Modelling terrestrial nitrous oxide emissions and implications for climate feedback


  • Xu-Ri,

    Corresponding author
    • Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author
  • I. Colin Prentice,

    1. Department of Biological Sciences, Macquarie University, North Ryde, NSW, Australia
    2. Division of Ecology and Evolution, Grantham Institute for Climate Change, Imperial College, Ascot, UK
    Search for more papers by this author
  • Renato Spahni,

    1. Climate and Environmental Physics, Physics Institute and Oeschger Centre for Climate Change Research, University of Bern, Bern, Switzerland
    Search for more papers by this author
  • Hai Shan Niu

    1. Graduate University of the Chinese Academy of Sciences, College of Resources and Environment, Beijing, China
    Search for more papers by this author

Author for correspondence:


Tel: +86 10 84097072



  • Ecosystem nitrous oxide (N2O) emissions respond to changes in climate and CO2 concentration as well as anthropogenic nitrogen (N) enhancements. Here, we aimed to quantify the responses of natural ecosystem N2O emissions to multiple environmental drivers using a process-based global vegetation model (DyN-LPJ).
  • We checked that modelled annual N2O emissions from nonagricultural ecosystems could reproduce field measurements worldwide, and experimentally observed responses to step changes in environmental factors. We then simulated global N2O emissions throughout the 20th century and analysed the effects of environmental changes.
  • The model reproduced well the global pattern of N2O emissions and the observed responses of N cycle components to changes in environmental factors. Simulated 20th century global decadal-average soil emissions were c. 8.2–9.5 Tg N yr−1 (or 8.3–10.3 Tg N yr−1 with N deposition). Warming and N deposition contributed 0.85 ± 0.41 and 0.80 ± 0.14 Tg N yr−1, respectively, to an overall upward trend. Rising CO2 also contributed, in part, through a positive interaction with warming.
  • The modelled temperature dependence of N2O emission (c. 1 Tg N yr−1 K−1) implies a positive climate feedback which, over the lifetime of N2O (114 yr), could become as important as the climate–carbon cycle feedback caused by soil CO2 release.


Nitrous oxide (N2O) is a long-lived, biogenic, greenhouse gas, contributing to both global warming and stratospheric ozone destruction (Wuebbles, 2009). Atmospheric N2O concentration is increasing by 0.2–0.3% yr−1 (Denman et al., 2007). The increase has been attributed primarily or even exclusively to human activities that increase the reactive nitrogen (N) supply to ecosystems, including anthropogenic N deposition (Liu & Greaver, 2009) and N fertilization of crops (Kroeze et al., 1999; Bernard et al., 2006; Hirsch et al., 2006; Davidson, 2009).

However, ice core records show rapid N2O concentration changes that closely parallel Northern Hemisphere temperature variations during the last glacial period and the glacial–interglacial transition (Smith, 1997; Flückiger et al., 1999; Schilt et al., 2010a,b), and this suggests a pervasive link between N2O emissions and climate. Although fertilizer use is presumed to have greatly stimulated global N2O emission, nonagricultural (especially tropical) soils still represent a source of N2O equal to all of the anthropogenic sources combined (Denman et al., 2007). This natural N2O source must be influenced by climate, as can be deduced from both field observations and manipulative experiments (Barnard et al., 2005). In particular, widespread enhancement of soil N2O emission by warming is to be expected, because both nitrification and denitrification are highly temperature-dependent processes with estimated optima as high as 38°C (Li et al., 1992; Smith, 1997; Kesik et al., 2006; Xu-Ri & Prentice, 2008). It has been proposed that the global soil source of N2O should increase as the atmosphere warms (Khalil & Rasmussen, 1989), and that this climate-induced increase could even become as important for the global N2O budget as projected increases in the anthropogenic sources of N2O (Khalil & Rasmussen, 1989). We use a combination of modelling and observations to revisit this hypothesis.

Progress in the understanding of the controls of natural N2O sources has been held back until recently by a lack of suitable modelling tools. Terrestrial soil N2O fluxes have been estimated in a ‘bottom-up’ manner by multiplying average measured fluxes by the extent of the vegetation and soil types from which the measurements were taken (Denman et al., 2007), but this approach does not allow either interannual variability or climate change effects to be quantified (Bouwman et al., 1993; Kroeze et al., 1999). Inverse modelling based on concentration measurements (Hirsch et al., 2006; Crutzen et al., 2008) is complicated by a marine source (although its magnitude has been downgraded: see Rhee et al. (2009)) and incomplete information on the variability of stratosphere–troposphere exchanges that influence the N2O sink (Nevison et al., 2007). Empirical and process-based N2O emission models have not addressed responses to historical climate (Li et al., 1992; Bouwman et al., 1993; Potter & Klooster, 1998). Our modelling approach (Xu-Ri & Prentice, 2008) has allowed us to isolate the natural terrestrial source of N2O, to quantify its transient response to environmental changes over the 20th century and to consider the implications for climate change during the 21st century.

Incubation experiments have shown increases in N mineralization, nitrification, denitrification and N2O emission with warming (Barnard et al., 2005). The situation is more complex in the field because warming can influence N2O emissions indirectly, by increasing plant transpiration and reducing soil moisture (Barnard et al., 2005), which tends to decrease N2O emission (Li et al., 1992). The response of soil N2O emissions to carbon dioxide (CO2) concentration depends on the availability of soil N (Ineson et al., 1998; Phillips et al., 2001; Billings et al., 2002; Mosier et al., 2002; Barnard et al., 2005; Reich et al., 2006). In Free Air Carbon dioxide Enrichment (FACE) experiments, some ecosystems have shown an increase in N2O emission (Ineson et al., 1998; Kammann et al., 2008), whereas others have shown a decrease (Billings et al., 2002; Mosier et al., 2002). This apparent contradiction reflects a balance between enhanced soil mineralization and increased N demand by plants (Mosier et al., 2002; Reich et al., 2006). Disentangling the long-term response of soil N2O emissions to CO2 and climate change requires the application of a model that represents the interactions of the carbon (C) and N cycles in ecosystems and the environmental responses of each component process. Our dynamic global vegetation model (Xu-Ri & Prentice, 2008) has this capability. We show here, using sensitivity experiments, that it reproduces key features of the observed complex response of N2O emission to environmental perturbations.


Model description

DyN-LPJ (Xu-Ri & Prentice, 2008) is based on the Lund–Potsdam–Jena (LPJ) dynamic global vegetation modelling framework (Sitch et al., 2003). In addition to the coupled C and water cycling and the vegetation dynamics processes simulated in LPJ, DyN-LPJ represents the uptake, allocation and turnover of N in plants, and soil N transformations: mineralization, N2 fixation, nitrification and denitrification, NH3 volatilization, N leaching, and the production and emission of dinitrogen (N2), N2O and nitric oxide (NO) (Xu-Ri & Prentice, 2008).

The model structure is based on the concepts of ubiquitous, microbial-mediated N turnover processes and mass balance, as summarized in Fig. 1. The microbial N transformation pathways considered are those generally considered to be dominant in terrestrial soils. N2O emissions are modelled as by-products of nitrification and denitrification. The water-filled pore space (WFPS) was used to regulate the relative importance of nitrification and denitrification(Xu-Ri & Prentice, 2008). Nitrification, a process conducted in two steps by consortia of chemoautotrophic ammonia-oxidizing and nitrite-oxidizing bacteria and archaea, is treated as a single process and assumed to occur in aerobic microsites within the top 50 cm of soil. The total rate of nitrification in the model is controlled by inline image availability and soil temperature, consistent with experimental evidence. Denitrification, conducted mainly by heterotrophic bacteria, is also treated as a single process and is assumed to occur in anaerobic microsites within the same soil layer following the reduction sequence (inline image). The rate of each transformation is regulated by the availability of the particular oxidized N species (inline image or inline image), labile C availability (the C source required by the denitrifiers) and soil temperature, again on the basis of experimental evidence. These rates respond to substrate concentration (inline image or inline image) following Michaelis–Menten kinetics (Li et al., 1992). N2O production is then represented as a fraction of the nitrification rate and a (larger) fraction of the denitrification rate. The fraction of gross nitrification lost from the soil as N2O is c. 0.1–0.2% according to the experimental results of Goodroad & Keeney (1983), Breuer & Butterbach-Bahl (2002) and Khalil et al. (2004), or c. 0.01–0.05% according to Davidson et al. (1993) and Ingwersen et al. (1999). The fraction of denitrification lost as N2O is 0.2–4.7%, with a mean value of c. 2% (Groffman et al., 2000; Well et al., 2003; Khalil et al., 2004). The model parameters RN2ON and RN2ODN determine the fraction of mineralized N that is released as N2O (at standard temperature, 20°C) during nitrification and denitrification, respectively. We applied representative values of 0.05% for RN2ON and 1.8% for RN2ODN. The actual modelled fractions are temperature dependent, based on experimental data for both processes reviewed by Xu-Ri & Prentice (2008).

Figure 1.

Schematic diagram of the dynamic global nitrogen model.

The model thus represents the terrestrial N cycle in a simple, semi-empirical way, disregarding many poorly quantified ‘bypass’ processes, such as anaerobic ammonia oxidation, dissimilatory nitrate reduction and organic N uptake by plants. Nevertheless, as we show, it can reproduce the broad features of N2O emission and its large-scale environmental controls, and thereby provide insight into the most fundamental global processes governing the dynamics of terrestrial N2O emission.

The model employs daily internal time steps for reasons of numerical stability, but it is driven by monthly data, interpolated to days. The input data include monthly wet days; precipitation is distributed stochastically among these days, with intervening dry periods (Sitch et al., 2003; Gerten et al., 2004). Thus, the model is not expected to reproduce daily time series of N2O emission, but does include a realistic degree of precipitation variability within each month, and should be able to simulate seasonal and longer variations.

Model evaluation against field measurements

Sixty-six measurements of annual N2O emission were compiled from the peer-reviewed literature (Table 1). These measurements were all conducted in the field, mostly using closed chamber techniques and gas chromatography. Data were compiled for 61 locations covering 20 biomes. We used only datasets from nonagricultural ecosystems where at least a full year of measurements was available.

Table 1. Observed annual nitrous oxide (N2O) emission data for comparison with model results
No.Long.Lat.Vegetation typeYear observedObserved N2O (kg N ha−1 yr−1)LocationSource
  1. Notes on the selection criteria. Multiple (and apparently equivalent) measurements from within a single half-degree grid cell were averaged. For example: Breuer et al. (2000), Queensland, Australia, 1998, gave measurements of 5.36, 1.15 and 3.75 from one grid cell. The average of 3.42 was entered.

  2. If there were several records from different soil or vegetation types in a grid cell in 1 yr, and the authors had indicated which vegetation or soil type was dominant in the region, we used the value from this dominant soil or vegetation type. For example: Jungkunst et al. (2004), southwest Germany, gave measurements of 0.93 and 1.86 from one grid cell. The former measurement was from the dominant soil type in the regions as stated by the author, and so we entered this value.

  3. Measurements on one grid cell in different years have been entered as distinct records. For example: Bowden et al. (1990), Harvard Forest, USA, gave values of 0.02 and 0.06 for 1988 and 1989, respectively. Both values were entered, and compared with simulation results for the two different years.

  4. In cases in which measurements were made over more than a year, but the individual years’ records were not given, we used the multiyear value and compared this with the corresponding multiyear mean from the simulation. For example: Schmidt et al. (1988), Bechenheim, Germany, gave just one measurement of 0.802 representing 1981 and 1982. This value was entered, and compared with the simulation results averaged over these 2 yr.

  5. In cases in which there were records over multiple years showing a large increase, and the increase was attributed by the authors of the study to increasing N deposition, we entered only the data from the earliest year. For example: Papen & Butterbach-Bahl (1999), Bavaria, Germany, gave three records of 0.4, 0.8 and 3.1 in the years 1994, 1995 and 1996, respectively, and explained the abrupt increase as a consequence of rapidly increasing atmospheric N deposition. In this case, we entered only the 1994 record.

1172.5−43.5Temperate grassland20000.255New ZealandMűller & Sherlock (2004)
223−31Savanna1992 0–0.6Kruger, South AfricaLevine et al. (1996)
328.5−24.5Savanna19940.28Nylsvley, South AfricaScholes et al. (1997)
4145.5−17.5Tropical rain forest19983.42Queensland, AustraliaBreuer et al. (2000)
5145.5−17Tropical moist forest20004.36Queensland, AustraliaKiese & Butterbach-Bahl (2002)
6−62.5−10.5Tropical forest1992–19931.9Brazilian AmazonMelillo et al. (2001)
7114.5−3.5Tropical peatland19985.698IndonesiaHadi et al. (2000)
8114.5−3.5Tropical forest19993.504IndonesiaVerchot et al. (1999)
9−47.5−3Tropical Forest19952.4Eastern AmazoniaLuizão et al. (1989)
10114.5−3.0Tropical forest19992.92IndonesiaHadi et al. (2000)
11−60−3Tropical rain forest19871.9Manaus, BrazilLuizão et al. (1989)
12−55.0−3.0Tropical moist forest1998–20002.6East-central AmazoniaDavidson et al. (2004)
13115.5−2.0Tropical forest19992.628IndonesiaHadi et al. (2000)
14−63.59.5Savanna20000.73VenezuelaSimona et al. (2004)
15−8510Tropical pastures19942.365Costa RicaVeldkamp et al. (1998)
16−8410.5Tropical forest19913.74Costa RicaKeller & Reiners (1994)
17−6718Tropical grassland1992–19931.51Puerto RicoMosier & Delgado (1997), Mosier et al. (1997a)
18−6618Subtropical moist forest19951.75Northeastern Puerto RicoErickson et al. (2001)
19−155.519.5Montane rain forest1990–19910.223HawaiiRiley & Vitousek (1995)
209335Alpine grassland20000.069Tibet, ChinaPei (2003)
21−107.535.5Temperate forest19860.03New Mexico, USAMatson et al. (1992)
22−107.535.5Temperate forest19870.09New Mexico, USAMatson et al. (1992)
23−7936Temperate forest1998–19990.066North Carolina, USAPhillips et al. (2001)
24−11637Desert20000.11Mojave desert, USABillings et al. (2002)
25116.539.5Temperate forest1997–19980.277Beijing, ChinaSun & Xu (2001)
26−104.540.5Temperate shortgrass steppe1992–19950.1667Colorado, USAMosier et al. (1996, 1997b)
27−104.540.5Temperate steppe1990–19920.184Colorado,USAMosier & Delgado (1997), Mosier et al. (1997a)
2812741.5Alpine tundra1994–19950.28ChangBai Mountain, ChinaChen et al. (2000)
29−10741.5Sagebrush steppe1986–19870.21South-central Wyoming, USAMatson et al. (1991)
30−72.042.5Temperate forest19880.02Harvard Forest, USABowden et al. (1990)
31−72.042.5Temperate forest19890.06Harvard Forest, USABowden et al. (1990)
32−72.543.0Temperate forest1990−0.098Northeastern, USACastro et al. (1992)
33−8943Temperate forest19790.59Wisconsin, USAGoodroad & Keeney (1983)
34−8943Temperate forest19801.45Wisconsin, USAGoodroad & Keeney (1983)
35116.043.5Temperate grassland19950.27Inner Mongolia,ChinaChen et al. (2000)
36116.543.5Temperate grassland19980.365Inner Mongolia,ChinaXu-Ri et al. (2003)
37−71.044.0Temperate forest1990−0.02Northeastern, USACastro et al. (1992)
38−68.544.0Temperate forest19900.0315Northeastern, USACastro et al. (1992)
39−74.044.5Temperate forest19900.1848Northeastern, USACastro et al. (1992)
40−7344.5Temperate forest19900.1708Northeastern, USACastro et al. (1992)
41847Temperate forest2000–20020.93Southwest GermanyJungkunst et al. (2004)
421047.5Temperate grassland19970.155Siggen, GermanyGlatzel & Stahr (2001)
4311.547.5Temperate forest1990–19910.081Innsbruck, AustriaHenrich & Haselwandter (1997)
441148.5Temperate forest19940.42Bavaria, GermanyPapen & Butterbach-Bahl (1999)
451049Temperate forest1981–19820.473Waldhausen, GermanySchmidt et al. (1988)
46−80.549Boreal forest19920.035Ontario, CanadaSchiller & Hastie (1996)
47849.5Temperate forest1981–19820.802Bechenheim,GermanySchmidt et al. (1988)
48850Temperate forest1981–19820.714Langenlonsheim, GermanySchmidt et al. (1988)
498.550Temperate forest1981–19820.298Ober-Olm, GermanySchmidt et al. (1988)
508.550.5Temperate grassland20000.292Giessen, GermanyMűller & Sherlock (2004)
51−4.052.0Temperate forest1994–19950.25Dunslair Heights, UKMacDonald et al. (1997)
52−10653Boreal forest1993–19950.02Saskatchewan, CanadaCorre et al. (1999)
5312.556Temperate forest1990–19920.8DenmarkAmbus & Christensen (1995)
5412.058.0Temperate forest1993–19940.071Southwestern SwedenKlemedtsson et al. (1997)
558.559.0Temperate pine forest19920.73NorwaySitaula et al. (1995)
56−14865Subarctic grassland19920.201AlaskaMosier & Delgado (1997), Mosier et al. (1997a)
57−11746.5Temperate grassland19940.15Washington, USAStehfest & Bouwman (2006)
58−91.529.5Marsh19800.416Louisiana, USAStehfest & Bouwman (2006)
599.554Temperate forest19950.4Kiel, GermanyStehfest & Bouwman (2006)
6012.555.5Temperate forest19860.52DenmarkStehfest & Bouwman (2006)
61−7441Temperate forest1979–19800.9NewYork, USAStehfest & Bouwman (2006)
62−81.526.5Marsh19801Florida, USAStehfest & Bouwman (2006)
63−63−10Tropical rain forest19983.21BrazilStehfest & Bouwman (2006)
64−7143Temperate forest19810.9New Hampshire, USAStehfest & Bouwman (2006)
65−355.5Temperate forest19940.158UKStehfest & Bouwman (2006)
66−11834Savanna19870.0California, USAStehfest & Bouwman (2006)

For comparison, annual N2O emissions were simulated at the geographical locations of each site following the transient simulation protocol as described below under the section ‘Historical simulations’. The model results were compared with the specific grid cell location and specific year of each measurement.

Sensitivity experiments

Effects of temperature on plant N uptake capacity, nitrification, denitrification and N2O emission are known from incubation experiments (BassiriRad, 2000; Barnard et al., 2005) and are implemented in DyN-LPJ, as are the effects of precipitation (mediated by soil moisture) and atmospheric CO2 concentration (mediated by soil N availability) on these processes (Xu-Ri & Prentice, 2008). Much less experimental information is available about the net effects of environmental changes on ecosystem-level N2O emission in the field, especially on long time scales. We analysed simulated responses of N2O emission to environmental treatments over annual to decadal time scales at the following locations, chosen to represent FACE sites at which some relevant measurements have been made: dry grassland in Colorado, USA (104.5°W, 41.0°N; Mosier et al., 2002); moist grassland in Switzerland (8.5°E, 47.5°N; Ineson et al., 1998); moist grassland at Jasper Ridge, California, USA (122.0°W, 37.5°N; Hungate et al., 1997; Barnard et al., 2006); the Mojave Desert, Nevada, USA (116.0°W, 37.0°N; Billings et al., 2002); and Duke Forest, North Carolina, USA (79.0°W, 36.0°N; Phillips et al., 2001). We also included a tropical humid forest in Brazil (62.5°W, 10.5°S; Melillo et al., 2001). Climates for each location were estimated from the relevant grid cell in the CLIMATE 2.2 monthly climatology, an update of Leemans & Cramer (1991). Model runs were made with fixed initial climate and atmospheric CO2 concentration (340 ppm). The simulations were started from bare ground (no plant biomass present) and spun up for 2000 model years to develop steady state in the C and N pools. Treatments, imposed as step changes, were as follows: warming by 2 K (year-round); CO2 doubling to 680 ppm; precipitation change by ± 14% (year-round: a 14% increase corresponds to the expected change in evaporation from the ocean associated with a global warming of 2 K; Trenberth et al., 2007); deposition of inorganic N (addition of 0.5 g N m−2 yr−1 as NH4+ and 0.5 g N m−2 yr−1 as inline image); and factorial combinations of these treatments.

Historical simulations

For the transient 20th century simulations, gridded (0.5 × 0.5) monthly mean values of temperature and cloud cover, and monthly precipitation totals, were taken from the Climatic Research Unit TS 2.0 dataset for 1901–2000 (Mitchell et al., 2004). Average annual atmospheric CO2 concentrations for 1901–2000 were taken from Etheridge et al. (1996) and Keeling & Whorf (2005). Soil texture information was taken from FAO (1991). Global atmospheric N deposition values for 1860 and 1993 were taken from Dentener (2006) and linearly interpolated (extrapolated after 1993). This simple approach allows us to obtain a model estimate of the approximate magnitude of the effect of increasing N deposition during the 20th century, although it does not realistically represent the decade-by-decade patterns of N deposition. The climate data from the years 1901–1930 were repeated during a model spin-up period of 2000 yr. Thereafter, the model was run with monthly climate data from 1901 to 2000.

Nine historical simulations were performed altogether: (1) all factors (CO2, climate and N deposition) varied through the 20th century; (2) CO2 and climate varied without N deposition; (3) climate varied but CO2 constant; (4) CO2 only varied; (5) temperature only varied; (6) precipitation only varied; (7) cloud cover only varied; (8) N deposition only varied; and (9) a reference simulation in which all factors were held constant (at 296 ppm for CO2 and 1901–1920 averages for climate variables).

Trend analysis

The Mann–Kendall test is a nonparametric technique (Mann, 1945; Kendall, 1948) widely used in hydrology and climatology. The Mann–Kendall τ-value (Hirsch et al., 1982; Hipel & McLeod, 1994) indicates the direction and magnitude of the trend in simulated natural N2O emission. We mapped the spatial pattern of this statistic for the historical simulation based on CO2 and climate, and, for comparison, the simulation with the CO2 effect removed.


Comparison of simulated N2O fluxes with field measurements

The DyN-LPJ model has been evaluated previously against worldwide observations on the soil pools of reactive N species, exchanges between these pools, rates of N uptake into plants and N limitation on net primary production (NPP) (Xu-Ri & Prentice, 2008). We extended the evaluation to soil N2O fluxes (Table 1, Fig. 2a). Simulated and observed annual N2O fluxes were highly correlated (R2 = 0.86, P < 0.001). N2O emission is highest in the humid tropics, but limited by low temperature in temperate and boreal ecosystems, and by low soil moisture in deserts. This pattern is reproduced in the global simulation (Fig. 3).

Figure 2.

(a) Observed vs simulated annual nitrous oxide (N2O) fluxes from nonagricultural soils: = 0.98x + 0.06, R2 = 0.86. (b) With atmospheric nitrogen (N) deposition included, y = 1.01x + 0.13, R2 = 0.86. This panel also shows the effect of changes in the values of the model parameters RN2ON (with a default value of 0.05%, a low value of 0.01% and a high value of 0.2%) and RN2ODN (with a default value of 1.8%, a low value of 1.0% and a high value of 4.7%). Note the different scale of the y axes in (a) and (b). Dashed line is the 1 : 1 line; solid line is the regression line.

Figure 3.

Simulated spatial pattern of annual nonagricultural soil nitrous oxide (N2O) emissions for the year 1990 (kg N ha−1 yr−1).

The effects of varying RN2ON and RN2ODN are shown in Fig. 2b. These model parameters determine the fractions of N gas emissions that appear as N2O (at 20°C) during nitrification and denitrification, respectively. We found that RN2ODN is the dominant parameter. N deposition effects were found to be negligible in the site-by-site comparison. The default values (1.8% for RN2ODN and 0.05% for RN2ON) are shown to fit the data closely, whereas extreme values (still consistent with the experimental range) are shown to give poor results.

Two multiple regressions were conducted to quantify the contribution of different factors to the variation in N2O emissions in space and time (Table 2). The analyses are ordinary least-squares regressions with all of the variables standardized to zero mean and unit variance. The first multiple regression concerns the spatial pattern of the observed annual N2O emission. As expected, both temperature and precipitation were highly significant predictors of N2O emission. We further compared the simulated response pattern of annual N2O emissions to mean annual precipitation (MAP) and temperature (MAT) with the corresponding response patterns derived directly from the field measurements, as shown in Fig. 4. The simulated patterns are consistent with the observed patterns, including a steep increase in the slope of the relationship between N2O emissions and annual temperature of c. 15–20°C and (albeit based on one high-rainfall data point) a flattening of the response of N2O emission to annual precipitation above 2000 mm (Fig. 4).

Figure 4.

Observed and simulated nitrous oxide (N2O) emissions vs spatial patterns of precipitation and temperature.

Table 2. Multiple regression results
ParameterEstimateSE of estimatet-valueProb>|t| 
  1. MAP, mean annual precipitation; MAT, mean annual temperature.

1. Spatial pattern
Observed soil N 2 O emissions vs climate
MAP 0.481 0.097 4.95 <0.0001*** R2 = 0.66
MAT 0.410 0.099 4.13 <0.0001*** P < 0.001
Cloud0.0190.0760.250.81df = 65
2. Temporal pattern
Simulated global soil N 2 O emissions vs climate
CO2−0.0480.120−0.400.69 R2 = 0.52
MAP−0.0280.088−0.320.75P < 0.001
MAT  0.7540.102   7.39<1010***df = 95
Cloud  0.0390.095   0.410.68 

Comparison of sensitivity experiments with field experiments

Effect of temperature

Increased temperature was predicted generally to increase NPP and plant N uptake (except at the forest sites), to decrease N limitation (by enhancing soil N mineralization), to enhance nitrification and denitrification, and to increase N2O emission on both annual and decadal time scales (Table 3). The positive responses of modelled NPP, plant N uptake and N mineralization rate are all consistent with the results of warming experiments conducted in grassland ecosystems (BassiriRad, 2000; Shaw & Harte, 2001; Wan et al., 2005). The response of forest NPP to warming is more complicated as the autotrophic respiration component is larger, and is also expected to respond positively to warming (Saxe et al., 2001).

Table 3. Sensitivity tests with dynamic global vegetation model (DyN-LPJ)
Ecosystem treatmentsNPP (g C m−2 yr−1)N uptake (g N m−2 yr−1)Total N mineralization (g N m−2 yr−1)Inorganic N (g N m−2)Nitrification (g N m−2 yr−1)Denitrification (g N m−2 yr−1)N2O emission (mg N m−2 yr−1)
  1. Simulated values (bold) are shown for the control case; changes from the control values are shown for the other treatments, after 1 yr and (10 yr) of simulation. Treatments, imposed as step changes, were: warming by 2 K (year-round), CO2 doubling to 680 ppm, precipitation change by ± 14% (year-round), N deposition (0.5 g N m−2 yr−1 NH4+ and 0.5 g N m−2 yr−1 inline image), and combinations of these.

Semi-arid grassland, 104.5°W, 41.0°N
Control     379   6.96     9.14    11.01    7.23     1.84    22.41
Temperature13 (1)0.21 (−0.01)1.44 (0.92)0.73 (1.74)1.68 (1.33)0.64 (1.16)11.06 (20.21)
CO2410 (201)7.44 (3.87)0.03 (1.73)−6.97 (−10.31)−1.66 (−6.41)−0.55 (−1.83)−0.85 (−19.75)
+ Precip41 (30)0.75 (0.57)0.39 (0.48)−1.07 (−3.15)0.01 (−0.39)0.32 (−0.46)5.86 (−4.73)
− Precip−47 (−47)−0.87 (−0.88)−0.42 (−0.63)1.00 (4.38)−0.04 (0.08)−0.37 (0.45)−5.41 (2.99)
+ N dep.0 (0)0 (0)0 (0)0.91 (4.03)0.30 (0.82)0.08 (0.82)0.14 (7.46)
+ N + C410 (289)7.44 (5.39)0.03 (2.32)−6.97 (−10.05)−1.33 (−5.15)−0.47 (−1.79)−0.71 (−19.28)
+ T + P65 (47)1.15 (0.84)1.9 (1.56)−0.34 (−0.83)1.75 (1.35)1.13 (0.81)20.08 (15.16)
+ T + P + C544 (342)9.86 (6.34)1.92 (4.24)−8.00 (−10.24)−0.52 (−5.64)0.02 (−1.81)16.98 (−20.61)
+ T + P + C + N544 (348)9.86 (6.42)1.92 (4.76)−7.12 (−7.24)−0.13 (−0.78)0.15 (−1.42)17.32 (−18.00)
Moist grassland, 8.5°E, 47.5°N
Control    556    7.25   12.75    6.84    8.12    2.99    28.21
Temperature85 (71)1.09 (0.95)2.12 (1.95)−0.14 (−0.2)2.01 (1.77)1.30 (1.14)17.28 (16.93)
CO2143 (118)1.82 (1.56)0.56 (1.26)−1.01 (−0.37)−0.63 (−0.11)−0.25 (−0.19)4.24 (−1.22)
+ Precip8 (7)0.12 (0.09)0.87 (0.58)−0.17 (−0.28)0.17 (0.21)0.53 (0.41)8.89 (4.48)
− Precip−17 (−14)−0.22 (−0.18)−0.83 (−0.61)0.43 (0.54)−0.13 (−0.18)−0.56 (−0.36)−7.15 (−3.56)
+ N dep.0 (0)0 (0)0 (0)0.51 (0.69)0.311 (0.64)0.28 (0.52)1.04 (4.84)
+ N + C143 (118)1.82 (1.56)0.56 (1.26)−0.53 (0.28)−0.30 (0.55)0.04 (0.33)5.59 (3.74)
+ T + P102 (84)1.31 (1.14)3.05 (2.61)−0.39 (−0.56)2.19 (1.97)1.94 (1.59)29.60 (22.68)
+ T + P + C285 (226)3.63 (3.02)3.73 (4.22)−1.26 (−0.71)1.21 (1.90)1.13 (1.20)30.88 (19.77)
+ T + P + C + N285 (226)3.63 (3.02)3.73 (4.22)−0.88 (−0.25)1.58 (2.55)1.52 (1.81)33.89 (26.68)
Moist grassland, 122.0°W, 37.5°N
Control    399    5.21     9.82    14.08    8.42     3.70    24.80
Temperature56 (52)0.73 (0.59)1.56 (1.48)0.14 (0.72)1.62 (1.49)1.14 (1.64)13.11 (17.92)
CO2299 (257)3.70 (2.49)0.13 (1.43)−4.25 (−4.08)−0.84 (−0.07)−0.04 (−1.22)2.55 (−9.32)
+ Precip22 (17)0.30 (0.18)0.35 (0.37)−1.37 (−3.34)0.03 (−0.21)0.60 (−0.55)5.29 (−3.46)
− Precip−27 (−38)−0.36 (−0.36)−0.38 (−0.53)1.42 (5.96)−0.06 (0.01)−0.67 (0.55)−5.04 (1.93)
+ N dep.0 (0)0 (0.01)−0.010.92 (2.67)0.34 (0.63)0.08 (0.80)0.06 (4.83)
+ N + C299 (257)3.7 (2.49)0.13 (1.43)−3.35 (−1.68)−0.50 (0.74)0.04 (−0.45)2.62 (−4.62)
+ T + P82 (72)1.07 (0.8)1.98 (1.92)−1.38 (−2.96)1.66 (1.29)1.94 (0.87)21.67 (12.00)
+ T + P + C444 (369)5.54 (3.58)2.14 (3.6)−5.99 (−5.01)0.44 (1.44)1.58 (−0.17)25.73 (2.40)
+ T + P + C + N444 (369)5.54 (3.58)2.14 (3.6)−5.16 (−3.41)0.83 (2.20)1.73 (0.63)25.89 (8.66)
Desert, 116.0°W, 37.0°N
Control     80    1.0     1.63    30.01    1.57     0.26     2.30
Temperature8 (−1)0.12 (0.03)0.25 (0.25)0.34 (4.23)0.26 (0.25)0.08 (0.12)0.99 (1.10)
CO2124 (87)1.56 (1.20)0.0 (0.63)−1.92 (−8.24)−0.01 (0.57)0.0 (−0.09)0.03 (−0.14)
+ Precip5 (5)0.06 (0.09)0.03 (0.07)−0.99 (−9.12)0.03 (0.06)0.05 (−0.04)0.51 (0.44)
− Precip−4 (−12)−0.06 (−0.18)−0.04 (−0.09)0.39 (4.97)−0.02 (−0.07)−0.05 (−0.02)−0.4 (−0.45)
+ N dep.1 (1)0.01 (0.01)0 (0)0.85 (9.61)0.35 (0.49)0 (0.07)0.02 (0.06)
+ N + C124 (87)1.56 (1.20)0 (0.63)−0.94 (2.12)0.33 (1.06)0.00 (−0.01)0.06 (−0.06)
+ T + P15 (20)0.22 (0.34)0.29 (0.39)−0.36 (0.87)0.29 (0.37)0.15 (0.16)1.76 (1.78)
+ T + P + C151 (114)1.95 (1.67)0.29 (1.25)−2.53 (−10.26)0.27 (1.13)0.16 (−0.01)1.89 (1.58)
+ T + P + C + N151 (114)1.95 (1.67)0.29 (1.25)−1.55 (−0.18)0.62 (1.63)0.16 (0.11)1.92 (1.71)
Forest, 79.0°W, 36.0°N
Control    743    8.88    16.13    4.87    11.87     6.16   107.22
Temperature−43 (−24)−0.48 (−0.25)2.12 (0.96)0.20 (−0.39)2.55 (1.22)2.52 (1.42)59.82 (42.21)
CO2585 (385)6.91 (4.64)0.19 (3.54)−2.79 (0.56)−4.24 (1.15)−4.0 (−1.19)−44.12 (−24.06)
+ Precip68 (44)0.80 (0.53)0.9 (0.89)−0.73 (−0.57)0.00 (0.16)0.37 (0.18)14.23 (2.88)
− Precip−82 (−51)−0.99 (−0.62)−0.93 (−1.01)0.73 (0.51)−0.03 (−0.17)−0.50 (−0.06)−17.62 (−0.63)
+ N dep.0 (0)0 (0)0 (0)0.45 (0.50)0.45 (0.67)0.52 (0.84)6.05 (14.70)
+ N + C585 (385)6.91 (4.64)0.19 (3.54)−2.32 (1.15)−3.55 (1.95)−3.54 (−0.37)−38.61 (−10.11)
+ T + P25 (17)0.32 (0.25)3.15 (1.9)−0.58 (−0.94)2.70 (1.42)3.04 (1.65)79.24 (45.71)
+ T + P + C632 (398)7.47 (4.85)3.41 (5.68)−2.2 (−0.17)−1.57 (2.88)−2.17 (0.63)−4.72 (21.59)
+ T + P + C + N632 (398)7.47 (4.85)3.41 (5.68)−1.89 (0.20)−0.93 (3.64)−1.57 (1.48)5.03 (37.92)
Tropical humid forest, 62.5°W, 10.5°S
Control   1027   25.72   41.01     2.35   24.66   13.22   282.88
Temperature−154 (−107)−3.75 (−2.65)3.91 (−2.01)0.32 (−0.41)6.65 (0.22)6.96 (0.86)187.44 (58.07)
CO2537 (501)13.19 (12.83)0.24 (12.94)−2.22 (0.22)−14.79 (3.65)−9.54 (−0.18)−177.59 (−4.88)
+ Precip32 (24)0.75 (0.60)0.84 (1.07)−0.06 (−0.02)0.04 (0.35)0.04 (0.33)3.54 (7.15)
− Precip−38 (−30)−0.89 (−0.76)−1.13 (−1.29)0.08 (0.04)−0.11 (−0.31)−0.15 (−0.36)−7.79 (−7.67)
+ N dep.0 (0)0 (0)−0.01 (0)0.1 (0.1)0.60 (0.68)0.81 (0.89)15.05 (19.05)
+ N + C563 (501)13.9 (12.83)0.24 (13)−2.21 (0.34)−14.41 (4.57)−9.32 (0.77)−172.84 (15.41)
+ T + P−124 (−85)−3.11 (−2.14)4.83 (−1.12)0.27 (−0.43)6.89 (0.52)7.13 (1.14)194.52 (65.16)
+ T + P + C563 (421)14.04 (10.74)5.03 (13.55)−1.51 (−0.07)−9.36 (6.16)−6.71 (2.45)−104.63 (93.90)
+ T + P + C + N580 (421)14.5 (10.74)5.03 (13.57)−1.41 (0.01)−8.71 (6.97)−6.35 (3.37)−97.09 (116.09)
+ T + C534 (387)13.38 (9.96)4.16 (12.08)−1.51 (−0.09)−9.01 (5.65)−6.88 (1.93)−108.28 (81.46)
+ P + C566 (537)13.88 (13.65)1.04 (14.48)−2.22 (0.26)−14.90 (4.16)−9.46 (0.36)−175.89 (6.31)
+ T + N−154 (−108)−3.75 (−2.65)3.91 (−2.04)0.39 (−0.32)7.2 (0.87)7.81 (1.76)205.82 (79.85)
+ P + N32 (24)0.75 (0.6)0.84 (1.07)0.04 (0.08)0.64 (1.03)0.84 (1.21)18.76 (26.04)
+ T + P + N−124 (−85)−3.11 (−2.14)4.83 (−1.12)0.35 (−0.34)7.44 (1.18)7.98 (2.04)212.98 (86.80)

Field data on the direct effect of warming on N2O emissions are still scarce (Barnard et al., 2005), but increases in nitrification and denitrification rates and N2O emissions have been shown in incubations (Schipper et al., 1993; Maag et al., 1997; Maag & Finn, 1999; BassiriRad, 2000) and in recent field experiments (Cantarel et al., 2011; Larsen et al., 2011).

Effects of CO2

Elevated CO2 was predicted to increase both NPP and N uptake initially. Increased N uptake tended to compensate or outweigh increased N mineralization, resulting in decreased or unchanged inorganic N, nitrification and denitrification (Table 3). These responses are consistent with FACE results (Hungate et al., 1997; Barnard et al., 2004, 2005; Hu et al., 2005; Holmes et al., 2006; Reich et al., 2006; Regan et al., 2011). Elevated CO2 is expected to enhance N2O emissions by denitrification under conditions of increased N mineralization and sufficient soil moisture (Hungate et al., 1997; Holmes et al., 2006). However, increased plant production also increases the demand for soil N, potentially reducing the microbial availability of soil N and thus decreasing N2O emission (Phillips et al., 2001; Mosier et al., 2002; Reich et al., 2006).

Both of these opposing effects have been seen in FACE experiments (Hungate et al., 1997; Ineson et al., 1998; Phillips et al., 2001; Billings et al., 2002; Holmes et al., 2006; Reich et al., 2006) and both occur in the model (Table 3). FACE experiments also confirm the modelled contrast between wet and dry ecosystems, with N2O emission enhanced, at least in the short term, by raised CO2 in humid grasslands (Arnone & Bohlen, 1998; Ineson et al., 1998; Robinson & Conroy, 1998; Regan et al., 2011), but reduced or unchanged in semi-arid short grass steppe (Mosier et al., 2002) and desert (Billings et al., 2002). On the decadal time scale, the modelled NPP increase is attenuated and the effect of increased N mineralization tends to overwhelm the plant N uptake rate, resulting in increased inorganic N concentration, as has been observed in moist grassland (Müller et al., 2009) and forests (Phillips et al., 2001; Mckinley et al., 2009; Rütting et al., 2010; Zak et al., 2011).

Effects of precipitation

Increased precipitation was predicted to initially increase NPP, plant N uptake, nitrification and denitrification, and N2O emission, whereas decreased precipitation had opposite effects, consistent with previous modelling results (Li et al., 1992). However, these effects declined over time. After a decade, the modelled enhancement of N2O emission at the temperate forest site was lowered by 80%, whereas the dry grassland site showed a net reduction in N2O emission. The modelled effect of increased N demand (as a result of faster plant growth) on microbial N availability eventually nearly or entirely outweighed the increase in N2O emission because of wetter soils in this simulation. A decrease in modelled N2O emissions from European forest soils under future (warmer and wetter) climate conditions has also been reported, together with increased denitrification, as a result of a decrease in the ratio of N2O to N2 emitted (Kesik et al., 2006).

Effects of N deposition

Climate and CO2 concentration are first-order controls on NPP in the model. The model assumes that biological N fixation can increase to support the increased N demand created by climate changes (when they favour increased growth) and increasing CO2. In the model, therefore, the addition of further inorganic N has a strong effect in increasing the inorganic N pools, the rates of nitrification and denitrification, and thus the soil N2O emission. These positive responses of N cycle components to N deposition are all consistent with generally observed patterns (Lu et al., 2011).

Modelled N2O emissions from grasslands increased by 0.01–0.70% because of N addition alone, or 0.03–0.69% in combination with other changes. These emission factors can be compared with recent measurements of 0.06–0.30% in a semi-arid grassland (Peng et al., 2011) and averaged 0.75% in European grasslands (Flechard et al., 2007). In temperate forests, the modelled enhancement was 0.6–1.5% with N addition alone, or 0.9–1.6% in combination with other changes. These ranges of values are similar to the range of 0.03–1.6% given in a recent review (Eickenscheidt et al., 2011). In tropical forests, the modelled enhancement was 1.5–1.9% with N addition alone, and 0.8–2.2% in combination with other changes. This simulated (higher) range of emission factors in a tropical forest is consistent with high emission factors of 0.2–2.8% as observed in tropical ecosystems (Steudler et al., 2002).

Multifactor interactions

The modelled effects of CO2 concentration on soil nitrification, denitrification and N2O emission are time scale dependent, and interact strongly with the effects of other environmental variables (Table 3). The simulated increased (decreased) soil N2O emissions caused by CO2 enrichment in wet (dry) soil conditions are consistent with ‘short-term’ FACE experiments (i.e. after 1 or 2 yr of CO2 enrichment). On a longer time scale (after a decade of CO2 enrichment), the CO2-only treatment generally resulted in a reduction in modelled N2O emission. However, when the CO2 treatment was combined with N addition and climate change (shown in Table 3 as treatments + N + C, + T + P + C and + T + P + N + C), its long-term effect again became positive. These simulated responses are consistent with those found in multifactorial interaction experiments by Barnard et al. (2006) (compare JRBP FACE, DEA, with treatments C only, NC, TWC and TWNC) and Kammann et al. (2008).

The amplified effect of CO2 when combined with N addition and climate change in the model appears most strongly in tropical ecosystems, where the positive effect on N2O emission of all factors combined is generally larger than the sum of single factor effects. At the location in Brazil, for example, the combination of CO2 increase with climate change enhanced N2O emission by 34 mg N m−2 yr−1 (+ T + P + C), much more than the effect of CO2 increase with N addition, but without climate change (1.2 mg N m−2 yr−1, + N + C). This amplification appeared in all two-way interactions, but the interaction of CO2 and temperature was consistently the strongest. Amplification of the CO2 effect by other environmental changes was also found by Niboyet et al. (2011) in field experiments.

The positive effect of N deposition on N2O emission was amplified by climate change and CO2 effects. Over a decade, our results suggest that the effect of N deposition on N2O emissions might be increased (by 30–43% in grasslands and 11–17% in forests) because of interaction with the effects of climate change and increasing CO2.

Simulated historical soil N2O emissions in relation to
climate and atmospheric CO2 concentration

Decadal-average N2O emissions from nonagricultural soils in the 20th century simulation without N deposition ranged from 8.2 to 9.5 Tg N yr−1, with individual years ranging from 7.6 to 10.0 Tg N yr−1 (Fig. 5), similar to the Intergovernmental Panel on Climate Change (IPCC) central estimate of 6.6 Tg N yr−1 (range, 3.3–9.0 Tg N yr−1; Denman et al., 2007) and previous model-based estimates of 6.8 Tg N yr−1 (Bouwman et al., 1993) and 9.7 Tg N yr−1 (Potter & Klooster, 1998). Interannual climate variability caused individual years’ emissions to deviate from the decadal mean by up to c. 1 Tg N yr−1.

Figure 5.

(top) Twentieth century simulations of total global nonagricultural soil nitrous oxide (N2O) emissions, as year-to-year variations (dashed lines) and smoothed with 10-yr splines (full lines). (bottom) Observed CO2 concentration and global climate variables: annual means over all land grid points, and tropical (30oS to 30oN) grid points, as 10-yr splines of standardized values.

The spatial pattern of global soil N2O emissions is illustrated in Fig. 3. Tropical ecosystems (30oS to 30oN) contributed the largest fraction (84%) and northern extratropical regions most of the rest (14%). Tropical rain forest, seasonal forest and dry forest/savanna were predicted to emit 2–5 kg N ha−1 yr−1. Humid subtropical forests were predicted to emit 1–2 kg N ha−1 yr−1, Mediterranean-type ecosystems, temperate grasslands and maritime humid forests 0.2–1 kg N ha−1 yr−1, and boreal forests, tundra and desert < 0.2 kg N ha−1 yr−1.

Simulated global N2O emission tracks variations in MAT across land grid points, as shown in Fig. 5. In a multiple regression based on individual years (Table 2), examining the effects of climate and CO2, land MAT emerged as the unique, highly significant predictor of simulated N2O emission (R2 = 0.52, P < 0.001). Annual mean precipitation, cloud cover and CO2 concentration changes had no significant effects. N deposition was not included in this analysis because its interannual to decadal variability was not realistically represented.

The simple regression relationship between simulated global N2O emission (FN, Tg N yr−1) and land MAT (°C) is:

display math(Eqn 1)

implying a sensitivity of c.1 Tg N yr−1 K−1 with respect to land MAT. Model runs with one variable changed at a time (Fig. 5) confirmed the predominant role of temperature in determining the simulated variability of global total N2O emission during the 20th century.

Most regions showed overall increases in simulated N2O emission over the 20th century (Fig. 6a). Individual grid cells in tropical and subtropical regions, coastal Europe and high-latitude North America reached significance at P < 0.1 with the Mann–Kendall test. Semi-arid temperate grassland grid cells showed nonsignificant decreasing trends. A comparison of trends with and without the CO2 effect (Fig. 6a,b) showed that, in many tropical regions, CO2 and climate change combined synergistically to increase N2O emission, whereas, in some temperate regions, the effect of increasing CO2 was a reduction in N2O emission, so that increasing CO2 produced a smaller increase in N2O emission than would have been predicted by climate change alone. As a result, the net effect of the CO2 increase in the 20th century simulation was slight.

Figure 6.

Spatial patterns of the Mann–Kendall τ statistic for simulated global 20th century trends of annual nonagricultural nitrous oxide (N2O) emissions, in simulations with (a) CO2 and climate change, and (b) CO2 concentration fixed at 296 ppm.

By the 1990s, the simulated increase in global soil N2O emissions as a result of climate and CO2 effects during the 20th century amounted to 0.85 ± 0.41 Tg N yr−1. The factorial sensitivity experiments (Table 3) suggest that the generally positive response of soil N2O fluxes to rising CO2 (Figs 5, 7) was possible only because temperature was increasing simultaneously with CO2.

Figure 7.

Global annual average increase of simulated nitrous oxide (N2O) emission during the 20th century (Tg N yr−1) as a result of CO2, nitrogen (N) deposition and climate change; singly (white bars) and interactions with other factors (red bars). Error bars are ± SE.

Simulated historical soil N2O emissions in relation to atmospheric N deposition

Decadal-average N2O emissions from nonagricultural soils in the 20th century simulation with atmospheric N deposition included were 8.3–10.3 Tg N yr−1, with individual years ranging between 7.7 and 11.0 Tg N yr−1 (Fig. 5). By the 1990s, the simulated increase in global soil N2O emissions as a result of increasing N deposition during the 20th century amounted to 0.80 ± 0.14 Tg N yr−1, similar to the IPCC estimate of 0.6 Tg N yr−1 (0.3–0.9 Tg N yr−1; Denman et al., 2007). The global N2O ‘emission factor’ for atmospheric N2O deposition on land during the 20th century was 1.15 ± 0.58% according to the model results – close to the IPCC default value of 1.25% (0.25–2.25%; Bouwman et al., 1995). Globally, c. 23% of the effect of N deposition on N2O emission during the 20th century was a result of its interaction with climate change and CO2.

The most important interactions of environmental factors in determining the modelled temporal pattern of global N2O emission over the 20th century are shown in Fig. 7. The effect of the CO2 increase alone is shown to be small and negative, but this changes to a positive effect when the interaction with climate change is included. (The single factor effect with interaction is calculated as the effect of CO2 and climate together, minus the effect of climate alone, according to the various sensitivity tests conducted in Fig. 5.) Similarly, the effect of climate change is large and positive, and further enhanced by interaction with the CO2 increase. The effect of N deposition is enhanced by its interactions with increasing CO2 and climate change.


Key processes determining annual N2O emissions
in space and time

Soil organic carbon (SOC) content has been considered to be one of the major determining factors for N2O emissions from agroecosystems (Bouwman et al., 2002; Li et al., 2004). It has also been suggested that soil N2O emissions by natural ecosystems should be proportional to either total N mineralization (Potter et al., 1996) or soil respiration (Xu et al., 2008). We examine the relationships between modelled N2O emissions and possible predictors of N2O emission: modelled SOC and inorganic N pool sizes (Fig. 8), and total N mineralization, soil respiration and denitrification rates (Fig. 9). These relationships are examined both spatially (across all grid points globally) and temporally (global average values throughout the 20th century).

Figure 8.

Simulated nitrous oxide (N2O) emissions compared with simulated soil organic carbon (SOC) and inorganic nitrogen (N) pool sizes: (a, b) spatial variation; (c, d) 20th century variation of global average values. The open squares in (c) and (d) are from simulations in which CO2 is constant and climate only is varied. Significance: **, P < 0.0001.

Figure 9.

Simulated nitrous oxide (N2O) emissions compared with simulated nitrogen (N) mineralization, soil respiration and denitrification rates: (a–c) spatial variation; (d–f) 20th century variation of global average values. The open squares in (d–f) are from simulations in which CO2 is constant and climate only is varied. Significance: **, P < 0.0001.

Spatially, the modelled soil N2O emissions are linearly related to both total N mineralization rate and soil respiration rate (Fig. 9). The slopes of these relationships (Fig. 9) are similar to those reported by Potter et al. (1996) and Xu et al. (2008): < 2% for the total N mineralization rate and ranging from 0.06 to 0.66 for soil respiration. However, the relationships with modelled SOC and inorganic N pools are weak, with modelled N2O emission related positively to SOC (with a broad range of slopes) and inversely to inorganic N (Fig. 8).

Over the 20th century, modelled global N2O emission is significantly positively correlated with N mineralization, soil respiration and SOC (Figs 8, 9) and negatively correlated with inorganic N levels. However, the relations to the SOC and N pool sizes disappear (Fig. 8), and the relations to N mineralization and soil respiration change slope markedly (Fig. 9), when the effect of CO2 changes is removed. Total denitrification, by contrast, is a very robust predictor of modelled soil N2O emission, with a slope of 3.0% for the spatial variation of emissions and 2.2–2.3% for the temporal variation of global emissions. The temporal relationship holds both with and without the effects of increasing CO2. These results indicate that the denitrification rate is a highly consistent predictor of N2O emissions. The other processes considered were less reliable as predictors because of their much stronger (indirect) responses to CO2 concentration.

Temperature vs precipitation effects

It is well documented that ‘pulses’ of soil N2O emission follow precipitation events (Li et al., 1992), and thus precipitation is often the dominant control of N2O emission on daily to seasonal time scales. However, little experimental information is available about the sensitivity of N2O emission to changes in environmental variables over periods longer than a year. One recent analysis, based on a 2-yr field experiment, found that ‘the N2O responses of temperate … grasslands to future climate change scenarios may be primarily driven by temperature effects’ (Cantarel et al., 2011). Although precipitation differences are responsible for much of the geographical variation in N2O emission by different ecosystems (Figs 3, 4, Table 2), our transient simulation indicated that temperature has been the dominant driver of simulated decadal variability in global N2O emission (Fig. 5). Effects of temperature variability on global N2O emission are also indicated by decadal-scale variations in the atmospheric N2O growth rate, as seen in ice core measurements (MacFarling Meure et al., 2006).

The dominant role of temperature variability could be, in part, simply because the long-term global trend in 20th century annual land precipitation was insignificant. Global land precipitation increased by only c. 3%, with some regions showing decreases (Trenberth et al., 2007). However, our model results also suggest that the enhancement of N2O emission by wetter conditions can be outweighed over multi-annual time scales by the more effective uptake of N by plants, in response to increased moisture availability, for plant growth (Table 3), thus allowing temperature effects to dominate. Long-term field experiments are required to test the hypothesis that temperature controls dominate over CO2 and precipitation controls of N2O emission on interannual to decadal time scales.

The CO2 effect

Only a small part of the increasing trend of simulated soil fluxes during the 20th century can be attributed to atmospheric CO2 concentration (Fig. 5). Increasing CO2 generally enhanced the N2O emission in tropical and temperate moist forests, whilst reducing the N2O emission in some other regions (Fig. 6). The simulated increase in N2O emission caused by the 73 ppm increase in CO2 concentration during the 20th century was c. 3%. This is comparable with the average effect size shown by van Groenigen et al. (2011) in a meta-analysis of experimental studies, showing an increase of 18.8% in N2O emissions for a CO2 increase of 317 ppm.

Soil N2O emission as a climate feedback

There is a considerable literature on the ‘climate–carbon cycle feedback’, whereby warming is predicted to lead to increased rates of soil organic matter decomposition and therefore to an enhancement of the rate of increase of atmospheric CO2 (Friedlingstein & Prentice, 2010). Here, we consider the possibility that a similar feedback exists involving soil emissions of N2O.

Modelled global soil emissions of N2O increased from 8.4 Tg N yr−1 in the 1900s to 9.2 Tg N yr−1 in the 1990s. We have shown land MAT to be the major control of this modelled increase and have estimated the sensitivity of soil N2O emission to land MAT to be c. 1 Tg N yr−1 K−1. As land temperatures have been rising 10–20% more rapidly than global temperatures, and 4.8 Tg N2O-N is equivalent to 1 ppb N2O in the atmosphere, Eqn (Eqn 1) implies a feedback strength of c. 30 ppb K−1 approached on a time scale of several N2O lifetimes (two to three centuries). The radiative efficiency of N2O is 3.7 W m−2 ppm−1; hence, this feedback amounts to 0.11 W m−2 K−1, a substantial amount, comparable in magnitude to the climate–carbon cycle feedback. For example, Frank et al. (2010) estimated a median climate–carbon cycle feedback strength of 7 ppm CO2 K−1, also equivalent to 0.11 W m−2 K−1, over a century.

A way to assess the magnitude of the climate–carbon cycle feedback over longer time scales is to consider the difference in equilibrium C storage in the terrestrial biosphere resulting from a warming of 1 K. Such computations were made by Gerber et al. (2004) using the LPJ model. It was found that a 1 K warming led to a loss of 131 Pg C, equivalent to 62 ppm CO2. Assuming that 20–35% of this CO2 remains in the atmosphere for more than two centuries (Archer et al., 2009), the implied feedback strength is 0.2–0.35 W m−2 K−1. Very roughly, the strength of the natural soil N2O feedback (when assessed over several centuries) may be about one-half to one-third of the climate–carbon cycle feedback.

We conclude, in agreement with Arneth et al. (2010), that a realistic assessment of biogeochemical feedbacks in the climate system must take into account the response of natural soil N2O emissions to climate. We have not attempted to simulate N2O emissions from fertilized agricultural soils, or any other anthropogenic effects associated with land use change or anthropogenic emissions. However, it seems likely that agricultural soils respond to climate variations in a qualitatively similar manner to nonagricultural soils. Indeed, there is experimental evidence for a strong effect of soil temperature on N2O emission from both unfertilized and fertilized grassland soils (Flechard et al., 2007). It follows that part of the observed variability in the growth rate of N2O concentration in the atmosphere could be a result of a climatic response in the large agricultural soil emissions of N2O (Grant & Pattey, 2008). The total feedback strength, in the presence of modern land use, may therefore exceed that predicted on the basis of emissions from ‘natural’ soils alone.


This research was funded by the National Natural Science Foundation of China (40975096, 40871032, 41175128), Strategic Priority Research Program-Climate Change: Carbon Budget and Related Issues of the Chinese Academy of Sciences (XDA05020402, XDA05050404-3-2), the UK Natural Environment Research Council under the QUEST (Quantifying and Understanding the Earth System) core team contract, the Oeschger Centre for Climate Change Research and the Swiss National Science Foundation. We thank Pru Foster for global mapping and Beth Holland and Sönke Zaehle for discussions and comments on earlier drafts. Xu-Ri gratefully acknowledges the support of the K.C. Wong Education Foundation, Hong Kong. Thanks are also given to the anonymous reviewers for their constructive comments. The authors declare no competing financial interests.