General Comments:

The authors present a comparative study of three ammonia surface exchange schemes, compared with NH₃ flux observations from a natural ecosystem. The topic is interesting, and the paper is written in a concise manner, but leaves some important details out of the main body of the text.

Specific Comments:

Paragraph at lines 46 – 54: The discussion of the NH₃ surface exchange schemes here is very abstract, and would benefit from including at least one illustrative figure, and/or one or more equations (for example the equation used for the compartment-specific compensation point; the parallel/in series resistances). This material is covered in the methods section as well, but should be introduced more clearly here.

Line 74 – 76: what is the temporal resolution of the sonic anemometer and the on/off-site meteorological measurements?

Section 2.2: This section discusses the parameters used to model the NH₃ fluxes in this study, but does not state how the modelling was done. I see that in the acknowledgements, thanks is given to Cor Jacobs for his assistance on the Fortran code of DEPAC 1-D, was all of the modelling done in Fortran? Did the modelling consist of setting up the system of equations from Table A-1, and calculating the net flux at each time step? Connecting with the comment for line 74 – 76, what time step was the model calculated at? The same as the fastest measurement, the slowest, or?

Additionally, I believe this section would benefit from moving table A1 into the main text, and more directly explaining the parameters in the text as well. (Which may simply involve appending the appropriate equations to the existing in-text descriptions).

Line 89 – 96: This paragraph might benefit from one or more equations or diagrams that summarize the relationships expressed in the text.

Line 97 – 99: I would suggest introducing the per-compartment compensation point (and as below, writing out the equation used in this study), and then extending this to the canopy compensation point, rather than the current order of introducing the canopy compensation point first and the per-compartment compensation point afterwards.

Line 105– 107: This might benefit from writing the equation used to calculate the canopy compensation point.

Line 117 – 118: Is Γ_s in the DEPAC scheme based on a meta analysis from Van Zanten et al (2010), or by a separate meta analysis by the authors?

Line 124 – 125: Is Γ_s calculated using an empirical equation & meta-analysis done by the authors, or reported by Massad (2010)? If the author’s work, more detail should be provided, otherwise add the appropriate citations.

Figure 1 (Line 141): I believe that X_(z0), and R_inc were not previously defined in the text

Line 220: Perhaps “Conversely” rather than “Contrarily”

Line 196 – 200: If the timing of the DEPAC scheme closely matches the measured fluxes, but is significantly offset, while the Zhang scheme closely matches the order of magnitude of the observed fluxes, but not the diurnal cycle, could better results be obtained by using the applicable inputs for the Zhang scheme as the inputs for the DEPAC scheme (or vice versa)?

Line 237: Would the DEPAC scheme result in similar results to the Massad scheme (and/or more closely match the observations) if the same parameter was used for Γ_s in each modelling scheme?

Section 4.1 (line 308): Do the poorly modelled emission episodes consistently occur only/usually at or around midday/afternoon? Are there factors (e.g., time of day, meteorological variables, SO₂) that can be used to explain the difference between the minority of emission episodes that are (somewhat) properly modelled, and the majority of emission episodes that are poorly modelled?

336 – 339: If the R_w parameterization is important to discuss in-depth, the equations used by each of the model schemes should be shown here, and not just in Table A1.

Line 443 – 446: It might be interesting to prepare a companion figure to Figure 6 (if possible) comparing the modelled deposition for each of the three schemes using their original parameterization methods (e.g. for Γ_soil, Γ_s, etc), but a version that uses the same input values (using your best estimate) for each model—presumably these schemes vary both in how emissions/deposition are modelled, and also what inputs are used, and comparing just on the modelling approach would be interesting.

Section 4.4: Something that stood out to me in this paper was the methodological choice to have each scheme (DEPAC, Zhang, Massad) encompass not only the functions that calculate whether/at what rate NH₃ is emitted or deposited (e.g. F_w, F_soil, R_s, R_w, etc); but also for each scheme to use its own parameterization for the model inputs as well (primarily for Γ_s judging by Table A1). I think this choice needs to be justified in this section, because it isn’t clear to me whether the conclusions—that the Zhang scheme accurately models the net flux, but not the diurnal cycle, while the DEPAC scheme captures the diurnal cycle, but over-estimates the net flux—are due to, for example, the DEPAC scheme over-estimating the mass transfer rate (e.g. Rs or Rw are too small), or because in the DEPAC the concentration gradient is too large (e.g. because Γ_s is too small).

The paper suggests that additional measurements of the underlying inputs (Γ_s, Γ_soil, Γ_w, etc) would improve the modelling approach (line 418), but this would also suggest that the models should be evaluated using the same inputs, rather than each model being evaluated with a separate set of inputs. Perhaps is not a significant factor as the different approaches for calculating the model inputs result in substantively similar values, but if so, that should be clarified in the text.

Technical corrections:

Line 21: The acronym DEPAC should be defined when used for the first time

Line 60: The acronym DEPAC should be defined when used in the body for the first time.

Line 117: Γs (the stomatal emission potential?) used without being defined the first time

Line 119: NH3 missing superscript

Table 1 (line 182): The text DEPAC scheme in the table seems to be right-justified, or is placed off-center.

Table A1(line 500 – 503): Parameters in the table should be defined, and where applicable, units should be given.

Table A2: Some, but not all parameters in the table are given with units; some of the parameters without (e.g. the emission potential) may be dimensionless, but others have previously been given with units (e.g. resistances). Not all of these parameters have been previously defined in the text.

This study assessed the performance of three ammonia bi-directional air-surface exchange schemes by (1) comparing model-estimated fluxes with one-year gradient measurement over a dune ecosystem, (2) comparing the dominant flux pathways between the models, and (3) conducting an error analysis to assess the uncertainties in modeled fluxes. The method used in this study is scientifically sound, some useful results are obtained, and the manuscript is generally well written. I have the following comments and hope the authors can consider when revising the manuscript:

• Quantifying surface flux using the concentration gradient method typically requires the measurements to be conducted over a homogeneous terrain. It is not clear to me if such condition was satisfied where and when the measurement was made. Heterogeneous terrain can cause large uncertainties in the estimated flux, even changing flux directions (e.g., from deposition to emission or the other way around). Even over homogeneous terrain flux measured using gradient method can differ significantly from those measured using eddy-covariance method (as has been seen for O3 flux measurements over a forest reported by Wu et al., 2015, ACP, 15, 7487-7496). What is the uncertainty magnitude (or confidence level) of the measured ammonia flux data? This directly affects the model-measurement comparison results.
• The model-measurement comparison split the flux data into different categories (Table 1), which is a very good practice. Can the analysis also be done by looking at different wetness conditions (as was done for O3 deposition in Zhang et al., 2002, Atmospheric Environment, 36, 4787-4799). Such an analysis may help identify emission sources (e.g., evaporation of morning dew or soil emission).
• Although using a bidirectional air-surface exchange scheme is more theoretically correct than using a traditional big-leaf dry deposition scheme, the former does not necessarily perform better than the latter in the simulated ammonia fluxes on seasonal to annual basis and in regional-scale air-quality modeling, as reported by several existing studies. This is because modelling the bi-directional flux requires additional model parameters such as the soil and canopy NH3 emission potentials, which may not be available at high spatial resolution on the reginal scale. Besides, more model parameters can introduce additional uncertainties. For example, Wen et al. (2014) compared the unidirectional model of Zhang et al. (2003) and the bi-directional exchange model of Zhang et al, (2010) against 53 sites in southwestern Ontario, Canada, and showed that the bidirectional scheme performed best at locations with high observed NH3 concentrations, but overestimated NH3 levels for locations with low observed NH3 concentrations, and the unidirectional dry deposition generally performed better than the bidirectional scheme at sites with low observed NH3 concentrations. Another example is by Zhu et al. (2015, Atmos. Chem. Phys, 15, 12823-12843) who introduced bidirectional fluxes into GEOS-CHEM and compared with surface AMoN observations, which resulted decreasing errors in July, but increasing them in April and October. Thus, similar to Wen et al., 2014 above, even though the bidi scheme incorporates more physics into the parameterization, it still might not perform the best under all conditions. Thus, I am wondering if the authors can recommend under what conditions the unidirectional model is acceptable and under what conditions the bi-directional exchange model should be used?
• Considering the large uncertainties in the modeled ammonia flux between the three bi-directional exchange models investigated in this study (and likely the case for several other existing ones in literature), would an ensemble approach reduce such uncertainties?