See the attached file.

This paper describes differences in BVOC emissions resulting from different versions of the MEGAN model, along with the effect on ozone and other trace gases over portions of China.  Differences due to the specification of vegetation types are also examined.  Parameterizations developed for specific processes in air quality models, such as online treatments of biogenic emissions, continually undergo modification as new information is available to constrain the relationships used by the parameterization.  So, it is not surprising that the estimated emission rates are different among the versions of MEGAN.  The other conclusion is that specification of the subgrid vegetation fraction is also important, which was already shown in Zhao et al. (2016) (and noted in lines 253-254) who is a co-author on this paper. 

The authors conclude that their results show there is still a large uncertain range in modeling BVOCs (line 61); however, this begs the question: What is uncertainty?  In terms of the paper presented, the range of results are from different versions of MEGAN.  This sort of model-to-model uncertainty might be even larger if one considers other treatments of BVOC emissions (listed in lines 88-91). But what is the value of using an older version of MEGAN when a newer version is available and has presumably been shown to perform better and/or have more physical processes represented than the older version?  What is lacking here is a comparison of predicted BVOCs with observations to truly understand uncertainty.  Maybe there are no observations of BVOCs such as isoprene or monoterpene in these regions for the simulation periods.  Perhaps the simulations could have been done for periods when such observations are available and/or for different parts of China is needed.  I would assume that at a minimum some air quality data, such as ozone and NOx, would be available to compare with the model results to indirectly assess the effect of BVOC emissions. Section 3.1.3 seems to suggest there are some observations, but they chose not to show any results. In terms of specification of vegetation, one can see differences in the vegetation classifications between the dataset and conclude that those will result in different BVOC emissions without even running MEGAN.  There should be satellite derived products that could be used to understand differences in the types and spatial distribution of vegetation between observations and datasets used by MEGAN. 

In addition to chemical observations, an evaluation of the predicted surface meteorological quantities (temperature, precipitation, soil moisture, radiation) using observations would have been useful to understand how those uncertainties would influence the predicted BVOCs.  A time series of observed and simulated meteorological quantities over the month-long period compared with predicted BVOC emissions would have been useful. The month-long period would also make simulating soil moisture challenging.  There are satellite products that could be used to assess soil moisture variability over the region.  As pointed by the authors, the major difference in MEGAN v3 is the inclusion of a drought activity factor (lines 239-240), but the importance of this factor for the 2 month-long simulations is discussed only briefly (line 486-495) in terms of seasonal changes.  The authors do not say whether or not April or July of 2015 are periods characterized by drought.  The authors note that the drought factor is constrained by limited observations (line 242) and may not be suitable for China (line 244).  So how is one supposed to assess the seasonal impacts of this parameter compared to other parameters in MEGAN (i.e. Figure 7)?  Perhaps some sensitivity simulations are needed with a range of values for the drought factor to get a better idea of its potential effect.   

In summary, while I appreciate having differences in the physical representation among the MEGAN versions discussed in one paper, my issues with this paper are 1) the primary main findings have been described previously and 2) there is a lack of any observations needed to fully understand model uncertainty. Section 3.1.3 implies there are observations.  If the authors choose to include only modeling analyses, the paper needs to be revised to frame the purpose of the paper better and improve Section 3.1.3 to put their results into the context of what might be observed in the real world, what is needed to assess components of that latest version of MEGAN, and what are the most missing or uncertain treatments – apart from the specification of vegetation. 

In addition, the abstract has numerous awkward phrases and I pointed out those in the specific comments below.  However, I did not point out other instances throughout the paper.

Specific Comments:

Line 37: The phrase “over East China” seems out of place and could be deleted.  The first part of the sentence actually applies to many places in the world, not just East China. The authors should true to put the paper into a context for a broad range of readers whenever possible. 

Lines 37-38: Change “are generally resulted from”, which is an awkward phrase, to “are primarily due to”.

Lines 38-39: Change “in model” to “in the model”.

Line 38-40: Change this sentence to “One originates from different treatments in the physical and chemical processes associated with the emission rates of BVOCs.”

Line 40: Change ”from the biased distribution of vegetation types” to “errors in the specification of vegetation types and their distribution”

Lines 47-48: Change “biogenic VOCs” to BVOCs which was already defined.

Line 59: The authors use “significantly” here, but that is a vague term.  Can they provide a number to quantify range?

Lines 61-62: This sentence just states that the three versions of MEGAN produced different results, which is not surprising since updates usually include new treatments based on observations or theoretical relationships. Presumably, older versions have out-of-date information and should be less accurate. But at this point, the abstract fails to say anything about how the simulated BVOCs compares to observations which is a better way to characterize uncertainty. 

Lines 166-182: This section does not list the land-surface parameterization used.  As noted by other studies MEGAN2.1 (called here MEGAN3.0) is coupled to the CLM land surface parameterization.  The other versions of MEGAN can use different land-surface parameterizations.  So it is not clear whether differences in the land-surface treatment, which will affect the surface energy budget and fluxes, will affect the BVOC emissions.  While Zhao et al. (2016) found that differences due to the land-surface parameterization were of secondary importance compared to specification of the vegetation, that point on the model configuration should be cleared up.  Also, I assume that meteorological predictions should be identical among the simulations?  That should be explicitly stated.

Lines 512-513: This is a very general statement. Which results are the authors talking about?  BVOC concentrations or some other quantities? 

Lines 514-515: This statement needs to be backed up by the evidence that is not shown.

Lines 583: There can be anthropogenic isoprene emissions so how did the author separate out the biogenic source of VOC concentrations versus the anthropogenic sources?

Line 551: The authors need to state how the calculation of ozone from biogenic and anthropogenic sources are separated out.  I assume this is a highly non-linear system and it is not easy to separate out these sources since biogenic emissions will influence the anthropogenic system and visa versa as emission sources mix.  This comment also applies to NOx sources in lines 562-563.

Line 580: At the end of Section 3, there needs to be some discussion regarding the meteorological conditions during the simulation periods in 2015.  Some the figures (6,7,9) plot differences between July and April.  Presumably the seasonal changes are dependent on meteorology which might change from year to year. The authors should discuss whether these differences would be representative of other years and why.

Adding the capability of MEGAN v3.0 into WRF-chem is a great advancement for the community. However, the paper needs to be restructured and more evaluation of the model updates against observations are needed. The main problem is that the overall conclusions of the paper (there are large uncertainties in the BVOC emission schemes) are not supported by the results of the paper. Differences between sequential updates of MEGAN v1, v2, and v3 does not demonstrate uncertainty as presumably the later versions are more accurate as explained further below. The paper could easily be restructured to show how the sequential updates of MEGAN v1, v2, and v3, which increase in process complexity, have led to improvements (or not) in BVOC emissions, ozone, and PM2.5. This restructuring should better emphasize the isoprene observations specified in Table 3 and compare to other observations such as ozone and PM2.5 that may be more readily available than isoprene as explained more below. Also as further explained below related to Figure 12, please confirm that soil NOx does not change between your MEGAN v1, v2, and v3 configurations?

Specific comments:

Line 35 and 61: Because the differences in BVOC emissions from MEGAN v1, v2, and v3 are sequentially improving, the differences between the versions do not demonstrate uncertainty in BVOC emissions, but instead model advances. Certainly, there is still a lot of uncertainty in BVOC emissions, but this paper does not really address these uncertainties. For example, testing the uncertainty in the MEGAN inputs like emission factors and vegetation types or comparing MEGAN to an entirely different BVOC emission scheme would be a way to evaluate uncertainty, but comparing MEGAN to older versions of itself without a more thorough comparison against observations than that provided in this work is not an appropriate way to demonstrate uncertainty.

Line 153, Please add a reference for MEGAN v3.0 here.

Line 211: Please be very clear in this section how your version of MEGAN v2.0 differs from the released version. For example, state what is available in the released version and then specify which options you use or adjust in this work.

Line 243: Is there an estimate on the uncertainty on this alpha value? Some sensitivity tests showing the range of possibilities of this alpha value and its impact on isoprene would demonstrate existing uncertainty in BVOC emission schemes.

Line 262: It appears this VEG-2015 is a land cover map calculated in this work. Has this or a similar approach been used in other papers even if for different regions of the world or for different models? If so, please reference them. If not, please include more references on the advantages to using this new approach to make it clearer to the reader, which land cover type is better.

Figure 4: Why is v2.0 here not separated by land cover type like v1.0 and v3.0 in this figure? And which land cover type is used here for v2.0?

Figure 5: For clarity, please list what PFT-1, -4, and -6 refer to in the figure or figure caption.

Line 414: Which MEGAN versions do you mean by “current versions of MEGAN”

Line 449: This section is quite interesting and demonstrates important improvements to MEGAN v3.0 versus v2.0 and why future model simulations should move to MEGANv3.0. Seems like these advances and results should be highlighted more clearly in the overall conclusions.

Line 507: Please provide references here. From Figure 10, the formaldehyde concentration looks to be quite different between the MEGAN versions especially in the northern part of the domain. These are not large enough differences to see on a satellite? Plotting formaldehyde from anthropogenic VOC emissions here would be useful to fully explain this.

Table 3: Please provide significantly more detail about the observations listed in Table 3. For example, are these observations averaged over some time period? Is the averaging in the model and observations the same? When (year, season, month) were the observations collected? If the time was different than the model runs, how would this impact the comparison? For the first observation, this is no reference listed. Please add more info about where this data came from.

Line 533: Please explain in more detail how you got these biogenic VOC and biogenic formaldehyde values. Did you run the model without anthropogenic VOC emissions or without anthropogenic VOC and NOx emissions or did you use a tag for formaldehyde production from biogenic VOCs? This is important to describe as anthropogenic NOx will impact the production rate of formaldehyde from BVOC emissions.

Line 535: Adding similar plots of the total VOC and total formaldehyde would make it clearer that BVOCs contribute significantly to total VOCs over East China. This statement is contradictory to your statement previously for why you could not use satellite formaldehyde to evaluate the changes in BVOC emissions, please explain further.

Figure 11: These are large differences in surface ozone. Please describe more what is meant by “ozone due to biogenic emissions” is this the ozone value for the simulation with all emissions on minus that without anthropogenic emissions? And in the simulation without anthropogenic emissions does this include removing NOx and VOC emissions? Because the combination of anthropogenic NOx and biogenic VOCs leads to ozone production. It is important to be very clear here what you mean by ozone due to biogenic emissions. Is this biogenic VOCs only or is this biogenic VOCs and anthropogenic NOx? Most of the model domain is maxed out in MEGAN v3.0 please increase the range of this color bar. Also given the large impact these MEGAN inventories have on surface ozone and possibly surface PM2.5 through SOA formation, comparisons against surface ozone and PM2.5 observations would be useful for evaluating the updates to the model. Since there are not a lot of isoprene measurements available, evaluating ozone and PM2.5 is a good second choice.

Line 568 and Figure 12: Does soil NOx change at all in these simulations between MEGAN v1.0, v2.0, and v3.0 or do you turn soil NOx off in these simulations? The differences in Figure 12 look more like differences in soil NOx than changes in the NOx lifetime? If soil NOx is different between these simulations, please calculate the total soil NOx emitted and make sure the values seem reasonable compared to other studies.

Line 644: I do not agree that differences in BVOC emissions calculated by MEGAN v3.0 compared to older versions of MEGAN especially very old versions like v1.0 that are not used in any current model demonstrates uncertainty. See first comment. Restructuring this paper toward evaluating the differences in the MEGAN versions and determining whether MEGAN v3.0 improves (or not) model performance of BVOCs, ozone, and PM2.5 would be more useful.

Technical corrections:

-In key point 4, do you mean “is sensitive to the MEGAN version”?

-Line 39 “coupled in the model”

-Line 39, do you mean “chemical processes” and I would restructure this sentence to be clearer.

-Line 74 “VOCs play”

-Line 97 “the impact of BVOCs on air pollutants” or rephrase in another way

-Line 150 “to investigate”

-Line 194 “loss and production”

-Line 263 “over all of China”

-Line 309 Remove extra spaces and “emissions are”

-Line 311: “More details”

-Line 336: “USGS, large differences”

-Line 379 “Over the southwest domain”

-Line 395 “when estimating”

-Line 475 and 478 “Light-dependent”

-Line 504: “for use in this study”

-Line 588: “MEGAN v3.0 includes”

-Line 595: “Physical and chemical”

-Line 1201 “dataset”