Impacts of Nitrogen Addition on Nitrous Oxide Emission : Model-Data Comparison

The contributions of long-lived nitrous oxide (N2O) to the global climate and environment have received increasing attention. Especially, atmospheric nitrogen (N) deposition has substantially increased in recent decades due to extensive use of fossil fuels in industry, which strongly stimulates the N2O emissions of the terrestrial ecosystem. Several models have been developed to simulate N2O emission, but there are still large differences in their N2O emission simulations and 15 responses to atmospheric deposition over global or regional scales. Using observations from N addition experiments in a subtropical forest, this study compared six widely-used N2O models (i.e. DayCENT, DLEM, DNDC, DyN, NOE, and NGAS) to investigate their performances for reproducing N2O emission, and especially the impacts of two types of N additions (i.e. ammonium and nitrate: NH4 and NO3, respectively) and two levels (low and high) on N2O emission. In general, the six models reproduced the seasonal variations of N2O emission, but failed to reproduce relatively larger N2O emissions due to 20 NH4 compared to NO3 additions. Few models indicated larger N2O emission under high N addition levels for both NH4 and NO3. Moreover, there were substantial model differences for simulating the ratios of N2O emission from nitrification and denitrification processes due to disagreements in model structures and algorithms. This analysis highlights the need to improve representation of N2O production and diffusion, and the control of soil water-filled pore space on these processes in order to simulate the impacts of N deposition on N2O emission. 25


Introduction
Nitrous oxide (N2O) is one of the most important greenhouse gases, and contributes 6.24 % to overall global radiative forcing as the third contributor after carbon dioxide and methane (Forster et al., 2007;WMO, 2011).N2O plays an important role in depleting stratospheric ozone, which decreases harmful ultra-violet radiation reaching the earth.A doubling of the atmospheric N2O concentration could decrease the ozone layer by 10 % (Crutzen and Ehhalt, 1977;Ravishankara et al., 2009).Since the industrial revolution, the atmospheric N2O concentration has increased nearly 21 % from about 270 ppbv during the pre-industrial era to 325.9 ppbv in 2013, with an average increase rate of about 0.82 ppbv yr -1 during the last decade (WMO, 2014).Terrestrial ecosystems can act as either sources or sinks for atmospheric N2O, depending on time and location (Potter et al., 1997;Ridgwell et al., 1999;Chapuis-Lardy et al., 2007;Xu et al., 2008).Globally, natural sources from terrestrial ecosystems contribute more than 50 % to the N2O releases to the atmosphere (Denman et al., 2007).
Quantifying N2O fluxes in global terrestrial ecosystems, therefore, is an urgent task for predicting future climate change (Sheldon and Barnhart, 2009).
Several process-based N cycle models have been developed and widely used for quantifying the spatial-temporal variations in N2O flux (Li et al., 1992;Engel and Prentice, 1993;Grant et al., 1993;Potter et al., 1996;Xu and Prentice, 2008;Zhuang et al., 2012).In general, these models usually integrate key biogeochemical processes, including nutrient mineralization, immobilization, nitrification, and denitrification.However, there exist substantial model disagreements in the estimated magnitude and spatial distribution of N2O at regional and global scales (Figure 1).For example, Xu and Prentice (2008) used the DyN model to estimate global terrestrial ecosystem N2O emission at 13.31 Tg N yr −1 , which is 3.94 times the estimate of 3.37 Tg N yr −1 arrived at by Zhuang et al. (2012) (Figure 1).Atmospheric nitrogen (N) deposition, which is closely related to N2O emission, has shown a strong increasing trend in recent decades due to the extensive use of fossil fuels in industry and transportation and the heavy application of fertilizers in agriculture (Galloway et al., 2004).It is estimated that global atmospheric N deposition has increased from 1 Tg N in the 1860s to 25-40 Tg N in the 2000s, and is projected to continuously increase to 210 Tg N by the year 2050 (Neff et al., 2002;Lamarque et al., 2005;Galloway et al., 2008;Lu et al., 2016).The forest ecosystem in eastern China was recognized as the region receiving the highest atmospheric N deposition in southeast China (Liu et al., 2013).The N deposition input into terrestrial ecosystems alters plant physiology and the soil microbial community (Litten et al., 2007;Treseder, 2008), thereby changing the soil biogenic N2O flux (Butterbach-Bahl, 1997;Allen et al., 2004;Bange, 2006;Chen et al., 2015).Based on a meta-analysis of N addition experimental data worldwide, Liu and Greaver (2009) concluded that N addition could increase N2O emission by up to 216 %.In general, chronic N deposition will increase ammonium (NH4 + ) and nitrate (NO3 -) availability in terrestrial ecosystems, thereby affecting N2O flux through changing the activity and composition of the microbial community (Smith et al., 2003;Bowden et al., 2004;Monteny et al., 2006).However, to our knowledge, few studies have evaluated model performance in simulating the impacts of N deposition on N2O emission.
In this study, six widely-used N2O models, i.e.DayCENT (the daily version of the CENTURY ecosystem model; Parton et al., 1996Parton et al., , 2001;;Del Grosso et al., 2001), DNDC (the Denitrification-Decomposition model; Li et al., 2000), DLEM (Dynamic Land Ecosystem Model;Tian et al., 2010), DyN (the global Dynamic Nitrogen model; Xu and Prentice, 2008), NOE (the algorithm of Nitrous Oxide Emission; Henault et al., 2005), and NGAS (a hybrid of a process-oriented model and a nutrient cycling model; Parton et al., 1996), were chosen for examination of their performance under different levels of N deposition in a subtropical forest in southeast China.The study aims to (i) examine performance of the models in a forest ecosystem, particularly for seasonal variations of N2O emission, (ii) investigate the ability of these models under different levels of N deposition as well as two N types, and (iii) identify the key issues in the application of these models and future model development so as to improve their simulation of N2O emissions.

Study Site
This model-data comparison is based on field experiments conducted at a subtropical evergreen forest station, the Qianyanzhou Ecological Station (26°44′39″ N, 115°03′33″ E).The station is in Jiangxi Province of southern China, which is one of the important regions subject to atmospheric N deposition.The study plots were located in the slash pine plantation < 0.002 mm (15 %).Soil bulk density, organic carbon, total N content, and pH of the surface part (0-40 cm) were 1.57 g cm - 3 , 7.2 g kg -1 , 0.55 g kg -1 , and 4.6, respectively.The study site has a humid monsoon climate with a mean air temperature of 17.9 °C and precipitation of 1469 mm per year.A large portion of the precipitation occurs in spring and early summer, but it is relatively dry in late summer and autumn with high air temperatures and low precipitation.

Field Experiments
The field experiments were conducted during April-December 2012.According to previously reported levels of atmospheric N deposition at the study area (Wang et al., 2011), two levels (low and high N of 0 and 120 kg N ha -1 yr -1 , respectively) of two different N fertilizers (NH4Cl and NaNO3) were applied to mimic two future scenarios of N deposition.
At the same time, a control experiment was carried out for comparison.Each level of N treatment was conducted in a plot of 20 m × 20 m with a space of 10 m between any two plots.The N fertilizer solutions were sprayed on the plots once a month in 12 equal applications, and the control plots received only equivalent deionized water.
Flux data of N2O were determined using a static opaque chamber and gas chromatography method (Fang et al., 2014), which were installed near an eddy covariance tower in the ecological station.Daily fluxes were collected from the measurements approximately every two weeks.The soil fluxes were calculated based on the rate of changes in their concentration within the chamber, estimated as the slope of the linear regression between concentration and time (Wang et al., 2011).Soil temperature at 5 and 10 cm depths were monitored at each chamber site, using portable temperature probes (JM624 digital thermometer, Living-Jinming Ltd., Tianjin, China).At the same time, soil samples were collected nearby the static chambers from a depth of 0-20 cm using an auger (2.5 cm in diameter).Volumetric soil moisture (m 3 m -3 ) was measured using a moisture probe meter (TDR100, Spectrum Technologies Inc., PlainField, IL, USA).Soil pH was also measured using the potentiometry method.Soil water-filled pore space (WFPS) was calculated using the methods reported by Fang et al. (2014).

N2O Models
Six N2O models were selected in this model-data comparison: DayCENT (Parton et al., 1996(Parton et al., , 2001;;Del Grosso et al., 2001), DNDC (Li et al., 2000), DLEM (Tian et al., 2010), DyN (Xu and Prentice, 2008), NOE (Henault et al., 2005), and NGAS (Parton et al., 1996).All six investigated N2O models are based on two major microbial processes: nitrification and denitrification, which are separately simulated from these two processes using the following equation: where   2  is the N2O emission from soil to air (g N m -2 day -1 ), and   and   are N2O emissions from nitrification and denitrification processes, respectively.Detailed model algorithms can be found from the Supplemental Online Materials.

Simulation Protocol and Parameter Inversion
The field observations of soil temperature, soil moisture, pH, soil respiration, dissolved organic carbon, soil NH4 + content, and soil NO3 -content were used to drive the six models.As one of the key drivers, WFPS was derived using the following equation (Fang et al., 2014): where VWC is soil volumetric moisture content (%), BD is soil bulk density (g cm -3 ), and 2.65 is soil particle density (g cm - 3 ).
The nonlinear regression procedure (Proc NLIN) in the Statistical Analysis System (SAS, SAS Institute Inc., Cary, NC, USA) was applied to optimize the model parameters using observed N2O emission for all five experiments.The calibrated parameter values were used to simulate N2O emissions (Table S1).
Three metrics were used to evaluate the performance of these models: (i) The coefficient of determination between observation and simulation (R 2 ).
(ii) Absolute predictive error (PE), quantifying the difference between simulated and observed values.
(iii) Relative predictive error (RPE), computed as: where  ̅ and  ̅ are mean simulated and mean observed values, respectively.

Results
All six models generally reproduced the seasonal variations of measured N2O fluxes for the control and four N addition experiments.The measurements showed the largest N2O emissions during April-July, and the lowest in winter (Figure 2).The simulated emissions showed some differences in estimates for various models.Although the simulated N2O emissions from different models decreased from spring and summer to autumn and winter, indicating the seasonal pattern of emissions (Figure 1), there were some abrupt changes in model estimates.Most models captured the peak and trough of N2O emission.Collectively, the six models explained 1 %-16 % of the variations in N2O fluxes across all experiment plots (Table 1).experiments, and high NO3 -and NH4 + additions led to higher N2O emission compared to low additions (Figure 3).145 Furthermore, larger increases of N2O emission occurred for NH4 +compared to NO3 --addition experiments (Figure 3).However, NGAS, DyN, DayCENT, and DNDC models simulated larger NO2 fluxes for low compared to high NH4 +addition treatments (Figure 3).NOE and NGAS did not correctly indicate the differences of N2O fluxes between high and low NO3 -treatments.In addition, the experiments also indicated higher simulations of N2O emission for NH4 + compared with NO3 -additions.However, only NOE and DLEM models reproduced larger impacts of NH4 + on N2O emissions 150 compared with low NH4 + level.Because N2O emissions are generally from two different microbial processes, i.e. nitrification and denitrification, the proportions of N2O emissions due to both processes were calculated to quantify their contributions to total emissions.All six models showed consistently negative correlations between the ratios of N2O emission from nitrification and WFPS (Figure 4a).The six models showed that nitrification contributed more than half of N2O emissions; however, there were large 165 differences in the ratios of N2O fluxes generated by nitrification and denitrification among the models (Figure 4b).On average, the DayCENT model simulated the lowest ratio (about 55.4 %) of N2O emissions generated by nitrification, and the largest ratio (about 89.5 %) was for the DyN model (Figure 4b).

Model Performance
Compared with ecosystem carbon dioxide emissions, few studies have evaluated model performance for simulating N2O emissions due to the relative scarcity of N cycle measurements (Henault et al., 2012).Notably, N depositions from the atmosphere have been documented to increase with industry processes (Galloway et al., 2004), which are believed to have significant impacts on soil N2O emissions due to their impacts on microbial processes.Therefore, the sixth IPCC report, which will be conducted in the next five years, requires Earth System Models integrate N cycle (IPCC, 2017).Therefore, process-based N2O models have been widely developed and applied in recent years (Li et al., 1992;Engel and Priesack, 1993;Parton et al., 1996;Potter et al., 1997;Del Grosso et al., 2001;Henault et al., 2005;Xu and Prentice, 2008;Tian et al., 2010).These models are now being used not only for the prediction of N2O emissions from different ecosystems, but estimation of N2O inventories on national, regional, and global scales, and for assessing climate change impacts and mitigation strategies (Del Grosso et al., 2006, 2009;EPA, 2006).However, it should be noted that these model predictions may not be reliable when applied to a new environment, and their performance should be first tested with different data streams from real world experiments.
Our comparison showed the general performance of six investigated models in reproducing seasonal variations and magnitudes of N2O emissions (Figure 2).This conclusion was supported by several recent model evaluations, which revealed unstable performance of N2O models (Senapati et al., 2016).For example, different studies with the DayCent model have found a range of correlations from weak to strong across different agroecosystems (Henault et al., 2012).Parton et al. (2001) found correlations between daily measured vs. simulated N2O emissions, with range 0-0.44, from a variety of five Biogeosciences Discuss., https://doi.org/10.5194/bg-2018-126Manuscript under review for journal Biogeosciences Discussion started: 28 March 2018 c Author(s) 2018.CC BY 4.0 License.different grassland sites in the USA.Other ecosystem models also face similar difficulties in simulation of daily N2O emissions, for example DNDC (Yeluripati et al., 2015) and CoupModel (He et al., 2016).
The comparison revealed the complexity in modeling the impacts of N addition on N2O emission.The field observations in the current study indicated larger N2O emissions for NH4 + compared with NO3 -additions at two addition levels (Figure 3).However, these impacts were not reproduced by all of the six models except for the DELM and NOE models.Previous study showed that the impacts of NH4 + addition on N2O emissions are, to some extent, larger compared with NO3 -addition (Wang et al., 2016).This is probably due to two primary reasons.One is that under favorable temperature and moisture, nitrification dominates N2O emission compared with denitrification if soil is acidic and rich in NH4 + .The addition of NH4 + can significantly increase the substrates for ammonia-oxidizers and the abundance of ammonia-oxidizing archaea, which give rise to increases in soil autotrophic nitrification rate (Gao et al., 2016a;2016b).The other reason is that additions of NH4 + fertilizers can have larger impacts on the acidification of soil compared with the additions of NO3 -, which is closely related to the accumulation of H + in soil solution and the leaching of NO3 -from soil (Tian and Niu, 2015).Soil acidification decreases availability of NH4 + , which is favorable to the growth of soil nitrifiers, i.e. ammonia-oxidizing archaea, but unfavorable to soil denitrifiers (Isobe et al., 2012).

Structure Differences among N2O Models
The performance of these N2O models strongly depends on model algorithms, and also on the major pathways of N2O emissions and their responses to environmental conditions.This is because the processes of N2O emissions are extremely competitive and are controlled by many drivers, e.g.soil temperature, moisture, soil redox potential, and the availability of substrates for microbes (Schmidt et al., 2000).In the present study, the models did not adequately capture the environmental regulation of N2O emission.Nitrification and denitrification are two major processes of N2O production.Numerous experiments have shown that nitrification and denitrification can occur simultaneously because of the coexistence of aerobic and anaerobic zones in soils (Henault et al., 2012;Hu et al., 2015); however, the availability of soil oxygen-determined by soil water content and other soil propertiesstrongly regulates the proportion of nitrification and denitrification (Li et al., 1992).Numerous studies have investigated the relationship between soil moisture and the contributions of nitrification and denitrification processes.In N fertilizer-amended soil, N2O emission has been found to be highly correlated with WFPS, with the highest emission at around 70 % WFPS, which was attributed to a combination of nitrification (35 %-53 %) and denitrification (only 2 %-9 %) (Huang et al., 2014).In sandy loam soils, when moisture status was sub-optimal for denitrification (50 % and 70 % WFPS), nitrification was the significant contributor (around 29 %) to N2O emissions (Kool et al., 2011); however, in wetter soils (-0.1 kPa) nitrification contributed less than 3 % (Webster and Hopkins, 1996).Well et al. (2008) attributed 88 % of total N2O emission to nitrification at 45 %WFPS.This suggests that favorable conditions for N2O production from nitrification occur within the range of 30 %-70 %, whereas denitrification dominates N2O production in wet soils with WFPS>80 % (Braker and Conrad, 2011;Huang et al., 2014).The values of WFPS in the current study were within the range of 30 %-70 %, which was favorable to the occurrence of nitrification in all of these models.In general, all six N2O models use soil water content to control the balance of two processes.NOE uses a simplified scheme to separate the nitrification and denitrification processes.Nitrification only occurs if WFPS < 80 %, whereas denitrification only occurs if WFPS > 62 %; within the range of 62 %-80 %, the two processes may occur simultaneously (Henault et al., 2005).For DLEM, denitrification and nitrification are simulated as a one-step process.Due to the effect of soil moisture, denitrification only occurs when soil moisture exceeds field capacity (Tian et al., 2010).For DyN and DNDC, aerobic and anaerobic microsites are assumed to simultaneously exist in most soils.Nitrification occurs in aerobic microsites, but denitrification is mainly in anaerobic microsites.The key factor affecting the ratio between aerobic and anaerobic microsites is soil redox potential, which controls the ratio between nitrification and denitrification (Li et al., 1992;Xu and Prentice, 2008).For NGAS and DayCENT, no specific threshold is applied for the occurrences of the two processes and they are assumed to occur simultaneously (Parton et al., 1996(Parton et al., , 2001;;Del Grosso et al., 2001).Thus, the differences in the algorithms of the six models are believed to be the key reasons for the differences in the model estimates of N2O emission.

Conclusions
We examined the performance of six N2O models for indicating the impacts of different levels of N addition on N2O emission.Results indicated that the investigated models can represent the general seasonal variations of N2O emissions under both N addition and non-N addition levels.However, additions of NH4 + rather than NO3 -could have more significant effects on N2O emissions from soils, which were not represented by most of the models.In addition, most of the models failed to reproduce larger N2O emissions at high level of nitrate additions compared with ammonia additions.Moreover, the analysis suggested that the disagreements in model structure and algorithms resulted in substantial differences in N2O emission and mediating processes (i.e.nitrification and denitrification).
Competing interests.The authors declare that they have no conflict of interest.Data availability.The data can be obtained upon request to the authors.

Figure 1 .
Figure 1.Comparison of global estimates of N2O emission from the terrestrial ecosystem.

Figure 2 .
Figure 2. Comparisons of N2O emission simulations and observations for five experiment treatments.Most models did not fully indicate the stimulations of N additions to N2O emission that were observed in field experiments.According to the observed N2O fluxes, NO3 -and NH4 + additions increased N2O emission for four addition

Note.Figure 3 .
Figure 3. Comparisons of N2O emission differences between N addition and control treatments from observation and model simulations.Lower-case letters indicate significant differences among the values for different N addition level for an individual model or observation.Capital letters indicate the significant difference among the values for the same levels of N 160 addition from different model simulations or observation.

Figure 4 .
Figure 4. Proportions of N2O emissions for different models.(a) Seasonal proportion of N2O emission from nitrification and seasonal water-filled pore space (WFPS).(B) Averaged proportion of emissions from nitrification and denitrification.