive to changes in the assumed water depth as the coefficient of diffusion is proportional to the water content to the 10/3 power (see Appendix A).
We conducted various sensitivities to synthetic fertilizer application.Early synthetic fertilizer application decreases NH 3 emissions due to their strong temperature dependence and increase the susceptibility of the TAN pool to runoff.An early fertilization date (set to 20 March in the Northern Hemisphere or 20 September in the Southern Hemisphere) decreases the NH 3 emissions by 23 % and increases the nitrogen runoff from the TAN pool by 62 % (EX18f).To investigate the sensitivity to the application rate of synthetic fertilizer, synthetic fertilizer was applied over 20 days as opposed to the single-day application assumed in the default version of the model (EX19f).This did not have a significant impact on the emissions.The assumed synthetic fertilizer type in the default version of the model (urea) was replaced with ammonium nitrate fertilizer in EX20f.Whereas urea is converted to NH 3 rather slowly, the conversion of ammonium nitrate is rapid (in the sensitivity test the resulting nitrogen is assumed to be instantaneously released into the TAN pool with no changes in pH).However, the emissions are not particularly sensitive to this change.This is in contrast to differences in volatilization rates of different synthetic fertilizers given in Bouwman (2002).Whitehead and Raistrick (1990) show that one of the primary differences between the addition of urea vs. ammonium nitrate fertilizer is in the effect of the fertilizer on the soil pH, an effect that we do not consider in this study.In particular urea increases the soil pH and thus the NH 3 emissions.
Finally, we test the impact of manure composition on the NH 3 emissions (EX18m).The composition of manure nitrogen excreted by animals depends in part on the digestibility of the feed, which can vary in both time and space.To investigate this uncertainty we varied the composition of the manure assumed in the default model version (50 % urine, 25 % available, 22.5 % resistant and 2.5 % unavailable) to the less soluble N excreta from dairy cattle in sensitivity simulation EX18m (41 % urine, 21 % available, 25 % unavailable and 13 % resistant; Smith, 1973).This decreased the NH 3 emissions by 21 % demonstrating an important sensitivity to the composition of manure and urine.
It is important to emphasize that these sensitivity simulations only test the parameter sensitivity within the imposed model.In particular, the sensitivities to various farming practices are generally extraneous to the model assumptions with some exceptions.The sensitivities to synthetic fertilizer or manure input assumptions are tested in simulations EX18m, EX18f, EX19f, and EX20f; sensitivities to the water depth which may crudely represent some of the impacts of plowing manure or synthetic fertilizer into the soil are examined in EX12 and EX13; and finally, modifications to soil pH are tested in EX5, EX6 and EX7.

Discussion and conclusions
In this paper we develop a process-oriented model that predicts the climate-dependent pathways of reactive nitrogen applied as manure or synthetic fertilizer onto the surface of the land.It is also capable of taking into account the resulting biogeochemical cycling of nitrogen.Continued population growth will likely result in an increased application of synthetic fertilizers with concurrent increases in manure production in the future (Davidson, 2012).Climate is an important determinant in the ultimate fate of this applied nitrogen, important in determining the resulting emissions of NH 3 and other reactive nitrogen gases, in the runoff of the applied nitrogen, its nitrification and its incorporation into the soil organic and inorganic nitrogen pools.The fate of the resultant applied nitrogen may act to acerbate climate change through the formation of N 2 O, or perhaps mitigate climate change through increased carbon fertilization and the increased formation of aerosols.
Both natural and anthropogenic processes control agricultural NH 3 emissions.Previous global NH 3 emission estimates have relied on detailed information on agricultural practices including domestic animal type, animal housing and the field application of synthetic fertilizer or manure (e.g., Bouwman et al., 1997) but have minimized the representation of meteorological processes.In some cases the resulting emission factors have implicitly included temperature dependence by using different emission factors for industrial and developing countries (e.g., Bouwman et al. 1997), although recently some inventories have explicitly included empirical emission factors that vary with temperature (Paulot et al., 2014;Huang et al., 2012).Here, we take the opposite tact by constructing a model where the N r pathways and in particular the NH 3 emissions following manure and synthetic fertilizer application are explicitly driven by climate but where the explicit representation of most agricultural practices is minimized.
In this study the global emissions of NH 3 due to manure and fertilizer nitrogen sources are similar to other recent inventories, with 21 Tg N yr −1 emitted as NH 3 from manure nitrogen and 12 Tg N yr −1 emitted from synthetic fertilizer for present-day conditions.Strong regional differences in emissions captured by the bottom up inventories are also simulated.Moreover, we are able to simulate the interannual, seasonal and diurnal changes in NH 3 emissions critical for air pollution applications (e.g., see de Meij et al., 2006).It is perhaps important to note that the impact of nitrogen emissions on the global carbon budget has generally made use of previous emission inventories without explicit seasonal or diurnal dependence of NH 3 emissions and with a rather minimal representation of the geographic meteorological dependence.
The model developed here uses a process-level approach to estimate nitrogen pathways from synthetic fertilizer and manure application.It is suitable for use within an Earth system model to estimate the resulting NH 3 emissions, nitrogen runoff, and the incorporation of the nitrogen into soil organic and inorganic matter.The modeled N r pathways dynamically respond to climatic variation; for example, (1) the breakdown timescale of manure and fertilizer into TAN depends on temperature; (2) the formation of NH 3 gas from the TAN pool is highly temperature-sensitive, with the rate of formation described by the temperature dependence of the thermody-namic Henry and dissociation equilibria for NH 3 (Nemitz et al., 2000); (3) the rate of nitrification of NH 3 within the TAN pool, determined by the rate at which ammonium ions are oxidized by nitrifying bacteria to form nitrate ions (Abbasi and Adams, 1998), is controlled by environmental factors such as soil temperature and soil moisture; and (4) the runoff of N r is determined by the precipitation.Predictions for direct nitrogen runoff from fertilizer and manure nitrogen pools and the incorporation of nitrogen into soil pools from applied fertilizer and manure nitrogen are some of the first made by a global process-level model.
Manure is not a new nitrogen source but rather contains recycled N r from soil nitrogen produced when animals eat plants.Therefore, to conserve nitrogen within an Earth system model, the application of manure determines the consumption of plant matter by animals.Specifically, the model calculates the amount of nitrogen and carbon needed for a given manure application and subtracts it from the plant leaf pools within the CLM.The manure production may speed up the decay and processing of plant biomass, releasing different N r products to the atmosphere than natural decay (Davidson, 2009).
The climate dependency incorporated into the model suggests that the pathways of nitrogen added to the land are highly spatially and temporally heterogeneous.An examination of nitrogen loss pathways at a point over Texas shows the variation in the nitrogen pathways on a variety of timescales with changes in temperature, precipitation and soil moisture.Globally, NH 3 volatilization is highest in the tropics.The percentage of manure nitrogen volatilized to NH 3 in this study show a large range in both developing countries (average of 20 %; maximum: 36 %) and industrialized countries (average of 12 %; maximum: 39 %).The model also predicts spatial and temporal variability in the amount of NH 3 volatilized from synthetic fertilizers, ranging from 14 % (maximum 40 %) in industrialized countries to 22 % (maximum 40 %) in developing countries.In comparison, the EDGAR database uses the emission factors based on Bouwman et al. (2002), where 21 and 26 % of manure is converted into NH 3 in industrialized and developing countries, respectively.Nitrogen runoff from the manure and synthetic fertilizer TAN pools is highest in areas of high N r application and high rainfall, such as China, North America and Europe.Despite high nitrogen input rates we simulate low nitrogen runoff in India and Spain, for example.We also simulate climate-dependent pathways for the diffusion of N r into the soil inorganic nitrogen pools and the nitrification of ammonium to nitrate.
Historically we predict emissions of NH 3 from applied manure to have increased from approximately 3 Tg N yr −1 in 1850 to 22 Tg N yr −1 in 2000, while the volatilization of fertilizer reaches 12 Tg N yr −1 in 2000.The NH 3 emission factors increase by approximately 3.4 % for manure applications and 11 % for synthetic fertilizer applications over this historical period (1930 to 2000 for fertilizer).Under cur-rent manure and synthetic fertilizer application rates we find that globally an additional 1 Tg NH 3 is emitted from the manure and synthetic fertilizer pools per degree of warming.Percentage-wise this increases manure emissions by 4 % and synthetic fertilizer emissions by 3 % per degree of temperature increase.
The NH 3 emissions appear reasonable when compared to other inventories on the global scale, as well as when compared to the local-scale measurements of manure and synthetic fertilizer (Figs. 2 and 3), although these latter comparisons highlight the difficulty in making global-scale assumptions about surface parameters and farming methodology.The biggest disagreement with the manure emission measurements is from beef cattle feedlots in Texas.On the whole the model performs best when estimating NH 3 manure emissions from cows on grassland.Despite the issues described above, this model gives reasonable NH 3 emission predictions given the limited global information available on the grazing land of agricultural animals.Measurements of nitrogen runoff from rivers heavily impacted by anthropogenic nitrogen input also compare favorably with simulated results using the river transport model within the CESM (Nevison et al., 2016).
The assumption of simplified farming practices made here may be acceptable in many locations as more complex farming methods are rarely employed in the developing world.While the Food and Agriculture Organization (FAO, 2005) suggests over 75 % of the global agricultural land uses traditional farming methods, one of the largest sources of uncertainty in this study is associated with the simplification of agricultural practices.A number of future model improvements are necessary in the next-generation model.
1.More realistic representation of manure management practices.While there is a wide range of variation in animal housing and storage practices, the unique set of emission factors entailed in animal housing and storage should be incorporated in the next-generation model.
2. A better representation of nitrogen transport throughout the soil column and the resulting NH 3 generation.This would allow a differentiation between NH 3 emissions resulting from grazing, where urine is rapidly incorporated into the soil column, vs. the emissions resulting from the spreading of manure slurry.It would also allow a representation of fertilizer injection or mixing into the soil column and the transport of nitrogen into the soil column in association with water transport.
3. Representation of NH 3 emissions from different synthetic fertilizer types.Different types of synthetic fertilizer have rather different emission factors.As shown by Whitehead and Raistrick (1990) many of these differences can be represented by the impact of the fertilizer on soil pH.
4. A full biogeochemical coupling of the FAN process model to the overall biogeochemistry within the CESM.This would allow the nitrogen introduced through agricultural practices to impact the overall model biogeochemistry and allow a more thorough investigation of the flows of agricultural nitrogen.
5. A full coupling between the NH 3 emissions represented by the FAN process model and the atmospheric chemistry model through a PFT-dependent compensation point approach.The applied nitrogen should be added explicitly to the CLM crop model or to pasture land where appropriate.
The increased use of synthetic fertilizer and growing livestock populations has increased N r flows to both the atmosphere and oceans to unprecedented levels with a marked effect on the environment.We have provided a first estimate of globally distributed temporal changes in nitrogen pathways from manure and synthetic fertilizer inputs in response to climate.This is relevant to current studies investigating the ecosystem effects of N r and, in particular, how adding synthetic fertilizer to farmland affects the ocean and the atmosphere and impacts climate.The model predicts vastly different nitrogen pathways depending on the region the inputs are applied.Scenarios predicting future synthetic fertilizer use and livestock populations suggest large increases in nitrogen added to the land surface from both sources (Tilman et al., 2001;Skjøth and Geels, 2013), with an important component sensitive to climate change.The volatilization of NH 3 increases exponentially with temperature, suggesting future emission increases are likely.However, increases in temperature may surpass the optimal temperature at which certain biological processes occur, slowing the process.Washout pathways are also likely to change, not only with climate but also with increases in nitrogen loading.Future applications of this model will investigate the tight coupling between nitrogen, agriculture and climate.The Supplement related to this article is available online at doi:10.5194/bg-13-3397-2016-supplement.

Figure 1 .
Figure1.Schematic of the addition of the FAN (flow of agricultural nitrogen) process model to the CESM nitrogen cycle.Some minor pathways are not shown.Soil nitrogen pools and plant nitrogen exist in CLM4.5.Urine nitrogen (N urine ) is directly input to the TAN pool, while fecal matter is split into three parts that decompose into the TAN pool at a rate determined by their C : N ratio (N available , N resistant , N unavailable ).Manure nitrogen that does not mineralize (N unavailable ) is added to the soil organic nitrogen pool.Nitrogen applied as synthetic fertilizer is added to the N fertilizer pool where it decomposes into the TAN pool.Losses from the TAN pool include ammonia (NH 3 ) emission (into CAM-chem), nitrogen runoff (into the RTM), nitrate (NO −3 ) formation and diffusion to the soil nitrogen pools.

Figure 2 .
Figure 2. Comparison of model to measurements for percentage of nitrogen lost as NH 3 emissions from manure for a range of studies (see TableS1).Symbol color indicates temperature at which the emissions occurred; shape gives the study.

Figure 3 .
Figure 3.Comparison of model to measurements for percentage of nitrogen lost as NH 3 emissions from synthetic fertilizer (see TableS2) for a range of studies.Symbol color gives the latitude at which the emissions occurred; symbol shape gives the study and type of synthetic fertilizer application.

Figure 4 .
Figure 4. Simulated NH 3 emissions from synthetic fertilizer application from 1995 to 2004 for the present-day control simulation.Simulated emissions (kg N ha −1 yr −1 ) as (a) an annual average, (c) January-February-March average, (d) April-May-June average, (e) July-August-September average, and (f) October-November-December average.Panel (b) shows simulated emissions as a percent of annual synthetic fertilizer application.

Figure 5 .
Figure 5.As in Fig. 4 but for manure application.

Figure 6 .
Figure 6.Simulated runoff of N r from synthetic fertilizer and manure TAN pools for the present-day (1995-2004) control simulation.Simulated runoff (kg N ha −1 yr −1 ) as an annual average for (a) synthetic fertilizer, (c) manure.Simulated as (b) percent of annual synthetic fertilizer application, (d) percent of annual manure application.

Figure 7 .
Figure 7.As in Fig. 6 but for simulated nitrification.

Figure 8 .Figure 9 .
Figure 8.As in Fig. 6 but for flux of TAN nitrogen to the soil.

Figure 10 .
Figure 10.As in Fig. 9 but for fate of TAN nitrogen by country and region.Countries are split between developed countries and developing countries.

Figure 11 .
Figure 11.Comparison of manure (red) and synthetic fertilizer (blue) NH 3 emissions or combined manure and synthetic fertilizer emissions (green) (Tg N yr −1 ) for various inventories (see text for details) and regions: (a) global, (b) China, (c) Europe and (d) US.Emissions from this study are labeled as "Riddick".

Figure 12 .
Figure 12.Site-specific pathways for nitrogen budget at 35 • N and 100 • W, near the Texas panhandle.Panels show (a) the temperature ( • C) and precipitation (mm s −1 ) used to force the CLM, (b) the manure (solid) and synthetic fertilizer TAN pools (dashed) (gN m −2 ), and the four major loss pathways (g N m 2 s −1 ) from (c) the manure TAN pool (d) the synthetic fertilizer TAN pool.Loss pathways: NH 3 emissions, red; runoff, orange; nitrification, green; and diffusion to the soil, blue.

Figure 13 .
Figure13.Applied nitrogen and nitrogen losses for the historical simulation in Tg N yr −1 for (a) manure and (c) synthetic fertilizer.Nitrogen losses from the TAN pool as a percentage of applied nitrogen for the historical simulation for (b) manure and (d) synthetic fertilizer.The losses from the TAN pool are divided into emission losses of ammonia to the atmosphere (gold diamond), runoff (green diamond) and loss to the soil.Loss to the soil is divided into that due to canopy loss (asterisk) (see Sect. 3.2.3),direct diffusive loss (cross) and nitrification (plus).
1 Control experiment. 2 Parameter changed from default values. 3New parameter value.