This paper explores some of the interactions and feedbacks of a fully coupled nitrogen cycle by examining the interactive coupling between nitrogen emissions from synthetic fertilizer, the resulting nitrogen deposition, the impact on climate, terrestrial ecosystems and crops. The paper first introduces a parameterization for nitrogen emissions from synthetic fertilizer, then evaluates it, and then finally examines various feedbacks with interactive nitrogen emissions.

While the topic is interesting, the paper requires substantial improvement in all aspects prior to publication as detailed below.

1. The explanation of the scheme for nitrogen emissions (section 2.2) needs to be more complete.

While there is a general reference to the DNDC model at the beginning of the derivation of NH3 emissions from synthetic fertilizer, it is somewhat of a mystery where the specific equations come from. Please explicitly include a rationale for the formulation of the specific equations, especially equations 1, 2 and 7.

Equation (1) is in terms of  f_vol , the fraction of the non-adsorbed aqueous NH3 that volatilizes into NH3 gas. This does not seem to include an explicit term for the partitioning between NH3(aq) and NH3(g). If not, why not? Other formulations include this term. Where does the formulation for f_vol come from (equation 7)? This equation seems to have the peculiar property that for a one layer model (l_max=l) there are no emissions, while for a very thin layer this fraction will be maximum.

Are the emissions sensitive to the vertical depth profile of fertilizer application? If yes, what is the depth profile of application?

Most formulations use the resistance approach for emissions from the surface which seems appropriate. The formulation presented here appears not to include resistances either for emissions into the canopy or from the canopy to the atmosphere. What is the evidence that this type of approach is valid? It also appears that NH3 from any soil layer is emitted directly into the atmosphere. What is the justification or rationale for this?

A number of models have incorporated a bidirectional flux of ammonia emissions. The results from the model presented here are indeed more complex than some of the simplest schemes used, but they seem somewhat less complex than the bidirectional approach in Pleim et al (2019) or Zhu et al (2015). Comparisons to these other more complex schemes should also be made in the text and these additional papers should be referenced.

2. I find the results from the model are rather minimally evaluated and in some cases the evaluation is questionable. This is a bit strange as in lines 72-86 the paper outlines various measurement techniques for evaluation of NH3 emissions, but these are not used in the paper. This seems a little incongruous, as certainly the paper could have used more detailed model-measurement analysis, particularly the N deposition. Better evaluation is needed. While I could understand a minimalistic evaluation if the section on the feedbacks in the system were presented with more depth (see comments below) this is not the case.

-The model evaluation is in part against other established inventories EDGAR, CMIP and MASAGE (although arguably not with state of the art inventories such as the HTAP.v2.2 inventory and the CEDS inventory, which include significant local information into their emission estimates). The trouble is the EDGAR and CMIP inventories do not separate out manure and synthetic fertilizer emissions from agricultural soils. The paper assumes 1/3 of these emissions are “fertilizer associated” (line 364) where I assume the authors mean synthetic fertilizers. This number may be roughly valid globally, but certainly not regionally valid. Regionally, there may be very different apportionments between manure and fertilizer emissions from soils. Consequently, the geographic comparisons and statistical analysis between the CAM4_CLM5 EDGAR and CMIP6 are likely subject to significant local errors and are therefore not suitable for a quantitative comparison of the inventories. The temporal comparison between these emission inventories is also suspect as NH3 emissions of manure from agricultural soils may have a different seasonality than those from synthetic fertilizer.

Two model simulations are compared against the IASI satellite measurements: CAM4_CLM5 and CAM4_CMIP6. It is not clear to me (although maybe I missed it) if the CAM4_CLM5 and the CAM4_CMIP6 emissions are identical except for the synthetic fertilizer NH3 emissions. Are the other sources of NH3 emissions identical so that the only difference in these simulations is from differences in the  synthetic fertilizer emissions? Differences in the simulations can only be attributed to the simulated ammonia emissions if the inventories are identical except for the emissions from agricultural soils. Not only do the NH3 emissions from other sectors besides synthetic fertilizer need to be the same between the simulations, but also the NOx and sulfate emissions as the partitioning of NH3 into the aerosol phase depends sensitively on these emissions also. Thus, unless the emission inventories are identical except for the ammonia emissions from fertilizer, it seems difficult, without more analysis, to quantitatively compare CAM4_CLM5 and CAM4_CMIP6.

3. The model simulations are not well explained.

A number of simulations were made with various feedbacks enabled. However, it is unclear how long any of these simulations were run for, whether ensemble simulations were made and the statistical significance of any difference between the simulations. In these coupled simulations meteorological variability can result in apparent differences. The regional changes in temperature (S6, for example), are quite large, probably larger than can be expected from rather small changes in radiative forcing.

Biogeochemical models of soils are notoriously difficult to deal with, as they are notoriously difficult to spin up to equilibrium. In these simulations this is not discussed. Please elaborate on the spinup of the biogeochemical part of these simulations including the extent to which the coupled system was spun up to equilibrium. What state was the model initialized from? How did this state change with the introduction of the new parameterization? Was the model spun up to equilibrium after making changes to the parameterization of ammonia emissions?

There are some feedbacks with the crop model which need to be addressed. It appears that only synthetic fertilizer is added to the crops whereas in reality manure is also be used to fertilize crops. Consequently, the crops are likely significantly under-fertilized in the model simulations.  The apparent under fertilization in these simulations would suggest that the crops take up a greater fraction of added nitrogen in the model-world than they would in reality. This would suggest the fraction of applied fertilizer volatilized is underestimated in the model. (In reality surplus nitrogen is added to agricultural systems, a fraction of which is lost.) Please address the extent to which this might impact the simulations.

It seems likely that with the under-fertilization the simulated crops may not show sufficient growth. This is likely to have climate impacts including changes in latent and sensible heat flux and changes in albedo. How does crop growth in the scheme documented here compare with that in the standard CLM model without the ammonia emissions? How does this impact the radiative budget?

4. I found the section regarding model sensitivities needs significant more in depth analysis. It seems to me this section could be the real novelty of the paper (schemes with bidirectional fluxes have been implemented previously as have prognostic equations for NH3 emissions). This is particularly true as the model evaluation is not comprehensive.

In section 3.3 the different responses of the system are simulated after a change in forcing (i.e., a change in added fertilizer).  If I understand correctly the authors are comparing the [CAM4_CLM5] with 2000-level fertilization with: (i) [CAM4_CLM5] with a 30% increase in fertilization, with (ii) CAM4_CLM5_CLIM with a 30% increase in fertilization but constant nitrogen deposition and with (iii) CAM4_CLM5_NDEP with a 30% increase in fertilization but constant aerosol forcing. I found the section somewhat confusing, perhaps in part due to the notation. It would probably be clearer if the authors distinguished in their notation the simulations with different emissions (e.g., the CAM4_CLM5 with standard emissions versus that with a 30% increase in fertilization). Also, how was the aerosol forcing kept constant in CAM4_CLM5_NDEP?

In CAM4_CLM5 emissions increase by 27% or to 2.4 Tg N/year. In CAM4_CLM5_CLIM with constant year 2000 deposition fluxes the emissions increase to 2.5 Tg N/year; in CAM4_CLM5_NDEP and constant aerosol forcing they increase to 2.7 Tg N/year. It is hard to interpret the significance of these differences as they seem small on the face of it. Are these differences really significant? What is the difference in radiative forcing? Substantial more analysis could be conducted here. As just one example the paper states  some changes are “likely a consequence of better vegetation growth driven by increased NHy deposition following higher NH3 emissions”. This can be evaluated by examining the model.

I am rather puzzled by the statement (Lines 601, 602): “We estimated that the effect of nitrogen deposition on NH3 emission is +2.7 Tg-N yr–1 globally” with a reference to Figure 5. Figure 5 shows the changes in emissions when fertilizer is increased by 30% compared to the case with no change in emissions. Shouldn’t the 2.7 Tg-N yr–1 increase in emissions in CAM4-CLM-NDEP be, to a large extent, attributable to the increase in fertilizer, not to the effect of nitrogen deposition. Maybe I have completely missed something here.

Other points:

-Lines 104-105: “Recent inventories…..”, but then the paper quotes Sutton (2013). There are in fact much more recent inventories than that. Other more recent inventories include the HTAP_v2.2 inventory and the CEDS inventory which are not mentioned.

-Lines 274- 295 Emissions of other reactive nitrogen compounds. As far as I can the emissions of species other than NH3 are not discussed in the paper or evaluated. This section can then be omitted. It seems to me what is pertinent here is the loss of ammonia through nitrification.

- Line 212: I assume that equation (1) should also include other loss terms: washout and nitrification for example. Please clarify.

-Lines 325-327: In equation (9) why is Vc set to a fixed deposition velocity instead of the deposition used in the chemistry model?

- The constants in a number of the equations in section 2.2 do not have defined units (e.g., equations 4 and 5). Please give explicitly where these equations come from as appropriate and the units for the constants.

-The change in ammonia is apparently calculated for each soil layer (equation 1), but I assume that equation (8) is in terms of all soil layers. Please clarify.

-Line 305: “In our coupled simulations, we omitted the portion of NH3 emission associated with synthetic fertilizer from the inventory input for CAM4-chem.”It is not clear where the inventory for this input comes from in CAM4-chem.

-Line 409: Some synthetic fertilizers have a much smaller ammonia volatilization loss than urea.

-Figure 5. Please include figure captions for the figure components: a, b, c and d. Also the caption on Figure 5b is misleading. It would be helpful if the two cases were distinguished.

REFERENCES

Pleim, J. E., Ran, L., Appel, W., Shephard, M. W., & Cady-Pereira, K. New Bidirectional Ammonia Flux Model in an Air Quality Model Coupled With an Agricultural Model. Journal of Advances in Modeling Earth Systems, 11(9), 2934–2957. https://doi.org/10.1029/2019MS001728, 2019.

Zhu, L., Henze, D., Bash, J., Jeong, G.-R., Cady-Pereira, K., Shephard, M., Luo, M., Paulot, F., and Capps, S.: Global evaluation of ammonia bidirectional exchange and livestock diurnal variation schemes, Atmos. Chem. Phys., 15, 12823–12843, https://doi.org/10.5194/acp-15-12823-2015, 2015.

This paper presents a parameterization for evaluating ammonia (NH3) volatilization from soils within the Commonity Eath System Model (CESM). The authors couple both the emission and deposition fluxes between the land and atmospheric components of the CESM, and use this capability to evaluate the emission of NH3 following fertilizer applications and the transport, deposition and possible re-emission of the emitted NH3. They perform further simulations to evaluate the effect of increased fertilizer usage on modeled ammonia emissions and crop harvests and the role of atmospheric feedbacks in these responses.

The two-way land-atmosphere exchange of ammonia has been simulated in a number of other models, although not within CESM. However, introducing the bi-directional NH3 exchange into the nitrogen cycle of an Earth system model does open up new possibilities for assessing the role of atmospheric transport of NH3 in the global nitrogen cascade. The coupled simulations in this manuscript present an interesting step towards this direction.

Thus, I find the manuscript in principle suitable for publication Biogeosciences. However, I also have several concerns related to the model formulation and the experiments, as detailed below. Addressing these will probably require major revisions before the paper can be published.

General comments

1. Model formulation and evaluation

The model equations (1-9) are supposed to be derived from the DNDC model. However, except for the chemical equilibria in Eqs (3)-(6), which are fairly standard, I haven’t found the equations in the DNDC literature cited. It is possible that I have missed something, since there exist multiple versions of DNDC. But if the DNDC is only a source of inspiration, then the authors should not write that their model is “derived from” the DNDC. In the current form, the model equations look like plausible but rather ad-hoc parameterizations for the volatilization process.

Most of the existing models for NH3 volatilization and exchange have been verified with at least some field data. Here, the model is evaluated by comparing with existing inventories and by comparing the simulated NH3 columns with the IASI data on yearly level. This makes sense, but as the only source of empirical evaluation, it has the issue that the geographic variation predicted by the model becomes conflated with the variation in the N input. In other words, some of the resolved variation is likely to originate in the geographic distribution of fertilizer use. Only a fraction of the global NH3 emission is from synthetic fertilizers, which further weakens the signal.

I understand that there is no easy way around this, but it would be good to see how the model predicts the geographic distribution of the ratio between NH3-N emitted and fertilizer-N applied. Also, for easier comparison with existing inventories, it would be useful to provide the regional emission totals.

Eq. (1): How do you define the potential emission rate? The left hand side is a time derivative of a NH3 concentration (?), which is generally not the same as the NH3 flux to the surface. The factor 1/(delta t) on the right hand side does not make sense: the flux per time unit cannot depend on the timestep of the model.

How does the deposited NH4+ enter the soil pools? Is 100 % of the wet deposition assumed to enter soil, or do you consider surface runoff? Figure 1 indicates that NOx deposition goes to the NH4+ pool, is this really the case?

Eq. (2): What is the source of this formula? Does it mean that f_ads is greater than 1 when f_clay is very small? Why?

Eq. (7): Please give a rationale for this equation. What is the role of T_soil here, given that it already appears in Eqs. (4) and (5)? Why does the flux from a given layer depend on the thickness of the soil column below? If you evaluate the emission from deeper layers, shouldn’t there be also exchange between the layers? Finally, why parameterize the exchange between the soil and the atmosphere using the wind speed instead of using the resistance formulation already present in CLM for calculating dry deposition?

How does the evaluation of the volatilization flux relate to the other NH4-consuming processes? Does it contribute to the N demand similar to the plants and microbial immobilization?

Eqs (8) and (9): Which equations in the DNDC document do you refer to? Also, I’m not sure of what Eq. (8) means – it gives a linear relation between the concentration in soil and in the canopy, but what about the flux? What is the rationale for the 1/s factor? What if the wind is calm? Finally, the constant 14 m-1 seems to go back to the the paper of Erisman et al. (1994), where it is given as an empirical constant in a certain resistance component. How does it correct for the effect of canopy thickness?

2. Experiment setup

More details are needed about the model setup. First, describe how the aerosol-radiation interaction was evaluated. Direct and/or indirect effects, nitrates, sulfates? Second, please describe whether the runs used some kind of nudging of the meteorological fields. This would be very important for understanding the comparisons between the runs which are presented later. Are CAM and CLM the only active components in the simulation?

If the model is driven or nudged by atmospheric reanalysis data (I think this is called the “offline” configuration in Lamarque et al. (2012)), the different simulations will share the same meteorological variability. However, if the simulations are run with fully prognostic atmospheric dynamics, the simulations will develop chaotic variations, and in this case, five years is unlikely to be long enough to obtain statistically significant differences. If the results shown are indeed from a free-running CAM simulation, all comparisons of means between the configurations should be tested for statistical significance to rule out the effect of the internal variability. The large differences in parameters like surface temperature over remote areas (Fig. S6) suggest that this might be an issue.

The experiments using modified fertilization rates (Section 3.3) should be introduced in the methods (Section 2.4). It might be worth noting that the future increases in agricultural production might involve also expansion of agricultural land area, and thus, the fertilizer application rate might on some areas change differently from the total fertilizer use. This would affect the response of nonlinear processes.

3. Results

Regardless of the statistical aspects, some of the findings seem non-trivial and should be backed up with more analysis. First, fairly large differences between the configurations are attributed to aerosol radiative effects. Please show the differences in the aerosol load (e.g. AOD) and in the aerosol radiative forcing, perhaps split by aerosol type if relevant. Are you able to rule out other atmospheric feedbacks, for example due to changed evapotranspiration?

Second, the results for grain production under increased N fertilization seem surprising. Generally increasing N fertilization would be expected increase harvest yield, even if not linearly. Here, the effect is negative for many regions, especially those in the southern hemisphere. What causes this? The authors should verify that the fertilization response in the CLM crop model is realistic (perhaps on a regional or per-crop basis) because otherwise the discussion of atmospheric feedbacks on crop production is not very meaningful.

Finally, I’m a bit surprised to see such a big difference in crop growth (Fig. 6) between CAM4_CLM and CAM4_CLM_CLIM (i.e. due to deposition) over areas like China or the U.S. corn belt, where N fertilization rates are known to be high. How large is the difference in the annual N deposition flux, and how does it compare to the annual N fertilization per crop area?

Specific comments

Introduction: the intro is not bad, but could be shortened to give a stronger focus on the present work. For example, the paragraph about in-situ observations seems excessive, since none of those data are used here.

L53: is the spending in USD a global total?

L142-145: there have been many (non-CESM) modeling studies using the resistance framework to simulate NH3 exchange, including the canopy capture.

L175-180: Lombardozzi et al., (2020) could be a useful reference about the CLM crop model.

Figure 1: Much of the litter N is first assimilated to the microbial biomass and then remains in the soil organic matter (SOM) before becoming mineralized to NH4+. Having a SOM N pool in the figure would make sense, perhaps instead of the microbial N, which is anyway only implicitly represented in CLM (see e.g. https://escomp.github.io/ctsm-docs/versions/master/html/tech_note/Decomposition/CLM50_Tech_Note_Decomposition.html). Also, N2O and NOx are produced by both nitrification and denitrification. Denitrification also produces N2.

Section 2.3: the purpose of this section is unclear, since the rest of the paper is only about NH3.

L297: Is this the setup described in Lamarque et al., (2012)?

Section 2.4: do you include any biomass burning emissions of NH3?. If not, aren’t you missing part of the N deposition in some regions?

L303: biogenic emissions...of isoprene?

L315: by boundary layer, do you mean the quasi-laminar layer resistance Rb?

L321: the Henry’s law applies to NH3, right?r.

L343: Manure N is a significant N source in many areas. What is the reason for omitting it, and how does this affect the model results?

L387: Boyland and Russell discuss a certain type of air quality models. I don’t think that their conclusion can be used as a universal standard for a quite different application.

L410: perhaps even more importantly, some fertilizers have typically much lower NH3 emission factor than urea (e.g. Bouwman et al., 2002). This includes for example anhydrous ammonia, which is common in the US.

L421: “high spatiotemporal correlation is” unclear, I guess you just mean the temporal correlation. Though, how interesting is it to correlate the model to the inventories, given that the inventories usually prescribe the monthly variation? Why not compare to the IASI data on a monthly basis?

L464: did Hu et al. really use CESM?

Figure 4: the IASI heatmap is saturated over large areas. Can you add more color levels to better show the variability of high-NH3 regions?

L515: “ammonium salts” maybe better “secondary aerosols”

L529: I’m not sure if this kind of nonlinearity is excepted, since the NH3 emission is usually evaluated with constant emission factors. The emission has sometimes been suggested to increase faster than linearly (Jiang et al., 2017). Slower than linear increase seems to imply that either plant or microbial N demand increases nonlinearly to the input. Have you tried to analyze this?

L540: I’m not sure if I follow the logic here. What drives such a gradient in plant uptake?

L555: is there any further evidence to show that this is a causal connection?

L566: is this in Tg of C, dry matter, or something else?

L589: more reliable...than what?

L610-615: Are the manure management emissions really easier to track? There is a huge diversity in manure management systems around the world. Not all facilities are confined. In many regions, collecting accurate information about farming practices is certainly not a trivial task.

L620: what data for validation do you have for fertilizer but not manure?

L621: By manure fertilizer, do you mean manure application on crops, or all manure-related sources? But Riddick et al (2016) only considered agricultural emissions, and did not consider different manure management processes.

L629-632: Does the effect of the initial NH4 pools still remain after five years of spinup? What do you mean with a “soil nitrogen map”? I thought that the soil N is evaluated prognostically in CLM.

L640: note that the N fertilization rate soybean is usually low, since it is a leguminous crop.

References