|I thank the authors for correcting the previous manuscript according to my comments. The procedure of the experiment becomes much clearer, and generally, I agree with the interpretation of the results. However, now I found some contradictions in the results and discussion. Authors need to carefully explain the results for ET and the estimated parameters related to the water stress. As the other reviewer suggested, the conclusions should be supported by the numerical results.|
I also suggest some improvements for the manuscript so that the readers can easily understand (notified with “Draft amendment”). They are just my recommendations for the improvements of the manuscript, and there may be misunderstandings. Therefore, the authors needed to reconsider the way of writings. I read these sentences, again and again, therefore, the difficulties should be the same for other readers.
The authors need to describe the below results carefully.
1) Improvement of springtime increase
GPP is greatly improved indeed, but improvement for ET is not so clear.
2) b and r^2
- Fig. 2 and S2 show that both b and r^2 are improved for GPP. However, b for ET is not so much improved or sometimes gets worse (especially the biases at FI-Ken and FL_Sod are apparent). I think it is better to show the results which get worse compared to the default using italic letters in Table 6. Then the bias increment for ET at FI-Ken, FL_Sod, and both for ET and GPP at RU-Zot and US-Prr will also be clear.
- r^2 for ET is improved compared to the default, but much lower than r^2 for GPP. These results are also needed to be explained in the result section. The bold letter may be better for, e.g. from 0.9 to 1.1 so that the readers easily recognize the higher values.
- For model validation, b and r^2 may better to be evaluated separately to prevent the overestimate, because sometimes r^2 is high but b gets worth. From this aspect, it may better to cumulate the number of the “good” result e.g. from 0.9 to 1.1 for b and 0.9 to 1.0 for r^2, respectively. Then b and r^2 is better to be explained and discussed for ET and GPP respectively.
3) q, θpwp, θtsp
The effects of these parameters are complicated. Therefore, a clear explanation is needed so that the readers can understand it easily. The optimised θpwp and θtsp for the general condition are smaller than the default; that makes β larger (which means, β became ineffective). The oprimised θ tsp for the dry condition is also smaller than the default but larger than that of general condition and θpwp varies between the sites; that makes βdifferent between the sites. Larger q makes β smaller, whereas smaller q makes β larger (because β is between 0 and 1). In my understanding, q changes An, which control GPP but not affect ET. Is that correct?
4) I understand that the authors used the mean of the site level cost function values for APIS. Then, the parameters of the 50 draw for each IS sampler are the same for all study site? Does Figure 1 show the “global” estimate by using the “mean” of the site level cost function? I think it is better to add these descriptions to P9 Lines 25-27 and Fig. 1 so that the readers can easily understand the procedure.
P1. The title is needed to be reconsidered. For example, "Parameter calibration for stomatal conductance and photosynthesis.."
P1. Lines 6-8: Draft amendment, also please reconsider the description with regarding general comments 1. “This modification enabled the model to correctly reproduce the springtime increase in GPP for conifers throughout the measurements sites used in this study. However, the improvement for ET was limited. The key parameters identified along with this modification were the parameters which control the soil moisture stress function and the overall rate of carbon fixation.”
P1. Lines 8-10: Please reconsider the description, "Overall, … models", concerning general comments 2.
P5. Line21: What is “new”? Is that mean new to the JSBACH model? The explanation for q (one of the key parameters) is also not found in the main text.
P10. Lines 4-11: I do not understand this paragraph. Please reconsider the descriptions. In my understanding, each IS sampler use just one parameter set for spin-up. Then the remaining 49 members use the same spin-up but the parameters are perturbed around the first parameter set. If so, such description may help the readers to understand the spin-up process easily. The initial parameter combination for each IS sampler is also selected randomly from the ranges in Table 2. This information is also needed. I also do not understand the procedure, “We also slightly scale (reduce) the importance weights based on the distance of ..”. Is it the same as procedure 3 or different procedure just for spin-up?
P11. Lines 6-8: I still do not understand “Since we also did not run the model spin-up for .. parameter values”. The authors described that the post procedure for APIS is used for the “correct spin-up” at P10. Lines 4-11. Is not enough? If so, why?
P11. Line 8: Does “the same datasets as APIS” mean, same calibration period (the first five years), same climate forcing data, and the same observation data? Is the initial state also generated by APIS, or same as spin-up for APIS? Please make these settings clear.
P12. Line 5: What are the "acceptable values"? Is that mean the final state of the parameter distribution range of the parameter optimisation? Please make it clear.
P14. Lines 6-8: The former sentence indicates that bud burst is not as critical, whereas the latter sentence indicates that the acclimation parameter dominates the phenology parameters. At first, I thought these sentences contradict, but later I understand the difference of these procedures. I think it is better to describe the different functions of these parameters clearly so that the readers easily understand.
P15. Lines 10-11: "The annual cycles of the Bethy model are more in line with the Ball-Berry variants than those of the Baseline model (see supplements S2 for the yearly cycles of the other models)." I think this is not always true. This result is not used in discussion, so I think it is better to remove this sentence. The authors can rewrite that "The results of other stomatal conductance models are shown in S2".
P15. Lines 15-18: Please reconsider this paragraph according to my general comment 1 and 2.
P16. Table 6: “best values” -> “N over threshold” (please refer general comment 2).
P17. Line 10-16: The authors do not show the results. Therefore, I do not understand the experimental setting and what is validated.
P18. Lines 24-26: “This is mostly a direct result of the normalisation of the cost function that inflates the target distribution and gives too much weight to the initial locations and draws.” I do not understand this sentence. I think the convergence is rather related to the parameter sensitivity to the observations.
P19. Lines 1-5: Draft amendment, also please refer general comment 1.
“This delay is also reflected in transpiration, and consequently in ET at FI-Hyy and FI-Sod to some extent. However, the effect at the other sites is not clear." (see general comment 1). The description for FI-Sod seems too much detailed, and discussion is rather needed to be done for general comment 2. Also, I wonder why ET is not improved greatly compared to GPP, although both the observations are used for the calibration. There may be some possible reasons for the mismatch. 1) interaction in the optimisaion process. For me, it seems that GPP is optimised by soil water parameters, and that affect ET estimate. 2) Parameter estimation bias (e.g. θhum strongly decreased and get to its lower limit). 3) Bias correction by q. In my understanding, GPP bias can be corrected using q, but q does not affect the bias of ET. Considering these issues are important for the study with multi observations and to improve ET. I recommend the authors to run some additional experiment to clarify the issues mentioned above for the discussion (only one site is enough). If it is beyond the scope, please discuss the possible reasons for future study.
P19. Lines 13-16: Please reconsider this paragraph according to my general comment 2.
P20. Lines 1-3: I do not find "the result of the site level estimates of g1”, so I do not understand this sentence. “not only” is needed at Line 1. I also do not understand what does “control” for Wang mean.
P20. Lines 18-28: Please reconsider this paragraph with general comments 3.
P20, Line 20: Not only θtsp but also θpwp is lower.
P21. Lines 3-4: "The parameters affecting the optimisation process the most were consistent for all stomatal conductance formulations." I do not understand this sentence.
P21. Lines 8-11: Please reconsider this paragraph with general comments 3.
- How did the authors evaluate the "importance of q for the Ball-Berry type model"? The same validation of importance in Table 5 may need for Table 7.
- “Overall, both optimisations strongly indicate that boreal forest transpiration is not limited by soil moisture stress under normal conditions.” In the discussion, the authors indicated two reasons (the other is the water retention capabilities of the soil). Also, ET is sometimes underestimated. Therefore, the authors can just indicate the possibility.
P21. Line 13: Please reconsider this paragraph with general comments 2.
P1. Lines 2-3: Draft amendment
“The parameter posterior distributions were generated by the adaptive population importance sampler (APIS), then the optimal values were estimated by a simple stochastic optimisation algorithm”
P1. Lines 3-5: Draft amendment
“Using the in-situ measurements of evapotranspiration (ET) and gross primary production (GPP), we calibrated three model parameter groups (**, **, and **), and identified the key parameters. “
P1. Lines 11-13: Draft amendment
“This optimisation improved the model behaviour, but the changes to the parameter values were significant except for the unified stomatal optimization model (USO). Interestingly, the USO model demonstrated the best performance during this event with only small changes to the parameter values."
P2. Lines 20-22: Draft amendment
"It can be hypothesised that the choice of the stomatal conductance model affects the ecosystem model parameters broadly. Because the stomatal conductance formulations vary in their responses to the different conditions. However, a holistic assessment of the performance of the stomatal conductance models together with other parameters (e.g. photosynthesis parameters) has been missing."
P3. Lines 6-7: Draft amendment
“The APIS algorithm samples the full parameter space (as do MCMC methods) and it can treat a mixture of parameter prior distributions. Therefore, APIS can estimate complicated multidimensional probability distributions.”
P3. Lines 11-14: Draft amendment
“First, we utilise APIS to sample the full parameter space with the different stomatal conductance formulations and to locate different modes of the target parameter distributions (peaks of high probability). Second, using the distributions generated by APIS as the prior distributions, the parameters are optimized using a simple stochastic optimisation method. Finally, we assess the inter-site variability and the robustness of the calibrated parameters together with different stomatal conductance formulations. Optimised parameters for a specific drought is also investigated and compared with the parameters for the general optimisation.”
P3. Line 20: Draft amendment
“The site level half-hourly measurements of eddy covariance (EC)”
P3. Lines 21-24: Draft amendment
“The gap-filled and low quality (based on FLUXNET data quality flags) measurements were masked, and the daily aggregates (usually means) were accepted as part of the calibration process if at least 60% of values between 4:00 and 20:00 (i.e. daytime measurements) for that day were unmasked. The daily aggregated data (ET and GPP) were used for the calibration and the validation, whereas all of the half-hourly data were used as the climate forcing data (as explained in section 2.4).”
P5. Line 32: Draft amendment
“However, coniferous evergreen trees do not shed all of their leaves for winter, and the original phenology model is not suitable for a boreal forest.”
P6. Table 2: The additional parameters for Friend and Kiang model is also needed to be included.
P7. Line 1: There are many “b” in this manuscript: photosynthetic acclimation, additional parameter for Friend and Kiang model, and the slope of the regression line. The authors should change them so that the readers can recognize these parameters are different.
P7. Line 17: Draft amendment
“In the original JSBACH formulation (i.e. the Baseline version),”
P9. Lines 5-8: Draft amendment
“Above i is the elements with each IS sampler (described later). Generally, Eq. (4) cannot be analytically solved, hence it is usually estimated numerically. Commonly this is achieved by one of the many Markov chain Monte Carlo (MCMC) methods, but in this study, we apply the adaptive population importance sampler (APIS) defined by Martino et al. (2015). APIS is a Monte Carlo (MC) method that utilises a population of importance samplers (IS) to jointly estimate the target pdf (p(θ|x)) and the normalising constant (Z(x)) by a deterministic mixture approach (Veach and Guibas, 1995; Owen and Yi, 2000), whereas the MCMC methods do not care about the value of Z. Importance sampling density q(θ) is also introduced in APIS algorism.”
Then P10. Lines 13-15 needed to be removed to here.
P11. Line 6: Draft amendment
“overshadows the calculations” -> I do not understand "overshadow". Appropriate word is needed.
P13. Line 17: I could not find Ball-Berry results in S1.
P13. Table 4: Draft amendment
“Parameter scale reduction ^R (at APIS iteration) and stability δ(threshold number of the iteration) estimates from the Bethy simulations.”
P14. Lines 2-3: Draft amendment
“There is an overall agreement on the values of the most prevalent parameters (see the bold and the italic letters in Table 5 between the models.”
P14. Table 5: The order of the parameters is different from Table 2. I think that the same order is better to be understood. “b” also should be reconsidered.
P15. Lines 26-27: Draft amendment
“The values of the relative humidity parameter θhum, the residual stomatal conductance g0, and fC3 have remained nearly unchanged,”
P15. Lines 28-29: Draft amendment
“Noticeably the USO optimisation only changes the value of θtsp and q, and leaves the rest of the parameters almost untouched.”
P15. Lines 30-33: Draft amendment
“For ET, the Baseline, Ball-Berry, and USO are greatly improved especially at the drought in summer 2006 when compared to more general optimisation, however too much drawdown was found for Bethy. The Baseline, Ball-Berry, Leuning, and to a lesser degree the Friend and Kiang formulations, now suffer from the too low ET values before the actual drought. GPP was greatly improved both for general and dry period optimisatons except for the drawdown for the Baseline and Bethy at the drought in summer 2006. Drawdown for USO is also clear but successfully reproduce the observed drawdown. The GPP of other formulations has remained roughly the same as with the more generally optimised parameter values. Overall, The Bethy model has a too strong drawdown for both ET and GPP during the drought.”
P17. Line 4: “Fig. 4, right”
P18. Line 9: Draft amendment
“reproducing the fluxes for the validation sites with low LAI (i.e. RU-Zot and US-Prr)”
L18. Line 17: Draft amendment
“We optimised the model for individual (calibration) sites as well (not shown).”
P20. Line 1: “The site level estimates of (g0 and) g1 are sensitive not only to”
P3. Figure 4: “Bethy (general)” seems better.