Global modeling of soil nitrous oxide emissions from natural processes

Authors

  • E. Saikawa,

    1. Center for Global Change Science, Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
    2. Department of Environmental Studies, Emory University, Atlanta, Georgia, USA
    Search for more papers by this author
  • C. A. Schlosser,

    1. Center for Global Change Science, Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
    Search for more papers by this author
  • R. G. Prinn

    1. Center for Global Change Science, Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts, USA
    Search for more papers by this author

Corresponding author: E. Saikawa, Department of Environmental Studies, Emory University, 400 Dowman Drive, Math and Science Center 5th Floor E512, Atlanta, GA 30322, USA. (eri.saikawa@emory.edu)

Abstract

[1] Nitrous oxide is an important greenhouse gas and is a major ozone‒depleting substance. To understand and quantify soil nitrous oxide emissions, we expanded the Community Land Model with coupled Carbon and Nitrogen cycles version 3.5 by inserting a module to estimate monthly varying nitrous oxide emissions between 1975 and 2008. We evaluate our soil N2O emission estimates against existing emissions inventories, other process‒based model estimates, and observations from sites in the Amazon, North America, Central America, Asia, Oceania, Africa, and in Europe. The model reproduces precipitation, soil temperature, and observations of N2O emissions well at some but not at all sites and especially not during winter in the higher latitudes. Applying this model to estimate the past 24 years of global soil N2O emissions, we find that there is a significant decrease (increase) in soil N2O emissions associated with El Niño (La Niña) events.

1 Introduction

[2] Nitrous oxide (N2O) is a major greenhouse gas with a global warming potential of approximately 300 in a 100 year time horizon [Forster et al., 2007]. N2O is also involved in stratospheric ozone depletion [Crutzen, 1970], and its emissions weighted by ozone‒depletion potential currently dominate those of ozone‒depleting substances following the decline of the chlorofluorocarbon emissions [Ravishankara et al., 2009]. Microbial production in soils is considered to be the largest producer of N2O [Davidson, 2009], contributing nearly 60% to the total [Werner et al., 2007]. Measurements of atmospheric N2O mole fractions collected at various stations in the world since the late 1970s show an increase (with a drop in 1992–1993) at a rate of 0.2%–0.3% per year [Weiss, 1981; Prinn et al., 1990; Nevison et al., 1996; Khalil et al., 2002], and the recent increase in the atmospheric mole fractions has led to an estimation of the anthropogenic source (including agricultural soil) to be approximately 1/3 of the total N2O source [Khalil et al., 2002; Hirsch et al., 2006; Nevison et al., 2007]. Despite a large number of studies in the last several decades examining the cause of this increase as well as estimating the magnitude and the source of N2O emissions, large uncertainties still remain [Nevison et al., 1996; Forster et al., 2007; Huang et al., 2008]. Understanding and quantifying N2O fluxes from global soil in long time series is therefore an urgent task for predicting the future climate change and stratospheric ozone depletion [Forster et al., 2007].

[3] The bacterial processes of nitrification and denitrification are considered to be the most important source of N2O emissions from soil [Davidson et al., 2000]. Microbial biomass decomposes in soil and creates ammonium ion (inline image), which is converted to nitrate (inline image) by the nitrification process in aerobic conditions. In this process, N2O is produced and a part of it is emitted to the atmosphere. During multistage redox reactions of denitrificaiton from inline image to N2, N2O is also generated and a small fraction of that can escape from the soil before further reduction to N2[Goreau et al., 1980; Bremner and Blackmer, 1981; Poth and Focht, 1985; Nevison et al., 1996]. These nitrification and denitrification processes have been stimulated further by the increasing use of synthetic nitrogen fertilizers for food production [Davidson, 2009; Park et al., 2012].

[4] The mechanism of N2O emissions from soil has been studied in several process models [e.g., Li et al., 1992, 2000; Bouwman et al., 1993; Potter et al., 1996; Xu et al., 2003; Werner et al., 2007], but so far, no model has been able to capture both the long‒term variability of soil N2O emissions as well as the details of seasonality and interannual variability at the global grid level. In this paper we present and evaluate an N2O emissions module added to the Community Land Model with coupled Carbon and Nitrogen cycles version 3.5 (CLM‒CN v3.5), in order to better understand the seasonality and interannual variability of global natural soil N2O emissions.

[5] CLM v3.5 is the land component of the Community Earth System Model, which is designed to study interannual and interdecadal variability, paleoclimate regimes, and projections of future climate change [Collins et al., 2006; Oleson et al., 2008]. With a coupled carbon‒nitrogen (CN) biogeochemical model [Thornton et al., 2007; Randerson et al., 2009; Thornton et al., 2009] based on the terrestrial biogeochemistry Biome‒biogeochemical cycle model [Thornton et al., 2002; Thornton and Rosenbloom, 2005], the CLM‒CN v3.5 model represents land terrestrial water and carbon (C) and nitrogen (N) balances, and it is nominally run at an hourly time scale [Lawrence et al., 2011].

[6] Here we add a new N2O emissions flux module within CLM‒CN v3.5 to create CLMCN‒N2O. CLMCN‒N2O includes all of the Denitrification‒Decomposition (DNDC) Biogeochemistry Model [Li et al., 1992] components to capture both the nitrification and denitrification processes that are important producers of N2O. CLMCN‒N2O estimates inline image produced by decomposition and calculates N2O production through nitrification and denitrification depending on soil temperature and moisture, utilizing the soil C and N concentrations in soil as calculated by CLM‒CN v3.5.

[7] The main objectives of this study were (1) to build and validate the soil N2O emissions module in CLM‒CN v3.5, (2) to quantify global natural soil N2O emissions between 1975 and 2008, and (3) to understand the effects of meteorology on seasonal and interannual natural soil N2O emissions. In this paper, the term “natural” soil N2O emissions refers to those from soil where we assume no application of artificial nitrogen fertilizers on the land. We therefore do not exclude any land areas from our model, but our emissions do not include agricultural emissions that are due to fertilizers. We first estimate the global natural N2O emissions from 1975 to 2008 and analyze the variability in annual and seasonal emissions in different regions. We use four separate forcing data sets (described in section 2.2) to compare our N2O emissions estimates due to given meteorological conditions. Next, we evaluate CLMCN‒N2O by comparing our estimated N2O emissions with observations from field measurements in the Amazon, the USA, Costa Rica, Indonesia, Australia, Kenya, and China. Finally, we analyze the impact of meteorology on regional emissions by paying special attention to the role of El Niño–Southern Oscillation (ENSO).

[8] The paper is organized as follows. Section 2 describes the methodologies we use, including the model development and the observational data for this study. Section 3 explains the model simulation and the comparison with existing emissions inventories. Section 4 provides the comparison of our model results with observations at 25 sites. Section 5 discusses the interannual variability of our modeled soil N2O emissions in relation to ENSO and some accounts on uncertainties in our model estimates. We present a summary of results and conclude in section 6.

2 Methods

2.1 Model

[9] CLMCN‒N2O is based on the CLM v3.5 model [Oleson et al., 2008; Stöckli et al., 2008] including the carbon‒nitrogen (CN) biogeochemical model [Thornton et al., 2007; Randerson et al. 2009; Thornton et al. 2009]. CLM‒CN v3.5 simulates terrestrial water and C and N budgets for each plant functional types (PFTs) from hourly to decadal time series. The model can be run on any regular grid and here we run at a horizontal resolution of 1.9° latitude and 2.5° longitude.

[10] There are 10 soil layers in CLM‒CN v3.5; from the top to the bottom, the layers are 0.007, 0.028, 0.062, 0.119, 0.212, 0.366, 0.620, 1.038, 1.073, and 2.865 m thick. The source of the soil data used in the model is the Global Soil Data Products by the International Geosphere‒Biosphere Programme (IGBP) [IGBP, 2000], and the scale of the soil data is 0.0833° latitude×0.0833° longitude. The data are aggregated to the model grids by using the dominant IGBP data within each of the coarse resolution grid cells.

[11] For each PFT and each spatial unit, CLM‒CN v3.5 balances soil C and N between four soil organic matter pools of differing decomposability (i.e., fast, medium, slow, and slowest), three litter pools (i.e., labile, cellulose, and lignin), and a soil mineral N pool. The dynamics of the soil C pool are calculated in CLM‒CN v3.5 using the converging cascade structure, where the litter and soil organic matter is decomposed in a cascade that converges at the end [Thornton and Rosenbloom, 2005]. The dynamics of the soil N pool are calculated in CLM‒CN v3.5 based on the C:N ratio specified by PFT. CLMCN‒N2O treats N inputs from atmospheric deposition, biological N fixation, N mineralization, and inline imageleaching as is modeled in CLM‒CN v3.5 and includes pools of N2O, inline image, NH3, and inline image. CLMCN‒N2O is added to CLM‒CN v3.5 in a one‒way coupling framework and simulates N2O emissions due to nitrification and denitrification at an hourly time step.

[12] inline image is produced via biomass decomposition. As both labile and resistant microbial biomass pools decompose, some new biomass is created, while others are transferred to resistant humads and the others produce CO2. In our module, following Li et al. [1992], the percentage of biomass decomposing to new biomass, resistant humads, and CO2are 60%, 20%, and 20%, respectively. In the process of decomposition to CO2, inline image is also produced but these do not feedback into the main model. Some of the produced inline image dissociates into NH3, a part of which then volatilizes. Nitrification is temperature and moisture dependent and during the nitrification process, inline image is produced from inline image, and in between, N2O is created by the following equation as described in Table 4 in Li et al. [1992]:

display math(1)

where NH4 is the inline image concentration in the soil liquid and T is the soil temperature.

[13] Denitrification, a process converting inline image into N2, is also soil temperature and moisture dependent and it takes place under the anaerobic state. In this module, we specify the anaerobic state when the water‒filled pore space is more than 41.5% in the soil layer. The specification of this threshold value can be dependent on factors relating to soil properties and the grid scale of the model. Herein, we select 41.5% rather than the 40% in Li et al. [1992] after a sensitivity analysis of this parameter to the resulting total soil N2O emissions from nitrification and from denitrification. We find that 40% leads to too frequent occurrence of anaerobic conditions in our model, whereas 41.5% results in approximately similar total N2O emissions both from nitrification and denitrification, which match the relative contributions found in the previous literature [e.g., Davidson et al., 2000].

[14] The growth rate (dB/dt)gof denitrifying bacteria is calculated proportional to the respective biomass, as explained in Table 6 in Li et al. [1992] as follows:

display math(2)

where μt,dn is the temperature reduction factor (inline image, if T < 60°C and 0 if T ≥ 60°C with T being soil temperature) and inline image is the pH reduction factor (the values are pHNO3= 1.002, pHNO2= 1.0, and pHN2O = 0.9985, taken from Table 6 in Li et al. [1992] and assuming the soil pH of 7.0). The relative denitrifier growth rate that affects the reduction rate of a specific nitrogenous oxide (inline image in a general form) is written as the following using the maximum growth rate of denitrifiers with different substrates (inline image), mineralized carbon concentration in the soil (C), half‒saturation value of soluble C (Kc,1/2), concentration of the NxOy in soil water (NxOy), and the half‒saturation value of the denitrifier (inline image), taken from Li et al. [1992]:

display math(3)

All the values except C (taken from the CLM‒CN v3.5) and NxOy (calculated in CLMCN‒N2O) are taken from Table 7 in Li et al. [1992].

[15] Consumption of these nitrogenous oxides is calculated by the following equation, taken from Li et al. [1992]:

display math(4)

where inline image is the maximum growth rate on NxOy, inline image is the maintenance coefficient on NxOy, and N is the total N as NxOy. The values for inline image and inline image are taken from Table 7 in Li et al. [1992].

[16] The N2O assimilation rate by plants and microbes is calculated as

display math(5)

where R is the C:N ratio in denitrifiers, given by Li et al. [1992].

[17] The net increase in N2O in the soil layer (ΔN2O) therefore is calculated as

display math(6)

[18] Adding this to the existing soil N2O concentration, the percentage of N2O emitted from soil is calculated as follows, based on the equation in Table 6 in Li et al. [1992]:

display math(7)

where AD is the adsorption factor, based on the clay content (clayfrac) in the soil (2.0 × clayfrac/max(clayfrac)), and PA is the air‒filled fraction of the total porosity.

[19] We use N deposition data as estimated by the Community Atmosphere Model for year 2000. Mineral N deposition is one of the pathways for N addition in the model. The flux is prescribed as an annual rate and is kept constant over the duration of the model run. It is applied daily to the soil mineral N pool. Our N deposition data would include the indirect effect of fertilizer use in agricultural lands, but we consider this to be a negligible effect based on Mosier et al. [1998] that the mean estimated emissions of N2O from atmospheric deposition is 0.3 (0.06–0.6) Tg N2O−N yr−1.

2.2 Measurement Data

[20] To test the model performance, we first compared the environmental variables (e.g., precipitation, volumetric water content (VWC), and soil temperature) that are inputs to our CLMCN‒N2O module at a couple of sites: (1) the Tapajós National Forest in east central Amazonia (2.90°S, 54.95°W), as described in Davidson et al. [2004, 2008], (2) the Fazenda Vitoriá forest located in eastern Amazonia, near Paragominas (2.98°S, 47.52°W) as described in Verchot et al. [1999], (3) the White Mountain National Forest in New Hampshire, USA (43.93°S, 71.75°W), as described in Groffman et al. [2006], and (4) the Golfo Dulce Forest Reserve in Costa Rica (8.72°N, 83.62∘W) as described in Wieder et al. [2011].

[21] In addition to the environmental variables, we compared our model results with the measurements at 25 sites including the four above (see Table 1). These sites cover a variety of spatial and temporal ranges on the soil without any (or with minimal) impact of the agricultural N fertilizers. The plant types from these measurements also include the following: rainforest, grassland, shrubland, plantation, steppe, and boreal forest. For the seven sites where the monthly means are available, we compared these values with our monthly mean calculated by using the four climate forcing data sets, as explained below. For the other 18 sites, we compared the seasonal or annual mean at each site with the equivalent mean values we calculated from our model results.

Table 1. Measurement Sites Used in This Study
SiteCountryMeasurement Year(s)Lat. [°N]Lon. [°E]Ecosystem TypeCitation
TapajósBrazil1998–2004−3306Tropical ForestDavidson et al. [2008]
White MountainUSA1997–200044289Hardwood ForestGroffman et al. [2006]
Golfo DulceCosta Rica20088277Tropical ForestWieder et al. [2011]
Muara EnimIndonesia2003–2004−3104Tropical ForestArai et al. [2008]
Pasir MayangIndonesia1997–1998−1102Tropical ForestIshizuka et al. [2002]
QueenslandAustralia2001–2002−17214Tropical ForestKiese et al. [2003]
Fazenda VitoriáBrazil1994–1996−3313Tropical ForestVerchot et al. [1999]
XishuangbannaChina200521101Tropical ForestWerner et al. [2006]
Inner MongoliaChina199843116GrasslandXu et al. [2003]
KakamegaKenya2005034Tropical ForestWerner et al. [2007]
QueenslandAustralia1997–1998−17145Tropical ForestBreuer et al. [2000]
AchenkirchAustria1998–1999, 2002–20034711Spruce ForestKesik et al. [2005]
CopenhagenDenmark19925512Spruce ForestKesik et al. [2005]
GlencorseUK2002–2003553Birch and Spruce ForestKesik et al. [2005]
Harvard ForestUSA19894272Hardwood ForestKesik et al. [2005]
HöglwaldGermany1994–1997, 2002–20034811Beech and Spruce ForestKesik et al. [2005]
HyytiäläFinland2002–20036124Boreal ForestKesik et al. [2005]
KlausenleopoldsdorfAustria1996–1997, 2002–20034816Beech ForestKesik et al. [2005]
MatrafüredHungary2002–20034820Spruce and Oak ForestKesik et al. [2005]
Parco TicinoItaly2002–2003459Hardwoods and Poplar ForestKesik et al. [2005]
San RossoreItaly2002–20034310Boreal ForestKesik et al. [2005]
SchottenwaldAustria1996–1997, 2002–20034816Beech ForestKesik et al. [2005]
SorøDenmark2002–20035512Beech ForestKesik et al. [2005]
SpeulderbosNetherlands2002–2003525Douglas‒Fir ForestKesik et al. [2005]
WildbahnGermany19975314Boreal ForestKesik et al. [2005]

2.3 Emissions Inventory Data

[22] We also compared the model results with two existing emissions inventory for global natural soil emissions. One is GEIA v1 in which an estimate of global N2O emissions from soils under natural vegetation and arable lands are calculated using the model by Bouwman et al. [1993]. Bouwman et al. [1993] use the “process pipe,” or “hole in the pipe,” concept [Firestone and Davidson, 1989; Davidson, 1991] and calculate the soil NO and N2O emissions flux simultaneously. In the model, total soil N availability determines the total N gas production, and the soil water content determines the ratio of N2O to NO emitted to the atmosphere. Because GEIA v1 only provides annual gridded global soil N2O emissions that combine both the natural and agricultural soil, we limit ourselves to comparing their total global estimate value for natural soil (taken from [Bouwman et al., 1995]) with our total estimate.

[23] Another inventory we compare our results with is the Carnegie‒Ames‒Stanford Approach (CASA) Biosphere model [Potter et al., 1996]. Similar to CLMCN‒N2O, CASA simulates natural soil N2O emissions as well as daily and seasonal patterns in C fixation, nutrient allocation, litterfall, soil N mineralization, and CO2 exchange. This model provides a monthly global soil N2O emissions with 1°×1° resolution for the year 1990, which allow us to analyze the spatial distribution of emissions.

3 Simulations and Soil N2O Emissions Flux Estimate

3.1 Simulations

[24] In order to estimate annual and monthly global soil N2O emissions flux, the CLMCN‒N2O was run with four climate forcing data sets: (1) Climate Analysis Section (CAS): a data set based on the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis [Qian et al., 2006], (2) Global Offline Land‒Surface Dataset (GOLD): a data set which combines the reanalysis with monthly observations [Dirmeyer and Tan, 2001; Betts et al., 2006], (3) Global Meteorological Forcing Dataset (GMFD): a data set based on the combination of global observation‒based data sets and the NCEP/NCAR reanalysis [Sheffield et al., 2006], and (4) NCEP Corrected by CRU (NCC): a data set based on the NCEP/NCAR reanalysis [Ngo‒Duc et al., 2005]. Each data set provides air temperature, humidity, wind speed, surface pressure, precipitation, and solar radiation to run CLM. Based on the availability of the data, CLMCN‒N2O was run for 1975–2004 for CAS, 1975–2002 for GOLD, 1975–2008 for GMFD, and 1975–2000 for NCC.

[25] To implement the CLMCN‒N2O simulation, we conducted the equilibrium run with the NCC data set, followed up by a transient run. For the equilibrium run, we used a 26 year repeating climate of 1949–1974 to drive the model to reach an equilibrium state. After running the model for 1300 years, we have determined that the model was at an equilibrium by confirming that an annual carbon storage change less than 1gC m−2yr−1was reached and the imbalance in the plant N cycle was approximately 1%, as specified in McGuire et al. [1992] and Lin et al. [2000], respectively. From the equilibrated state, we ran the model from the years available for CAS (1948), GOLD (1958), GMFD (1948), and NCC (1949), using the respective forcing data of matching years. We used the first years of the transient run to be a spin‒up and only analyzed years between 1975 and the respective end years. Annually varying meteorology with a constant N deposition for the year 2000 was used for all simulations.

3.2 Emissions Estimate

[26] Figure 1 shows the estimated global natural soil N2O emissions flux for the year 2000 using the GMFD forcing data set. The global spatial distribution of seasonal N2O emissions flux from the CLMCN‒N2O model illustrates that the emissions are the highest in the Northern Hemisphere summer and the lowest in the Northern Hemisphere winter. The global total emissions in the summer is more than twice higher than those in the winter. One of the reasons for the large seasonal variation is due to a clear seasonality in the Northern Hemisphere, where there are only little emissions in the winter, whereas high emissions are visible in the summer. We find no emissions in the high latitudinal regions in the winter because DNDC assumes no soil N2O emissions when the soil temperature is below 0°C.

Figure 1.

Global seasonally averaged soil N2O emissions flux for the year 2000 estimated by the CLMCN‒N2O model using the GMFD forcing data set in mgN m−2 month−1.

[27] We compare our annual total model estimates using the four forcing data sets for 2000 with CASA (supporting information Figure S1). From Figure S1, we notice that the model results using different forcing data sets lead to substantial differences in the spatial distribution of the emissions. However, our model results all tend to estimate lower emissions than those of CASA in the eastern part of South America, eastern part of Africa, and southern Europe. On the other hand, a part of Asia, a part of equatorial Africa, and the higher latitudinal regions in the Northern Hemisphere tend to have higher estimated emissions compared to those of CASA. Looking closely at the seasonal emissions, difference between the model results using the GMFD data set and the CASA model results, as shown in Figure S1, we find that CLMCN‒N2O predicts larger emissions in the higher latitudinal regions compared to CASA, especially in the summer.

3.3 Interannual Flux Estimate

[28] When using CAS, GOLD, GMFD, and NCC forcing data sets, the CLMCN‒N2O model estimates global average soil N2O emissions (for the period from 1975 through 2000) to be 9.10±0.49, 10.6±0.57, 7.55±0.54, and 7.42±0.35 TgN2O‒N yr−1, respectively. It is important to note that the differences among the forcing data sets is up to 3.5 TgN2O‒N yr−1, and it highlights that our model is sensitive to the input forcing data set, which leads to large uncertainty. Figure 2 shows the interannual variability of global total natural soil N2O emissions flux as estimated by the model using the four different forcing data sets. When we calculate the regional distribution based on seven regions (Figure 3) for year 2000, the largest emissions are always from Southern Asia with approximately 25%–35% of the global total, and the remaining emissions originate mainly in Central/South America and Africa/Middle East, sharing 23%–26% and 15%–16% of global total, respectively (see Figure 3). The lowest emissions originate in Europe and Oceania, with approximately 2.4%–3.1% and 1.9%–2.8% of the shares, respectively. One of the reasons for the difference in the magnitude of emissions is simply due to the area of the covered region. Europe and Oceania only has 5.5 and 8.1 km2, respectively, whereas the three largest regions Africa/Middle East, Northern Asia, and North America have 35.9, 23.4, and 23.3 km2, respectively. This regional emissions anomaly is mainly due to the differences in (1) temperature, leading to differing soil temperature and (2) precipitation, leading to differing VWC in the model estimates.

Figure 2.

Interannual variability of global soil N2O emissions derived from CLMCN‒N2O using four forcing data sets: CAS, GOLD, GMFD, and NCC.

Figure 3.

Regional total soil N2O emissions between 1975 and 2008 using four forcing data sets: (a) CAS, (b) GOLD, (c) GMFD, and (d) NCC.

[29] The interannual variability and spatial distribution of the emissions are similar among the results using different forcing data sets. Based on the multivariate ENSO index, we decide that the El Niño years are those with the index above 0.5 and La Niña years are those with the negative index values. Based on this criterion, 1977, 1979, 1980, 1982, 1983, 1987, 1991, 1992, 1993, 1994, 1997, 1998, and 2002 are the El Niño years and 1975, 1976, 1984, 1985, 1988, 1989, 1996, 1999, 2000, 2001, 2007, and 2008 are the La Niña years. We find decreased emissions in the beginning of the El Niño events and increased emissions in most of the La Niña as well as after the dip in El Niño events, especially in Southern Asia as shown in Figure 3. We discuss the reasons for this oscillation in the discussion section below.

[30] The global total soil N2O emissions from our model are in alignment with the existing bottom‒up model estimates. Bouwman [1990] cites Seiler and Conrad [1987] that the estimated natural soil global budget of N2O is 6 ± 3 Tg N2O‒N yr−1. Bouwman et al. [1995] estimate global natural soil N2O emissions flux in 1990 to be 6.6–7.0 Tg N2O‒N yr−1, and Potter et al. [1996] estimate it to be 6.1 Tg N2O‒N yr−1. There are other estimates made in 1980s and 1990s including 7‒16 Tg [Bowden, 1986], 3‒25 Tg [Banin, 1986], and 6.7 Tg [Kreileman and Bouwman, 1994]. Schlosser et al. [2007] estimate natural soil emissions from the Global Land System to be 6.1 Tg N2O‒N yr−1. Conducting a top‒down inversion study and assuming that oceanic flux has not changed over time, Hirsch et al. [2006] estimate the preindustrial terrestrial source to be 3.9–6.5 TgN yr−1. Global average soil N2O emissions using CAS and GOLD forcing data sets (9.10±0.49 and 10.6±0.57 TgN2O‒N yr−1, respectively) are on the higher end of the estimates, whereas those using GMFD and NCC forcing data sets (7.55±0.54 and 7.42±0.35 TgN2O‒N yr−1, respectively) are equivalent to the recent estimates.

4 Model‒Observation Comparison

[31] In this section, we evaluate the model performance at various time scales as well as input variables with the observations available. First, we compare several important environmental variables that we use as inputs for our model. Second, we compare our modeled natural soil N2O emissions with various sites depending on the data available at each site: (1) monthly average time series, (2) seasonal/monthly average, and (3) annual average. We summarize our key insights and uncertainties in the model in section 5 below.

4.1 Environmental Variables

[32] In order to evaluate our model, we first compare several forcing parameters for both observations and model inputs for CLMCN‒N2O. There are three important parameters to estimate soil N2O emissions, which are precipitation, VWC, and soil temperature. Precipitation is essential to achieve anaerobic condition for denitrification to start in the model. VWC is directly linked to soil moisture used in the model to determine whether the soil layer is at the anaerobic state for denitrification to take place. Soil temperature affects the amount of soil N2O emissions emitted to the atmosphere in both nitrification and denitrification processes.

[33] We first conduct comparisons with two Amazon measurements. One is from the Tapajós National Forest in east central Amazonia between 1999 and 2000 [Davidson et al., 2004, 2008], and the other is from Fazenda Vitoriá between 1995 February and 1996 May [Verchot et al., 1999]. Figures 4a and 5a illustrate that our meteorological forcing inputs from the four data sets capture the variability and the seasonality of observed precipitation. Although there are some large discrepancies of over 200 mm month−1in Fazenda Vitoriá, correlation coefficients at the Tapajós National Forest of the observations and the four forcing data sets are 0.88 (CAS), 0.70 (GOLD), 0.83 (GMFD), and 0.82 (NCC), respectively, and the values at Fazenda Vitoriá are 0.87 (CAS), 0.88 (GOLD), 0.87 (GMFD), and 0.81 (NCC).

Figure 4.

Comparison of (a) precipitation in mm month−1, (b) VWC in cm3 cm−3, (c) soil N2O emissions flux in ngN cm−2 h−1 between the observations and modeled values using four forcing data sets (CAS, GOLD, GMFD, and NCC) in CLM‒CN v3.5, and (d) the relationship between N2O fluxes (model (blue) and observations (red)) in ngN cm−2 h−1 and VWC of the top 30 cm soil with nitrous oxide in cm3 cm−3 at the Tapajós National Forest, taken from Davidson et al. [2008]. Dashed lines for observations indicate that the data are not continuous.

Figure 5.

Comparison of (a) precipitation in mm month−1 and (b) soil N2O emissions flux in ngN cm−2 h−1 between the observations and modeled values using four forcing data sets (CAS, GOLD, GMFD, and NCC) in CLMCN‒N2O at the Fazenda Vitoriá, taken from Verchot et al. [1999]. Dashed lines for observations indicate that the data are not continuous.

[34] VWC calculated in CLM‒CN v3.5 as water content in a volume of soil in cm3cm−3, used as an input to CLMCN‒N2O, is also able to reproduce the variability of measured values at the Tapajós site (Figure 4b) and the Costa Rica sites (Figure 7a) well but not at the White Mountain National Forest in New Hampshire, USA (Figure 7a). Correlation coefficients of the measurements and the model inputs using the four forcing data sets at Tapajós are 0.70 (CAS), 0.82 (GOLD), 0.73 (GMFD), and 0.83 (NCC). In Costa Rica, the correlation coefficients are 0.70 (CAS), 0.82 (GOLD), 0.73 (GMFD), and 0.83 (NCC), and the modeled VWC are within the observational uncertainty in 7 out of the 9 months.

[35] The average correlation coefficients, on the other hand, at the White Mountain National Forest between observations (both sugar maple and yellow birch) and the CLM‒CN v3.5 model outputs are 0.36 (CAS), 0.26 (GOLD), 0.31 (GMFD), and 0.21 (NCC). At the White Mountain National Forest, CLM‒CN v3.5 overestimates the observed VWC most of the time, and this discrepancy is, in large part, due to the shallow depth of the soil moisture considered (10 cm), as well as to the fact that CLM tracks the macroscale variations of the soil hydrothermal profile—allowing for multiple vegetation types to compete with the same soil‒column water. This may also be due to this forest being northern hardwood vegetation with the soils that are shallow (75–100cm) and acidic (pH 3.9), which diverges greatly from the module assumption of the pH being 7. These mismatches are expected at the local scale in a global model with a relatively coarse resolution. Overall, we find that the model overestimates at some sites and underestimates at other sites. This indicates a degree of subgrid scale heterogeneity that the coarse‒grid level cannot resolve. We hope to focus on this subgrid scale heterogeneity issue in our future work, as more comprehensive observations become available to do so.

[36] CLM‒CN v3.5, however, represents the variability of the soil temperature well at the White Mountain National Forest, as seen in Figure 6b. The average correlation coefficients between the measurements (both sugar maple and yellow birch) and the estimates calculated in CLM‒CN v3.5 are 0.97 (CAS), 0.97 (GOLD), 0.98 (GMFD), and 0.96 (NCC). CLM‒CN v3.5 slightly overestimates measured temperature in the summer, but it is able to reproduce the seasonal cycle well, as the correlation coefficients illustrate. However, it is worth noting that the model‒estimated soil freezes at this depth in the winter months in most simulations, while the measurements do not. This has important implications as 0°C is the threshold below which soil N2O is not emitted in our model.

4.2 N2O Emissions

[37] We compare our modeled natural soil N2O emissions with observations at 25 sites as listed in Table 1. First, we compare our model results with observations that have monthly mean values for more than 9 months. These sites include (1) two Amazon sites (the Tapajós National Forest and Fazenda Vitoriá), Brazil, (2) the White Mountain National Forest, USA, (3) Golfo Dulce, Costa Rica, (4) two Indonesian sites (Muara Enim and Pasir Mayang), and (5) Queensland, Australia. Second, we compare our model results with observations that have limited monthly and seasonal averages at the following sites: (1) Xishuangbanna, China, (2) Inner Mongolia, China, (3) Kakamega, Kenya, and (4) Queensland, Australia. Finally, we compare our model results with annual averages at 14 sites listed in Table 5.

4.2.1 Monthly Average Time Series

[38] Figure 4c compares the model results using four forcing data sets (NCC until the end of 2000 and GOLD until the end of 2002, while CAS and GMFD are shown until August 2004) and observed natural soil N2O emissions in the Tapajós National Forest. Correlation coefficient between the measurements and the model results is the highest (0.78) when using NCC. On the other hand, the correlation coefficient between the measurements is the lowest (0.54) using GOLD. The root‒mean‒square error is the lowest (1.06 ngN cm−2h−1) using CAS and is the highest (1.57 ngN cm−2h−1) using GOLD due to a large underestimation of emissions when high observations are recorded.

[39] At the Tapajós National Forest, we also compare the relationship between VWC and soil N2O emissions flux between the observations and the model. As depicted in Figure 4d, we find that there is a nonlinear yet robust relationship between VWC and soil N2O emissions for both model results (blue) and observations (red). The model results are clustered around VWC between 0.29 and 0.38, whereas the observations are more scattered, and as for the reference case, there are VWC values greater than 0.407—the model maximum value in this grid cell. Observed soil N2O emissions vary from less than 0.2 to more than 6.6 when VWC is approximately 0.41. The model also shows a similar behavior, where there is a variation in soil N2O emissions in the same order of magnitude when the VWC reaches higher than 0.37. There are many more data points for the model compared to the observations, and there is a need for more measurements to better understand the relationship between VWC and the soil N2O flux.

[40] Figure 5b compares the modeled (with four forcing data sets) and observed (primary forest, secondary forest, active pasture, and degraded pasture) soil N2O emissions in Fazenda Vitoriá. Although located at the same site, soil N2O emissions from different vegetations vary as large as 8.2 ngN cm−2h−1in the spring and these observations illustrate a significant difficulty in modeling emissions, especially at the coarse resolution in a global scale. Our model results estimate measurements best from the primary forest, and the correlation coefficients of these measurements and the model results are 0.70 (CAS), 0.62 (GOLD), 0.67 (GMFD), and 0.69 (NCC). The model is not able to capture the variability of the emissions from other ecosystems as well as it does for the primary forest, and the model results lie between the emissions from primary forest and from pasture in most years. Considering the scale difference between the measurements and the model grids as well as the temporal scale difference between the model and instantaneous measurements, the model captures the characteristic features seen in the observations at this site.

[41] In addition to the scale issue, one of the reasons the model is unable to reproduce the natural soil N2O emissions from a certain ecosystem is due to the PFT considered at this specific grid cell. We have four PFT types on the grid cell that we analyze in Figure 5b, and they include C4 grass, broadleaf evergreen tropical tree, broadleaf deciduous tropical tree, and corn. Primary forest is the forest that has never gone through human intervention, whereas secondary forest has, and no PFT is included for the pasture lands for this grid cell. It is therefore very possible that, under these PFT types, we are unable to resolve the emissions exactly as estimated from the measurement at a specific ecosystem type.

[42] Figure 6c illustrates the modeled (using four forcing data sets) and observed (sugar maple and yellow birch branches) soil N2O emissions flux in the White Mountain National Forest. In our current model setup, there are no soil N2O emissions when the soil temperature is below 0°C, based on the assumption in the DNDC model. Therefore, we find a clear seasonality in all four model results where there are maximum emissions in the growing season and zero or negligible emissions in the winter. In the measurements, however, we find noticeable emissions from December through March, and this might be due to the discrepancy in the soil temperature between the observations and model results as discussed above. The model tends to have lower soil temperature values (below 0°C) which result in no emissions, when observations show positive temperature (Figure 6b). Still, the model results, especially using the GFMD data set, are able to reproduce the variability of the emissions, with the correlation coefficients between the observed and the modeled as 0.63 for Sugar Maple and 0.62 for Yellow Birch.

Figure 6.

Comparison of (a) VWC in cm3 cm−3, (b) soil temperature in K, and (c) soil N2O emissions flux in ngN cm−2 h−1 between the observations and modeled values using four forcing data sets (CAS, GOLD, GMFD, and NCC) in CLMCN‒N2O at the White Mountain National Forest, taken from Groffman et al. [2006]. Dashed lines for observations indicate that the data are not continuous.

[43] Groffman et al. [2006] find that increasing soil freezing enhances soil N2O emissions flux and that winter fluxes are important. Kielland et al. [2006] have, on the other hand, suggested that labile substrate production might be more important than temperature for soil N2O production. Moreover, the observations also indicate that there are conditions resulting in N2O uptake, which most N2O emission models would be unable to capture. More research is needed to enhance our understanding of the winter fluxes and to insert an uptake process in the model, as this might be an important source of N2O exchange.

[44] Figure 7b compares the measured soil N2O emissions and the modeled flux using the GMFD forcing data set in the Golfo Dulce Forest Reserve in southwest Costa Rica. The correlation coefficient between the observed and the modeled is 0.45, and the model does a poor job of reproducing emissions. The largest discrepancy is in November 2008, and even when taking the standard error into account, the modeled flux is overestimating the observations by 75%. This yet again illustrates that soil N2O emissions do not simply depend on soil moisture, as we find agreement between the observations and the CLM‒CN v3.5 model outputs of VWC at this site (Figure 7a). This large discrepancy in soil N2O emissions may be due to the difference in modeled atmospheric temperature to the observed as the monthly average of the model results does not compare well with the average statistics in 2008, resulting in a correlation value of −0.52. However, no observational data at the measurement site are available to conduct further analysis at this time.

Figure 7.

Comparison of (a) VWC and (b) soil N2O emissions flux in ngN cm−2 h−1 between modeled values using four forcing data sets (CAS, GOLD, GMFD, and NCC) and observations in the Golfo Dulce Forest Reserve in Costa Rica, taken from Wieder et al. [2011]. Dashed lines for observations indicate that the data are not continuous.

[45] Figure 8 illustrates the range in soil N2O emissions based on the type of ecosystems ((a) Acacia Forests and (b) Secondary Forest) measured at Muana Enim, Indonesia, taken from Arai et al. [2008], and how our model results lie in between the two. As the difference in the scale in y axis indicates, Acacia Forests emit substantially higher emissions, of more than an order of magnitude, than from the Secondary Forests. The timing of the maximum emissions within a year also differ between the two. Model results using GMFD forcing data set have high correlation coefficients of larger than 0.7 with measurements from two Acacia Forest sites (AM1 and AM2) and 0.80 with those from a Secondary Forest site (SF3). With the measurements at the secondary forest site 2 (SF2), all the modeled results have an anticorrelation of approximately −0.3. Those large variations in soil N2O emissions found simply with different forcing data sets reconfirm that the model is sensitive to the input forcing data set. Figure 8 illustrates the need for more high‒frequency measurements as well as finer‒resolution modeling using the best available forcing data sets to resolve these issues in the global scale.

Figure 8.

Comparison of soil N2O emissions flux in ngN cm−2 h−1 between modeled values using two forcing data sets (CAS and GMFD) and observations at (a) Acacia Forests and (b) Secondary Forests at Muara Enim, South Sumatra Province, Indonesia, both taken from Arai et al. [2008]. Model results using NCC and GOLD forcing data set are not shown, as NCC and GOLD are only available for years before 2000 and 2002, respectively.

[46] Figure 9a shows the comparison between the modeled estimates using the four forcing data sets and the measured soil N2O emissions from Pasir Mayang Research Site, Indonesia for two different ecosystem types (Primary forest for P1 and P2 and Logged‒over forest for L1 and L2). There are also striking differences between the four modeled results at this site, and the variability is captured the best by the model results using the NCC forcing data set for all measurements except for measurements at L2, which is represented better by the model using CAS. The correlation coefficients between the model results using NCC and observations are 0.57 (P1), 0.65 (P2), 0.59 (L1), and 0.32 (L2). The correlation coefficient between the model estimate using CAS and the L2 measurements is 0.68. Most of the correlation coefficients between measurements and the other model results are negative, but the model results using GOLD and GMFD most often lie in between the observations for the primary forest and for the logged forest. Even when these model results overestimate, they are closer to the measurements compared to the values estimated by the model using CAS. What this illustrates, therefore, is the fact that soil N2O emissions is complex and forcing data sets play a large role in our model.

Figure 9.

Comparison of soil N2O emissions flux in ngN cm−2 h−1 between (a) modeled values using four forcing data sets (CAS, GOLD, GMFD, and NCC) and observations at the Pasir Mayang Research Site in Jambi province, Sumatra, Indonesia, taken from Ishizuka et al. [2002] and (b) modeled values using three forcing data sets (CAS, GOLD, and GMFD) and observations at the Tropical Forests at Queensland, Australia, taken from Kiese et al. [2003]. Model result using NCC forcing data set is not shown, as NCC is only available for years before 2000.

[47] Figure 9b compares the observed soil N2O emissions with the model results using three forcing data sets (CAS, GOLD, and GMFD). At this site, the model produces very flat estimates without large seasonality. Measurements, however, show large fluxes between November and May—in the Southern Hemisphere summer—which the model is not able to reproduce. This discrepancy is most likely due to the model inputs for precipitation, where all the four forcing data sets are problematic at reproducing Southern Hemisphere precipitation in the summer. Kiese et al. [2003] reports the mean annual precipitation at the site to be 4360 mm, but the forcing data sets provide the values almost four times less: 1017 (CAS), 791 (GOLD), and 919 mm (GMFD) for November 2001 to October 2002. Although we realize that the soil N2O emissions is not a linear process, as found earlier, the difference between the model‒estimated emissions in a large grid cell and the measured values also has the same order of difference as the precipitation, which reconfirms the importance of the input data to our CLMCN‒N2O model.

4.2.2 Seasonal/Monthly Average Emissions

[48] In addition to the longer time series data of monthly measurements, we compared seasonal and monthly averages of soil N2O emissions at four sites, as shown in Tables 2-4. First, we compared the modeled fluxes using the GMFD forcing data set with observations at Xishuangbanna, China (Table 2), taken from Werner et al. [2006]. The model is within the error range for the emissions from the primary forest in March, although it overestimates the observations (both from primary and secondary forest) in February and underestimates in April, both approximately by 30%.

Table 2. Model‒Observation Comparisons of N2O Flux in ngN cm−2h−1From Primary and Secondary Forests at Xishuangbanna, Chinaa
TimePrimary ForestSecondary ForestModel (GMFD)
  1. a

    Measurements taken from Werner et al. [2006].

5 Feb0.33±0.09 0.54
5 Mar0.50±0.110.71±0.130.42
5 Apr0.74±0.160.84±0.180.34

[49] In Table 3, we compare the model produced flux estimates using the four forcing data sets with the observations from three sets of ecosystem types (Leymus chinensis steppe, free‒grazing area, and Stipa grandis steppe) at Inner Mongolia, China, taken from Xu et al. [2003]. We compare seasonal averages and find that our models, especially the results using CAS and GOLD, are able to reproduce the measurements from Leymus chinensis steppe and free‒grazing area in the summer of 1998. However, emissions from Stipa grandis steppe are lower than those from the other ecosystem types by 54%–61%, and none of our model estimates can reproduce them. In the spring, the emissions from Leymus chinensis steppe are larger than those of the other ecosystems by 80%–120% and models underestimate even compared to those from Stipa grandis steppe, which has the lowest emissions. In autumn, on the other hand, models overestimate emissions by 200%–680% depending on the ecosystems. For these months, Xu et al. [2003] find that the version 7.2 of the DNDC model had significantly underestimated emissions and they applied modification to the nitrification by including the impact of low temperature and soil frost on N2O emissions to correct them. Because we underestimate in spring in a similar manner, we may also try to include thawing impacts in our model in the future. It is, however, also possible that more modifications may be necessary as we do not see the same issue in autumn.

Table 3. Model‒Observation Comparisons of N2O Flux in ngN cm−2h−1From Leymus chinensis Steppe (LC), Free‒Grazing Area (GLC), and Stipa grandis Steppe (SG) at Inner Mongolia, Chinaa
 MeasuredModel
LCGLCSGCASGOLDGMFDNCC
  1. a

    Measurements taken from Xu et al. [2003], Table 3.

Spring 19980.350.190.160.060.050.000.04
Summer 19980.590.520.320.530.590.630.64
Autumn 19980.100.110.050.310.350.390.34

[50] Table 4compares the model results using GMFD and the measurements at Kakamega, Kenya, taken from Werner et al. [2007]. The model notably underestimates emissions in April and May, at the beginning of the rainy season. Furthermore, the variability we find is the opposite between the measurements and the model results: decreasing emissions in the measurements and the increasing emissions in the model during the 3 months. These observations are based on the automated measurement system, taking 10–16 measurements per day. Werner et al. [2007] observed large pulse emissions of N2O after the first rainfall events in April, which led to the high monthly average value. CLMCN‒N2O is unable to take into account these specific conditions which result in large pulsed emissions. There is a need for model improvement to be able to reproduce these characteristics in soil emissions in the future.

Table 4. Model‒Observation Comparisons of N2O Flux in ngN cm−2h−1at Kakamega, Kenyaa
TimeMeasuredModel (GMFD)
  1. a

    Measurements taken from Werner et al. [2007].

5 Apr7.39±1.810.14
5 May3.37±0.440.43
5 Jun1.06±0.280.74

4.2.3 Annual Average Emissions

[51] Finally, we compare the model results with annually averaged observations found in Kesik et al. [2005] in Table 5. A large range in emissions flux was often visible at the same site among the model results using different forcing data sets. For example, in Achenkirch, Austria, model estimates varied from 0.31 to 1.80 ngN cm−2h−1. The model result using GOLD reproduces the results best for 1998 and 1999, whereas the model using CAS and GMFD reproduces the measurements better for 2002 and 2003. In many sites with high observed values such as Höglwald, Matrafüred, and Schottenwald, the model estimates using GOLD tends to reproduce the highest values that are closest to the observations. However, it is important to remember that some annually averaged measurements are calculated with a limited number of observations, whereas the average for the model uses all the data for that specific year.

Table 5. Model‒Observation Comparisons at 14 Sitesa
SiteCountryYearMeasuring DaysN2O flux [ngN cm−2 h−1]
From Kesik et al. [2005]This Study
MeasuredPnET‒N‒DNDCCASGOLDGMFDNCC
  1. a

    Measurements and PnET‒N‒DNDC model results are taken from Table 2 in Kesik et al. [2005]. Bold italic and italic model results illustrate that the model estimate is within 1‒ and 2‒sigma standard error from the measurements mean, respectively. The bold model results indicate that the measured values lie within the model results using four forcing data sets.

AchenkirchAustria1998181.67±0.250.96±0.170.391.800.310.44
AchenkirchAustria1999191.00±0.210.96±0.170.391.360.370.45
AchenkirchAustria20021220.46±0.041.21±0.040.500.750.47 
AchenkirchAustria20031940.33±0.041.25±0.040.360.380.37 
CopenhagenDenmark1992170.96±0.210.21±0.040.320.320.380.32
GlencorseUK2002870.21±0.040.54±0.000.220.070.42 
GlencorseUK20031220.08±0.000.25±0.080.160.050.38 
Harvard ForestUSA1989100.04±0.080.08±0.040.980.950.920.84
Höglwald (Beech)Germany19941051.13±0.043.17±0.210.401.390.360.40
Höglwald (Spruce)Germany19943450.46±0.001.00±0.040.401.390.360.40
Höglwald (Beech)Germany19953414.17±0.214.71±0.170.371.040.390.42
Höglwald (Spruce)Germany19953590.88±0.041.17±0.380.371.040.390.42
Höglwald (Beech)Germany19963077.58±0.426.54±0.330.351.090.350.37
Höglwald (Spruce)Germany19963433.54±0.331.92±0.080.351.090.350.37
Höglwald (Beech)Germany19973482.29±0.134.00±0.130.371.810.350.38
Höglwald (Spruce)Germany19973460.75±0.080.96±0.040.371.810.350.38
Höglwald (Beech)Germany20021030.96±0.080.67±0.080.500.750.47 
Höglwald (Spruce)Germany20023430.79±0.041.00±0.040.500.750.47 
Höglwald (Beech)Germany20031581.63±0.131.63±0.170.360.380.37 
Höglwald (Spruce)Germany20033400.42±0.040.83±0.210.360.380.37 
HyytiäläFinland2002170.04±0.040.21±0.040.130.120.11 
HyytiäläFinland2003110.08±0.000.08±0.040.110.110.10 
KlausenleopoldsdorfAustria1996162.75±0.421.58±0.170.411.480.420.43
KlausenleopoldsdorfAustria1997212.17±0.461.29±0.210.451.080.400.52
KlausenleopoldsdorfAustria20021190.79±0.081.29±0.080.501.160.48 
KlausenleopoldsdorfAustria20031780.63±0.041.58±0.040.380.390.41 
Matrafüred (Oak)Hungary2002101.67±0.580.58±0.080.571.790.53 
Matrafüred (Spruce)Hungary2002112.00±0.630.17±0.080.571.790.53 
Matrafüred (Oak)Hungary2003272.75±0.332.21±0.210.500.510.41 
Matrafüred (Spruce)Hungary2003282.00±0.290.54±0.040.500.510.41 
Parco Ticino (BoscoNegri)Italy2002570.58±0.130.67±0.080.601.110.77 
Parco Ticino (Poplar)Italy2002480.21± 0.080.42±0.040.601.110.77 
Parco Ticino (BoscoNegri)Italy20031770.21±0.000.50±0.040.420.490.47 
Parco Ticino (Poplar)Italy2003170.29±0.080.25±0.040.420.490.47 
San RossoreItaly2002650.38±0.041.71±0.040.620.830.71 
San RossoreItaly20031830.08±0.000.88±0.040.570.560.37 
SchottenwaldAustria1996166.63±1.173.75±0.500.411.480.420.43
SchottenwaldAustria1997215.58±1.383.46±0.460.451.080.400.52
SchottenwaldAustria20021624.75±0.383.33±0.130.501.160.48 
SchottenwaldAustria20032524.08±0.212.67±0.080.380.390.41 
SorøDenmark2002191.00±0.210.71±0.080.350.370.36 
SorøDenmark20031560.63±0.041.04±0.040.300.320.29 
SpeulderbosNetherlands20021070.33±0.040.75±0.040.470.440.49 
SpeulderbosNetherlands20032160.17±0.001.13±0.040.520.490.49 
WildbahnGermany1997100.71±0.080.50±0.080.280.250.290.31

[52] Kesik et al. [2005] used these measurements at the 14 sites to evaluate their PnET‒N‒DNDC model that is modified slightly from Li et al. [2000]. A couple of the sites and years where their simulated means lie within the 1‒sigma measurement uncertainty are Achenkirch in 1999, Harvard Forest in 1989, Höglwald (Spruce) in 1995, Höglwald (Beech) in 2003, Hyytiaälä in 2003, Matrafüred (Oak) in 2003, Parco Ticino (BoscoNegri) in 2002, and Sorø in 2002. From Table 5, the sites and years that meet the same requirement for CLMCN‒N2O are quite different: Achenkirch in 1998, 2002, and 2003, Glencorse in 2002, Höglwald (Spruce) in 2002 and 2003, Matrafüred (Oak) in 2002, Matrafüred (Spruce) in 2002, and Parco Ticino (BoscoNegri) in 2002.

5 Discussion

[53] Based on the rigorous comparisons between the model estimates and observations, we find that the CLMCN‒N2O emissions module is able to reproduce emissions for certain ecosystems as a global model at a relatively coarse resolution. In this section, we ask two questions related to the model results. The first is on the strong interannual variability from our estimated emissions. We try to analyze what might be causing this variability, focusing on the El Niño and La Niña events. The second is the uncertainty analysis on our estimates. Although it is beyond the scope of this paper to assess the uncertainty in all aspects of the simulations, we focus on a few important aspects of the uncertainty in our model associated with the four forcing data sets and our model setup.

5.1 Interannual Variability

[54] From Figure 2, we find that our model results, regardless of the forcing data sets used, illustrate decreased emissions in the beginning of the El Niño events and increased emissions in most of the La Niña, as well as after the dip in El Niño events. 1992—one of the El Niño years—shows the lowest of all the estimated emissions over the simulated years when using GOLD, GMFD, or NCC. 1992 has the second lowest emissions when using CAS. This result matches well with the study that finds the growth rate of N2O in 1992 to be half that in the previous decade [Smith, 1997], despite the continued growth before and after 1992.

[55] Bouwman et al. [1995] suggest that this low growth rate in 1992 was possibly due to the observed global cooling associated with the eruption of Mount Pinatubo (located in the Philippines, 15°N and 121°E) in 1991, which caused lower N2O emissions in soils. Figure 3 illustrates that there was a large decrease in emissions in 1992 in Southern Asia, where the eruption took place. It is therefore possible that the relatively low emissions we find in 1992 may not simply be due to the El Niño event but also due to the eruption leading to low atmospheric temperature, affecting our temperature dependence in the model.

[56] To illustrate the emission anomaly, we analyze the difference between the N2O emissions estimates in El Niño years and those in La Niña years in the summer months of June, July, August, and September (Figure 10) and in the winter months of December, January, February, and March (Figure 11). Here and beyond, because GMFD covers the longest time period for the years we have analyzed, we compare the soil N2O emissions results using the GMFD forcing data set. Figure 10 illustrates that when comparing emissions in El Niño years to La Niña years, negative emissions anomalies are found in a northern part of North America, a northern part of South America, Southern Asia (most visibly in South and Southeast Asia), and the central and the equatorial part of Africa. On the other hand, there are positive anomalies in a part of China and the southern part of South America. This spatial anomaly is similar in winter, but the anomalies are not visible in the higher latitudinal regions. We find similar spatial patterns in all four model results using different forcing data sets, suggesting that the emissions anomaly is due to the model characteristics and is not because of the difference in our forcing data sets.

Figure 10.

Average of the soil emissions flux anomalies in the Northern Hemisphere summer (June, July, August, and September) using the GMFD forcing data set between the El Niño years (1977, 1979, 1980, 1982, 1983, 1987, 1991, 1992, 1993, 1994, 1997, 1998, and 2002) and the La Niña years (1975, 1976, 1984, 1985, 1988, 1989, 1996, 1999, 2000, 2001, 2007, and 2008) in gN m−2 month−1.

Figure 11.

Average of the soil emissions flux anomalies in the Northern Hemisphere winter (December, January, February and March) using the GMFD forcing data set between the El Niño years (1977, 1979, 1980, 1982, 1983, 1987, 1991, 1992, 1993, 1994, 1997, 1998, and 2002) and the La Niña years (1975, 1976, 1984, 1985, 1988, 1989, 1996, 1999, 2000, 2001, 2007, and 2008) in gN m−2 month−1.

[57] Precipitation, VWC, and soil temperature are important parameters for determining soil N2O emissions in the CLMCN‒N2O model. By analyzing the anomaly of these variables in El Niño and in La Niña years, we find high correlations between soil N2O emissions and the three environmental variables within some regions (Figure 12). For the equatorial regions, stronger correlations between emissions and precipitation/VWC are visible. However, in the upper Northern Hemisphere regions, there is a higher correlation between emissions and soil temperature.

Figure 12.

Anomalies of (a) soil N2O emissions, (b) soil temperature, (c) precipitation, and (d) VWC using GMFD forcing data set between the annual average of El Niño years (1977, 1979, 1980, 1982, 1983, 1987, 1991, 1992, 1993, 1994, 1997, 1998, and 2002) and the La Niña years (1975, 1976, 1984, 1985, 1988, 1989, 1996, 1999, 2000, 2001, 2007, and 2008).

[58] To analyze the general impact of environmental variables on soil N2O emissions in the model, we also calculated the correlations between soil N2O emissions and the following variables: (1) soil temperature, (2) precipitation, (3) VWC, (4) GPP, (5) net N mineralization rate, and (6) soil inline image concentration (Figure 13). As expected, we find high correlations between emissions and VWC in the equatorial regions and between soil N2O emissions and soil temperature in the upper Northern Hemisphere regions.

Figure 13.

Correlation coefficients between the CLMCN‒N2O calculated N2O emissions and the following variables for 1975 and 2008: (a) Soil Temperature, (b) Precipitation, (c) VWC, (d) GPP, (e) net N mineralization rate, and (f) Soil inline image concentrations. The model calculation is done using GMFD forcing data set.

[59] What is interesting, however, is the high correlations between soil N2O emissions and GPP in the nonequatorial regions. Ciais et al. [2005] find that the European heat wave and drought in 2003 caused a reduction in GPP, due to stomatal closure. Their finding not only supports our high correlation values between these two variables, but also the reduction in soil N2O emissions in El Niño years. We furthermore find large correlations between emissions and net N mineralization rate as well as soil inline image concentrations. This high correlation is reasonable considering that GPP is directly related to the amount of available C in soil, which is correlated to plant N demand in the model. Net N mineralization rate affects nitrification, producing soil inline image concentrations, and thus, these two variables also have high correlations to soil N2O emissions. Our model indicates that direct climate impact from drought and heat waves, as well as the nonlocal influence due to El Niño, have impacts on soil N2O emissions.

5.2 Uncertainty Analysis

[60] A detailed study on the uncertainty of our model results is beyond the scope of the paper, but we believe that our simulation results with the four forcing data sets suggest where the uncertainty may lie in the simulation results. We find lower N2O emissions in the beginning of El Niño and higher N2O emissions in La Niña regardless of the data sets. However, we also find a larger range in absolute values when using different forcing data sets (as large as 3.5 TgN yr−1), as indicated in Figure 2, and it highlights that our model is highly sensitive to the input forcing data set. Each simulation, with a different forcing data set, provided a different magnitude in emissions from different regions as well. As illustrated in Figure 13, multiple environmental variables are responsible for high correlations with soil N2O emissions in different parts of the world.

[61] We also find that a large range of soil N2O emissions is possible when the VWC is around 0.36 and 0.38 (Figure 4d). This means that the little change in the calculated VWC in this range leads to a large difference in the emissions estimates. However, as Schlosser and Gao [2010] find, it is more likely that model structure would play a more important role in improving global soil N2O emissions estimates. For example, changing the threshold value for denitrification itself impacts not only the total emissions estimates but also the regional variation.

6 Concluding Remarks

[62] In this study, we have linked the Community Land Model (CLM) with a process‒based model of N2O soil emissions. When comparing with available measurements, we find that the model reproduces the observations, especially at the two Amazon sites, the Tapajós National Forest and Fazenda Vitoriá, but not so well at some of the other sites. For example, it was clear that the model was unable to reproduce the emissions at the White Mountain National Forest in USA, partly due to the discrepancy in the soil temperature between CLM‒CN v3.5 and observations, as well as the lack of winter activity in the DNDC model. Our model results suggest that the improvement in the winter dynamics within the upper Northern Hemisphere regions will most likely lead to better capturing the soil N2O exchange. Further research is needed to explore the possibility of including the winter soil biological processes from soil freezing and thawing as well as the observed N2O uptake events in the model framework.

[63] An analysis of annual and seasonal variation of global soil N2O emissions reveals some interesting insights. We find significant interannual variations in the global natural soil N2O emissions in our model simulation. Our composite analysis indicates strong changes to N2O emissions associated with ENSO, especially in Southern Asia, and there is a known interannual variability in atmospheric mole fractions of N2O [Nevison et al., 2007]. Therefore, together, these suggest a need for better understanding of the impact soil N2O emissions has on the atmosphere.

[64] This study indicates a clear relationship between the climate and soil N2O emissions in our model, and we find significant negative precipitation anomalies in these El Niño years in the aforementioned regions. We furthermore observe the impact of other environmental variables such as GPP, net N mineralization rate, and soil inline image concentrations, where we find high correlations between these variables and soil N2O emissions. It is thus possible that future climate change will have a large impact on global soil N2O emissions and vice versa. More study is necessary to understand this feedback mechanism as well as to improve the model to better predict soil N2O emissions.

Acknowledgments

[65] This research was supported by NASA Upper Atmosphere Research Program grants NNX11AF17G and NNX07AE89G to MIT and the federal and industrial sponsors of the MIT Joint Program on the Science and Policy of Global Change. We thank Klaus Butterbach‒Bahl (Institute of Meteorology and Climate Research, Atmospheric Environmental Research Karlsruhe Institute of Technology, Germany), Eric Davidson (The Woods Hall Research Center, USA), Peter Groffman (Cary Institute of Ecosystem Studies, USA), Louis Verchot (Center for International Forestry Research, Indonesia), Shigehiro Ishizuka (Forestry and Forest Products Research Institute, Japan), Christian Werner (Biodiversity and Climate Research Centre (BiK‒F), Germany), and Will Wieder (National Center for Atmospheric Research, USA) for providing their observational data. Analysis of data from Hubbard Brook were supported by U.S. National Science Foundation grant NSF DEB‒0919131. We thank the two anonymous reviewers for their thorough and constructive comments.

Ancillary

Advertisement