Integration of nitrogen dynamics into the Noah-MP land surface model v1.1 for climate and environmental predictions

Abstract. Climate and terrestrial biosphere models consider nitrogen an important factor in limiting plant carbon uptake, while operational environmental models view nitrogen as the leading pollutant causing eutrophication in water bodies. The community Noah land surface model with multi-parameterization options (Noah-MP) is unique in that it is the next-generation land surface model for the Weather Research and Forecasting meteorological model and for the operational weather/climate models in the National Centers for Environmental Prediction. In this study, we add a capability to Noah-MP to simulate nitrogen dynamics by coupling the Fixation and Uptake of Nitrogen (FUN) plant model and the Soil and Water Assessment Tool (SWAT) soil nitrogen dynamics. This model development incorporates FUN's state-of-the-art concept of carbon cost theory and SWAT's strength in representing the impacts of agricultural management on the nitrogen cycle. Parameterizations for direct root and mycorrhizal-associated nitrogen uptake, leaf retranslocation, and symbiotic biological nitrogen fixation are employed from FUN, while parameterizations for nitrogen mineralization, nitrification, immobilization, volatilization, atmospheric deposition, and leaching are based on SWAT. The coupled model is then evaluated at the Kellogg Biological Station – a Long Term Ecological Research site within the US Corn Belt. Results show that the model performs well in capturing the major nitrogen state/flux variables (e.g., soil nitrate and nitrate leaching). Furthermore, the addition of nitrogen dynamics improves the modeling of net primary productivity and evapotranspiration. The model improvement is expected to advance the capability of Noah-MP to simultaneously predict weather and water quality in fully coupled Earth system models.


space
Abstract. Climate and terrestrial biosphere models consider nitrogen an important factor in limiting plant carbon uptake, while operational environmental models view nitrogen as the leading pollutant causing eutrophication in water bodies. The community Noah land surface model with multiparameterization options (Noah-MP) is unique in that it is the next-generation land surface model for the Weather Respace in capturing the major nitrogen state/flux variables (e.g., soil nitrate and nitrate leaching). Furthermore, the addition of ni-trogen dynamics improves the modeling of net primary pro-ductivity and evapotranspiration. The model improvement is expected to advance the capability of Noah-MP to simulta-neously predict weather and water quality in fully coupled Earth system models. spacesearch and Forecasting meteorological model and for the op-erational weather/climate models in the National Centers for Environmental Prediction. In this study, we add a capabilspaceity to Noah-MP to simulate nitrogen dynamics by coupling the Fixation and Uptake of Nitrogen (FUN) plant model and the Soil and Water Assessment Tool (SWAT) soil nitro-gen dynamics. This model development incorporates FUN's state-of-the-art concept of carbon cost theory and SWAT's strength in representing the impacts of agricultural manage-ment on the nitrogen cycle. Parameterizations for direct root and mycorrhizal-associated nitrogen uptake, leaf retranslo-cation, and symbiotic biological nitrogen fixation are em-ployed from FUN, while parameterizations for nitrogen min-eralization, nitrification, immobilization, volatilization, at-mospheric deposition, and leaching are based on SWAT. The coupled model is then evaluated at the Kellogg Biological Station -a Long Term Ecological Research site within the US Corn Belt. Results show that the model performs well 1 spaceIntroduction Over the past several decades, eutrophication -high concen-trations of nutrients in freshwater bodies leading to severe oxygen depletion from the resultant algal bloomshas be-come a worldwide problem facing river, lake, and coastal wa-ters (Conley et al., 2009;Howarth et al., 2006). As one of the greatest threats to freshwater and coastal ecosystems, eu-trophic conditions lower biotic diversity, lead to hypoxia and anoxia, increase the incidence and duration of harmful algal blooms, and change ecological food webs that reduce fish production (Diaz and Rosenberg, 2008;National Research Council, 2000). These eutrophic conditions are attributed to excessive fertilizer leaching in river basins Boyer et al., 2006). To complicate this further, climate variation and climate change also determine the variation of space Published by Copernicus Publications on behalf of the European Geosciences Union. space spacehypoxia extent (Donner and Scavia, 2007): higher tempera-tures may extend the thermal stratification period and deepen the thermocline, thereby resulting in the upwelling of nutri-ents from sediment and increasing the concentration of nu-trients in the bottom layer of water in lakes (Komatsu et al., 2007). Further, higher precipitation produces more runoff, and very likely more nutrients are delivered to the ocean as well (Donner and Scavia, 2007).
Nitrogen (N) is recognized as the leading nutrient causing eutrophication. Without human interference, N cycling is relatively slow, as most ecosystems are efficient at retain-ing this in-demand nutrient. N enters soil regularly either through atmospheric wet and dry deposition or through at-mospheric N2 fixation by microorganisms (occurring mostly in legume plants). N taken up by plants is confined to rel-atively slow processes (e.g., growth, decay, and mineraliza-tion); in some regions or during the growing season, N may also limit plant growth, which reduces carbon sequestration over land (Fisher  et al., 2012). In addition, N cycling pro-duces nitrous oxide (N2O), which is considered one of the important greenhouse gases responsible for climate warm-ing. These facts make the N cycle important for studying the response of the climate to the elevated greenhouse gas con-centrations. With human tillage of soils, mineralization and nitrification of N are amplified, which results in the reduc-tion of N storage in soil (Knops and Tilman, 2000;Scanlon et al., 2008). In addition, a large amount of N fertilizer is applied in specific areas within a short period of time; as a result, a massive excess of N is leached to the aquatic sys-tems through discharge and erosion, which contributes to the eutrophication in aquatic systems. Many of these N processes have been included in land sur-face, hydrologic, and water quality models developed partic-ularly for environmental, climate, and agricultural applica-tions (Bonan and Levis, 2010;Dickinson et al., 2002;Fisher et al., 2010;Kronvang et al., 2009;Schoumans et al., 2009;Thornton et al., 2007;Wang et al., 2007;Yang et al., 2009). These developments are still in their infancy, and large-scale climate models lack N leaching parameterizations that are comparable to those used in water quality models. Thus, large-scale models are not feasible for inherently fine-scale applications such as agricultural fertilization management and water quality prediction. Therefore, the present study improves these weaknesses by incorporating the strength of agriculturebased models into large-scale land surface mod-els (LSMs).
The community Noah LSM with multi-parameterization options (Noah-MP) Yang et al., 2011) is used as an exemplar of LSMs because it is the nextgeneration LSM for the Weather Research and Forecasting (WRF) meteorological model (Rasmussen et al., 2014) and for the operational weather and climate models in the spacelight, temperature, and soil moisture -it is logical to aug-ment this scheme with N limitation and realistic plant N up-take and fixation. The state-of-the-art vegetation N model is the Fixation and Uptake of Nitrogen (FUN) model of Fisher et al. (2010), which is embedded into the Joint UK Land Environment Simulator (JULES) (D. B.  and the Community Land Model (CLM) (Shi et al., 2016). Modeling the impacts of agricultural management (e.g., fer-tilizer use) on N leaching is the strength of the Soil and Water Assessment Tool (SWAT) (Neitsch et al., 2011). Therefore, this study incorporates into Noah-MP both FUN's strength in plant N uptake and SWAT's strength in soil N cycling and agricultural management.
Our objective is to develop and utilize a land surface mod-eling framework for simultaneous climate (carbon) and en-vironmental (water quality) predictions. We first describe the nitrogen dynamic model which combines equations used in FUN and SWAT. We then focus on evaluating the new inte-grated model at a cropland site, because fertilizer application on croplands globally contributes approximately half of the total N input to soil, with the other half coming from natu-ral processes (i.e., atmospheric deposition and biological N fixation) (Fowler et al., 2013;Gruber and Galloway, 2008). Furthermore, cropland is a major source of N loading in wa-ter bodies. We evaluate the new model against observed soil moisture content, concentration of soil nitrate, concentration of nitrate leaching from soil bottom, and annual net primary productivity (NPP). We then analyze the impacts of the addi-tion of N dynamics on the carbon and water cycles. To guide the use of this model on regional scales, we also analyze the impacts from different fertilizer application scenarios. Fi-nally, we discuss other model behaviors, i.e., N uptake from different pathways and the major soil nitrate fluxes.
2 Models, data, and methods

Noah-MP
The Noah-MP model was augmented from the original Noah LSM with improved physics and multiparameterization op-tions Yang et al., 2011), based on a state-of-the-art multiple-hypothesis framework (M. P. . Noah-MP provides users with multiple options for parameterization in leaf dynamics, canopy stomatal resis-tance, soil moisture factor for stomatal resistance, and runoff and groundwater. Until this work, Noah-MP did not include any N dynamics. The only N-related parameterization is in the calculation of the maximum rate of carboxylation (Vmax, Eq. 1) -an important factor in estimating the total carbon assimilation (or photosynthesis) rate : spaceNOAA/National Centers for Environmental Prediction. Be- (1) spacecause Noah-MP has an interactive vegetation canopy option -which predicts the leaf area index (LAI) as a function of spacewhere Vmax25 is the maximum carboxylation rate at 25 • C (µmol CO2 m −2 s −1 ), avmax is a temperature-sensitive paspace spacerameter, f (Tv) is a function that mimics the thermal break-down of metabolic processes, f (N) is a foliage nitrogen fac-tor (f (N) 1), and β is the soil moisture controlling factor. Since there were no N dynamics in the model, f (N) was set as a constant 0.67, which translates to a constant 33 % of Vmax down-regulation due to N stress. This factor was origi-nally used in Running and Coughlan (1988) and adapted into LSMs by Bonan (1991).
Our modifications to the original Noah-MP mainly concern the sub-models dealing with dynamic leaf and subsurface runoff. The dynamic leaf option is turned on to pro-vide NPP and biomass to the newly coupled N dynamic submodel. In the original Noah-MP model, subsurface runoff from each soil layer was not an explicit output, but it is now a new output in the updated model. However, N concentrations are different among soil layers, which affects the amount of N removed from each soil layer by subsurface runoff. There-fore, in conjunction with the runoff scheme options 1 (TOP-MODEL with groundwater) and 2 (TOPMODEL with an equilibrium water table), the lumped subsurface runoff for all four layers is first calculated, and then the water is removed from each soil layer weighted by hydraulic conductivity and soil layer thickness.

Nitrogen dynamics
In Noah-MP, the soil N model structure is the same as in SWAT, which includes five N pools consisting of two inorganic forms (NH + 4 and NO − 3 ) and three organic forms (active, stable, and fresh pools). The N processes employed from SWAT are mineralization, decomposition, immobilization, nitrification, denitrification, and atmospheric deposition. The N processes employed from FUN are uptake and symbiotic biological N fixation, which can be further divided into active and passive soil N uptake, leaf N retranslocation, and sym-biotic biological N fixation. Figure 1 shows the flow chart of the nitrogen dynamic model. In this section, we describe the core equations. The full description for plant N uptake and soil N dynamics is available in Fisher et al. (2010) andNeitsch et al. (2011), respectively. Table 1 shows the model input variables and parameters. Most of these parameters use the values recommended by Fisher et al. (2010) andNeitsch et al. (2011), while some of them are adjusted to best repre-sent the site condition and hence match site observation. The important adjusted parameters include the γsw,thr (threshold value of soil water factor for denitrification to occur), βmin (rate coefficient for mineralization of the humic organic nitrogen), and βrsd (rate coefficient for mineralization of the fresh organic nitrogen in residue).
Noah-MP calculates the NPP or its available carbon, CNPP (kg C m −2 ), following FUN. To maintain the prescribed carbon-to-nitrogen (C : N) ratio (rC N), the N demand, Ndemand (kg N m −2 ), is calculated: r C : N where rC N is the C : N ratio for the whole plant, which is computed for each component (leaf, root, and wood) of the plant proportionally to the biomass. C : N ratios for each component of the plant for each vegetation type are from Oleson et al. (2013).
Because no extra energetic cost is needed, passive uptake, Npassive (kg N m −2 ), is the first and preferred source of N that a plant depletes: where Nsoil is the available soil N for the given soil layer (kg N m −2 ), ET is transpiration rate (m s −1 ), and sd is the soil water depth (m). This pathway is typically a minor contributor except under very high soil N conditions.
If Npassive is less than Ndemand, then the remaining required N must be obtained from retranslocation (Nresorb, kg N m −2 ), active uptake (Nactive, kg N m −2 ), or biological N fixation space space(Nfix, kg N m −2 ), all of which are associated with energetic cost and hence require C expenditure (C cost). The C costs of fixation (Costfix, kg C kg N −1 ), active uptake (Costactive, kg C kg N −1 ), and resorption (Costresorb, kg C kg N −1 ) are space spacecalculated as follows: spacewhere Costactive,ly is the C cost for active N uptake of soil layer ly and n is the total number of soil layers. Using Ohm's law, N acquired from C expenditure (Nacq, C root spacekg N m −2 ) is analogous to current and thus is calculated as follows: spacewhere a, b, and c ( 3.62, 0.27 and 25.15, respectively) are empirical curve-fitting parameters (dimensionless) from Houlton et al. (2008); s is a scaling factor ( 62.5; use kg C kg N −1 • C for unit consistency); Tsoil is soil temperature ( • C); kN and kC are both 1 kg C m −2 ; kR is 0.01 kg C m −2 ; Croot is total root biomass (kg C m −2 ); and Nleaf is the amount of N in the leaf (kg N m −2 ). Active uptake is typ-ically a dominant form of N uptake in natural ecosystems, consuming large quantities of NPP (that would otherwise go to growth or other allocations) in exchange for N. Similar to parallel circuits, each carbon cost is treated as a resistor, and the integrated cost (Costacq, kg C kg N −1 ) is calculated (Brzostek et al., 2014): spaceTherefore, plant N uptake and fixation are computed and are updated for each N pool. In addition, the effect of N limitation on CO2 sequestration is represented in the model through the theory of C cost economics.

Mineralization, decomposition, and immobilization
Fresh organic residue is broken down into simpler organic components via decomposition. The plant-unavailable organic N is then converted into plant-available inorganic N via mineralization by microbes. Plant-available inorganic N can also be converted into plant-unavailable organic N via immobilization by microbes. space spaceImmobilization is incorporated into mineralization calcu-lation (net mineralization). Mineralization and decomposi-tion, which are only allowed to occur when soil temperaspacecalculated: spaceture is above 0 • C, are constrained by water availability and temperature. The nutrient-cycling temperature factor for soil space N nit,ly space fr nit,ly + fr vol,ly fr vol,ly spacenit|vol,ly spacelayer ly, γtmp,ly, is calculated as follows: space N vol,ly = ( fr nit,ly space + fr vol,ly space ) · N nit|vol,ly , Cost active,ly ( )  (14) as fol-lows: spacewhere Tsoil,ly is the temperature of soil layer ly ( • C).
The nutrient-cycling water factor for soil layer ly, γsw,ly, is calculated as follows: θ ly γ sw,ly = θ s,ly , where θly is the water content of soil layer ly (mm H2O) and θs,ly is the water content of soil layer ly at field capacity (mm H2O).
The mineralized N from the humus active organic N pool, Nmina,ly (kg N m −2 ), is calculated as follows:

Denitrification
Denitrification is the process of bacteria removing N from soil (converting NO − 3 to N2 or N2O gases). Denitrification rate, Ndenit,ly (kg N m −2 ), is calculated as follows: spacewhere βmina is the rate coefficient for mineralization of the humus active organic nutrients and Naon,ly is the amount of N in the active organic pool (kg N m −2 ).
The mineralized N from the residue fresh organic N pool, Nminf,ly (kg N m −2 ), is calculated as follows: where δntr,ly is the residue decay rate constant, and Nfon,ly is the amount of N in the fresh organic pool (kg N m −2 ). The decomposed N from the residue fresh organic N pool, Ndec,ly (kg N m −2 ), is calculated as follows: N dec,ly = 0.2 · δ ntr,ly · N fon,ly . (13)

Nitrification and ammonia volatilization
Using a first-order kinetic rate equation, the total amount of ammonium lost to nitrification and volatilization in layer ly, Nnit|vol,ly (kg N m −2 ), is calculated as follows: where NH4,ly is the amount of ammonium in layer ly (kg N m −2 ), ηnit,ly is the nitrification regulator, and ηvol,ly is the volatilization regulator. The calculation of ηnit,ly and ηvol,ly is described in Neitsch et al. (2011). Nnit|vol,ly is then partitioned to nitrification and volatilization. The amounts of N converted from NH + 4 and NO − 3 of the ammonium pool via nitrification and volatilization are then spaceβdenit is the rate coefficient for denitrification, and γsw,thr is the threshold value of γsw,ly for denitrification to occur.

Atmospheric deposition
While the mechanism of atmospheric deposition is not fully understood, the uncertainty is parameterized into the concentration of nitrate/ammonium in the rain for wet deposition, and the nitrate/ammonium deposition rate for dry deposition.
The amounts of nitrate and ammonium added to the soil through wet deposition, NO3,wet (kg N m −2 ) and NH4,wet (kg N m −2 ), are calculated as follows: NH4,wet = 0.01 · RNH 4 · P , where RNO 3 is the concentration of nitrate in the rain (mg N L −1 ), RNH 4 is the concentration of ammonium in the rain (mg N L −1 ), and P is the amount of precipitation. The values for RNO 3 and RNH 4 used in this study are listed in Table 1.

Fertilizer application
The N fertilizer application process is included in the new model as well. If real fertilizer application data (timing and amount for a specific year) are available, they can be used as model inputs. Otherwise, a fixed amount of N fertilizer (e.g., 7.8 g N m −2 yr −1 in this study) is applied at a fixed time of a year (e.g., 20 June in this study). space

spaceLeaching
N leaching from land to water bodies is a consequence of soil weathering and erosion processes. In particular, organic N attached to soil particles is transported to surface water through soil erosion. Therefore, the modified universal soil loss equation (USLE) (Williams, 1995) is used to determine N mina,ly = β mina,ly γ tmp,ly · γ sw,ly where orgC ly is the amount of organic C in the layer (%), · Naon,ly,  (2011). N in nitrate form can be transported with surface runoff, lateral runoff, or percolation, which is calculated as follows:   NO 3,lat,ly = β NO 3 spaceat 1.2 m of soil depth, is available from 1995 to 2013. These two measurements are used to evaluate model-simulated con-centrations of soil nitrate for the top 25 cm and nitrate leach-ing from the soil bottom. Soil N mineralization, which mea-sures the net mineralization potential and is available from 1989 to 2012, is compared with the modeled mineralization rate qualitatively. In addition, soil moisture content is sampled from the sur-face to 25 cm soil depth and is calculated on a dryweight basis. In order to compare with model output, it is converted to volumetric soil moisture by applying the soil bulk density. Annual NPP is converted from annual crop yields   space NO 3,perc = conc NO 3 ,mobile · w perc,ly , where NO3,surf, NO3,lat,ly, and NO3,perc are the soil nitrates removed in surface runoff, in subsurface flow, and by percolation, respectively (kg N m −2 ); βNO 3 is the nitrate percolation coefficient; concNO 3 ,mobile is the concentration of nitrate in the mobile water in the layer (kg N mm H2O −1 ), and wperc,ly is the amount of water percolating to the underlying soil layer (mm H2O), Qsurf is the generated surface runoff (mm H2O), and Qlat,ly is the water discharged from the layer by lateral flow (mm H2O).

Description of evaluation data and model configuration
At the regional scale, N-related measurements are very limited. Even at site level, measurements are limited with respect to plant and carbon dynamics. The Kellogg Biological Sta-tion (KBS) -a Long Term Ecological Research (LTER) site -is unique in its long-term continuous measurements of N related variables (soil nitrate, N leaching, mineralization, ni-trification, and fertilizer application) in an agricultural setting with multiple crop and soil controls. Even within the LTER Network, we cannot find a second site that conducts this in-tegrated suite of measurements. Therefore, the new model is evaluated at this site.
KBS is located in Hickory Corners, Michigan, USA, within the northeastern portion of the US Corn Belt (42.40 • N, 85.40 • W, elevation 288 m). Mean annual temperature is 10.1 • C, and mean annual precipitation is 1005 mm, with about half falling as snow. This study uses two treatments from this site: T1 cropland with conventional tillage and T2 cropland without tillage. Both treatments are rainfed and are planted with the same crops: corn, soybean, and winter wheat in rotation.
This site features multiple N-related measurements. Soil inorganic N concentration, which is sampled from the surface to 25 cm soil depth, is available from 1989 to 2012. Concentration of inorganic N leaching at bedrock, which is sampled spaceWest et al., 2010). Although N uptake cannot be evaluated directly at this site, by evaluating the annual NPP, we can see the model's performance in representing the N limitation effect on plant growth.
Noah-MP requires the following atmospheric forcing data at least at a 3-hourly time step: precipitation, air temperature, specific humidity, surface air pressure, wind speed, incom-ing solar radiation, and incoming longwave radiation. The weather station at the site measures all of these except for incoming longwave radiation, but it does not cover the entire period from 1989 to 2014 (e.g., hourly precipitation data are only available since 2007), when the N data are available. Therefore, atmospheric forcing data are extracted from the 0.125 • 0.125 • gridded forcing data from the North American Land Data Assimilation System (NLDAS; Xia et al., 2012). Table 2 compares the atmospheric forcing data between NLDAS and site measurements for 2008-2014. We can see that the differences in precipitation and air temperature -the two most important forcing fields for N cyclingare very small, with relative biases 1.4 and 4.2 %, respectively.
Finally, the site management log records the detailed oper-ational practices such as soil preparation, planting, fertilizer application, pesticide application, and harvest. N fertilizer application data include the date of application, rate, fertil-izer type, and equipment used. The fertilizer application date and rate are used as model inputs.

Evaluation of soil moisture
Modeled volumetric soil moisture, which is important for nu-trient cycling and plant growth, is compared to measured soil moisture (Fig. 2). The model performs reasonably well on both treatments (i.e., with and without tillage) in terms of capturing the mean and seasonal variation, which is consisspace  spacetent with previous study by Cai et al. (2014b). The model-simulated multiple year averages are both 0.243 for the two treatments. These are very close to observations, which are 0.238 and 0.264 for T1 and T2, respectively. The correlation coefficient is 0.78 for T1 and 0.76 for T2, which are consid-ered high skills, especially on a daily scale. However, differences between modeled and observed soil moisture are also found. From observation (Fig. 2), we can see that the treatment without tillage (T2) has slightly higher soil moisture than the treatment with tillage (T1). Therefore, tillage practice reduces soil moisture. However, the differ-ence in modeled soil moisture is negligible between the two treatments (both are 0.243). This is because Noah-MP does not consider water redistribution due to tillage, although N redistribution is considered in the soil N dynamic sub-model. N is redistributed by mixing a certain depth (i.e., 100 mm) of soil with a mixing efficiency (i.e., 30 %) (Neitsch et al., 2011). In addition, observed soil moisture has higher varia-tions. As we can see from Fig. 2, observation tends to have ei-ther higher peaks or lower valleys than model simulation. We also notice that some values from observation are extremely low, which may not be necessarily true in reality. Consider-spaceing the wide spread of the observational ranges defined by up to six replicating plots, Noah-MP provides a reasonable water environment for the N cycling.

Evaluation of soil nitrate
Soil nitrate concentration is the outcome of all N-related pro-cesses that occur in soil such as decomposition, mineraliza-tion, nitrification, denitrification, and uptake. It determines the available N that plants can use. The skills in modeling the soil nitrate concentration reflect the overall performance of the model in simulating the N cycle. Figure  3 shows the comparison of the model-simulated soil nitrate concentration with site observations for both T1 and T2. The model cap-tures the major variations of the soil nitrate. N fertilizer ap-plication is responsible for the high peaks. These high peaks drop very fast at first and then drop slowly, which can sustain crop growth for a few months.
The multi-year average of modeled soil nitrate concentration is 5.77 mg kg −1 (4.90 mg kg −1 ) for T1 (T2), which is consistent with the observed 5.61 mg kg −1 (4.81 mg kg −1 ). Correlation coefficients are 0.58 and 0.56 for T1 for T2, respectively. From the wide spread of the range error bars, we space  ., 9, 1-15, 2016 www.geosci-modeldev.net/9/1/2016/  space can see that soil N dynamics may be affected by a variety of complicated factors, which makes it difficult to model. Therefore, although the correlation coefficients are not considered high skills relative to the soil moisture statistics, they are still reasonable. While both treatments show very similar patterns (Fig.  3), T1 with conventional tillage tends to have higher soil nitrate concentration. This is understandable because tillage prac-tices redistribute water and nutrients in soil, which acceler-ates the N cycling. Table 3 shows annual averages of major N fluxes for both treatments. T1 has higher rates of humus min-eralization and residue decomposition, but, at the same time, it also has higher rates of denitrification and leaching. There-fore, it produces more N2O (a greenhouse gas) and more N runoff to rivers. Particularly, with higher N leaching, less soil nitrate is available for passive uptake by plants. As a result, plants need to acquire more N through active uptake. space 3.3 Evaluation of nitrate leaching from soil bottom N leaching can be transported to rivers through surface and subsurface runoff and to groundwater through percolation from soil bottom. Only the last pathway is measured at this site. Figure 4 shows the comparison of concentrations of the leached solution from the soil bottom between model simulation and observation. The averaged concentration of N leaching from the soil bottom for T1 (T2) is 12.84 mg kg −1 (8.86 mg kg −1 ) from model simulation and 13.57 mg kg −1 (9.26 mg kg −1 ) from observation. The corre-lation coefficients are 0.43 and 0.40 for T1 and T2, respec-tively. Although these skills may not be considered satisfac-tory, the model can still produce comparable results with ob-servation. The peak in 2003 is extremely high and long lasting. This is probably due to the abnormal pattern of precipitation distribution in 2003. In a normal year, storms higher than 50 mm usually occur in either summer or fall. However, in 2003, there was an early storm on 4 April which reached 61 mm in 1 day. As we can see from Fig. 3, the soil nitrate concentraspace tion is also high. The combination of high water infiltration (due to the storm) and high soil nitrate concentration resulted in a large amount of soil nitrate being brought to the bottom soil layer. A few months following that, there was no large storm, which was again different from a normal year. As a result, the high-concentration nitrate solution was drained slowly out of the bottom layer of soil. The modeled nitrate leaching also shows a peak over this period, but the values are only close to the lower bound of the observed range. This suggests that improvement is needed so the model can better capture peaks under this situation. We also notice that, without tillage, N leaching is about one-third lower than that with tillage. Without tillage, the temporal variation is also smaller.

Evaluation of annual NPP
NPP indicates the amount of C that is assimilated from the atmosphere into plants and thus is important in studying not only crop and ecosystem productivity but also climate change feedbacks. NPP is mainly determined by plant photosynthesis and autotrophic respiration. It is also affected by water and nutrient stresses. In this study, N stress on plant growth is calculated by the reduction of NPP due to N acqui-sition, which can be considered another form of plant respi-ration. Figure 5 shows the comparison of simulated annual NPP against observation. Since the original Noah-MP with-out N dynamics also simulates NPP, its results are also shown here as a reference. The mean annual NPP simulated by the original Noah-MP is 544 gC m −2 (the same simulation for both treatments as original Noah-MP does not distinguish space tillage and no tillage). By including the N dynamics, simu-lated annual NPP is reduced to 432 gC m −2 (441 gC m −2 ) for T1 (T2), which is more consistent with observed 437 gC m −2 (471 gC m −2 ). The correlation coefficient increased to 0.77 for T1, and from 0.30 to 0.72 for T2, which is a significant improvement. This improvement is due to the better charac-terization of the amount of carbon allocated to N acquisition instead of growth.
The modeled rate of NPP down-regulation -the fraction of NPP reduction due to N limitation -is 35.4 and 34.7 % for T1 and T2, respectively. These rates are close to the 33 % of down-regulation rate used in the default Noah-MP. By dy-namically simulating the demand and supply of N with time, these become even closer to the observations. Surprisingly, even with slower N cycling, T2 produces slightly higher NPP (Table 3), which is consistent between model and observation. If this is the case, except for drying up soil, releasing more N2O gas, and producing more N leaching, is there any benefit from tillage? The answer is yes. Less N fertilizer is needed for cropland with tillage. Based on the site management log, in total there was 194.8 gN m −2 of N fertilizer applied to T1 from 1989 to 2013, which is less than the amount (210.7 gN m −2 ) applied to T2 during the same period.

Impacts of nitrogen dynamics on carbon cycle
The coupling of the N dynamics into Noah-MP not only adds N-related modeling but also affects other components of the model, i.e., the carbon and water cycles. This is be-cause the change in NPP affects leaf biomass and hence LAI. space The change in LAI can affect photosynthesis, which in return affects NPP. Figure 6 shows the comparison of the simulated C-related state and flux variables between the default and N dynam-ics enhanced Noah-MP. We can see that NPP is decreased from 544 to 432 gC m −2 . Most of the decrease occurs before the peak growing season, which results in a slight decrease in LAI. However, the peak LAI has very minor increase. After the peak, LAI decreases more slowly than the default, which is due to the decreased turnover rate proportional to the NPP down-regulation rate. If the turnover rate is not down-regulated accordingly, the peak LAI will be cut in half. Due to the slower turnover rate, more leaf biomass (indicated by LAI) is involved in photosynthesis. Therefore, compared to the default, Noah-MP with N dynamics generates higher gross primary production (GPP) during the second half of the growing season, although it is lower during the first half of the growing season. Annual mean GPP is decreased by about 28 gC m −2 .
Net ecosystem exchange (NEE) has a similar change. The annual NEE is 179 gC m −2 ( 183 gC m −2 ) from Noah-MP with N dynamics (default Noah-MP), which is comparable to the NEE in West et al. (2010) for this region. Its absolute value is decreased by 4 gC m −2 , which means that the C sink is slightly decreased. This decrease is small compared to the GPP decrease, probably because soil respiration is also decreased. All annual peaks of NPP, LAI, GPP, and NEE are de-layed for about half a month. This is probably due to the fact that the primary N fertilizations (> 10 gN m −2 ) were mainly space  ., 9, 1-15, 2016 www.geosci-modeldev.net/9/1/2016/ applied after late June and thus plants encountered high N stress during the first half of the growing season.

Impacts of nitrogen dynamics on water cycle
Through the changes in LAI and soil organic matters (SOMs), the addition of N dynamics affects not only the car-bon cycle but also the water cycle. The change in SOM is not currently considered, and therefore the impacts on the water cycle are from the change in LAI only, as shown in Fig. 7. These impacts are most pronounced on plant tran-spiration, which is increased by 33 mm yr −1 . The increase mostly occurs during and after the peak growing season. In Cai et al. (2014a), Noah-MPsimulated evapotranspiration (ET) over croplands increases too fast during the first half of the growing season and reaches peak about 1 month ear-lier than observation. The delayed peaks of LAI and ET can partly mitigate this issue. As there is more water extracted from soil by transpiration, soil moisture further decreases during the second half of the growing season. Therefore, less water is available and thus soil evaporation is decreased by 9 mm yr −1 . With the increase in ET, runoff is decreased by 13 mm yr −1 . Therefore, besides the great implications for C modeling and the potential for being used in environmental predictions, the addition of N dynamics can improve the hydrological simulations as well. space  space(2) N fertilizer is applied on 15 April every year. The first experiment is designed because in this site a large amount of N fertilizer is applied mostly during mid-June and early July. Other dates are also reported in the literature; therefore, we use 15 April as another example. Both experiments use the same amount of N fertilizer as in the management log, which on average is 7.8 g N m −2 yr −1 . Figure 8 shows comparison of some of the most relevant results between the two experiments and the one (real) with space June experiment is much closer to the real case; even the seasonal variation is identical. The largest discrepancy is in 1993 and 1996. Based on the management log, in these two years, a large amount of N fertilizer was applied, which resulted in much higher NPP than results from the two experiments. Since 15 April is much earlier than the primary fertilizer application dates, NPP from this experiment is flattened out through the year. This also confirms the statement in Sect. 3.5 that later N fertilizer applications delay plant growth. Sim-ulated N uptake from both experiments shows exactly the same story as NPP. The simulated N leaching shows the opposite pattern to NPP. The default simulation produces the highest leaching, followed by the 20 June experiment and then the 15 April experiment. This is very likely because the fertilizer application dates are closer to the flood season for the former two cases and the chance of fertilized N being flashed out is higher. The difference in N fertilization dates also clearly affects the simulations of total soil nitrate. In the 20 June experiment, soil nitrate reaches the lowest level in May because no N fertilspace izer is applied before 20 June. In the default case, N fertilizer can actually be applied as early as April, but with a smaller amount before mid-June, which prevents the soil nitrate con-centration from getting too low. Besides a large amount of N fertilizer applied in later months, the other reason that the default simulation reaches the highest concentration of soil nitrate is because it produces higher NPP, which can be re-turned to soil for decomposition. Overall, the default simulation grows better plants (higher NPP) because N fertilizer is applied based on expert judg-ment of plants' demand. At the same time, however, it pro-duces more N leaching than the two experiments, which is significant (insignificant) with respect to the 15 April (20 June) experiment at 90 % confidence level. The experi-ment with closer dates of N fertilizer application to reality can better reproduce the N dynamics in observation. There-fore, although we cannot always know the exact dates of N fertilizer application, a survey on this can help to improve model simulation. space

Analysis of nitrogen uptake
As described in Sect. 2.2.1, plants can get N for growth from four pathways: passive uptake, active uptake, fixation, and  ., 9, 1-15, 2016 www.geosci-modeldev.net/9/1/2016/ retranslocation, and the last three require C costs. Figure 9 shows the actual N uptake from these pathways and their percentages of contribution to the total N uptake. Passive up-take is the dominant pathway, which contributes 57.7 % of the total N uptake. Fixation, active uptake, and retransloca-tion contribute 28.6, 8.7, and 5.0 %, respectively. This con-trasts the results from the study by Brzostek et al. (2014) for non-fertilized trees, in which passive uptake only accounts for a small contribution. This is understandable because the purpose of fertilization is to minimize active uptake so that more NPP can be retained for crop growth. As demonstrated in Timlin et al. (2009), a higher fertilization rate results in a higher ratio of N uptake in transpiration to total N uptake. On the one hand, fertilization maintains soil nitrate concen-tration at high level. On the other hand, higher NPP for crop growth in turn results in higher LAI and thus higher transpi-ration. During peak growing season, therefore, plants receive a large amount of N under the combination of high transpiration and high soil nitrate concentration. During other periods, biological N fixation dominates.

Analysis of major soil nitrate fluxes
The soil nitrate storage with time is an outcome of the variations in incoming and outgoing fluxes. Besides N fertilizer and atmospheric deposition, humus mineralization and residue decomposition are the two major incoming fluxes. Because N fertilizer is a jumping behavior and atmo-spheric deposition is a relatively small fraction in this study space Figure 10. Daily climatology of the soil nitrate (blue solid line) and some major fluxes (color label bars) going in (humus mineralization and residue decomposition) and out (plant uptake, nitrate leaching, and denitrification) of the soil nitrate pool.
( 1.5 gN m −2 yr −1 ), they are not analyzed here. The major outgoing fluxes are denitrification, leaching, and plant uptake. Figure 10 shows the seasonal variation of the above ma-jor fluxes. During the growing season, N fertilizer provides an important role in meeting the plant N demand; however, residue decomposition still makes the largest contribution and is the dominant factor responsible for the increase in total soil nitrate. During the non-growing season, a large amount of N is lost through denitrification and N leaching. However, when it reaches the peak growing season, plants consume a large fraction of soil nitrate, which leaves very little for denitrification and leaching. N leaching is mostly associated with the timing and intensity of precipitation. Denitrification is also associated with precipitation, but it is directly related to the soil water content. High denitrification rate occurs dur-ing high soil water content, especially during water logging.

Conclusions
In this study, a dynamic N model is coupled into Noah-MP by incorporating FUN's strength in plant N uptake and SWAT's strength in soil N cycling and agricultural management.
We evaluated the new model at KBS that provides goodquality, long-term observed N and ecological data. The model-simulated soil moisture is consistent with observation, which shows that Noah-MP provides a good water environment for the N cycling. The simulated concentrations of soil nitrate and N leaching from soil bottom also compare well with observations. Although the model does not simu-late some peaks well, especially for N leaching, the averages are very close to the observed values and the correlation co-efficients are reasonable. Considering the wide spread of the range error bars defined by the measurements at the six repli-cates, the model shows high skills in capturing the major N flux/state variables. The significant improvement of annual NPP simulation demonstrates that the N limitation effect on plant growth is well represented in the model. space spaceMoreover, the addition of N dynamics in Noah-MP im-proves the modeling of the carbon and water cycles. Com-pared to the default Noah-MP, NPP simulations are improved significantly, in terms of consistent averages and much higher correlation coefficients with observation. The temporal pat-tern of simulated ET is also improved, featuring lower ET during spring and delayed peak.
This enhancement is expected to facilitate the simultaneous predictions of weather and environment by using a fully coupled Earth modeling system.

Code availability
Noah-MP is an open-source land surface model. The model is being developed by a community led by The University of Texas at Austin. The code is archived at both http://www. ral.ucar.edu/research/land/technology/noahmp_lsm.php and http://www.jsg.utexas.edu/noah-mp. The new code implemented in this study will be made available and may be obtained by contacting the corresponding author via email.