Dear Amelie Saunier et al.,

The revised manuscript has improved significantly. I only have a few comments and the clarifications of these will make the discussion and conclusions clearer. Most of the comments are of technical nature.

Scientific comments:

Page 6 lines 196-197: “The emissions of all BVOCs followed during this experimentation were reduced under amplified drought compared to natural drought, especially in spring and summer (Table 1) except for acetaldehyde emissions”. Was this statistically significant? Can you give p-values e.g. in Table 1.

Page 7, line 223: “…a slight underestimation of emissions (sl = 0.84, P < 0.05)…” how can the algorithm, if fitted seasonally to the data, underestimate the emission? If I understand correctly the EF to calculate the emission was obtained by fitting the algorithm to the exactly same dataset. If I misunderstood, others may also and thus this needs to be written more clearly.

Also, is the p-value for the correlation or have you tested the deviation from 1:1 line? You should have the latter to be not misleading the readers.

These two questions concern also at least the following claims:
-Page 7, line 231: “…a slight underestimation was observed (sl = 0.87, P < 0.05)”.
-Page 7, lines 236-237: “…underestimations were observed in spring and summer (sl = 0.72, and sl = 0.57, P < 0.05, respectively)”
-Page 7, line 238: “…underestimation was only observed in summer (sl = 0.80, P < 0.05)”.
-Page 8, lines 263-264: “…slight underestimations were observed (sl = 0.88, P  < 0.05 and sl = 0.69, P  < 0.05, respectively)”.
-Page 8, line 265: “…line overestimation of modelled emissions (sl = 1.27, P < 0.05)”.



Technical comments:

In many locations: “BVOCs emissions” should be “BVOC emissions”

Page 2, lines 68-70: “These studies reveal that there are still some misunderstandings at the level of emission mechanisms since some works showed increases (Funk et al. 2004; Monson et al. 2007) or decreases of isoprene emissions (Brüggemann & Schnitzler 2002; Fortunati et al. 2008)”. Would this indicate rather lack of understanding rather that misunderstanding?

Page 3, lines 71-72: “Moreover, the sensitivity of isoprene and highly volatile BVOCs emissions to recurrent water stress (few years) under in situ conditions is clearly missing”. Is it sensitivity or rather understanding of the sensitivity that is missing?

Page 4, line 116: “air generator”. Is this zero air generator?

Page 4, line 132: “Exchanges of CO2 and H2O from the enclosed branches were…” should be “Exchange of CO2 and H2O from the enclosed branches was…”

Page 4, line 144: “catalyzer”. What is catalyzer? It has not been described before. Is it air generator mentioned in line 116?

Pages 4-5, lines 144-145: “Each sample was analysed every hour, with 15min of analysis”. How can this be as there are five samples according to the previous sentence.

Page 5, line 148: “A calibration gas standard” Which compounds were in gas standard?

Page 5, Equation (1) expresses gas exchange per foliar mass. However, on page 4, lines 138-139 you state “Gas exchange were hence expressed in a leaf surface basis”, leading to inconsistency.

Page 6, lines 212-213: “All six BVOCs monitored showed daytime light and temperature dependencies (isoprene, degradation products of isoprene and acetaldehyde),…” The beginning of this sentence is confusing as it mentions six BVOCs but then only gives three in parentheses.

Page 7, lines 247-248: “However, some observed phenomena suggested that methanol emissions was sustained by temperature alone at certain moment of the day”. This sentence is a bit confusing and vague. Could it better as “However, some observed phenomena suggested that methanol emission was sustained by temperature alone in the absence of light”.

Table 1: Too many significant numbers in part of the values. E.g.  107.7± 18.6 should be expressed as 110 ± 20 (as the two last digits in the original form contain no information).

Dear Amelie Saunier et al.,

The manuscript has improved considerably. I have one detail that I find confusing still in the analysis. In your reply you didn’t really answer to my question on it.

In your reply you write:

“Indeed, EF was determined with the same dataset as used to modelled emissions. EF was the slope of the correlation between emission rates and Cl*Ct for L+T algorithm and the correlation between emission rate and Ct for T algorithm (line 172-176)”.

If you determine the EF as the slope between the emissions and Cl*Ct or Ct, and then plot the data modelled with these EFs against the very same data the EFs were derived with, I would expect 1:1 line, unless non-linear data transformations have been done. If that is not the case I would doubt the fitting procedure.

Furthermore, you write:

“In our study, temperature, light, season and the water stress were factors taken into account to modeled BVOCs emissions”.

Actually, if I understand correctly, only temperature and light are taken into account explicitly (in algorithm). Drought effect is only implicitly included in the seasonal variation of EF.

And further:

“An under (or –over) estimation between measured and calculated emission rates highlights the fact that the driving parameters considered in the algorithm (temperature, light, season, drought) did not allow to explain 100% of emissions. Thus, it seems that other factors which are not taken into account in our study influenced emissions”.

This seems to be misinterpretation, as you have used the very same data to create the EFs that you use to evaluate them, and thus the EF implicitly is influenced by all parameters in seasonal timescale. To me it looks like there is something strange in the procedure of obtaining EFs, calculating the emission and then comparing them back to the original data.

If we look for example methanol emission in autumn in Figure 5, we see that the L+T algorithm constantly underestimates the emission. This is very strange if the EF used in the calculation is obtained from this same dataset. One would expect part of the data be above, part below the algorithm. Were the algorithms fitted to original data and not the averages shown here? Could the distribution of the data be very skewed to cause this?

I would like to have a short explanation on how the systematic deviation of the modeled emission and measured one is possible in this case.

Sincerely,
Janne Rinne
Co-Editor

Dear Saunier et al.,

Thank you for your detailed reply. Now I understand where the discrepancy between modeled and measured emission originate from. As I said, one would expect modeled the modeled versus measured comparison to follow 1:1 line (of course with R2 below unity) when the emission factor is obtained with the same data that is used for comparison. In your study, you have used only the slope to calculate the fluxes, disregarding the intercept. In the case of your example, methanol in summer, natural drought, this leads to an underestimation of around 0.05 gC gdw-1 h-1. We can also see from Figure 3-5 of the manuscript that the systematic underestimation occurs for those compounds which exhibit considerable night time emission, and thus intercept on fitting. For compounds with zero nigh time emission the algorithm works much better. The proper way of fitting emission algorithms is to force the intercept to zero, or to describe the intercept, i.e. the light independent emission, with an additional parameterization, as done e.g. in the case of monoterpene emission from boreal trees by Ghirardo et al. (2010).

You should modify the manuscript to reflect the origin of the underestimation in the night time emission of certain compounds.

Reference
Ghirardo, A., K. Koch, R. Taipale, I. Zimmer, J.-P. Schnitzler & J. Rinne: Determination of de novo and pool emissions of terpenes from four common boreal/alpine trees by 13CO2 labeling and PTR-MS analysis. Plant, Cell & Environment, 33, 781-792, 2010.

Janne Rinne
Co-editor

Dear Amelie Saunier,

Your modification to the manuscript does not yet fully satisfy. In matter of fact, if I plot the (visually estimated) points in the first panel of Fig 3, I get slope of 1.06 (close to one, (intercept of -0.80, Matlab polyfit function), not 0.84 as in the manuscript. So, this makes the data processing still look a bit suspicious and the one sentence added to the revised manuscript does not fully cover this.

If I understand correctly, the only difference in correlating emission against CTCL, as you do for obtaining EF, and plotting emission against modeled emission, as you do to obtain slopes in Fig. 3-5, is that in the latter one you multiply the x-axis by EF (as E_model=EF*CTCL). Am I correct in this?

Thus, in the case when you have no nigh-time emission (intercept = 0) the slope between modeled emission and measured emission should be close to unity. However, in several cases this is not the case (Fig 3, spring natural drought; Fig 4, several of the plots). Have you looked at how similar are the emission against CTCL plots to the emission against model emission plots? Could these be added as supplementary material?

Because of the above mentioned possible discrepancy in the data processing, I feel the following sentence, “The modelled emissions were very representative of measured emissions except in spring for natural drought when we obtained a slight underestimation of emissions (sl = 0.84, P < 0.05) maybe, because light and temperature, in spring, were not the only parameters driving isoprene emissions”, may be over-interpretation.

Page 8, lines 283-285: “Predicting emissions rates of these 3 compounds (methanol, acetone and formaldehyde), during the night, seem to require other parameters such as a temperature threshold, below which methanol, acetone and formaldehyde synthesis and so emissions do not occur (Ghirardo et al. 2010)”. You should remove citation Ghirardo et al. from this as that paper does not deal with methanol, acetone or formaldehyde.

Janne Rinne
Co-Editor

Dear Saunier et al.,
From you answers and manuscript I understand now that you have obtained the EF as slope of the CTCL or CT (on x axis) against measured emission (y axis) without forcing line through origin. Then you compare the algorithm performance by slope of measured emission (x axis) against modeled emission (y axis). In the case of non-linear relations this will lead to slopes below 1, even though modelled emission (x axis) vs measured emission (y axis) would have slope of unity. Thus, the result concerning the slopes being below unity may be partly due to non-linear relation of modeled and measured emission. The nonlinearity of relation for isoprene is actually interesting, as it can be due to light penetration deeper to canopy in high light levels. The algorithm used here is anyway a big leaf approximation.

Page 1, lines 22-13: “these 3 last compounds detected under the same ion”. Actually, not the same ion but same mass-to-charge ratio.

Page 2, lines 69-70: “…impact of water stress on highly BVOCs emissions.” What does “highly” mean here? Please reformulate.

Pages 5-6, lines 181-182: The slope of those correlations indicate if there was an under- or over- estimation of modelled emissions when sl < 1 and sl > 1, respectively. There is also under/over estimation if significant intercept, even if slope is 1. See fig 5, autumn methanol AD.


Janne Rinne