General response:

We thank you for all your comments and suggestions on the manuscript. We now realize that the results section and discussion should be made more compact to help the reader focus on the important and interesting aspects of the article. The results section will be made shorter and clearer. In particular, we will focus on how we present and discuss the results related to different years and chambers, since the amount of specific results related to a particular year or chamber is sometimes larger than necessary.

Specific comments:

Lines 158-159: Can you add in when measurements began for each variable? I had assumed that all environmental variable measurements began at the same time as the N2O flux measurements (although the start date for those is also not mentioned), but the next paragraph about water table level suggests otherwise.

1. Response: Thank you for pointing this out. The measurement period times for each environmental variable will be added. Exact dates for the chamber measurement period will be added as well. We will also more clearly state how the lack of soil moisture measurements at the end of the study period was considered in the analysis.

Lines 171-172: Can you add in what variables were used to model WTL before WTL measurements began in 2015?

2. Response: We apologize for being unclear. We will add a clearer explanation about the modelling of WTL, including the explanatory variables used in the model.

Lines 175-180: How close were the nearest weather stations to the study site? Can you please add in that detail?

3. Response: We will add the distance to the nearest weather station (10 km) in the methods section.

Lines 477: I know that most automated N2O flux datasets are restricted to small areas within an ecosystem due to how far chambers can extend away from a single gas analyzer. Nevertheless, I am not comfortable with characterizing “N2O budgets of the study site” based on a small area that may not be representative of the entire study site. Perhaps this can be more carefully worded (e.g., “study area” instead of “study site”) to avoid the reader assuming inference beyond what is valid.

4. Response: This is a good point since the automatic chambers indeed cover only a small area and the word “study area” can be understood as a much larger area around the chambers. More careful wording referring to the area the N2O budgets represent is used in the revised manuscript.

Lines 490-498: Here a contrast is drawn between the relatively less important spring N2O fluxes at this study compared compared to other boreal and temperate studies in which spring N2O fluxes have accounted for a greater proportion of the annual N2O budget. Rather than simply pointing out this difference, it would be more meaningful to explore what could have caused this difference because it then allows readers to hypothesize whether spring N2O fluxes might be important in their study systems. What might be different about environmental conditions during spring at this site compared to the other studies? Or what might be different about soil properties at this site (perhaps in combination with site differences in weather) that would make mechanisms for high spring N2O fluxes not in play here?

5. Response: Small spring N2O fluxes are an interesting phenomenon, for which we have not figured out any specific explanation. We will look at this deeper and discuss the possible causes in the final manuscript. We will also point out more clearly that the phenomenon is something that can largely vary between sites.

Line 616-618: I do not understand the logic behind this explanation of long lags between peak soil moisture and peak N2O fluxes. If peat retains moisture and therefore anaerobic microsites, then it seems that this would allow high denitrification rates to occur long after a rainfall event or after the water table level drops. Could you clarify the reasoning that peat retaining moisture would lead to a delay in N2O fluxes responding to high soil moisture?

6. Response: We are sorry for being unclear. We hypothesize that the properties of peat allow moisture to stay in the surface soil for longer time, allowing denitrification to occur in the anoxic soi pores for longer time. This could explain long high-flux periods after rain events. N2O production by nitrification in the draining soil contributes to the total N2O flux. We hypothesize that the maximum flux is reached when denitrification and nitrification co-occur in the draining soil when the soil still has anoxic microsites but is increasingly oxic. The slow drainage of peat soil extends the time before optimal conditions for co-occurring nitrification and denitrification are reached. We will add clearer reasoning about this in the revised manuscript. However, we admit that the specific conclusion made here related to soil moisture are relatively weak due to the lack of chamber-specific soil moisture data. We will also indicate this more clearly in the revised manuscript.

Lines 643-645: Again, I caution against using data from a 20 m x 20 m area of the study site to characterize boreal peatland forest N2O emissions or non-agricultural N2O emissions in general. Given the high spatial variation within this small area, can we really assume that the results can be extrapolated across larger spatial extents?

7. Response: We agree with you that the fluxes of the small study area do not represent fluxes on a larger scale. Fluxes and their spatio-temporal dynamics will likely vary also between relatively similar sites as the one used here. We believe that information about the processes and their controls, however, can contribute to the understanding of spatio-temporal variation of N2O even on a relatively large scale. We will be more careful with the conclusions made and make a more clear distinction between general conclusions and site-specific conclusions.

Technical comments:

Line 170: Should this state that chambers 4-5 shared at WTL sensor? Based on Figure 1, chambers 3-4 are on opposite ends of the sampling area so it seems unlikely that they would share a WTL sensor.

8. Response: We apologize for the mistake in the chamber number. The chamber number will be corrected. The corrected sentence: “Chambers 1–2 and 4–5 shared a WTL sensor…”

Lines 506: Missing a comma after “especially during summers”

9. Response: A comma will be added.

The responses are numbered from 10 to 44. The responses to Referee 1 are numbered 1-9.

10.  General response:

We thank you for the valuable feedback. We will add a more detailed explanation of the used chamber technique in the material and methods section (See response 11). We will shorten the results section and focus more clearly on the main results instead of chamber- or year-specific details. 

The daily resolution was chosen for two reasons. Firstly, the temporal variation in N2O flux was dominated by weekly, monthly and seasonal variation instead of hourly variation. Hourly peaks that are not captured by daily means are not typical. When N2O peak starts, the N2O fluxes steadily increase day by day. Secondly and most importantly, the automatic chamber system seems to create an artificial diurnal cycle of N2O from which the natural diurnal cycle is impossible to separate. The fans inside the chambers run at a low but constant speed, while the wind and turbulence of the surrounding environment vary. During calm periods, especially during summer nights, the transfer of N2O from soil pores to the atmosphere is slowed down and leads to increased N2O concentration outside chamber closure times. When the chamber closes and the turbulence increases because of the fan, the N2O from the soil pores is blown to the chamber headspace air. This increased transfer of N2O from soil pores to chamber air creates artificially increased N2O flux in non-turbulent conditions. The phenomenon likely occurs also the other way around in windy conditions such as during daytime, when the fan inside the chamber creates less turbulence than there is in the ambient air, although this is more difficult to spot from the data. The solution to this turbulence problem would be fans that adjust their speed to the ambient wind speed. We considered filtering out problematic hours but setting criteria to define problematic fluxes would have been very difficult and led us possibly to delete large amounts of good quality data as well. We used daily averages since they capture the main temporal variation seen in N2O flux (daily to yearly) and are less affected by the over or underestimation of fluxes caused by the artificial diurnal cycle. We admit that the turbulence problem creates an additional source of uncertainty to the reported annual N2O budgets, and we will mention this in the article. Based on our experience, automatic chamber fluxes measured in drained peatlands with dry and porous peat soil are especially sensitive to this phenomenon. More information and discussion about the artificial diurnal cycle can be found from Korkiakoski et al. (2017) for methane. As a response to your concerns, we will explain the choice of the temporal resolution better in the methods section and include a better description of the turbulence problem and issues related to it. 

The lack of chamber-specific or close-to-chambers soil moisture data is an unfortunate limitation in our study. We will be careful with the conclusions made from soil moisture results alone and support findings with WTL and precipitation that better describe conditions at the chamber site. See also the responses nr. 29 and 39. 

Understanding machine learning models is often a challenge, although tools like variable importance (VI) scores or accumulated local effect (ALE) plots help with interpreting the results. Regarding VI scores we want to clarify one thing here and in the revised manuscript. Variables and their lagged versions are highly correlated, which is considered in model and method selection (See lines 233-237) (Strobl et al., 2007, Strobl et al., 2009, Apley and Zhu, 2020). Out of the set of correlating explanatory variables (unlagged + lagged), the one that gets the highest VI score out of all the correlating unlagged and lagged variables has the highest ability to predict N2O flux, for one reason or another. The reason can be something very data-specific (typical for many machine learning models) or explained by real phenomena (for example, the unlagged variable is truly more important than others). For this reason, the interpretation of the VI score Figure 8 must be done carefully and we have not based on our conclusion about the importance of lag times on VI scores alone (See response nr. 29). We are sorry for not making this clear in the manuscript. We will explain the concept of VIs more carefully in the revised manuscript. We will also replace the current VI Figure 8 with a figure showing only the total summed VIs (unlagged + lagged) for each variable. The current Figure 8 will be moved to supplements. Summed VIs show the overall importance of each variable, including the contribution of unlagged and lagged variables in the model performance, with less risk for very data-specific conclusions. 

Regarding the comment about ALE plots, ALE plots visualize the mean predicted flux in a range of environmental conditions, while VI scores tell about the overall importance of the variable in model predictions. A high VI score of a variable does not always translate into a high mean predicted flux in ALE plot, or the other way around. In our case, ALE plots showcase dynamics in model predictions between lagged and unlagged variables, while the importance of these dynamics for model prediction are spread on several variables in the VI scores (interactions). Also, a certain lag is unlikely to get a high VI score if the lag-time varies depending on other environmental conditions, as it seems to be the case in our data (See lines 627-633). See a more detailed explanation about the interpretation of lag times in response nr. 29. We will explain the concept and interpretation of ALE plots better in the methods section of the revised manuscript to help the reader interpret the results. We thank you for bringing this issue up. 

Specific comments:

L139-140: Could you give more background information on the measurement procedure, such as flush time and/or time given to get the air circulated to reach a chamber-analyzer equilibrium? It must have taken some time for these big chambers. Is there an internal or external multiplexer involved to switch between chambers? Was there any ambient air measured? Was there any issue with drift (I do not know if these analyzers have overcome this due to the method)? What is the detection limit or threshold of the system (e.g. noise level of the system while sampling ambient air from an open chamber. When you are certain this is a small flux rather than noise? Did you adjust for temperature and air pressure fluctuations over the year for the flux calculations? How did you deal with snowy conditions? Did the chambers get buried in the snow or did you dig them out or lift them up with extensions? Etc.

• 11.    Response: We will add more detailed information about the automatic chamber measurement system in the revised manuscript, including minimum detectable flux, flow rate and information about the ambient concentration measurement. The drift of the analyzer is very minimal (maximum drift 0.1 ppb in 24 h) in these kinds of analyzers, does not affect fluxes, and is therefore not considered in flux calculation. The effect on air temperature and pressure is considered in the flux calculation (See equation for flux calculation in Korkiakoski et al., 2017). The mean air temperature of each chamber closure and air pressure measured at the site were used and we will mention this in the manuscript. Information about chamber maintenance in winter (extension collars, clearing snow, measuring snow depth of chambers) will also be added.

L146-148: This information cannot be understood easily without reading the references, which interrupts the reading flow. Please provide a bit more explanation here on how a fan can create a diurnal flux dynamic.

• 12.    Response: We will add a clearer explanation of the turbulence problem in the methods section. See response nr. 10 for the explanation.  

L150: Could you add the analyzer position into the Figure 1b plot?

• 13.    Response: We will add it to the plot. 

L167-173: I would have thought that soil moisture varies much more between chambers rather than WTL. Did you see statistically significant differences between the WTL measurement locations?

• 14.    Response: Water table level can be relatively variable in drained peatland forests due to for example variable distance to closest ditches and varying amounts of transpiring trees. The differences in WTL between all the loggers are statistically significant, with chambers 3 and 6 having the highest WTL and chambers 4-5 having the lowest WTL. 

L194-206: This concept of identifying high-flux periods is not quite clear to me in the way it is described here and why you want to do this numerically. N2O fluxes can be incredibly short lived and do not always continue over several days or even hours for that matter. Is this categorization needed for your interpretation and correlation to seasonal dynamics? Could you state here why you need to identify these periods numerically on the daily resolution? (And please reconsider the order of sentences, this is very hard to follow. E.g. The second sentence of the paragraph is starting with ‘finally’, which implies that this statement is now finished but then the whole explanation on how you get to the 70% comes after it. I thing a few sentences could be deleted to make this shorter and clearer. E.g. ‘High fluxes were measured less frequently (30%) compared to the more common low fluxes (70%), which was apparent from flux histograms (Fig. S2).’ Then you can delete the next three sentences. And finish with the earlier mentioned sentence from above, adjusted to the context here: ‘This histogram analysis concluded the 70% percentile threshold to define high-flux periods.’)

• 15.    Response: We thank you for the feedback regarding the explanation of high-flux period identification. We will clarify the explanation of high-flux period identification in the revised manuscript and pay attention to the order of the sentences. We wanted to identify high-flux periods to explore their length and timing (Fig. 5). High-flux periods were also used to understand the linkages between environmental conditions and N2O peaks (Fig. 7) besides random forest. Identification of high-flux periods or characterization of the temporal variation could also be done entirely manually by looking at the timeseries, but this method is very sensitive to biases caused by our own thinking, especially in large datasets like ours. For these reasons, high-flux periods were determined numerically. Temporal variation in our data set is dominated by daily and seasonal variation instead of hourly variation, and very short-lived peaks are not typical. High-flux periods identified on daily resolution are enough to capture the main characteristics of temporal variation seen at the study site. 

L209-211: I must admit that I do not understand how and why the ‘N2O flux of each chamber can be explained by flux of one other chamber’ within a multiple linear regression. Could you maybe include some visual output in the supplemental material to makes this easier understood? And why do you want this? What does this correlation highlight?

• 16.    Response: We admit that the explanation about the multiple linear regression test was not clear enough and will add a clearer explanation about the use of linear regression. We will also add model results in the supplements for interested readers. The idea was that the correlations between fluxes of different chambers tell about the similarity of temporal variation in different chambers, a high correlation between chambers meaning that temporal variation is similar despite the possibly different range in flux variation between part of the chambers. This was done to strengthen the visual interpretation made about the similarities in temporal patterns of the fluxes across chambers with very different ranges of N2O flux variation. Linear regression between chamber pairs does the same as correlation if only one explanatory variable, like flux from other chambers, is used. By adding also environmental variables in the regression as explanatory variables, we can test if some of the environmental variables can explain temporal variation of flux (response variable) that was not explained by the other chamber (explanatory variable). This was done to provide additional information about the reasons explaining differences in temporal patterns (meaning poor correlation of fluxes) between some chambers. What we did was we used linear regression, where the N2O flux of one chamber (for example Chamber 1) was a response variable and N2O flux of one other chamber (in the case of Chamber 1: Chamber 2, 3, 4, 5 and 6 tested in separate models) together with environmental variables (soil moistures, soil surface temperature, WTL) were as explanatory variables. This was repeated until the explanatory power of all other chambers to explain fluxes in other chambers was tested (in total 15 different chamber pairs, 15 models). Second-order polynomial effects in environmental variables were considered as well. Then, the ability of different environmental variables to explain variation in the models was checked. For example, if fluxes in Chamber x cannot be explained very well by fluxes in Chamber y, but soil temperature is able to explain the variation left unexplained by Chamber y. In this case, we can conclude that differences in the way fluxes respond to soil temperature might explain differences in the temporal variation between Chambers x and y. Correlation or linear regression is not an essential part of the study and could also be left out. We will also consider this option.

L243: I have a general concern for the flux resolution you are using. You have such great data of 24 measurements per day and chamber. Why do you bring it to the daily resolution first before running the models to see which environmental parameters are driving the fluxes? You might get incredible insight from the short-lived emissions that happen besides freeze-thaw or dry-wetting events. Is there a time during the day where the fluxes are higher? Is, for example, in the morning of very dry days the dew a factor? If you want to figure out the environmental drivers, why exclude the diurnal dynamic in your data. You certainly have the resolution for it. Could you gap-fill the hourly rates? And the lag might be even less than a day worth too.

• 17.    Response: We thank you for the critical comment regarding the choice of temporal resolution. Please see the response number  10. 

L250: How did you analyze the prediction accuracy based on the R2 and RMSE? Are these parameters just calculated and stated or did you select the model based on the best R2 and lowest RMSE? Please a little more detail here.

• 18.    Response: We did not perform model selection and we will add a mention about this in the article. R2 and RMSE were only used to evaluate model performance on OOB data, evaluation data within the training period and outside the training period (See lines 248-252). R2 and RMSE allowed us to evaluate model performance outside training data, which is an important part of modeling. R2 and RMSE are commonly used metrics in model evaluation and therefore chosen.

L254-255: ‘…to visualize the response …, by illustrating how the predicted N2O flux values deviate from the mean predicted flux (ALE = 0) along the gradients of the measured environmental parameters’ You describe it in the Figure 9 caption but not in the material and methods. Not everyone has the same knowledge than you. Please describe briefly or just copy from your caption.

• 19.    Response: We will add more information about ALE and its interpretation in the methods section. 

L269-270: Is the code somewhere published? And the data made available?

• 20.    Response: Yes, data is made available. You can find the data here (Rautakoski et al., 2023a and 2023b):  https://doi.org/ 10.5281/zenodo.8142188 and https://doi.org/ 10.5281/zenodo.8141569. See lines 698-670 in the article. I’ll be happy to share the code with any interested readers. I’ll also consider publishing the code openly.

L276-280: Here you describe the temperature variation and refer to Figure 2. Yet in Figure 2, you show anomalies only, nothing about this in the text.

• 21.    Response: We will change the position of the figure reference and put it after the next sentence. The sentence starting in the end of line 276 is about the seasonal weather anomalies and it is a better position for the figure reference.   

L288-290: This sentence is very confusing, consider rearranging it. E.g. Soil moisture at 7cm and 20 cm were continuously lower in summer 2018 until 2019 (xx and yy respectively) compared to the entire measurement period mean (xx and yy respectively). Please consider this for the entire results, please always keep one statement with their numbers clearly separated and then compare them to something else. Next sentence as well, WTL was deeper in the summer and autumn 2015 (xx) than the mean of the study period (yy). I am getting lost in every sentence to which statement the value belongs to.

• 22.    Response: We appreciate your comments related to the readability of the text. We will make the sentences including references to multiple chambers, environmental variables or years clear. We will also consider removing some of the details from the results section to make it easier to read and to highlight the main findings. 

L296: Again, I cannot follow here what you want to say about the snow cover periods, reaching the maximum in 2018-2019 and lowest in 2016-2017? What does this mean about the ‘all winters had a period or periods of snow cover with the maximum measured… Or do you just mean ‘All winters had a period with permanent snow cover. The maximum snow cover was measured…’

• 23.    Response: Yes, you understood correctly. We will split the sentence into two and make it more readable.  

L310: Figure 3: Very nice graph. The water table varies a fair bit spatially but does not really go closer to the surface than -20cm, while the 20cm soil moisture is not really picking up that dynamic. I would suggest including precipitation in this graph as well, to give more insight into the 7cm soil moisture dynamic, since there are no replicates for soil moisture, right? Overall I think this experimental setup (if continuing) would greatly benefit from more 7cm soil moisture replicates (i.e. per chamber), since N2O is highly correlated to the small-scale moisture dynamic!

• 24.    Response: I strongly agree that having more soil moisture measurements and having them closer to chambers would be important. We were very unlucky with the soil moisture sensors installed near the chambers, which were not working as they should have been. We will also add precipitation to the figure. 

L334: I do not understand why flux time series correlations where done on chamber pairs. What is this information supposed to tell me? Their dynamic relates to one another? Is this something new?

• 25.    Response: Please see the response nr. 16.

L340: Figure 4: Very nice. If you like, you could make the figure even more powerful by highlighting the high-flux periods in another color.

• 26.    Response: Thank you. Yes, that would be nice. We will try that and update the figure in case the change doesn’t make the figure too messy. 

L390: Figure 6: Again, very nice and you could improve it by extending the ‘soil surface <0 °C’ shaded area to the other two plots in the graph, to make the N2O and moisture responses even easier to follow.

• 27.    Response: We will change the soil moisture & WTL panel (now panel c) above the temperature panel (now panel b) to make it easier to compare the moisture-N2O dynamics. We will highlight the “soil surface <0 °C” periods also in the soil moisture & WTL panel so that the timing of freezing periods can still be easily compared to N2O flux dynamics.

L400: Figure 7: Not sure if I read this figure right, high-flux period density in different environmental conditions. How does the baseline period fit into this? Do you mean the density of high N2O fluxes, which are happening in the no-high-flux periods? Do you want to compare those? There is nothing in the text about it. Maybe add again to the caption that the scaling of 0-1 was done to compare the chambers. Not sure what I should take away from this graph. There is not much in the text other than referring to the temperature plot. Could you add an introducing sentence when bringing in figure 7 on what you want to compare/highlight here, with a take-away message that you want to discuss later?

• 28.    Response: We will modify the caption to help the reader with the interpretation of the figure and discuss the important messages of the plot in the results section. The aim of Figure 7 is to visualize in which kind of condition the identified high-flux periods started and compare those to the conditions during baseline periods. The density is a way to visualize distribution. Histograms visualize the number of observations in each range of x-axis values, while density plots show the proportion of observations on different values of x-axis. In this case, the absolute proportions are scaled between 0 and 1 to simplify the interpretation of y-axis values between seasons and variables. The high-flux period density 1 points to the conditions when the highest proportion of high-flux periods started in each season for each environmental variable. For example in spring, most high-flux periods started when soil surface temperature was close to 0°C, soil moisture at 7 cm about 0.2 m-3 m-3, soil moisture at 20 cm about 0.62 m-3 m-3 and WTL at about 55 cm depth. The idea with the baseline period densities is to show what kind of conditions there are in each season outside high-flux periods. If the conditions at high-flux period start are very distinct compared to the overall variation in conditions outside high-flux periods (See for example Fig 7b for summer), we could assume that only certain conditions trigger high-flux periods in that season. If the conditions triggering high-flux periods are the same as the conditions outside high-flux periods, there must be some other cause for the high-flux periods and the variable in question is not able to explain why high-flux periods start (See for example Fig 7d for winter). We will clarify this in the revised manuscript. 

L 405: Could you please explain VI scores. Because without explanation, this whole part sound to me that the importance of the unlagged variables always scored higher than the lagged one, except precipitation. But in the abstract, you mention that the N2O responses to soil wetting with a lag period. Then it should be rain events rather then soil wetting, since the water table can contribute to wetting in 20 cm as well. Overall, I would be careful of the conclusions you get from the measurements of the weather station, since it is not representative for each chamber individually! Soil moisture can vary significantly on the small scale. This might explain why precipitation is behaving quite ‘smooth’ in your analysis. One would assume that all chambers received the same amount of rain. As sad as it is, I would consider dialing down the take-home message here, since I am not confident in the representation of soil moisture for the site’s soil wetting overall.

• 29.    Response: We will introduce the concept and interpretation of VIs in more detail in the revised manuscript. We also deeply understand the critics related to the soil moisture data and we will be more careful with the conclusions made only from it.
• The ability of VIs to show the importance of a certain lag is affected a) by the fact that lack-times can vary in time (see lines 627-633), b) lag-effects may not be always present (for example, only after soil wetting) and c) the effect depends on the values of other variables (interactions). In addition, the interpretation of VIs of correlated versions of the variable (unlagged and lagged) is difficult and sensitive to false interpretation (See response nr. 10). The fact that lags for soil moisture (7 cm and 20 cm) received increased VI scores (see mean-row in Fig. 8), tells about the overall importance of lags in responses of N2O flux to soil moisture. The increasing VIs of precipitation with increasing lag-time indicates the importance of lags more clearly, which could suggest a stronger link to time after precipitation instead of the soil moisture that is now available. We will discuss this in the revised manuscript.
• The conclusions about the length of lag-times are made from ALE plots that enable comparing mean predicted fluxes between unlagged and different lagged variables for each environmental variable. The lag-times were interpreted from the ALE plots in the following way: Out of the unlagged and lagged versions of a variable, the one for which the model had predicted the highest N2O flux compared to the mean prediction of the model (y=0), was considered to best represent typical lag-time between environmental condition and peak flux. In the case of soil moisture, only soil moisture values higher than 0.3 m-3 m-3 (See lines 425-429) were considered when interpreting lag-dynamics in response to precipitation events on thawed soil since very low soil moisture values on ALE plots tell about the response to freezing. We will include this explanation also in the methods section. See also response nr. 10 for more information about VI. Overall, we will be careful in making conclusions based on soil moisture due to its limitations. 

L409-410: I do not understand this sentence, ‘and increased VI scores also for lagged variables with the mean across lags 0.25.’ So far everything with lag decreased VI scores, so what does the ‘also’ refer to? And actually, in this case too, the VI score decreased from 0.27 to 0.25. Was that just a typo? Maybe write shorter sentences or use commas but I am not able to follow the information in this sentence.

• 30.    Response: We will split the sentences and make the text easier to read. 

L424: I am curious here, the ALE curves highlight that lagged moisture is more predictive when the moisture is already high. Why do you think that is? This is interesting and could be explored more. I haven’t seen it in the discussion being mentioned again.

• 31.    Response: ALE curves can be thought of as response curves with the predicted flux increasing towards the top of the y-axis and values of environmental condition on the x-axis. In our case, we have also added a curve for each lagged variable in the same plot. Interpretation goes like this (See also lines 424-433): If, for example, 7-day lagged moisture variable gets the highest y-axis value on high moisture conditions, this tells that the model predicts high N2O flux when soil moisture was high 7 days ago. If at the same time, unlagged soil moisture gets the highest y-axis values on intermediate soil moisture conditions, we can interpret that the highest flux is predicted when soil moisture is on intermediate level but was high 7 days ago. A high y-axis value for a lagged variable on certain conditions does not tell about the predictive power of that variable on those conditions, although a variable that has a strong response to N2O is likely to have predictive power in the model. VI scores tell about the predictive power of each variable. We hope that once we make the concepts of VI and ALE clear in the methods section and open up the results more clearly in the results section, confusion related to the interpretation of modeling results can be avoided.

L473-474: Maybe consider using other comparisons in the discussion chapter rather than numbers, e.g. the annual budget for three of the chambers was double in 2016 and 2017 compared to 2018. This makes it easier for the reader to follow your line of thought and to pick up your take-away message.

• 32.    Response: We will pay attention to this and change the wording in cases where numbers can be replaced with better options. 

L480: During harvest, were the roots left behind in the soil? The decomposition of finer and medium sized roots might fuel N2O emissions by substrate contribution. But of course, you are right, that doesn’t explain the increase in the control site.

• 33.    Response: Yes, the roots were left in the soil, although relatively few trees were cut close to the chambers (See line 117). The effects of harvesting cannot be completely excluded but the fact that emissions increased similarly also in the control site after harvesting, implies that environmental conditions explain the increased fluxes in post-harvest years 2016-2017 (Korkiakoski et al., 2020).

L490: I agree, you should highlight more your findings of winter fluxes! More than a third of the annual budget was emitted during winter. That is worth stressing a bit more since many annual measurements do not cover this season.

• 34.    Response: Thank you for the constructive feedback. Winter fluxes are an important part of the annual N2O budgets, and we will expand the discussion related to winter fluxes. 

L520-525: You could play around here way more with what you got from your results! These winters with discontinuous snow cover are increasing with processing climate change due to warmer temperatures. This trend will then also most likely increase N2O budgets. Highlight more and in the abstract too! On the other hand, summers are predicted to be drier too, for the same reason of rising temperatures. Would this balance the annual budgets? Have you thought of doing some calculations for this, as a potential outlook for your study site? Discuss more the bigger picture conclusion from your results.

• 35.    Response: Conclusion and abstract now both include a careful sentence about the effects of drying summers and warming winters in annual N2O budgets. We will discuss this issue more in the revised version of the manuscript since this is, indeed, an important conclusion that can be made. We will keep the focus on the spatio-temporal variation of N2O and not go very far with projections related to climate change in this article. 

L528: No need to repeat yourself. Rather use an introductory sentence on the spatial focus now. The next sentence is sufficient.

• 36.    Response: We will remove the repetitive sentence. 

L533: Could you remind me here of how big the difference was with a comparison of magnitude? E.g. on average double/triple the annual budget.

• 37.    Response: We will remind here about the magnitude.

L591: Not even in the hourly resolution?

• 38.    Response: Not at least in the scale that it could be spotted from the data and linked to thawing. When the soil surface started to reach > 0°C temperatures during the daytime in spring, the N2O flux started to increase slowly during the next weeks and months with some diurnal variation. Temporal variation in spring is strongly dominated by day-to-day or week-to-week variation rather than hourly variation. 

L615-632: Please change the focus. Your soil moisture is not fully representative of the chamber conditions and have no replicates! General conclusions here are inappropriate due to the limited amount of soil moisture measurements. However, precipitation shows a quite consistent gradual VI score increase towards the lag time of 7 days. Precipitation is less likely to vary on the micro-scale as much as soil moisture and has therefore, in my opinion, a stronger message here. Please consider revising throughout the manuscript.

• 39.    Response: The lack of chamber-specific or close-to-chambers soil moisture data is an unfortunate limitation in our study. The measured soil moisture is most likely able to describe the general temporal variation in soil moisture relatively well, but we cannot say anything about the absolute level of soil moisture in different chamber locations or inside chambers. We admit that varying peat properties, evapotranspiration patterns and tree cover can create variation also in temporal dynamics of soil moisture, and these dynamics are not captured here. For example, part of the soil may have dried faster after rain due to variations in peat properties, which can affect the lag dynamics seen in response to precipitation events and explain differences in lag-times between chambers. We will be careful with the conclusions made from soil moisture results and if conclusions are made, we will support the findings with WTL and precipitation that better describe conditions at the chamber site. We appreciate your critical comment. 

Appendix B L664: How was the RMSE used to analyze the prediction accuracy? This is all about the R2. A bit more context would be helpful to make the data processing as transparent as possible.

• 40.    Response: RMSEs of the model in different evaluation datasets were compared. RMSE values depend on the values of the data, so comparisons about the model performance between chambers were made using R2. We will clarify this in the methods section. 
  
   

Technical comments:

L38: I would suggest using the GWP with feedback effect accounting for 298 CO2-eq.

• 41.    Response: We will revise this according to the recent literature.

L38-41: A major part of the N2O emissions of N2O originates from soils (Butterbach-Bahl et al., 2013; Davidson and Kanter, 2014) and human impact through altered nitrogen (N) cycle. Land use and climate change affect the soil N2O emissions both in natural and managed ecosystems (Tian et al., 2018, 2020).

• 42.    Response: We will simplify the sentence.

L424: Write ALE out, no abbreviations at the beginning of the sentence, and it is a good reminder here what it means for everyone not too familiar with it.

• 43.    Response: Yes, we will use the full word here. 

L591: Thawing. Ice melts, soil thaws.

• 44.    Response: Thank you, for the reminder.

References: 

Apley, D. W., and Zhu, J.: Visualizing the effects of predictor variables in black box supervised learning models. Journal of the Royal Statistical Society Series B: Statistical Methodology, 82(4), 1059-1086, https://doi.org/10.1111/rssb.12377, 2020

Korkiakoski, M., Ojanen, P., Penttilä, T., Minkkinen, K., Sarkkola, S., Rainne, J., Laurila, T., and Lohila, A.: Impact of partial harvest on CH4 and N2O balances of a drained boreal peatland forest, Agric. For. Meteorol., 295, 108168, https://doi.org/10.1016/j.agrformet.2020.108168, 2020

Rautakoski, H., Korkiakoski, M., Aurela, M., Minkkinen, K., Ojanen, P., and Lohila, A.: 4.5 years of peatland forest N2O flux data data measured using automatic chambers, Zenodo, https://doi.org/ 10.5281/zenodo.8142188, 2023a

Rautakoski, H., Korkiakoski, M., Aurela, M., Minkkinen, K., Ojanen, P., and Lohila, A.: Supplementary material to the article "Exploring temporal and spatial variation of nitrous oxide flux using several years of peatland forest automatic chamber data", Zenodo, https://doi.org/ 10.5281/zenodo.8141569, 2023b

Strobl, C., Boulesteix, A.-L., Zeileis, A., and Hothorn, T.: Bias in random forest variable importance measures: Illustrations, sources and a solution, BMC Bioinform., 8(1), 25, https://doi.org/10.1186/1471-2105-8-25, 2007

Strobl, C., Hothorn, T., & Zeileis, A. (2009). Party on!

Our responses are numbered from 45 to 50. Responses to Referee 1 and 2 are numbered 1-44. 

45.    General response:

We thank you for your comments. We will modify the manuscript to better highlight the important results and include also all the row numbers in the revised version. 

 

Specific comments:

L 190-193: There certainly is a point to not only being interested in the “hottest” hot moments like those that you cite, but perhaps you can provide a sentence or too that succintly describes your method so that it’s easy to cite/use in the future.

• 46.    Response: We thank you for the idea. We will consider this.

L199: However, it does seem that you are more-or-less simply just choosing “days above mean flux” as having high fluxes: “The mean N2O flux of the study period was close to the chosen 70 % percentile threshold in all chambers. Days with the mean flux above the 70 % percentile were classified as high-flux days.” What inherent value is that besides being high?

• 47.    Response: The values of chamber-specific means and 70 % percentiles are shown in table 2. As the use of chamber-specific 70% percentile suggests, the definition of high-flux is here proportional to the range of the temporal variation measured in each chamber. In chambers with small fluxes and small range of flux variation, the days classified as high-flux periods can have fluxes that were considered low in other chambers.

L207,209: Not exhaustive, but N2O is not subscript on these lines. Please doublecheck throughout.

• 48.    Response: We will add the subscript here and doublecheck the text.

L610-626: This may likely be the case, but so far datasets collected can only infer this, could you potentially provide some advice on what measurements need to be taken in the future to further our understanding of this microsite variability. Can we predict this variability?

• 49.    Response: Soil properties likely play a role here since they strongly determine the availability of substrate, soil oxygen conditions and drainage after rainfall. We will add general advice to analyze the soil properties from sites where N2O fluxes are measured. To predict variability in lag-times, more information about the variability of lag-times in different ecosystems and soils is needed. With our data, we are only able to hypothesize possible linkages to environmental conditions and spatially varying soil properties. 

L643-645: While understanding the variability is important (and should be condensed into a less extensive part of the manuscript), I think this is one of the most important takeaways and should be expanded: how do these compare to other forest peatland measurements? Were less extensive measurements still producing similar mean flux values?

• 50.    Response: We will expand the take-away message section in the conclusions and abstract as Reviewer 2 also suggested (See response nr. 35). We will strengthen the conclusion related to N2O budgets in the changing climate and highlight the importance of carefully considering the large variation in N2O budgets between seasons and years when planning measurement campaigns, reporting N2O emissions and modeling N2O fluxes in the current and future climates. We will also mention the importance of analyzing soil properties to better understand the spatial variation of N2O and the interactions between soil and temporal variation of N2O.