Enhancing the soil and water assessment tool model for simulating N2O emissions of three agricultural systems

Abstract Nitrous oxide (N2O) is a potent greenhouse gas (GHG) contributing to global warming, with the agriculture sector as the major source of anthropogenic N2O emissions due to excessive fertilizer use. There is an urgent need to enhance regional‐/watershed‐scale models, such as Soil and Water Assessment Tool (SWAT), to credibly simulate N2O emissions to improve assessment of environmental impacts of cropping practices. Here, we integrated the DayCent model's N2O emission algorithms with the existing widely tested crop growth, hydrology, and nitrogen cycling algorithms in SWAT and evaluated this new tool for simulating N2O emissions in three agricultural systems (i.e., a continuous corn site, a switchgrass site, and a smooth brome grass site which was used as a reference site) located at the Great Lakes Bioenergy Research Center (GLBRC) scale‐up fields in southwestern Michigan. These three systems represent different levels of management intensity, with corn, switchgrass, and smooth brome grass (reference site) receiving high, medium, and zero fertilizer application, respectively. Results indicate that the enhanced SWAT model with default parameterization reproduced well the relative magnitudes of N2O emissions across the three sites, indicating the usefulness of the new tool (SWAT‐N2O) to estimate long‐term N2O emissions of diverse cropping systems. Notably, parameter calibration can significantly improve model simulations of seasonality of N2O fluxes, and explained up to 22.5%–49.7% of the variability in field observations. Further sensitivity analysis indicates that climate change (e.g., changes in precipitation and temperature) influences N2O emissions, highlighting the importance of optimizing crop management under a changing climate in order to achieve agricultural sustainability goals.


Introduction
Increasing greenhouse gas emissions have raised growing concerns about their potential warming impacts on the global climate system (Lashof andAhuja 1990, Solomon et al. 2009). Although the concentration of N 2 O in the atmosphere is much lower than that of CO 2 and CH 4 (Flückiger et al. 1999), N 2 O plays a disproportionately important role in contributing to global warming due to a long atmospheric lifetime (Ko et al. 1991) that contributes to its high global warming potential (Lashof and Ahuja 1990). In addition, N 2 O is the primary ozone-depleting gas in the stratosphere (Ravishankara et al. 2009). The agriculture sector is the major source of anthropogenic N 2 O emissions due to excessive fertilizer use (Reay et al. 2012).
N 2 O emissions are regulated by numerous factors including soil nitrogen contents, soil temperature, soil water, and quality of organic residues (Firestone et al. 1980, Novoa and Tejeda 2006, Butterbach-Bahl et al. 2013. Production of N 2 O through reduction of nitrate (NO 3 − ) and oxidation of ammonia (NH 4 + ) is directly controlled by levels of the two inorganic nitrogen species. Excessive nitrogen input via chemical fertilizer application has been considered as a key driver for the high N 2 O emissions from agricultural ecosystems (Thomson et al. 2012). However, non-linear correlations between fertilizer application and N 2 O emissions suggested that additional factors, such as soil temperature and moisture, may add variability Manuscript received 13 November 2016; revised 30 December 2016; accepted 3 January 2017. 7 E-mail: xuesong.zhang@pnnl.gov to the response of N 2 O production to fertilizer addition (Kim et al. 2013b). Microbial activities during nitrification and denitrification tend to be more active under higher temperatures (Kätterer et al. 1998), suggesting that air temperature plays an important role in the seasonal patterns of N 2 O fluxes (Rezaei Rashti et al. 2015). Soil water content is another factor with significant role in regulating N 2 O emissions. Water-filled pore space (WFPS) determines the reduction and oxidation environment in soils and thus controls the relative contribution of nitrification and denitrification to total N 2 O emissions (Bateman and Baggs 2005). Other factors, such as soil pH, soil carbon (Shcherbak et al. 2014), and soil texture, also impact N 2 O emissions, either through regulating microbial activities or through affecting soil water content (Weier et al. 1993).
Investigating the confounding impacts of multiple environmental factors on N 2 O emissions is critical for enriching understanding of N 2 O production, emission, and mitigation (Deng et al. 2016, Liu et al. 2016. Numerical modeling investigations are important in complementing and extrapolating field observations. While model simulation experiments are useful in disentangling the complex interactions among different environmental factors and ecological processes (Yang et al. 2015, Yang andZhang 2016), process-based algorithms have been developed and applied to quantify contributions of multiple processes and factors to N 2 O emissions, as well as to project N 2 O emissions under alternative climate and management scenarios (Del Grosso et al. 2008, Abdalla et al. 2010, Rafique et al. 2014. There is an urgent need to enhance regional-/watershed-scale agricultural models to simulate N 2 O emissions to complement their existing strengths in assessing impacts of cropping practices on soil quality, soil erosion, and water quality. The Soil and Water Assessment Tool (SWAT, Arnold et al. 1998) has been widely applied to assess impacts of crop cultivation on biogeochemical cycling (El-Khoury et al. 2015), hydrological dynamics , Leta et al. 2015, and environmental pollutions (Holvoet et al. 2008). Recent efforts (Zhang et al. 2013) incorporated the CENTURY model (Parton et al. 1994) into SWAT to simulate residue-soil organic matter (SOM) dynamics. N 2 O production and subsequent emissions are, however, not represented in the model, limiting application of SWAT to provide comprehensive assessment of agricultural activities on nitrogen cycling.
Our primary objective of this study was to improve SWAT's representation of soil nitrogen cycling by modifying its nitrification and denitrification algorithms and adding N 2 O emission algorithms. Specifically, we integrated the DayCent model's nitrification, denitrification, and N 2 O production modules (Del Grosso et al. 2000) with the existing widely tested crop growth, hydrology, and nitrogen cycling processes in the SWAT. We tested this new tool for simulating N 2 O emissions at three cropping sites (i.e., a continuous corn site, a switchgrass site, and a reference site dominated by smooth brome grass) located in the Great Lakes Bioenergy Research Center (GLBRC) scale-up fields in southwestern Michigan. A local parameter sensitivity analysis was conducted to understand how N 2 O estimates respond to changes in key parameters. We also analyzed how changes in precipitation and temperature affect N 2 O emissions. This work strengthens SWAT's capability to provide comprehensive assessment of sustainability of agricultural ecosystems under a changing climate.

Integrating DayCent's N 2 O emission algorithms into SWAT
The SWAT N 2 O emission algorithms are based on Parton et al. (2001) that simulate N 2 O production from both nitrification and denitrification. Specifically, soil ammonia oxidation is simulated with the following equations: where N nit is soil nitrification rate (g N·m −2 ·d −1 ); f moist represents impacts of soil water on nitrification, and f st represents soil temperature impacts; f pH refers to the pH impacts on nitrification; SW is soil water content (mm H 2 O); STH is soil depth (mm); WP is soil moisture at wilting point (mm H 2 O); FC is soil moisture at field capacity (mm H 2 O); SW mim is minimum volumetric soil water content (unitless); SW del is minimum volumetric soil water content below wilting point (0.042, unitless); ST is soil temperature (Celsius degree); SPH refers to soil pH; NH4 is soil ammonia content (g N/m 2 ); N nit_max is maximum nitrification rate (0.4 g N·m −2 ·d −1 ); N nit is denitrification rate (g N·m −2 ·d −1 ); N nit_base is minimum nitrification rate (0.00001 g N·m −2 ·d −1 ); f nit_max is maximum fraction of ammonia that is nitrified during nitrification (unitless). N 2 O production from nitrification is calculated as a fraction of nitrified ammonia: (6) f pH = 0.56 + 1 π × atan (π×0.45 × (SPH − 5)) where E N2O_nit is N 2 O production from nitrification (g N·m −2 ·d −1 ); f N2O_to_nit is the ratio of N 2 O to nitrified ammonia (unitless); adj fc is maximum ratio of N 2 O production to nitrified N at field capacity (calibrated parameter, unitless); adj wp is minimum ratio of N 2 O production to nitrified ammonia at wilting point (calibrated parameter, unitless); dDO fc and dDO wp are normalized diffusivity in soil at field capacity and wilting point, respectively (unitless); dDO sf refers to the normalized diffusivity of the top soil layer (unitless). More details about calculation of the diffusivity factors are provided in the supporting information.
Following Parton et al. (2001) andDel Grosso et al. (2000), we also revised SWAT to simulate N 2 O production from denitrification, which is influenced by soil nitrate content, temperature, soil water, and soil respiration: where E N2O_den is N 2 O production rate through nitrification on a given day (g N·m −2 ·d −1 ); E N2O is denitrification rate (g N·m −2 ·d −1 ); Rn2n2o is ratio of N 2 to N 2 O (unitless); fRno3_co2 represents CO 2 effect on the ratio of N 2 to N 2 O (unitless); wfps is water-filled pore space (unitless); nppm is soil nitrate content (ppm N/m 2 ); co2 ppm is CO 2 concentration in soils (ppm); C unit is a conversion coefficient to change unit from ppm to g/g (10 −6 ); ρ soil is soil density (g soil/cm 3 ); Dtotflux is the denitrified nitrogen (ppm N/d); fRwfps represents effect of wfps on the ratio of N 2 to N 2 O (unitless); fDwfps represents effect of wfps on denitrification; fDco2 is denitrification rate due to CO 2 concentration (ppm N/d); fDno3 is denitrification flux due to soil nitrate (ppm N/d); x inflextion denotes impacts of CO 2 concentration on fDwfps (unitless); co2 correction is corrected CO 2 concentration (ppm); min_nit is minimum nitrate concentration required in a soil layer for trace gas calculation (ppm N); respc is soil respiration (g C·m −2 ·d −1 ); wfps threshold is a threshold value for water-filled pore space (unitless); wfps_adj is the adjustment on inflection point for water-filled pore space effect on denitrification curve (unitless); aa denotes impacts of soil diffusivity on soil CO 2 concentrations (unitless); M is an intermediate parameter in calculating x inflextion (unitless); dD0 fc is normalized soil diffusivity at field capacity (unitless). Details about calculation of this variable are introduced in the supporting information.
Nitric oxide (NO) is a byproduct of the nitrification process and is also produced during the denitrification reaction sequence (Robertson and Groffman 2015). Because the DayCent algorithm does not explicitly represent all of the biochemical steps that occur during nitrification and denitrification, NO is calculated based on modeled N 2 O production and a NO/N 2 O ratio function. The function is based on the assumption that higher gas diffusivity and increased O 2 availability will lead to higher NO emissions. We used the following equations to simulate NO emission following the DayCent model (Parton et al. 2001): where E NO_N2O is the NO flux converted from N 2 O (g N·m −2 ·d −1 ); R no_n2o is the ratio of NO to N 2 O (unitless); dD0 is the normalized soil diffusivity (unitless).

Data collection
We collected observational data from three GLBRC sites, namely continuous corn, switchgrass, and smooth brome 3.14 ×Arctan 3.14 × 0.0022 × nppm − 9.23 grass (reference site) from the Marshall Farm scale-up experimental fields (Fig. 1). These cropping systems were established in 2009 to study how production of different biofuel crops affects biodiversity and biogeochemistry in this region (Zenone et al. 2011 We first simulated N 2 O fluxes at the three sites with default parameters from the DayCent model. Then, we adjusted key model parameters regulating N 2 O production through nitrification and denitrification manually to minimize the discrepancies between model estimates and field observations. The optimized parameters with least bias in N 2 O simulations were used to generate calibrated model estimates for the test sites (Table 1).
We evaluated model performance at multiple temporal scales. First, we examined model simulations of soil moisture over the selected sites for those days with field observations. Next, we compared model estimates with observed N 2 O fluxes at the monthly scale. Observed N 2 O fluxes from 2010 to 2014 were linearly interpolated to obtain daily fluxes, and then, we aggregated the gap-filled data to the monthly scale for model performance evaluation. We also evaluated model-simulated multiple-year average crop yields for the harvested corn and switchgrass sites.

Sensitivity analysis
We conducted a local parameter sensitivity analysis for five key parameters (Table 1). Here, we assumed that all the selected parameters are normally distributed. We increased and decreased, respectively, the calibrated optimum values of these parameters by 20% to assess the sensitivity of all five parameters. Results of such an analysis were expected to provide valuable information for future calibration and application of the algorithms.
We also evaluated how SWAT N 2 O estimates respond to changes in precipitation and temperature to understand how possible climate scenarios would affect N 2 O emissions. We increased and decreased daily precipitation by 20% to represent future wet and dry climate scenarios, respectively; we increased daily air temperature by 1° and 2°C to represent future warming scenarios.

Model performance evaluation
Previous investigations demonstrated that soil moisture has significant impacts on N 2 O emissions. Reasonable simulation of soil moisture is an important prerequisite for reliably simulating N 2 O fluxes. For most days with available field observations, simulated soil moisture was close to the mean or within one standard deviation of observation (Fig. 2), indicating that SWAT-estimated soil moisture matches well observations. Discrepancies between model estimates and observations, particularly for days with intensive rainfall events, should be further reduced through more comprehensive parameter calibration in the future.
With the default parameter values, SWAT simulated well the magnitude of average N 2 O emissions of the three cropping systems (Fig. 3). Specifically, the estimated growing-season N 2 O emission rate during 2010-2014 at the corn site was 1.08 ± 0.82 kg N·ha −1 ·month −1 (mean ± standard deviation), which was very close to the observed magnitude of 1.10 ± 2.58 kg N·ha −1 ·month −1 . At the switchgrass site, model-estimated and observed average N 2 O fluxes were 0.22 ± 0.10 and 0.16 ± 0.13 kg N·ha −1 ·month −1 , respectively. At the unmanaged reference site that had much lower N 2 O emissions than the corn and switchgrass sites, modeled N 2 O emissions of 0.08 ± 0.05 kg N·ha −1 ·month −1 also corresponded well to the observed fluxes of 0.05 ± 0.04 kg N·ha −1 ·month −1 . Overall, the default parameter settings could generally reflect the differences in the magnitude of N 2 O emissions across the three sites. The default parameterization also captured well temporal patterns in N 2 O fluxes at the two managed sites (corn and switchgrass sites), for which modeled and observed N 2 O fluxes were significantly (P < 0.1) correlated. At the reference site, the default simulation failed to reproduce seasonal patterns of N 2 O emissions at a significance level of 10% (P > 0.1).
Calibration of key parameters substantially improved the model performance (Figs. 4 and 5), in particular for further reducing biases in estimated magnitude of N 2 O fluxes at the reference and switchgrass sites. Specifically, parameter adjustment further decreased the bias at the switchgrass site to 15.1%. For the reference site, discrepancies between observations (0.05 ± 0.04 kg N·ha −1 ·month −1 ) and simulations (0.04 ± 0.02 kg N·ha −1 ·month −1 ) were reduced to 23% (Fig. 5), as compared to a 54% bias in the default simulation.
Apart from matching the magnitude, parameter adjustment achieved better representations of the seasonal patterns in N 2 O emissions than default simulations. Correlations between simulated and observed monthly N 2 O fluxes were improved and significant at the monthly scale across all sites (P < 0.05). N 2 O emissions were much higher during growing season, particularly from May to August, than during non-growing season. At the corn and switchgrass sites, both modeled and observed N 2 O fluxes increased rapidly from April to May and reached peak values in May and June. Then, N 2 O emissions decreased substantially from July to November. At the reference site, model simulations corresponded well with observations regarding the decreasing trend of N 2 O fluxes from June to November.
Across the three sites with different levels of management intensity, we attained a significant correlation between simulated N 2 O fluxes and field observations (Fig. 6). The model simulations explained 22.53% of the variability in N 2 O emissions across three sites, confirming the feasibility of employing the new algorithms Notes: adj fc is maximum fraction of N 2 O to nitrified N at the field capacity; adj wp is minimum fraction of N 2 O to nitrified nitrogen at the wilting point; wfps_adj is adjustment on inflection point for water-filled pore space effect on denitrification curve (unitless); min_nit is minimum nitrate concentration required in a soil layer for trace gas calculation (ppm N); f nit_max is maximum fraction of ammonia that is nitrified during nitrification (unitless). Range of wfps_adj was obtained through personal communication with Dr. Del Grosso.
to evaluate influences of agricultural activities on N 2 O emissions across diverse agricultural ecosystems. Note that when the month with extremely high N 2 O emissions was excluded in the model-data comparison, the model would explain 49.7% of the variability in N 2 O fluxes (Fig. 6). SWAT simulated well crop yields at the corn and switchgrass sites. Specifically, at the corn site, our estimate of crop yields during 2010-2014 was 7.96 ± 1.85 Mg/ha, which was comparable to the observations of 7.51 ± 3.25 Mg/ha. Model-estimated switchgrass yields of 7.89 ± 1.36 Mg/ha agreed well with the observations of 7.50 ± 2.56 kg/ha during 2011-2014.

Sensitivity analysis
We selected five parameters that are closely related to N 2 O emissions to examine the responses of simulated N 2 O fluxes to a 20% change (increase or decrease) of each parameter at the three sites (Table 2). N 2 O emissions positively correlated with maximum fraction of N 2 O to nitrified N at the field capacity (adj fc ) and minimum fraction of N 2 O to nitrified nitrogen at the wilting point (adj wp ), but had negative correlations with adjustment on inflection point for water-filled pore space effect on denitrification curve (wfps_adj). Specifically, with a 20% reduction of adj fc , N 2 O emissions were reduced by 9.41%, 12.19%, and 12.68% for corn, switchgrass, and reference sites, respectively; in contrast, a 20% increase in adj fc increased N 2 O fluxes by 9.21% at the corn site, 12.14% at the switchgrass site, and 12.69% at the reference site. In response to changes (±20%) in adj wp , simulated N 2 O emissions varied from a reduction of 0.19% to an increase of 0.17% at the corn site. Similarly, responses at the switchgrass site to this parameter ranged from 0% to 0.38%. At the reference site, N 2 O emissions were reduced by 0.72% with a 20% decrease in adj wp , but increased by 0.72% in response to a 20% increase in this parameter. Adjustments (±20%) of minimum nitrate content (min_nit) in soil for denitrification had insignificant influence on N 2 O emissions at the two managed sites (changes are less than 0.1%), but induced more sensitive responses (−0.72% to 0.2% changes) at the reference site.
Simulated N 2 O emissions were sensitive to changes in wfps_adj as well. At the corn site, a 20% increase in wfps_adj reduced N 2 O emission estimates by 40.48%, whereas a 20% decrease in this parameter increased model-estimated N 2 O fluxes by 86.79%. At the switchgrass site, model estimates varied from −33.65% to +18.14% with ±20% changes of this parameter. At the reference site, a 20% increase in wfps_adj decreased modeled N 2 O emissions by 3.9%, whereas a 20% reduction substantially increased N 2 O emissions by 195.1%. Responses of N 2 O emissions to the maximum fraction of ammonia that is nitrified during nitrification (f nit_max ) varied across the selected sites. At the corn and reference sites, a 20% increase in f nit_max boosted increases in N 2 O emissions by 2.35% and 0.53%, respectively, whereas a 20% reduction resulted in decreases of 3.62% and 0.77%, respectively. At the switchgrass site, the response of N 2 O emissions was less sensitive to ±20% changes in f nit_max , ranging from −0.19% to 0.18%.

Climatic influences
Changing climate conditions affected N 2 O emissions (Table 3). Our sensitivity analysis suggested that N 2 O emissions had positive responses to changes in precipitation. With a 20% increase in precipitation, N 2 O emissions would increase by 1.39%, 1.50%, and 1.44% at the corn, switchgrass, and reference sites, respectively, whereas under the drier scenario (a 20% reduction in precipitation), N 2 O fluxes would be reduced by 3.66%, 3.12%, and 1.52%, respectively. Higher temperatures would generally increase N 2 O emissions. With a 1°C increase in air Comparison of simulated and observed soil water content across the three sites. Soil moisture data were collected twice a year from 2009 to 2013 at the three sites. For days with available data, average soil moisture and its standard deviation were obtained from ten replicates at each site. temperature, N 2 O emissions would be enhanced by 1.94-3.69% across the three sites. A 2°C increase would further increase N 2 O emissions by 14.4% and 5.74% at the corn and reference sites, respectively, but the switchgrass would only increase by 0.59%.

Discussion
Enhanced SWAT for simulating N2O emissions As a potent GHG, increasing emissions of N 2 O from terrestrial ecosystems to the atmosphere has raised concerns about its potential impacts on the climate system (Butterbach-Bahl et al. 2013). Significant efforts have been devoted to investigating N 2 O fluxes from cropland since agricultural land has been identified as a key contributor of the anthropogenic N 2 O emissions (Del Grosso et al. 2009). Numerical simulation of N 2 O fluxes is critical for predicting N 2 O emissions under different management scenarios, and provides valuable information for the mitigation practices (Del Grosso et al. 2009). By integrating the DayCent N 2 O emission algorithms with SWAT's existing crop growth, hydrology, and biogeochemical cycling algorithms, we created a new modeling tool that allows us to include N 2 O emissions as an important dimension in watershed-scale assessment of sustainability of agricultural ecosystems.
Model evaluation shows that the new module provided reasonable estimates of N 2 O fluxes across sites with divergent management intensities, as well as reproduced the seasonal patterns of N 2 O emissions. Accuracy of model prediction in this study is close to the previous modeling efforts based on the DayCent model and the DeNitrification-DeComposition (DNDC) model (Parton et al. 2001, Abdalla et al. 2010, Rafique et al. 2013, Grant et al. 2015, indicating feasibility of applying the new tool, along with the existing capability of SWAT, to conduct comprehensive assessments of farming impacts on the environment.

Difference in N 2 O emissions between managed and unmanaged sites
Our simulations indicate that the corn and switchgrass sites had much higher N 2 O emissions than the unmanaged reference site. The difference further confirms the dominant impacts of nitrogen inputs on N 2 O emission. Annual average N 2 O emissions from the corn site reached 8.48 kg N·ha −1 ·yr −1 during 2009-2014. This emission rate fell within previous observations (approximately 3.29-8.76 kg N·ha −1 ·yr −1 ) in the U.S. corn belt (Iqbal et al. 2015), with the emission factor (fraction of N 2 O emission to fertilizer use) at the corn site (5.3%) being at the upper end  of the range (0.17%-21%) from previous studies (Novoa andTejeda 2006, Signor et al. 2013). The high emission factor at the corn site may result from the high precipitation in this region (Dobbie et al. 2003).
Investigations of nitrogen cycling in switchgrass cultivation have increased since this species has been identified as a promising cellulosic bioenergy crop (Vogel et al. 2002, Demissie et al. 2012. Previous studies found significant variability in N 2 O emissions from switchgrass sites with different fertilizer use rates, soil types, and climate conditions . Our estimate of 1.96 kg N·ha −1 ·yr −1 at the switchgrass site is consistent with a synthesis (ranging from 1.37 to 2.07 kg N·ha −1 ·yr −1 ) across multiple field sites (Oates et al. 2016). We derived a lower emission factor at the switchgrass site (3.3%) than at the corn site (5.3%), which may be explained by the high nitrogen-use efficiency at the switchgrass site (Monti et al. 2012).
For the unmanaged reference site, N 2 O emissions reached 0.35 kg N·ha −1 ·yr −1 , which is lower than the average of synthesis data (1.75 kg N·ha −1 ·yr −1 ) over more than 200 grass land sites (Kim et al. 2013a), indicating that the reference site may have relatively tighter nitrogen cycling. Although the managed sites had much higher emissions than the unmanaged site, their seasonal emission patterns were consistent, with much higher emission rates in summer (May-July) than other seasons, reflecting the fundamental influences of temperature on the seasonality of N 2 O emissions (Butterbach-Bahl et al. 2013, Liu et al. 2013).

N 2 O emissions in response to climatic changes
Responses of simulated N 2 O emissions to changes in precipitation and temperature provide valuable insights into projecting N 2 O emissions under a changing climate. All three sites demonstrated positive responses in N 2 O emissions to changes in precipitation. Increased soil moisture after rainfall induces elevated emissions mainly through stimulating microbial activities or enhancing the anaerobic conditions (Signor et al. 2013. Historical data indicate that growing-season precipitation has been increasing since the 1980s in most areas of the Midwest United States (Dai et al. 2016). As a result, this changing rainfall pattern may further stimulate N 2 O emissions in summer in this region. In contrast, other studies reported that plant growth following elevated rainfalls may deplete the soil inorganic nitrogen pool and thus reduce N 2 O emissions (Xu-Ri et al. 2012). Different response rates of N 2 O emissions to changes in precipitation at sites with different plant species and management activities, as demonstrated in our analyses, call for further investigations on confounding processes determining N 2 O emissions to better predict how future precipitation changes can affect N 2 O fluxes.
Model-simulated N 2 O fluxes generally increased under higher temperatures across the three sites. Positive responses of N 2 O emissions to higher temperatures may be caused by more active microbial activities and increased soil organic matter decomposition (Reth et al. 2005, Signor et al. 2013

Uncertainties and future work
Although the new modeling tool provided reasonable estimates of N 2 O emissions over the three sites, the  Notes: adj fc is maximum fraction of N 2 O to nitrified N at the field capacity; adj wp is minimum fraction of N 2 O to nitrified nitrogen at the wilting point; min_nit is minimum nitrate concentration required in a soil layer for trace gas calculation; wfps_adj is adjustment on inflection point for water-filled pore space effect on denitrification curve (unitless); f nit_max is maximum fraction of ammonia that is nitrified during nitrification (unitless); "−" indicates changes less than 0.01%.
unexplained variability in the observed N 2 O fluxes suggests that further improvement is needed to better represent processes regulating N 2 O emissions. For example, current model simulation is highly sensitive to parameters such as the adjustment on inflection point for waterfilled pore space effect on denitrification curve (wfps_adj), which represents soil properties affecting soil diffusivity other than soil water, soil texture, and soil bulk density. Process-based algorithms or spatially explicit datasets are needed to better model underlying mechanisms represented by this parameter to enhance N 2 O simulation in the future.
Notably, both nitrification of soil ammonia and denitrification of soil nitrate contribute to N 2 O production (Bateman and Baggs 2005). However, field observations at the three sites did not differentiate the relative contributions of each process to total N 2 O emission. As a result, N 2 O fluxes produced by nitrification and denitrification were lumped together to calibrate and evaluate simulated total N 2 O fluxes from soil columns. As a result, future model improvement should focus on the model simulation of the individual processes in N 2 O production, the soil inorganic nitrogen stocks, etc., to further strengthen the model's capability in modeling N 2 O fluxes.
In addition, our analysis indicated that extremely high N 2 O fluxes observed after fertilizer use dramatically affected model performances. Therefore, more frequent observations, in particular following fertilizer use, are needed to improve model performance by incorporating observational information through calibration.
Although manual calibration of the parameters directly controlling N 2 O production improved model performances, more comprehensive parameter optimization is needed to further enhance model simulations. Parameter sensitivity analysis in this study identified impacts of individual parameters on model estimates of N 2 O emissions. However, interactions among these parameters may jointly affect model responses (Kim et al. 2013b). As a result, further analysis targeting the interplay among multiple parameters should be conducted in the future. In addition, calibration of parameters that indirectly regulate N 2 O production, such as carbon-to-nitrogen ratio for structural litter, leaching coefficient of soil nitrogen, and water limitation coefficient on nitrification, together with the parameters identified in this study, would improve model representation of seasonal variability of N 2 O emissions (Rafique et al. 2013).

Conclusions
As a watershed-scale model, SWAT has been widely used to evaluate impacts of agricultural activities on the quality of the aquatic ecosystems (Gassman et al. 2007). However, N 2 O emissions were not included in previous SWAT modeling efforts, limiting its use for assessing and identifying best agriculture management practices under climate change. Here, we integrated DayCent's N 2 O emission module with the existing crop growth, hydrology, and biogeochemical processes in SWAT, and achieved a new tool (SWAT-N 2 O) that reasonably captured the magnitude and seasonality of N 2 O emissions from three diverse agricultural systems with different management intensities. Modeled N 2 O emission responses to climate change scenarios demonstrate that N 2 O emissions may increase under a warmer and wetter climate. Overall, the model development and application efforts enhanced SWAT to represent N 2 O emissions as a dimension in assessing sustainability of agricultural ecosystems and to explore climate-smart agricultural solutions under a changing climate.