Plant and Soil

, Volume 309, Issue 1, pp 169–189

N2O emissions from agricultural lands: a synthesis of simulation approaches

Authors

    • School of Resource Management, Faculty of Land and Food ResourcesThe University of Melbourne
  • Yong Li
    • School of Resource Management, Faculty of Land and Food ResourcesThe University of Melbourne
  • Peter Grace
    • School of Natural Resource SciencesQueensland University of Technology
  • Arvin R. Mosier
    • School of Resource Management, Faculty of Land and Food ResourcesThe University of Melbourne
Regular Article

DOI: 10.1007/s11104-008-9634-0

Cite this article as:
Chen, D., Li, Y., Grace, P. et al. Plant Soil (2008) 309: 169. doi:10.1007/s11104-008-9634-0

Abstract

Nitrous oxide (N2O) is primarily produced by the microbially-mediated nitrification and denitrification processes in soils. It is influenced by a suite of climate (i.e. temperature and rainfall) and soil (physical and chemical) variables, interacting soil and plant nitrogen (N) transformations (either competing or supplying substrates) as well as land management practices. It is not surprising that N2O emissions are highly variable both spatially and temporally. Computer simulation models, which can integrate all of these variables, are required for the complex task of providing quantitative determinations of N2O emissions. Numerous simulation models have been developed to predict N2O production. Each model has its own philosophy in constructing simulation components as well as performance strengths. The models range from those that attempt to comprehensively simulate all soil processes to more empirical approaches requiring minimal input data. These N2O simulation models can be classified into three categories: laboratory, field and regional/global levels. Process-based field-scale N2O simulation models, which simulate whole agroecosystems and can be used to develop N2O mitigation measures, are the most widely used. The current challenge is how to scale up the relatively more robust field-scale model to catchment, regional and national scales. This paper reviews the development history, main construction components, strengths, limitations and applications of N2O emissions models, which have been published in the literature. The three scale levels are considered and the current knowledge gaps and challenges in modelling N2O emissions from soils are discussed.

Keywords

Nitrous oxide emissionsAgroecosystem modellingNitrificationDenitrificationSoil nitrogenGreenhouse gases

Introduction

N2O has been recognized as a potent and long-lived greenhouse gas (Prather et al. 1995), contributing a Radiative Forcing of +0.16 ± 0.02 W m−2 of the atmospheric greenhouse effect (IPCC 2007). The Intergovernmental Panel for Climate Change (IPCC) Fourth Assessment Report (IPCC 2007) states N2O concentration in the atmosphere continues to rise at a rate of 0.26% year−1 and as of 2005, the concentration in the atmosphere was 319 ppbv. N2O is very effective in absorbing infrared radiation and its global warming potential (GWP) is 310 times greater than CO2 for a 100-year time horizon (Ramaswamy et al. 2001). Although N2O is stable, it is susceptible to photolysis with the subsequent formation of NO linked to ozone depletion (Crutzen 1970). Anthropogenic activities contributing to N2O emissions include the application of N fertilizers, crop biological N fixation, tillage, irrigation, animal manure, aquifers, sewage, industry, automobiles, biomass burning, land clearing and trash incineration. Two thirds of N2O emissions come from soil-based processes (Prather et al. 1995). Although, on a global basis, the N2O budget is well constrained (Crutzen et al. 2008), N2O production rates from regions, landscapes, and individual fields are highly uncertain (Robertson and Grace 2003).

For the few cropping systems for which we have reliably collected data, N2O loss is frequently the major source of GWP (Robertson et al. 2000). In high input agricultural systems, N2O may contribute from 10–50% of all greenhouse gas emissions (on a carbon (C) equivalent basis). In Australia, agriculture is the second largest contributor of greenhouse gases after the energy sector, accounting for an estimated 16% of Australian total greenhouse gas emissions, of which approximately 19% are emitted as N2O from agricultural soils (Australian Greenhouse Office 2007). Globally, agricultural lands contribute about 35% of all N2O emissions (FAO/IFA 2001). The large amount of fertilizer N used in agriculture is a major contributor to the terrestrial N cycle and hence N2O emissions (Matthews 1994).

The production of N2O from soils is primarily from the microbially-mediated nitrification and denitrification processes. N2O flux from soils, relatively small compared to other N fluxes, is dependent on soil temperature, soil water content, O2 availability, N substrate availability (nitrate and ammonium), and organic C substrate availability (Davison 1991). All these regulators are strongly influenced by climate, vegetation, soil properties (bulk density, organic C, pH and clay content), and land-use management or agricultural practices. N2O production is also influenced by other complex interacting N processes in the plant–soil N cycle, such as plant N uptake, ammonia volatilization and nitrate leaching. All these influencing variables and processes contribute to high spatial and temporal variability of N2O emissions (Folorunso and Rolston 1984; Aulakh et al. 1992). Therefore it is impractical to estimate regional and global N2O emissions from agricultural lands from field measurements alone. Computational models, which simulate N2O emissions from soils by integrating all the influencing variables and interacting processes, provide a means of assessing gas fluxes at field-to-regional scales.

Focht (1974) published an N2O simulation model, based on zero-order kinetics, and since that time, models of varying complexity have been constructed to predict N2O production from nitrification and denitrification. These include NGAS (Mosier et al. 1983; Mosier and Parton 1985; Parton et al. 1988a, b), the Hole-In-the-Pipe (HIP) (Firestone and Davidson 1989), DNDC (Li et al. 1992a, b; Li 2000), Expert-N (Engel and Priesack 1993), NASA CASA (Potter et al. 1996, 1997), an un-named mechanistic model of Müller et al. (1997b), the IPCC method based on Bouwman (1996), NLOSS (Riley and Matson 2000), ecosys (Grant 2001), DAYCENT (Parton et al. 1996, 1998, 2001; Del Grosso et al. 2000), the REML-based model (FAO/IFA 2001), WNMM (Li et al. 2005, 2007), FASSET (Chatskikh et al. 2005), and CERES-NOE (Gabrielle et al. 2006). These simulation models can be classified into three categories: laboratory, field and regional/global levels.

This paper reviews most of the known N2O emissions models, which have been published in the literature, at the above three levels. This review focuses more on the field scale models as most of simulation applications were carried out at this level, and on direct N2O production processes and related environmental regulators rather than on other agronomic issues, such as plant growth and management. The main objective of this review is to summarize the history, application, strength and limitation of a range of commonly used N2O simulation models, so that researchers can make informed decisions in selecting the model which best suits their own goals.

Model descriptions

Laboratory scale

Under the assumption that N2O is an obligatory precursor in nitrate reduction to N2 in the soil, Focht (1974) developed a zero-order kinetic model for denitrification and N2O emissions from laboratory incubations. In this model, the proportion of N2O emitted in relation to N2 is not greatly affected by soil temperature changes, but is more directly related to soil aeration and pH. Although the denitrification rate increases with increasing pH and decreasing aeration, the ratio of N2O/N2 declines.

Later models (e.g. Mehran and Tanji 1974; Smith 1980; Tanji 1982; McConnaughey and Bouldin 1985; Leffelaar and Wessel 1988; Arah and Smith 1989; Arah 1990; Smith 1990) focused on the various pathways of denitrification and the diffusion of gases (O2, N2O and N2) and substrates (ammonium and nitrate) into soil aggregates, where soil denitrification primarily occur under anaerobic conditions. The Leffelaar and Wessel model (1988) described microbial growth and denitrification as reductive pathways of mineral N in a homogenous soil layer as NO3 → NO2 → N2O → N2. The model simulated the growth and maintenance of the microbial biomass in response to glucose (soluble C) addition and the concomitant reduction of nitrate to N2, via the intermediates nitrite and nitrous oxide. Growth of the microbial biomass was calculated by a first-order kinetic equation in which the relative growth rate was described by a double Monod equation consisting of rate-limiting factors for C and N substrates. The Pirt equation was used to calculate the consumption rates of substrates. Several recent field-scale N2O emission models (e.g. DNDC, NLOSS, ecosys and Expert-N) have adopted this philosophy for simulating microbial growth and denitrification from soils.

The laboratory models explicitly simulate the denitrification reaction, but only for singular incubation experiments with limited applications outside of these strict boundaries. In addition, the N2O contribution from soil nitrification is not simulated.

Field scale

In more recent times, a number of process-oriented simulation models have been developed which attempt to describe the complex interactions of the water, C and N cycles of terrestrial ecosystems, including agroecosystems. The Langeveld and Leffelaar’s review (1996) outlined several common features within the process-oriented N2O emission simulation models: soil–air atmosphere and climate interactions, plant growth, C and N cycling, and land use management. In the N cycling component of each model, the contributions of N2O from both denitrification and nitrification are estimated. These models reviewed in this paper include NGAS-DAYCENT (Mosier et al. 1983; Mosier and Parton 1985; Parton et al. 1988a, b, 1996, 1998, 2001; Del Grosso et al. 2000), DNDC (Li et al. 1992a, b; Li 2000), NLOSS (Riley and Matson 2000), ecosys (Grant 2001), Expert-N (Engel and Priesack 1993), WNMM (Li et al. 2005, 2007), FASSET (Chatskikh et al. 2005) and CERES-NOE (Henault et al. 2005; Gabrielle et al. 2006). The differences in structure and functionality of these models are summarized and listed in Table 1.
Table 1

The structure and functionality of major N2O simulation models at the field scale

Model

NGAS-DAYCENT

DNDC

ecosys

NLOSS

Expert-N

WNMM

FASSET

CERES-NOE

Time step

Daily

Daily

Seconds–centuries

Daily

Daily

Daily

Daily

Daily

Plant growth

Own

Own and MACRO

Own

CERES

CERES, SUCROS, LEACHM, PLAGEN and SPASS

EPIC, SUCROS, CERES and GRASSGRO

Own

CERES

Water dynamics

Water balance

Water balance

Richards and Green-Ampt equations

Richards equation

Richards equation/water balance

Richards equation/water balance

Water balance

Richards equation

C cycling

8 C pools

8 C pools

6 organic states, 4 organic matter-microbe complexes and 6 biological organization

8 C pools

3–7 C pools

5 C pools

6 C pools

6 C pools

Gases: CO2 and CH4

Gases: CO2 and CH4

Gases: CO2 and CH4

Gases: CO2

Gases: CO2 and CH4

Gases: CO2

Gases: CO2

Gases: CO2

N cycling

Processes: mineralization, immobilization, ammonia volatilization, nitrification, denitrification and nitrate leaching

Processes: mineralization, immobilization, ammonia volatilization, nitrification, denitrification and nitrate leaching

Processes: mineralization, immobilization, ammonia volatilization, nitrification, denitrification and nitrate leaching

Processes: mineralization, immobilization, ammonia volatilization, nitrification, denitrification and nitrate leaching

Processes: mineralization, immobilization, nitrification, denitrification and nitrate leaching

Processes: mineralization, immobilization, ammonia volatilization, nitrification, denitrification and nitrate leaching

Processes: mineralization, immobilization, ammonia volatilization, nitrification, denitrification and nitrate leaching

Processes: mineralization, immobilization, ammonia volatilization, nitrification, denitrification and nitrate leaching

Gases: NH3, NO, N2O and N2

Gases: NH3, NO, N2O and N2

Gases: NH3, N2O and N2

Gases: NH3, N2O and N2

Gases: N2O and N2

Gases: NH3, NO (not tested), N2O and N2

Gases: NH3, N2O and N2

Gases: NH3, N2O and N2

N2O emissions

Nitrification

Nitrification: first-order kinetics

Nitrification: nitrifier dynamics

Nitrification: Nitrifier dynamics

Nitrification: nitrifier dynamics

Nitrification: zero or first-order kinetics

Nitrification: first-order kinetics

Nitrification: first-order kinetics

Nitrification: first-order kinetics

N2O: fixed proportion (2%)

N2O: fixed proportion (0.25%)

N2O: dynamic

N2O: fixed proportion (0.25%)

N2O: fixed proportion (0.5%)

N2O: fixed proportion (0.1–0.5%)

N2O: fixed proportion (calibrated)

N2O: fixed proportion (calibrated)

Denitrification

Denitrification: WFPS threshold driven and first-order kinetics

Denitrification: denitrifier dynamics and “anaerobic balloon” driven

Denitrification: denitrifier dynamics

Denitrification: denitrifier dynamics and “anaerobic balloon” driven

Denitrification: WFPS threshold driven and first-order kinetics

Denitrification: WFPSthreshold driven and first-order kinetics

Denitrification: WFPS threshold driven and first-order kinetics

Denitrification: WFPS threshold driven and first-order kinetics

N2O/N2O/N2 ratio division

N2O: dynamic

N2O: dynamic

N2O: dynamic

N2O/N2O/N2 ratio division/first-order kinetics

N2O/N2O/N2 ratio division

N2O/N2O/N2 ratio division

N2O/N2O/N2 ratio division

Gas diffusion

Soil diffusivity based on soil texture

Diffusion proportion

Dynamic

Dynamic

Dynamic

No

No

No

Source code

Availability

By request

By request

No

By request

No

By request

By request

By request

Language

C++

C++

FORTRAN

FORTRAN

FORTRAN and Visual Basic

Visual Basic

C++

FORTRAN

Landuse

Crops, pastures and forests

Crops, pastures and forests

Crops, pastures and forests

Crops

Crops

Crops and pastures

Crops and pastures

Crops

Applications

USA, Canada, Australia, New Zealand and Europe

USA, Canada, Australia, New Zealand, Europe, China and India

USA and Canada

Mexico

Germany, UK, USA and Canada

China, Australia, Korea and Mexico

Europe

France

NGAS-DAYCENT

The development of NGAS can be traced back to early N2O simulation efforts by Mosier et al. (1983), Mosier and Parton (1985) and Parton et al. (1988a, b), in which a simple mechanistic model accounting for the effects of soil water, nitrate and ammonium contents was developed to predict daily N2O losses from semi-arid grasslands and irrigated soils in northeastern Colorado, USA.

NGAS-DAYCENT (Parton et al. 1996, 1998, 2001; Del Grosso et al. 2000) is the daily-time-step version of the CENTURY ecosystem model (Parton et al. 1988a, b, 1994) which can simulate trace gas fluxes of NO, N2O and N2 from soils as well as terrestrial CH4 formation/oxidation. The finer time scale is used in DAYCENT because trace gas fluxes are often short term, episodic events in response to rainfall, snow melt or irrigation. Trace gas emissions have also been found to respond to changes in soil water levels in a non-linear manner, hence the need for fine temporal scales. DAYCENT and CENTURY both simulate exchanges of C, N, P and S among the atmosphere, soil, and plants and use identical files to simulate plant growth and events such as fire, grazing, cultivation, harvest, and organic matter or fertilizer additions. In addition to modeling decomposition, nutrient flow, soil water and soil temperature on a daily time scale, DAYCENT also uses increased spatial resolution for soil layers. DAYCENT includes submodels for plant productivity, decomposition of dead plant material and soil organic matter (SOM), soil water and temperature dynamics, and trace gas fluxes.

In DAYCENT, flows of C and N are controlled by the amount of C in the various pools (plant, SOM and microbial biomass), the N and lignin concentrations of the pools, abiotic factors (soil temperature and soil water content) as well as soil physical properties. SOM is divided into three pools based on decomposition rates (Parton et al. 1994). The decomposition of SOM and external nutrient additions supply a mineral N pool that is available for plant and microbial uptake and therefore subject to gaseous loss. The land surface sub-model of DAYCENT simulates water flow and evapotranspiration for the plant canopy, litter and soil profile, as well as snow melting and soil temperature through the profile.

In the nitrification submodel, N2O emissions are simulated as a function of soil ammonium content, soil water content, temperature, pH and soil texture (Parton et al. 1996). Nitrification is limited by moisture stress when soil water-filled pore space (WFPS) is too low and by O2 availability when WFPS is too high. The optimum WFPS for nitrification is around 55%, with a higher optimum for clay than sandy soils. N2O emissions from nitrification are estimated using a fixed fraction of the soil nitrification rate, e.g. 2%.

The denitrification submodel simulates N2O and N2 emissions as a function of soil nitrate, soil water content, labile C availability, and soil physical properties related to texture that influence gas diffusion rates (Del Grosso et al. 2000). Denitrification (N2O + N2) is triggered when soil WFPS exceeds a texture-related threshold, and then it increases exponentially as WFPS increases and levels off as the soil approaches saturation. Simulated heterotrophic respiration rate from organic matter decomposition components is used as a proxy for labile C availability. The N2/N2O ratio, as a function of soil water content, soil nitrate content, and soil heterotrophic respiration rate, is used to separate total denitrification into N2O and N2.

In DAYCENT, NO emission is also calculated using total N2O emissions from nitrification and denitrification, a NO/N2O ratio function based on soil gas diffusivity, and a pulse multiplier based on rainfall frequency and amount (Parton et al. 2001). As soil gas diffusivity decreases, a smaller proportion of total N gas fluxes are assumed to be in the form of NO. The pulse multiplier equations were developed by Yienger and Levy (1995) and account for the observed high NO emission rates following precipitation events onto soils that were previously dry.

Input data for DAYCENT includes daily weather variables (maximum/minimum air temperature, precipitation, solar radiation, relative humidity (optional), wind speed (optional)), site-specific soil properties (mechanical composition, bulk density, soil water contents at wilting point, field capacity and saturation, and saturated soil hydraulic conductivity), and current and historical land use.

For the simulation of N2O emissions from soils, the NGAS-DAYCENT model is relatively simple and more empirical, compared to more detailed ecosystem models (Grant and Pattey 2003). In USA systems, it has been shown to accurately simulate mean annual N2O emissions, however its ability to replicate daily emissions is less reliable (Parton et al. 1996, 2001; Del Grosso et al. 2000). In urine-rich pastures of New Zealand (Stehfest and Müller 2004) the model overestimated annual N2O emissions by 318% and nitrification was severely underestimated. Li et al. (2005) used NGAS coupled with WNMM to simulate N2O emissions from an intensive wheat–maize cropping system on loam-textured arable soils in the North China Plain. The results indicated that the original ammonium effect function on nitrification in NGAS severely restricts the nitrification process in these soils. The removal of this specific function improved the predictability of daily and annual N2O emissions by NGAS in comparison with DNDC and WNMM gas modules. DAYCENT has been used to simulate national N2O emissions in USA from major cropped soils (Del Grosso et al. 2006) and is used in the “Inventory of U.S. Greenhouse Gas Emissions and Sinks: 1990–2005” (USEPA 2007).

DNDC

DNDC (DeNitrification–DeComposition) is a complex simulation model which describes many of the C and N cycling processes in soils (Li et al. 1992a, b, 1994, 1996; Li 2000). It was specifically developed to predict daily N2O fluxes through the nitrification and denitrification pathways, CO2 production from decomposition of organic matter and root respiration, as well as anaerobic CH4 production within agro-ecosystems. The DNDC model consists of two components. The first component, includes soil climate, crop growth and decomposition submodels, predicts soil temperature, water content, pH fluctuation, redox potential (Eh), and substrate concentration profiles based on ecological drivers (climate, soil properties, vegetation and anthropogenic activity). The second component, consists of nitrification, denitrification and fermentation submodels, predicts NO, N2O, CH4 and NH3 fluxes based on soil environmental variables.

The soil climate submodel of DNDC requires daily meteorological data to simulate hourly soil temperature, water content and Eh profiles, soil water vertical flux and plant water uptake. The decomposition submodel divides SOM into four C pools: litter, labile humus, passive humus, and microbial biomass, following that of Molina et al. (1983). Each pool has a fixed C:N ratio with a first-order decomposition rate, modified by soil texture, temperature and water functions and mineral N limitations. Ammonium-N is nitrified to nitrate with loss of NO and N2O to the atmosphere, or via plant uptake, leaching, transformation to ammonia (and subsequent volatilization) or adsorption onto soil clay minerals. N immobilization for soil mineral N pools occurs when C is transferred from high C:N ratio pools to low C:N ratio pools. The soluble C level, which provides for denitrification, is related to the fraction of C re-released by the decomposition of litter, labile humus and dead microbial biomass that is re-assimilated in microbial biomass each day. The decomposition submodel operates on a daily time step.

The hourly-time-step denitrification submodel of DNDC is activated by three conditions: rain events (when soil water content increases and or soil O2 availability decreases), flooding/irrigation, and cold temperatures (new feature). Air temperature below −5°C is assumed to freeze the soil and thus inhibits O2 diffusion into soils. The DNDC model simulates relative growth rates of nitrate, nitrite, NO and N2O denitrifiers based on soil Eh, pH, dissolved organic C (DOC), and N oxides. An innovative concept called ‘anaerobic balloon’ was developed in the model to divide the soil matrix into aerobic and anaerobic parts. Tracking O2 diffusion and consumption in the soil profile, DNDC simulates swelling and shrinking of the anaerobic balloon. Only the substrates located in the anaerobic zone are engaged in the denitrification process.

Following Leffelaar and Wessel (1988), DNDC assumes that the relative growth rates of denitrifiers with different substrates are independent, and that competition among the bacteria takes place via the common DOC substrate. Denitrification is simulated via the basic laws of sequential chemical kinetic reactions to calculate NO, N2O and N2 fluxes (NO3 → NO2 → NO → N2O → N2). A fraction of the pool is directly converted to N2O. A simplified equation derived from laboratory data is used in the DNDC model to calculate diffusion rates of NO and N2O in the soil matrix. The predicted diffusion rate for a gas is a function of soil porosity, soil water content, soil temperature, and soil clay content. The exchange of N gases between soil layers is not simulated.

The DNDC model has been widely used in many countries for estimating both site and regional N2O emissions from agricultural lands (Li 1995; Li et al. 1996; Plant 1999; Wang et al. 1997; Zhang et al. 2002; Brown et al. 2002; Xu-ri et al. 2003; Cai et al. 2003; Smith et al. 2004; Saggar et al. 2004; Pathak et al. 2006), with some site-specific modifications.

NLOSS

Riley and Matson (1998, 2000) developed the NLOSS model (Nitrogen LOsses from Soil Systems) to simulate N losses from agricultural systems and to examine impact of management systems on N use efficiency and losses. The NLOSS model predicts soil water dynamics, soil C and N transformations, crop growth and N uptake, N leaching loss, and the production and efflux of NO, N2O and N2 (induced by nitrification and denitrification). The model uses a finite-difference method to simulate water, heat and vapour transport in soil, and estimates energy transfer across the soil-atmosphere boundary using either a Penman-Monteith or Kondo approach. Crop growth is simulated using selected sub-routines of the CERES model (Ritchie 1991). Soil C cycling is based on the decomposition submodel of DNDC (Li et al. 1992a, b; Li 2000).

NLOSS simulates decomposition using three organic matter pools: residues, microbial biomass, and humus. These pools are further divided into labile and resistant fractions. The decomposition rate constant for each pool is a function of the pool’s potential decomposition rate, soil temperature, and soil water content. Soil N cycling is tied to C transformations. Only advection of soluble N is considered to be available for N transport. Although few comparative applications are in the literature, NLOSS is very similar to the DNDC model.

ecosys

ecosys (Grant 2001) is a comprehensive mathematical model for natural and managed terrestrial ecosystems to simulate ecosystem behaviour under different environmental conditions (soils, climates, and land use practices). It integrates temporal scales from seconds to centuries, with validation data ranging from short-term laboratory incubations to long-term field studies, and multiple spatial scales (i.e. millimeter to kilometer), allowing representation of complex biomes. The ecosys model simulates the transport and transformations of water, heat, C, O2, N, P, and ionic solutes through soil–plant–atmosphere systems with the atmosphere as the upper boundary and soil parent material as the lower boundary. A full range of land use management practices, including tillage, fertilization, irrigation, planting, and harvesting, are simulated to allow this model to differentiate the impacts of a wide range of disturbances on ecosystem functions. The ecosys model is constructed in such a way that all flux equations are solved in three dimensions for each cell of a matrix defined by row, column, and layer position, and therefore it addresses the need to scale high-resolution ecosystem processes up to the landscape level for the issues of soil productivity and atmospheric gas exchange.

ecosys comprises seven submodels (Grant 2001): ecosystem–atmosphere energy exchange, canopy water relations, canopy C fixation, canopy respiration and senescence, nutrient uptake, plant growth, and soil microbial activity. Production of gaseous C and N products is associated with the microbial activity submodel. The concentration of the soluble hydrolysis products and C oxidation by the decomposer population is also constrained by O2 and nutrients. This oxidation is coupled to the reduction of O2 by all aerobic populations, to the sequential reductions of nitrate, nitrite and N2O by heterotrophic denitrifiers (Grant et al. 1993a, b; Grant and Pattey 1999, 2003), and to the reduction of organic C by fermenters and heterotrophic methanogens. All soluble and gaseous reductants and products undergo convective–dispersive transport through the soil profile. In addition, autotrophic nitrifiers conduct ammonium oxidation and nitrate production, and N2O evolution (Grant 1995). Autotrophic methanogens and methanotrophs conduct CH4 reduction and oxidation, the energetics of which determine autotrophic growth yields and hence autotrophic biomass activity. Microbial populations in the model seek to maintain equilibrium ratios of biomass C/N/P by mineralizing or immobilizing mineral N and P, thereby regulating solution concentrations used to drive N and P uptake by roots and mycorrhiza. Microbial populations undergo first-order decomposition, the products of which are partitioned between humus and microbial residue according to a function of soil clay content.

Compared to other models, ecosys does have a relatively large input data requirement, e.g. hourly or daily climate variables, site geographic information, soil properties by layers, plant characteristics (descriptors, CO2 fixation kinetics, phenology, morphology, grain characteristics, root characteristics, plant water relations, organ characteristics and temperature sensitivity) and land use management (tillage, fertilizer application, irrigation and plantation).

Whilst ecosys is regarded as highly mechanistic compared to other ecosystem simulation models, it has the capability of simulating N2O emissions from soils at any temporal and spatial scales. However, the parameterization of such an extremely comprehensive model is very difficult for inexperienced users, and thus, its application is limited.

Expert-N

Expert-N is an integrated modular model for simulating water, C and N dynamics in the soil–plant–atmosphere system on a daily time step (Engel and Priesack 1993; Priesack et al. 2001). The modules describe soil water flow, soil heat transfer, soil C and N turnovers (including N2O emissions from nitrification and denitrification), crop growth and land use management. Sub-modules of distinct algorithms can be selected to model each process. Expert-N utilizes published concepts and algorithms from the literature and other detailed models (Hutson and Wagenet 1992; Ritchie 1991; Simunek et al. 1998; van Laar et al. 1992; Nicolardot and Molina 1994; Hansen et al. 1991; Stenger et al. 1999; Priesack et al. 2001). Expert-N is relatively unique in allowing user-developed algorithms to be easily incorporated into the model (Kaharabata et al. 2003).

Daily soil water content and flux are solved by either the tipping-bucket approach or one-dimensional Richards equation. Crop growth is simulated using either an empirical growth model of LEACHN (Hutson and Wagenet 1992) or CERES (Ritchie 1991) models. The processes of soil C and N transformations involving decomposition, N mineralization/immobilization, urea hydrolysis, nitrification and ammonia volatilization are simulated according to either the LEACHN or CERES models. The Expert-N denitrification process follows the concepts of Johnsson et al. (1987), but is modified to estimate the production of N2O governed by half saturation kinetics with respect to nitrate, and the subsequent reduction from N2O to N2 by zero- or first-order kinetics. N2O production from nitrification is accounted for by assuming that induced-N2O levels represent a fixed fraction of the nitrified ammonium. The convection–diffusion equation is used to model soil solute (urea, ammonium and nitrate) and gas (O2, CO2, N2O and N2) transport within the soil layers.

Expert-N requires detailed soil properties that are segmented into at least three soil horizons or layers, detailed crop properties during the growing season, pooled or carry-over soil C and N from the previous year, and detailed daily meteorological data (Engel and Priesack 1993). Users generally need an advanced knowledge of agroecosystem modelling to apply this type model because module selection is critical. It has been applied to several studies to simulate N2O emissions from USA, Scottish and German soils (Frolking et al. 1998) and Canadian soils (Kaharabata et al. 2003).

FASSET

The FASSET soil–plant–atmosphere model simulates N turnover, including nitrification/denitrification and N2O emissions as well, and crop production affected by daily weather and availability of water and N (Olesen et al. 2002).

In FASSET, the growth of grass-clover and animal grazing are simulated according to Berntsen et al. (2005). The soil submodel has a one-dimensional vertical structure and simulates daily movement and plant availability of water and N. The soil profile is divided into horizontal layers, and the transport of water and N between the soil layers is calculated using the concepts of the Solute Leaching Intermediate Model, SLIM (Addiscott and Whitmore 1991).

According to Chatskikh et al. (2005), the submodel of soil denitrification and N2O emission in FASSET is based on the HIP scheme (Davidson et al. 2000), where the N intermediates from the nitrification and denitrification processes are assumed to be sources for N2O production. Basically, two steps are used to calculate N2O emission. First, for each soil layer the N2O production potential is determined from the simulated nitrification and mineralization by applying semi-empirical functions to estimate effects of environmental factors (e.g. soil temperature, water content and clay content). Second, the potential N2O emission is divided into N2 and N2O emission using semi-empirical relations to estimate the effects of soil physical properties and soil diffusion (layer depth) on the efficiency by which denitrification reduces N2O to N2. The actual N2 and N2O emissions are then calculated by summing the contributions from each soil layer.

FASSET has been used to estimate year-round N2O emissions from Danish, Finish and UK soils (Chatskikh et al. 2005). The model predicted daily N2O fluxes fairly well and responded to fertilizer applications and rainfall events. However, most of the large measured daily N2O peaks were not matched with the FASSET simulations. Other concerns related to this model are that the function parameters describing the impacts of soil temperature, water, mineral N, clay and depth on nitrification, denitrification and N2O emissions were derived from site-specific field and laboratory experiments. It may lead to the need for local calibrations when the FASSET model is applied on different soils, landuse, or climatic conditions.

WNMM

The Water and Nitrogen Management Model (WNMM) (Li et al. 2005, 2007) is a spatially referenced biophysical model coupled with a Geographic Information System (GIS) to simulate the temporal dynamics of soil water and soil-crop C and N cycling. Its was specifically developed to identify strategies for optimal management of water and fertilizer N application in intensive cropping systems, initially for the wheat and maize rotation in the North China Plain.

WNMM simulates the key processes of water and C and N dynamics in the surface and subsurface of soils, crop growth and agricultural management practices using a daily time step. Data required by WNMM are input as GIS layers (soil type, land cover, and administrative boundary); database-formatted source data (soil physical and chemical properties, land use types, and agricultural management practices); referenced data (climatic reference data and crop biological data); and control data (starting date, period of simulation, initial land surface and soil conditions, and agricultural management scenarios). The first two data categories are needed to convert to ARC GRID ASCII format from other formats and sources in the GIS environment.

N2O loss via the nitrification pathway is estimated as a function of the nitrification rate and WFPS. Denitrification is a function of soil temperature, soil water content and soil organic C content, and its main products are N2O and N2. Currently, two options of threshold of the fraction of WFPS for initiating denitrification are used: constant 0.80 and the ratio of water content at field capacity to water content at saturation. It assumes that denitrification occurs only within the 20 cm topsoil. Emissions of N2O from the denitrification process are estimated using the Xu et al. (1998) approach under saturated and unsaturated conditions. The ratio of N2O to N2 produced is fully controlled by soil saturation status. Gas diffusion between soil layers is not simulated, but the fraction of gases diffused from given layers to soil surface is predicted using the DNDC method by considering clay content, WFPS and soil temperature. In the latest version of WNMM, N2O emissions from soils can be predicted using its own N gas module, or the DNDC dynamic microbial growth approach, or DAYCENT empirical approach, or the FASSET approach.

The WNMM applications in two loam-textured arable soils in the North China Plain under intensive wheat–maize cropping (Li et al. 2005) indicated that the three N2O production modules employed in WNMM generated comparable results of N2O emissions from soils. The simply-structured WNMM gas module appeared superior compared to the DNDC and DAYCENT gas modules for the light texture soils. Using WNMM, Li et al. (2008) reported that on a rain-fed, wheat-cropped soil in southeastern Australia the nitrification-induced N2O emissions accounted for 34–45% of annual N2O emissions, and that annual N2O emissions were significantly correlated with annual N rate, annual rainfall and annual average maximum air temperature.

CERES-NOE

The CERES family of crop models include a relatively simple soil N cycling routine with a single humus pool (Godwin and Jones 1991). Gabrielle et al. (2006) replaced the original soil N module with the NCSOIL model (Molina et al. 1983). NCSOIL contains three organic matter pools, each with a fixed potential decomposition rate, with C and N recycling into the microbial biomass. In addition, a N2O emission (NOE) module (Henault et al. 2005) was included to estimate nitrification and denitrification as well as their N2O contributions. NOE is a semi-empirical algorithm that simulates the production and reduction of N2O in agricultural soils through both the denitrification and nitrification pathways. The denitrification component of NOE is based on NEMIS (Henault and Germon 2000), a model that expresses total denitrification of soil nitrate as the product of a potential rate with three dimensionless factors related to soil water content, nitrate content, and temperature. The fraction of denitrified nitrate that evolves as N2O is then considered as constant for a given soil type. In a similar fashion, nitrification is modelled as a Michaëlis–Menten reaction. The nitrification rate is similarly modified by dimensionless factors related to soil water content and temperature. The quantity of N2O produced during nitrification is calculated as a fraction of the nitrification rate, which is further modified by the fraction constant used in the denitrification process for reducing N2O to N2 when the soil is partially anaerobic (e.g. WFPS > 0.62).

Gabrielle et al. (2006) applied CERES-NOE to simulate N2O emissions from three contrasting soils under conventionally-managed wheat in the Beauce region of Central France from 1998–1999. The simulation results indicated that CERES-NOE performed more favourably when predicted N2O fluxes for three soils compared with CERES-NGAS (i.e. CERES coupled with the NGAS module).

NOE is a relatively simple model in estimating soil N2O emissions, similar to both WNMM and FASSET, and only simulates the top 20 cm of the soil profile. NOE requires five site-specific parameters to characterize soil microbial activities for N2O emissions: the potential denitrification rate, slope and intercept of the linear regression between soil nitrification rate and soil gravimetric water content, and proportions of nitrified and denitrified N emitted as N2O, which can be measured according to the procedures defined by Henault and Germon (2000) and Garrido et al. (2002). NOE may have potential to be used for regional N2O estimation (Henault et al. 2005).

Regional scale

The high temporal and spatial variability associated with N2O production make the regional assessment of annual N2O emissions from soils a challenging task (Groffman et al. 2000). The estimates of annual N2O emissions from soils are not only related to the fertilizer N applications (IPCC 2006), but also a function of soil management, cropping systems and precipitation (Smith et al. 1998). A number of methods have been developed to estimate annual N2O emissions at the regional scale, scaling up from the site to the landscape. Some studies estimate annual N2O emissions by extrapolating site/field measurements to larger areas using regression and scaling approaches (Kroeze 1994; Clayton et al. 1994; Reiners et al. 1998; Corre et al. 1996, 1999; Sozanska et al. 2002; Lilly et al. 2003; Freibauer 2003); some apply empirical models relating N2O emission flux to few environmental and land management variables (Conen et al. 2000); and some use mechanistic models to produce more refined extrapolations by assuming that rates of gross N mineralization account for major differences in N2O emissions among ecosystem types, and the variability of the emitted N2O/NO ratio over large spatial gradients is closely related to seasonal patterns of soil wetting and drying (Firestone and Davidson 1989; Davidson et al. 2000; Potter et al. 1996, 1997; Müller et al. 1997b; Henault et al. 2005). These models were developed from field level flux information and ancillary data of gross and net N mineralization and nitrification rates, as well as key soil, climate and management data. GIS is now a well developed tool for assembling key input data for estimating N2O emissions from agricultural land at large scales.

Regression models

Clayton et al. (1994) developed a regression model to relate N2O flux from non-grazed grassland to air temperature, soil nitrate content and recent rainfall events.

Sozanska et al. (2002) established a regression model of N2O emissions from British soils through multivariate regression analysis of 59 field measurement data from temperate climates in Europe and the USA:
$$\ln \left( {{\text{N}}_2 {\text{O}}} \right)\left( {{\text{kg}}\,{\text{N}}\,{\text{ha}}^{ - 1} {\text{year}}^{ - 1} } \right) = - 2.7 + 0.60\,\ln \left( {{\text{N}}\,{\text{input}}} \right)\left( {{\text{kg}}\,{\text{N}}\,{\text{ha}}^{ - 1} \operatorname{y} ^{ - 1} } \right) + 0.61\,\ln \left( {{\text{WFPS}}} \right)\left( \% \right) + 0.035\left( {{\text{soil}}\,{\text{temperature}}} \right)\left( {^\circ {\text{C}}} \right) - 0.99\left( {{\text{land}}\,{\text{use}}} \right)$$
in which landuse type, N inputs to soils (applied plus atmospheric deposition), WFPS and soil temperature at a depth of 30 cm are considered to be the four main controlling factors of N2O emissions. As a result, this simple model, based on the estimates of four-season N2O emissions, offered a first attempt to produce a spatial inventory of N2O emissions from agricultural soils for every 5 km2 grid in ARC/INFO GRID format in Great Britain.

Freibauer (2003) presented a detailed methodology compatible to the Guidelines of IPCC to assess the annual direct biogenic emissions of N2O released from European agriculture for 1995. Using stepwise multivariate linear regression analysis, empirical models of N2O emissions were established to allow a regionally disaggregated estimate of N2O emissions at the subnational, national and continental scales in the major climate regions of Europe. In arable soils of the temperate oceanic climate zone, N2O emissions from agricultural soils are estimated as a function of the N inputs from inorganic and organic N fertilizers, the organic C and the sand content in the topsoil. In arable soils with severe frost, N2O emissions are estimated using fertilizer N inputs and the total N content in the topsoils as key variables. For European grasslands, N2O emissions are estimated using a fertilizer-based approach similar to the Bouwman (1996) method. For farmed organic soils, two constant N2O emission rates (7 and 20 kg ha−1 year−1) are applied to grasses and cereal crops, and vegetables and root crops, respectively. Compared with the existing methods for large scale inventories (IPCC 1997; Koch et al. 2000; Brown et al. 2001), the regression models provide a better regional fit to measured gas flux data because they integrate the key driving variables controlling N2O emissions (water, temperature and substrate availability) as determined by soil C and N contents, and sand content (as a surrogate in predicting anaerobiosis).

Scaling models

Although it is well known that N2O emissions from soils occur via the microbially-mediated processes of nitrification and denitrification, the mechanisms of how environmental factors control these processes are very complicated. In scaling the very detailed, field-based measurements of N2O emissions to regional scales, it is necessary to identify simple attributes that reflect the dominant processes which control N2O emissions from the soil and for which spatial data are available (Lilly et al. 2003). These simple attributes are typically topography, land use, soil type and climate.

Topography plays a very important role in N2O emissions as it regulates the regional water regime through landscapes. Mosier et al. (1983) reported that the mean annual N2O fluxes for a swale are much higher than for a midslope location in a shortgrass steppe landscape in Colorado, USA. Reiners et al. (1998) divided a pasture landscape in Costa Rica into three slope positions (hilltop, mid-slope and swale) based on topographic terrain analysis by GIS, and estimated regional N2O fluxes by multiplying N2O emission rates observed at these three slope positions by their areal fraction. The result of this study indicated that hilltops, comprising 26% of the land area, contributed only 11% of the total N2O flux. In contrast, swales contributed 38% of total N2O emissions from soils, but comprised only 20% of the land area. Corre et al. (1999) also illustrated the relationship between topography and N2O fluxes in Saskatchewan, and used soil texture and land use to estimate annual N2O fluxes over a large area.

It is well known that soil water content is one of the main environmental factors controlling N2O emissions from soils (Davison 1991). Land use is also important controlling factor because it largely determines the N fertilizer inputs and management practices. Over the long term, SOM content is also determined by land use. Lilly et al. (2003) demonstrated, with intensive N2O measurements, that a spatial and temporal scaling model of annual N2O emissions from the field to the regional scale in two areas in Scotland can be developed using land use and soil wetness as the two key factors. They divided land use into nine types, with six levels of soil wetness in each land use type and created 54 potential N2O emission classes. Annual emission values were assigned to each class to generate annual N2O fluxes using a 100 m grid resolution, in which the large, localized hotspots of emissions were identified.

The current IPCC methodology for producing national inventories of N2O from agricultural land assumes a default emission factor (EF) of 1.25% (IPCC 1997) or 1% (IPCC 2006) of all N added to the soil, based on Bouwman (1996) and Bouwman et al. (2002). Agricultural N2O emissions are assumed to be derived from three principle sources: direct emissions from soil N (applied organic and inorganic N forms, N deposited by grazing animals, mineralization of crop residues, biological N fixation and the cultivation of high organic content soils), emissions from animal waste management systems, and indirect emissions from N lost to the agricultural system through leaching, runoff, or atmospheric deposition. For each of the three sources there are a number of sub-components with N input derived from national statistics data together with default estimation values. For example, direct emissions from soils and the N inputs are based on activity data which include livestock numbers, area of major crops, N fertilizer use, crop yields, the area of N-fixing crops, the N content of leguminous and non-leguminous crops, and the amount of N excreted by animals. Following the summation of N in each component of the sector, and the subtraction of the N that is emitted as ammonia and NOx, an EF is applied. Each sector has at least one EF, which estimates the proportion of the total N input that is emitted as N2O. EFs have a defined default value and a specified range, derived by IPCC from published information. The IPCC EFs are essentially production-based and do not account for climate, landuse management practices, soil types, and other controlling variables. Apart from EF1 (direct emissions from soils), there is also a large uncertainty in determining the fraction of fertilizer and manure N lost by leaching and surface runoff and the faction of N2O loss, EF5. Compared with UK-derived inventories, Brown et al. (2001) suggested that the IPCC methodology may overestimate N2O emissions from agricultural land. An over-estimation of N2O emissions from soils has also been found by many EU Member States using the IPCC methodology (Freibauer 2003).

Empirical models

Conen et al. (2000) developed an empirical model of daily N2O emissions from intensive agricultural soils in a temperate climate in Scotland, based on the relationship between N2O and three directly-measured soil parameters: soil mineral-N (ammonium and nitrate) content in the topsoil (0–10 cm), soil WFPS and soil temperature.

This model is based on the boundary line approach, which describes the upper limit of a response that can be expected over a range of values for a driving parameter when no other parameters are limiting. In this case the driving variables are WFPS, soil temperature, and the mineral N content in the topsoil. Emissions of N2O were categorized into low (1–10), medium (10–100), and high (100–1,000 g N ha−1 day−1) emission ranges. Emission estimates tended to be within the low range and increased to medium or high only after N fertilizer application, but dependent on temperature or WFPS limitations. This model gave satisfactory predictions of seasonal N2O fluxes from grasslands as well as cereal and oilseed rape crops over a wide flux range (<1 to >20 kg N ha−1 year−1), but failed in potato and broccoli crops due to either low WFPS/soil temperature combinations or other factors becoming important, e.g. increased availability of labile organic C after harvest and sheep grazing. In general, this model can be regarded as a relatively simple model to apply to soil data commonly acquired in conventional studies of crop response to N fertilizers, to estimate the likely scale of associated N2O emissions from grasslands and arable lands except for vegetable crops.

If estimates of these three soil parameters are available on a daily basis, this model may be suitable for producing a spatial distribution of daily N2O emissions. It has the potential to be applied as a submodel in more detailed process-oriented models for estimating N2O fluxes from agricultural soils at field and regional scales.

Mechanistic models

The mechanistic models herein refer to detailed conceptualizations and algorithms specifically developed for simulating soil N2O production (in contrast to whole ecosystem simulation).

The HIP model (Firestone and Davidson 1989) depicts N gas fluxes as by-products of gross inorganic N fluxes, a direct result of the soil microbial activity regulated by soil environmental conditions and which in turn: (a) control nitrification and denitrification processes dictating N fluxes through the ‘process pipe’, (b) control the partitioning of N gases via the size of holes in the pipe through which NO, N2O or N2 leak, and (c) control the diffusion of trace gas across the aqueous-air interface. Gaseous flow of N through the ‘process pipe’ is analogous to the rate of soil N cycling.

The HIP concept has been used in the daily time step version of NASA CASA (Carnegie–Ames–Stanford approach) model (Potter et al. 1997). NASA-CASA simulates seasonal patterns in C fixation, nutrient allocation, litterfall and soil N mineralization, net CO2 exchange and soil NO, N2O and N2 gaseous production. These processes are driven by the spatial data sets of climate, radiation, and a remotely sensed vegetation index, which are used for scaling up to regional estimates of vegetation production which can drive N trace gas fluxes. However, the default value of 2% of total mineralized N as the potential loss of either N2O-N or NO-N is empirical and tends to overestimate emissions in most agroecosystems.

As mentioned earlier in this review, Henault et al. (2005) produced a detailed N2O emission algorithm, NOE, for simulating the daily emission of N2O from agricultural soils. This model was tested using a database of 64 N2O fluxes measured on the field scale with corresponding environmental parameters collected from five agricultural soils in France. For 80% of the tested points, measured and simulated fluxes agreed. NOE can be coupled with agroecosystem models to simulate N2O emissions from cropping soils at the field scale, e.g. CERES (Gabrielle et al. 2006) and provided site specific biological parameters are available, would appear to be a suitable candidate for predicting mean annual N2O emissions at the regional scale.

KNOM, a process-oriented model, was developed to explain and predict daily N2O emissions from urine-affected intensive grassland in New Zealand (Müller et al. 1997a, b). The two main processes for N2O emissions, nitrification and denitrification, were modelled separately with functions based on Michäelis–Menten kinetics. The relevant Michäelis–Menten parameters of each process were estimated from soil temperature, soil water content and soil mineral N concentrations. The N2O emission model was coupled with GIS, specifically ARC/INFO, to carry out a regional N2O emission prediction for a 10 × 10 km area near Lincoln University, Canterbury, New Zealand (Müller et al. 1997a). The maximum data resolution was 25 m. The predicted N2O emissions were consistent with values expected for this region and for that time of the year.

Global scale

A monthly version of the expanded NASA CASA biosphere model (Potter et al. 1996) was applied to estimate the global N2O emissions from agricultural land on a 1° × 1° resolution. Except for the monthly time step, the functionality of this model is similar to the daily version of NASA CASA (described above in the section for regional scale simulations), in which N trace gas fluxes are a fixed fraction, i.e. 2% of the gross N mineralization of the decomposition of litter and SOM. In this model, it is hypothesized that N mineralization and soil water dynamics are the two major soil processes controlling N trace gas fluxes from soils. Using a gridded global database of climate, radiation, and remotely sensed vegetation index in satellite image format, the model predicted 6.1 Tg N year−1 of global annual N2O emissions from the soil surface, including 2.1 Tg N year−1 from agricultural lands.

The IPCC methodology (IPCC 1997, 2006) can also be used to produce the global estimates of N2O emissions from agricultural land. However, the most obvious weakness of this approach is the singular use of the emission factors of 1.25% or 1% of added fertilizer N without considering the differences of soil types and land management as well as climate.

Bouwman et al. (1993) developed a simple global model of the N2O production potential in natural soils. The model analyzed the relative importance, both geographically and seasonally, of the different controls on N2O production at the global scale, based on the hypothesis that N2O emissions are directly related to the rate of N cycling through the soil–plant–microbial system. Five major controls on N2O production from the top 30 cm of soil were included: (a) input of organic C, (b) soil fertility, (c) soil water status, (d) temperature, and (e) soil O2 status. Indices for these controls were derived from the global grided (1° × 1° resolution) database of NDVI, soil type, soil texture and climate in the rank of 1–10 or 1–5. Calibrated with available field data, the model produced a monthly N2O production index and estimated the global annual N2O emissions, 6.9 Tg N year−1 from natural soils. However, the global N2O emissions from agricultural soils were not specified in this approach.

FAO/IFA (2001) used 896 field measurements of N2O emissions from agricultural lands from 139 studies around the world to develop a model for calculating global N2O accounting for the main emission controlling factors, by applying the Residual Maximum Likelihood (REML) technique rather than the commonly used multiple linear regression approach. The REML is considered more appropriate for handling problems of controlling factors in case of scant measurement and ancillary data or limited spatial and temporal coverage. The controlling factors selected in this approach include: climate, crop type, fertilizer type, application rate, method and timing of application, soil organic C and N content, soil pH, soil texture and drainage, measurement technique, frequency of measurements and length of the measurement period. As a result, only those factors having a significant influence (Wald statistic, p < 0.005) on N2O emissions were included in the models: environmental factors (climate, soil organic C content, soil texture, drainage and soil pH), management-related factors (N application rate and crop type), and factors related to the measurements (length of measurement period and frequency of measurements). The REML-based model was used to calculate global emissions and fertilizer-induced N2O emissions from crops and grasslands by applying GIS with 0.5° × 0.5° resolution data sets on soil properties, climate types, rice-growing area, leguminous crops, other (upland) crops and grasslands. These were updated with statistical data for 1995 on land use, combined with country data on mineral N fertilizer use, animal populations for dairy and non-dairy cattle, pigs, poultry, sheep and goats, and animal N excretion and animal-waste management systems. The estimated global annual N2O emissions from fertilized agricultural lands are 2.9 Tg N year−1, which was updated to 2.8 Tg N year−1 by Bouwman et al. (2002), and the global mean fertilizer-induced emissions for N2O amount to 0.9% of the N applied, which is lower than EF1 of IPCC (1997), but close to the 1% value used by IPCC recently (IPCC 2006). Comparing with the IPCC method, the FAO/IFA model certainly incorporated more spatial variability of biophysical variables as well as management practices.

Discussion

A significant number of models for predicting N2O emissions from agricultural lands have been developed. Each of them has been developed for a specific purpose, with their own view and interpretation of the processes and controlling variables for N2O emissions, with its unique strengths, limitations and for different scales, ranged from laboratory, field, regional to global scales. Increased emphasis on upscaling processes at various scales has been evident in the more recent literature.

Laboratory to field models

The laboratory-scale models simulate microbial growth, substrate dynamics, solute and gas transport through the soil profile and aggregates, and mostly are used to test the researchers’ hypothesis of nitrification, denitrification and associated mechanisms of N trace gas emissions from soils. The major factors considered at this level are soil water content, soil pH, soil organic C content, ammonium and nitrate concentrations and soil temperature. Because of the complexity across natural and agricultural ecosystems, the application of these types of models is very limited. However, the laboratory-scale models serve as the theoretical basis for more complex field-scale models (e.g. DNDC, DAYCENT, ecosys and WNMM) which incorporate soil water dynamics, plant growth, C and N cycling and the impacts of site-specific soil properties, climate and agricultural management practices.

Field to regional models

Field-scale models have received the most attention in last two decades due to the need to simulate ‘real world’ conditions by integrating generic processes in plant–soil C/N and water cycling and anthropogenic influences (i.e. land use change and management practices). A field-scale model can also be used to develop a decision support system for mitigation (Chen et al. 2005). Field-scale models such as DNDC, DAYCENT, ecosys and WNMM have also been coupled with GIS or have the built-in spatial capability for application at regional scales, to estimate the spatial and temporal variability of N2O emissions in different landscapes if the detailed spatial information (specifically climate and soil) required by these models is available.

With spatially explicit input data, it is possible to explore inter-daily, interseasonal, and inter-annual variation of N2O emissions in direct response to soil, climate and sub-regional management practices. Whilst the use of very simplistic approaches such as the IPCC methodology for identifying, climate or annual differences across a region is not possible, regression models such as those developed by Clayton et al. (1994) and Sozanska et al. (2002) may work.

If users are only interested in the spatial pattern of annual N2O emissions from agricultural soils, e.g. for regional or national inventories, it may not be essential to use a detailed field-scale models to simulate specific management interactions in fine detail because of the intensive data requirements. A calibrated regional-scale models may be adequate for such a task whereby a select series of pre-determined simulations are run using the field-scale approach but then articulated as simpler N2O responses based on regressions (Grace et al. 2005). The landuse, soil type, topography and climate are generally recognized as the major environmental controlling factors to contribute to the spatial and temporal variations of N2O emissions from landscapes, and, thus such information is often used for constructing regional-scale models.

Regional to global models

The regional/global-scale models can produce the monthly, seasonal or annual pattern of N2O emissions for large landscapes or globally by using a few parameters through regression models, scaling approaches, empirical models, or in rare cases, simplistic mechanistic or processed-oriented models such as NASA CASA. In contrast to field and regional scales models, there are only four published modelling approaches at the global level: the monthly version of the NASA CASA model (Potter et al. 1996), Bouwman’s scaling model (Bouwman et al. 1993, 2002), the IPCC scaling model (IPCC 1997, 2006) and REML-based FAO/IFA model (FAO/IFA 2001). The total amount of global annual N2O emissions from soils estimated by Bouwman’s model is 6.9 Tg N year−1, while NASA CASA predicted 6.1 Tg N year−1 N2O emissions globally. In terms of global estimates of N2O emissions from agricultural lands, the prediction by REML-based FAO/IFA model (2.9 Tg N year−1) FAO/IFA (2001) is higher than that by the NASA CASA model (2.1 Tg N year−1) (Potter et al. 1996). The FAO/IFA estimation was developed using higher resolution land use and management data than the other approaches.

Some field and regional models have the potential for upscaling to provide global annual N2O emissions, but the underlying uncertainty is potentially large due to the fact that field and regional models were normally calibrated for localized areas, e.g. several to thousands hectares. Even though many of the processes underpinning the soil C, N and water cycles are generic, when applying some models to different climatic zones or latitude regions in which they were developed, the originally boundary conditions in these models may be exceeded, For example, the findings by Sabey et al. (1959), Mahendrappa et al. (1966), Thiagalingam and Kanehiro (1973), Myers (1975) and Malhi and McGill (1982) indicated that the optimal temperature for soil nitrification process tends to decrease from tropical to cool temperate regions. It is for this reason that the model of choice must be shown to work across a wide range of environments. The DNDC and DAYCENT models are probably the two most prolific models in use today across a variety of scales and environments and continue to be used by a diverse group of users around the globe.

Model comparisons

To date, there are only few studies (Frolking et al. 1998; Li et al. 2005) that have compared field-scale models for estimating N2O fluxes from agricultural soils using the field observed data. Frolking et al. (1998) examined the performances of DAYCNET, DNDC, Expert-N and the daily version of the NASA CASA models using ancillary data from an unfertilized semi-arid rangeland in Colorado, fertilized grassland in Scotland, and two fertilized cultivated fields in Germany. Simulations were run for two consecutive years and compared with the field observations, and all models simulated daily and annual N2O emission fluxes. In general, all models produced similar results in terms of soil N cycling, but the simulations of N trace gas fluxes were quite different. Both DAYCENT and Expert-N predicted N2O emissions close to the observed emissions for all sites, and in particular DAYCENT tended to perform better than Expert-N. However, DNDC and the CASA models were less consistent in their predictions, with overestimates of emissions in the Colorado site, underestimates of emissions in the Scottish site, and comparable to the measured emissions in the German sites. The lessons from this comparison indicated that accurate partitioning of gaseous N loss into NO, N2O and N2 is a challenge for all models and that accurate simulation of soil water content and an appropriate linking of soil water content dynamics with denitrification and N2O flux is a key component of each model (Frolking et al. 1998). Since the model comparison by Frolking et al. (1998), considerable efforts have been made to improve the performances of DNDC and DAYCENT. The updated comparison for the above models is needed.

Li et al. (2005) used WNMM with three different N gas modules to simulate soil denitrification fluxes and N2O emissions from two irrigated wheat–maize systems at two locations in the North China Plain. The three gas modules were WNMM, DAYCENT (or NGAS) and DNDC. Gas module refers to the actual model components which simulate N2O emissions from the processes of soil nitrification and denitrification. Soil water, temperature, organic matter decomposition, other N transformations, such as mineralization and immobilization, and crop growth are simulated by the WNMM platform. When compared to 2-year field data from one site and 1-year data from another site, all three gas modules generated similar soil mineral N dynamics in the 0–20 cm topsoil, but the relatively-simpler daily time-step WNMM gas module consistently out performed both DAYCENT and DNDC approaches for predicting soil denitrification fluxes and N2O emissions.

New modelling approach

An artificial neural network (ANN) has been used to simulate N2O emissions from an intensive grassland ecosystem in New Zealand (Ryan et al. 2004). Daily N2O fluxes were simulated as a function of six input variables: daily rainfall, soil water content and temperature, soil nitrate, ammonium and total inorganic N content. Results showed that the ANN was able to calibrate itself to within ±0.77% of measured N2O fluxes in the training data set, and within ±2.0% of measured N2O fluxes used in the validation data set. This first attempt demonstrated that an ANN is a viable tool for simulating complex and highly variable biological systems.

However, the use of ANN should not diminish the role of mechanistic models and their necessary role in developing an explicit understanding of soil C/N dynamics and N2O emissions. If the mechanism of a system is well characterized, the mechanistic modelling approach will guide the user in the development of abatement strategies for reducing global warming. A process-oriented mechanistic model is essential for examining controlling factors over a wider range of changing conditions, and when tested against suitable validation data sets, it is not limited by the conditions the model was originally developed. In addition the application of the ANN approach to different environments is extremely impractical because it would require large training data sets that would encompass all the possible different environmental scenarios. However, ANN may have the potential to be used in modifying the standard IPCC methodology for the N2O emissions inventory at large scales to account for the variations of climate, soil type, landuse, agricultural practices and topography.

Conclusions

Significant progress has been made in the simulation modelling of N2O at all scales, from the laboratory to the globe. N2O simulation models at the field scale are the most widely used models, not only because they bridge the knowledge gaps between the laboratory level and regional/global level, but also they simulate the entire agroecosystems, including water dynamics, C and N cycling and agricultural management, therefore it can be used to develop N2O mitigation strategies which align closely as best management practices within the particular industry. Among them, DNDC and DAYCENT are the most widely used simulation models and have been widely tested by a variety of research groups, across both native (natural) and managed (agricultural) ecosystems. This approach still has many uncertainties, and significant knowledge gaps still exist in the construction of suitable models and need to be addressed: (a) limited temporal resolution of experimental data, particularly N2O emissions as a function of the quantity, quality and timing of N applied; (b) limited unequivocal data, particularly using isotopes, to separate the N2O contribution between nitrification and denitrification; (c) the need for a detailed examination of N2O leakage in the nitrification process; (d) limited data for partitioning N2O and N2 in denitrification; (e) a need for development of N2O emissions from chemo-denitrification and nitrifier denitrification; and (f) detailed analysis of gaseous diffusion of N2O from depth and its assimilation in the soil profile.

The challenging task is how to scale up the relatively more robust field-scale models to catchment, regional and national scales, specifically the interaction of biophysical processes between fields. Much emphasis has been placed on vertical flow of soil water and less has been placed on lateral sub-surface flows across landscapes and their influence on soil water balance and N gas emissions. Upscaling is an essential requirement, not only for more accurate regional and national inventories, but also for development of site-specific mitigation practices with the increased interest in full greenhouse gas accounting and emissions trading in the agricultural sector. Spatially-referenced and mechanistic models, such as ecosys and WNMM, may increase in popularity with more testing.

Whilst there is a growing number of models for estimating soil N2O emissions, the final choice of model is ultimately dependent on both the specific question being asked by the user and the availability of sufficiently detailed input data. For example, to develop sub-regional mitigation strategies based on N management would normally require high resolution soil, climate and management data and a fully functional mechanistic or advanced empirical soil–plant simulation approach. In contrast, a regional or national emissions inventory may only require simple regression approaches with coarse resolution soil and climate data.

Acknowledgements

This work was sponsored by Australian Centre for International Agricultural Research project (LWR-2003-039) and Australian Cooperative Research Centre for Greenhouse Accounting.

Copyright information

© Springer Science+Business Media B.V. 2008