An Improved Sap Flow Prediction Model Based on CNN-GRU-BiLSTM and Factor Analysis of Historical Environmental Variables

Round 1

Reviewer 1 Report

The paper by Li et al., titled: “An Improved Sap Flow Prediction Model Based on CNN-GRU-BiLSTM and Factor Analysis of Historical Environmental Variables”, is original work, fitting to the scope of Forests journal.

The authors used network of deep learning models to simulate sap-flow of Agathis australis based on environmental variables (soil water content, atmospheric humidity, incoming radiation, temperature etc.). The different partial networks were compared with the final model and also the different environmental feature methods were compared to achieve most robust final model. The paper significantly expands the ecohydrological knowledge and is interesting for the readership of the Forests journal. It would be interesting to see in future studies by the authors how the model could be used for climate change scenarios predictions.

 

I suggest minor revision of the paper before publishing in Forests.

My comments:

Abstract

I would suggest authors to shorten their abstract and focus on the most relevant findings to attract broader readership.

Introduction

I think authors should mention the most frequently used model for sapflow simulations: Penman-Monteith: https://doi.org/10.1007/s00468-016-1513-3, https://doi.org/10.1061/JIDEDH.IRENG-9887

Line 70: Stone et al. (missing year). Use proper citation throughout manuscript, see authors guidelines for more info.

Line 74: Vapour pressure deficit!

Line 75: All species names in Latin should be in cursive.

Line 83: “Guillén et al investigated the sap flow velocities in different species, which shown that the oaks had lower sap velocity rated than the maples [34].” How is that relevant to your paper? You are working with single species and you are not comparing it to the others.

Line 115: You could mention more recent papers that found the strong correlation of sapflow with environmental parameters such as: https://doi.org/10.3390/w14193015, https://doi.org/10.1007/s10342-023-01549-w

Materials and methods

Line 163: Agathis australis should be in cursive.

Line 165: Full names for VPD, Ta, WS etc.

Results

Figure 7: The y axis names are too small to read, please enlarge them. Moreover, isn’t it a problem that air temperature, relative humidity and vapour pressure deficit are highly correlated and can cause multicollinearity of the model?

Discussion & Conclusion

The deep learning model is parametrized and optimized for the Agathis australis tree species! I think that authors should address this in their paper, the same model would not show the same performance for other species, different age, stand structure etc. and would need to be parametrized from the beginning. The model would be also very site specific and not universally applicable across wider geographic range. In conclusion the very high precision and low error of the model approach has trade-off with generality.

Comments for author File: Comments.pdf

Author Response

Yane Li, Ph.D.

College of Mathematics and Computer Science

Zhejiang A&F University

Hangzhou 311300

P.R. China

Email: [email protected]

Jun 5, 2023

 

 

Re: Manuscript ID forests- 2368182

 

 

 

Dear Editor,

 

Thank you very much for your letter dated May 27, 2023. We would like to thank the reviewers for their helpful comments and suggestions regarding our manuscript entitled “An Improved Sap Flow Prediction Model Based on CNN-GRU-BiLSTM and Factor Analysis of Historical Environmental Variables”. We have addressed the issues raised by the reviewers point-by-point and revised our manuscript accordingly. Please find our responses to the reviewers’ specific comments and suggestions below.

 

We hope our manuscript is now suitable for publication in Forests. Thank you very much for your time and consideration. We look forward to hearing from you soon.

 

 

Yours sincerely,

 

 

Yane Li, Ph.D.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Response to Reviewer 1 Comments

 

Point 1: Abstract. I would suggest authors to shorten their abstract and focus on the most relevant findings to attract broader readership.

Response 1: Thanks for the reviewer’s valuable comment. We have shorten the abstract and focus on the most relevant findings in the revised manuscript.

 

Thanks again.

 

Point 2: Introduction. I think authors should mention the most frequently used model for sapflow simulations: Penman-Monteith: https://doi.org/10.1007/s00468-016-1513-3, https://doi.org/10.1061/JIDEDH.IRENG-9887

Response 2: Thanks for the reviewer’s helpful comment. We have added researches related to Penman-Monteith Equation in section of “Introduction” in the revised manuscript.

Details are as follows:

Evapotranspiration can be described as a process consisting of a pair of adiabatic cooling and diabatic heating which leads to the energy state of the ambient air to chage in easily quantifiable ways. The Penman-Monteith equation is used as a standard method for modeling evapotranspiration by the United Nations Food & Agriculture Organization (FAO) [49], which can estimate the rates of heat and vapor transfers from less-than-saturated surfaces [50]. The Penman-Monteith equation requires daily mean temperature, wind speed, relative humidity, and solar radiation to predict net evapotranspiration [49]. Zerihun et al derivated the Penman-Monteith equation with the thermodynamic approach, which shown two steps of evapotranspiration, incluing evaporation from a wet source/sink surface into a quiescent ambient air, and the dynamic influence of wind-surface interaction and canopy system on evaporation [51]. Jiří et al presented an improved approach for directly parameterizing the Penman–Monteith equation for calculating the diurnal variation process of stand canopy conductance (gc). This method was used to calculate gc and establish a sap flow model, which can describe canopy conductance and stand sap flow with subhourly resolution under both day and night conditions [52].

49.  Lennox Raphael Eyvindr, Penman–Monteith Equation, Plor, 2011, 978-613-7-82645-4
50.  Penman-Monteith Equation. In: Gliński, J., Horabik, J., Lipiec, J. (eds) Encyclopedia of Agrophysics. Encyclopedia of Earth Sciences Series. Springer, Dordrecht. 2011. DOI: 10.1007/978-90-481-3585-1_758.
51.  D. Zerihun, C. A. Sanchez, A. N. French. Derivation of the Penman–Monteith Equation with the Thermodynamic Approach. I: A Review and Theoretical Development [J]. Journal of Irrigation and Drainage Engineering, 2023,149 (5). DOI: 10.1061/JIDEDH.IRENG-9887.
52.  Jiří Kučera, Brito Patricia, Jiménez, María Soledad, Urban Josef. Direct Penman–Monteith parameterization for estimating stomatal conductance and modeling sap flow[J]. Trees, 2017, 31(3):873-885. DOI:10.1007/s00468-016-1513-3.

Thank you very much.

 

Point 3: Line 70: Stone et al. (missing year). Use proper citation throughout manuscript, see authors guidelines for more info.

Response 3: Thanks for the reviewer’s valuable comment. We have revised the citation throughout manuscript according to the authors guidelines.

 

Thanks again.

 

Point  4: Line 74: Vapour pressure deficit!

Response 4: Thanks for the reviewer’s helpful comment. We have corrected “vapor pressure” in original manuscript to “vapour pressure deficit” in the revised manuscript.

 

Thank you very much.

 

Point 5: Line 75: All species names in Latin should be in cursive.

Response 5: Thanks for the reviewer’s valuable comment. We have corrected all species names in Latin in the revised manuscript.

Thanks again.

Point 6: Line 83: “Guillén et al investigated the sap flow velocities in different species, which shown that the oaks had lower sap velocity rated than the maples [34].” How is that relevant to your paper? You are working with single species and you are not comparing it to the others.

Response 6: Thanks for the reviewer’s helpful comment. We reread this paper which shown sap flow is related to tree species, it really not relevant to our paper. As a result, we have deleted this reference in the revised manuscript.

 

Thanks again.

 

Point 7: Line 115: You could mention more recent papers that found the strong correlation of sapflow with environmental parameters such as: https://doi.org/10.3390/w14193015, https://doi.org/10.1007/s10342-023-01549-w

Response 7: Thanks for the reviewer’s valuable comment. We have added more references which strong correlation of sapflow with environmental parameters in section of “Introduction” in the revised manuscript.

Details are as follows:

  Petrík et al analyzed the effects of environmental conditions and seasonality on the allocation of transpiration and evapotranspiration in pure European beech forest ecosystems, showing that air temperature is the main environmental factor affecting the dynamics of daily and monthly transpiration and evapotranspiration [46]. In addition, results of this study shown that the response of mature European beech transpiration to soil water content is non-significant and the response to VPD is linear. [46]. Dukat et al investigated the effects of drought on forest function by analysing ecosystem evaporation under normal and dry conditions, and identified key drivers of these processes in Pinus sylvestris [47].

46. 4Petrík, P.; Zavadilová, I.; Šigut, L.; Kowalska, N.; Petek-Petrik, A.; Szatniewska, J.; Jocher, G.; Pavelka, M. Impact of Environmental Conditions and Seasonality on Ecosystem Transpiration and Evapotranspiration Partitioning (T/ET Ratio) of Pure European Beech Forest. Water.2022, 14, 3015.DOI:10.3390/w14193015
47. 4Dukat, P., Ziemblińska, K., Räsänen, M. et al. Scots pine responses to drought investigated with eddy covariance and sap flow methods. Eur J Forest Res.2023,142, 671-690.DOI:10.1007/s10342-023-01549-w.

 

Thanks again.

 

Point 8: Materials and methods

Line 163: Agathis australis should be in cursive.

Line 165: Full names for VPD, Ta, WS etc.

Response 8 : Thanks for the reviewer’s helpful comment. We have corrected “Agathis australis” in cursive in revised manuscript. In addition, we also added full names for VPD, Ta, WS etc in revised manuscript. As shown in the section of “2.1.1 Dataset” in “Materials and Methods”.

 

Thanks again.

 

Point 9: Results. Figure 7: The y axis names are too small to read, please enlarge them. Moreover, isn’t it a problem that air temperature, relative humidity and vapour pressure deficit are highly correlated and can cause multicollinearity of the model?

Response 9: Thanks for the reviewer’s valuable comment. We have enlarged y axis names for figure 7 in the revised manuscript.

After original data processed by normalization, nine environment variables were analyzed and composed by method of factor analysis, a statistical technique to extract common factors from variables. In the case of no loss or less loss of the original data information as far as possible, nine environment factors are aggregated into a few independent common factors, which can reflect the main information of the original nine environmental factors. While reducting the number of variables, it also reflects the internal relationship between variables. The prerequisite for using factor analysis method is to satisfy Kaiser-Meyer-Olkin (KMO) test, which is an index to compare simple and partial correlation between variables. KMO values range from 0 to 1. When the sum of squares of correlation coefficients among environmental variables is larger than the sum of squares of partial correlation coefficients, the larger the KMO value is, indicating the the stronger the correlation between variables is, and the more suitable for factor analysis of original variables. In this study, the KMO value is 0.6896, greater than 0.6, indicating there is multi-collinearity of environment factors, such as air temperature, relative humidity and vapour pressure deficit. As a result, it is reasonable to adopt method of factor analysis for these nine environment factors. Multiple variables with strong collinearity are classified as potential factors. In addition, some environment factors such as Ta, RH and VPD are all important to sap flow, so factor analysis is adopted to eliminate the collinearity of each variable by variance rotation technology to construct potential factors and improvethe explanatory ability of factors.

These explanations have been added to section of “3.2 Dimensionality reduction method based on factory analysis” in “Results and Analysis” in the revised manuscript.

 

Thank you very much.

 

Point 10: Discussion & Conclusion

The deep learning model is parametrized and optimized for the Agathis australis tree species! I think that authors should address this in their paper, the same model would not show the same performance for other species, different age, stand structure etc. and would need to be parametrized from the beginning. The model would be also very site specific and not universally applicable across wider geographic range. In conclusion the very high precision and low error of the model approach has trade-off with generality.

Response 10 : Thanks for the reviewer’s valuable comment. Yes, This model is trained by using the environmental factors of tree under the condition that species, age, stand structure and other important parameters affecting sap flow are fixed. The idea of this study can be used as a reference to construct sap flow prediction models for trees of different species, ages and geographical locations. In addition, universality of this model we established in this paper still need to be further studied.

We have added it in the revised manuscript, as shown in section of “Conclusion” in the revised manuscript.

 

Thanks again.

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript is novel in the use of factorial models and neural networks for the prediction of sap flow, traditional methods tend to be complicated and with an implicit bias that can be minimized by methods with those proposed. Although the idea has great potential, However, it has many errors, gaps and doubts to be published, at times it feels more focused only on the computational part and forestry has a secondary role (rare if a forestry journal is presented), apart from that it does not seem like an ecophysiologist or a person Use this data from your vision and connect what is proposed with reality. I attach my general observations since reading all the documents in their current state was exhausting

1 . Abstract is extremely long, at some points repetitive, normally a summary should not be longer than 300 words (the WoS magazines I have published put 250 and even 150), it is a point that must be simplified and be more punctual.

2. The introduction requires more depth, although I see it is very focused on the proposed techniques, I do not see an adequate depth of what sap flow modeling implies, there is classic literature as McDowell, Allen, Granier, studies with Eucalyptus, Pinus, Populus, among others that is not reflected and gives me an idea that they focused only on the method but not on how its use improves modeling more in a context of climate change. This point is a critical weakness and as an ecophysiologist it raises doubts.

3. Methodology, again I feel the forest part was suppressed and with many gaps, such as the geographic location of the sites that collect SAPFLUXNET data, they mention the number of measurements but I do not clearly see a table or graph that shows me the variation of the climatic variables in the study time, nor a characterization with means and standard error of the study trees, if you see a study of this type, they are key factors to model sap flow, it is still not clear to me, how they estimated the sapwood, they used an equation, which Was it the selection criteria?

3. I recommend a restructuring of the methodology (see subtitles), a strange shape is cut, first I see dataset (it should be edited), then they say method (of what?) and they mention them more as references (without references) and it does not give a clear idea as It can be reproduced in future works, they need to be clearer.

4. The results are strange again, why not start saying the environmental variables to focus on (usually the VPD and PAR are better to model sap flow), then talk about the best model and later there should be a valuation stage which I did not find and generates Doubts about how I can reproduce your results. Another aspect was to see aspects of discussion in results (Example lines 521 to 528). This is rare and makes me doubt your weight.
5. The discussion is not clear, it is not connected to the results and I do not see a clear justification for what was obtained. I would expect to see a point that they talk about considerations, points for improvement or problems of using the proposed methods or how their use improves the modeling?

6. many acronyms are used, it would be good at the beginning of the article to make a table where all acronyms are listed so that the reader will enter the context quickly

 

 

The article is deficient, difficult to read and at times it seems like a thesis that was adapted without due care. Personally, this work is not publishable since they focused only on computational and forestry and the use of these data was secondary.

Author Response

Yane Li, Ph.D.

College of Mathematics and Computer Science

Zhejiang A&F University

Hangzhou 311300

P.R. China

Email: [email protected]

Jun 5, 2023

 

 

Re: Manuscript ID forests- 2368182

 

 

 

Dear Editor,

 

Thank you very much for your letter dated May 27, 2023. We would like to thank the reviewers for their helpful comments and suggestions regarding our manuscript entitled “An Improved Sap Flow Prediction Model Based on CNN-GRU-BiLSTM and Factor Analysis of Historical Environmental Variables”. We have addressed the issues raised by the reviewers point-by-point and revised our manuscript accordingly. Please find our responses to the reviewers’ specific comments and suggestions below.

 

We hope our manuscript is now suitable for publication in Forests. Thank you very much for your time and consideration. We look forward to hearing from you soon.

 

 

Yours sincerely,

 

 

Yane Li, Ph.D.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Response to Reviewer 2 Comments

 

Point 1: Abstract is extremely long, at some points repetitive, normally a summary should not be longer than 300 words (the WoS magazines I have published put 250 and even 150), it is a point that must be simplified and be more punctual.

Response 1: Thanks for the reviewer’s valuable comment. We have shorten the abstract and focus on the most relevant findings in the revised manuscript.

 

Thanks again.

 

Point 2: The introduction requires more depth, although I see it is very focused on the proposed techniques, I do not see an adequate depth of what sap flow modeling implies, there is classic literature as McDowell, Allen, Granier, studies with Eucalyptus, Pinus, Populus, among others that is not reflected and gives me an idea that they focused only on the method but not on how its use improves modeling more in a context of climate change. This point is a critical weakness and as an ecophysiologist it raises doubts.

Response 2: Thanks for the reviewer’s helpful comment. We have added more references which strong correlation of sapflow with environmental parameters in section of “Introduction” in the revised manuscript.

Details are as follows:

“Petrík et al analyzed the effects of environmental conditions and seasonality on the allocation of transpiration and evapotranspiration in pure European beech forest ecosystems, showing that air temperature is the main environmental factor affecting the dynamics of daily and monthly transpiration and evapotranspiration [46]. In addition, results of this study shown that the response of mature European beech transpiration to soil water content is non-significant and the response to VPD is linear. [46]. Dukat et al investigated the effects of drought on forest function by analysing ecosystem evaporation under normal and dry conditions, and identified key drivers of these processes in Pinus sylvestris [47].”

46.  Petrík, P.; Zavadilová, I.; Šigut, L.; Kowalska, N.; Petek-Petrik, A.; Szatniewska, J.; Jocher, G.; Pavelka, M. Impact of Environmental Conditions and Seasonality on Ecosystem Transpiration and Evapotranspiration Partitioning (T/ET Ratio) of Pure European Beech Forest. Water. 2022, 14, 3015. DOI: 10.3390/w14193015.
47.  Dukat, P., Ziemblińska, K., Räsänen, M. et al. Scots pine responses to drought investigated with eddy covariance and sap flow methods. Eur J Forest Res. 2023, 142, 671–690. DOI:10.1007/s10342-023-01549-w.

“Evapotranspiration can be described as a process consisting of a pair of adiabatic cooling and diabatic heating which leads to the energy state of the ambient air to chage in easily quantifiable ways. The Penman-Monteith equation is used as a standard method for modeling evapotranspiration by the United Nations Food & Agriculture Organization (FAO) [49], which can estimate the rates of heat and vapor transfers from less-than-saturated surfaces [50]. The Penman-Monteith equation requires daily mean temperature, wind speed, relative humidity, and solar radiation to predict net evapotranspiration [49]. Zerihun et al derivated the Penman-Monteith equation with the thermodynamic approach, which shown two steps of evapotranspiration, incluing evaporation from a wet source/sink surface into a quiescent ambient air, and the dynamic influence of wind-surface interaction and canopy system on evaporation [51]. Jiří et al presented an improved approach for directly parameterizing the Penman–Monteith equation for calculating the diurnal variation process of stand canopy conductance (gc). This method was used to calculate gc and establish a sap flow model, which can describe canopy conductance and stand sap flow with subhourly resolution under both day and night conditions [52].”

49. 49.  Lennox Raphael Eyvindr, Penman-Monteith Equation, Plor, 2011, 978-613-7-82645-4
50. 50.  Penman-Monteith Equation. In: Gliński, J., Horabik, J., Lipiec, J. (eds) Encyclopedia of Agrophysics. Encyclopedia of Earth Sciences Series. Springer, Dordrecht. 2011. DOI: 10.1007/978-90-481-3585-1_758.
51. 51.  D. Zerihun, C. A. Sanchez, A. N. French. Derivation of the Penman–Monteith Equation with the Thermodynamic Approach. I: A Review and Theoretical Development [J]. Journal of Irrigation and Drainage Engineering, 2023,149 (5). DOI: 10.1061/JIDEDH.IRENG-9887.
52. 52.  Jiří Kučera, Brito Patricia, Jiménez, María Soledad, Urban Josef. Direct Penman–Monteith parameterization for estimating stomatal conductance and modeling sap flow[J]. Trees, 2017, 31(3):873-885. DOI:10.1007/s00468-016-1513-3.

Thanks again!

 

In the revised manuscript, we think the section of”Introduction” more reasonable and  thoughtful than manuscript before modification. In section of “Introduction” we first described the important of accurately prediction sap flow and explained advantages and disadvantages of using sensor tools to measure sap flow directly in the first paragraph of “Introduction”. Then, through literature review, it is clarified that environment is significantly correlated with sap flow, and provide rationality to use environmental factors to predict sap flow in the second paragraph. Next, the Penman-Monteith Equation, the standard method for modeling evapotranspiration by the United Nations Food & Agriculture Organization (FAO) related studies was described in the third paragraph. In the fourth paragraph, we described studies of sap flow model established with environmental factors by using traditional machine learning methods. In the fifth paragraph, studies for time series prediction model built with the deep learning based method.

 

Thank you very much.

 

Point 3: Methodology, again I feel the forest part was suppressed and with many gaps, such as the geographic location of the sites that collect SAPFLUXNET data, they mention the number of measurements but I do not clearly see a table or graph that shows me the variation of the climatic variables in the study time, nor a characterization with means and standard error of the study trees, if you see a study of this type, they are key factors to model sap flow, it is still not clear to me, how they estimated the sapwood, they used an equation, which Was it the selection criteria?

Response 3: Thanks for the reviewer’s valuable comment. We have added graphes of the variation of the climatic variables in the revised manuscript, as shwon in Figure 1.

In this paper we used environment factors to predict sap flow value with 17569 observation records for one tree which under the condition that species, age and other important parameters of tree itself affecting sap flow are fixed. The idea of this study can be used as a reference to construct sap flow prediction models for trees of different species, ages and geographical locations. In addition, universality of this model we established in this paper still need to be further studied.

 

Thank you very much.

 

Point 4: I recommend a restructuring of the methodology (see subtitles), a strange shape is cut, first I see dataset (it should be edited), then they say method (of what?) and they mention them more as references (without references) and it does not give a clear idea as It can be reproduced in future works, they need to be clearer.

Response 4: Thanks for the reviewer’s valuable comment. We have restructured section of “Materials and Methods” in ths revised manuscript. Dimension reduction was adjusted to the section of “Results”.

After section of “Introduction”, we described “Materials and Methods” which including two parts of materials in section of 2.1 and methods in section of 2.2.

For part of materials, we first described dataset in section of “2.1.1 dataset” , including data source, raw data and details of tree. Then, it is necessary and important to process the raw dataset with normalization. Thus, we introduced the normalization method in details in section of 2.1.2.

For part of method, including typical deep learning network algorithm introduction, model construction method and model evaluation method in section of 2.2.1, 2.2.2 and 2.2.3 respectively.

 

Thanks again.

 

Point 5: The results are strange again, why not start saying the environmental variables to focus on (usually the VPD and PAR are better to model sap flow), then talk about the best model and later there should be a valuation stage which I did not find and generates Doubts about how I can reproduce your results. Another aspect was to see aspects of discussion in results (Example lines 521 to 528). This is rare and makes me doubt your weight.

Response 5: Thanks for the reviewer’s valuable comment. 

Reasons of not start saying the environmental variables to focus on (usually the VPD and PAR are better to model sap flow) are follows. First, there are different environmental factors that affect sap flow for different tree species. The main environmental factors affecting the species we used in this paper are unclear. Second, many environmental factors have a greater or lesser effect on the sap flow. In order not to lose information to build higher performance model, nine environmental factors what we can get from this dataset are adopted. Third, due to collinearity among some environmental factors, such as air temperature, relative humidity and vapour pressure deficit are highly correlated and can cause collinearity, may influence model performance, dimension reduction method of factor analysis was carried out to grouped nine environmental factors to independent implicit factors, which each implicit factors are grouped by nine environmental factors with different weights. The important implicit factors contain most information of this dataset were selected and used to build sap flow prediction model.

After implicit factors selected, ten sap flow prediction models were built with ten different algorithms of CNN-GRU-BiLSTM, multiple linear regression, support vector regression, random forest, LSTM, GRU, BiLSTM, CNN-GRU, CNN-BiLSTM and CNN-GRU-LSTM respectively. Performance evaluation indices including the mean absolute error (MAE), mean squared error (MSE), mean absolute percentage error (MAPE) and coefficient of determination (R2) were used to evaluate and compare models. In order to see the effect of the model more directly, we visualized the predicted and measured values of the test set, as shown in Figure 7, as shown in the revised manuscript. This is also the main content of this article.

In addition, results of models are reproducible. With the same data processed method and the same establish model method we described in this paper can reproduce the results we reported. Further more, the data we used and model code we studied are fully available to readers who need them.

In order to further analyze whether the dimensionality reduction method which is one part of building the model is reasonable, different dimensionality reduction methods of including principal component analysis (PCA), singular value decomposition (SVD) as well as all factors are used. Then different models were established with CNN-GRU-BiLSTM network.

 

Thanks again.

 

Point 6: The discussion is not clear, it is not connected to the results and I do not see a clear justification for what was obtained. I would expect to see a point that they talk about considerations, points for improvement or problems of using the proposed methods or how their use improves the modeling?

Response 6: Thanks for the reviewer’s valuable comment. Discussion all comes from the experimental results of this paper. All conclusions are objectively illustrated by data of experimental results.

Parts of “First”, “Second” and “Third” are to discuss performance of models estabilished with different types of algorithms including traditional machine learning methods such as multiple linear regression, support vector regression and random forest used in this paper, single network of deep learning method such as LSTM, GRU and BiLSTM we adopted in this paper, integrating two basic networks of deep learning method such as CNN-GRU and CNN-BiLSTM,  integrating three basic networks of deep learning method such as CNN-GRU-BiLSTM and CNN-GRU-LSTM. In addition, we discussed performance of models built with network including BiLSTM.

Part of “Fourth” discussed performance of different models with different dimensionality reduction method which is one part of building the model. These models are built with algorithm of CNN-GRU-BiLSTM.

Part of “Fifth” discussed specific details of factor analysis results, in order to explain why this method is suitable for sap flow prediction model.

 

Thank you very much.

 

Point 7: many acronyms are used, it would be good at the beginning of the article to make a table where all acronyms are listed so that the reader will enter the context quickly

 Response 7: Thanks for the reviewer’s valuable comment. We have added an abbreviated list in section of “List of abbreviations and full names used in this paper.

“ in the revised manuscript, as shown in section of “List of abbreviations and full names used in this paper”.

 

Thanks again.

 

Point 8: Comments on the Quality of English Language

The article is deficient, difficult to read and at times it seems like a thesis that was adapted without due care. Personally, this work is not publishable since they focused only on computational and forestry and the use of these data was secondary.

Response 8: Thanks for the reviewer’s valuable comment. For the revised manuscript, We have made a comprehensive and detailed revision of the paper, as shown in the revised manuscript.

 

Thank you very much.

 

Author Response File: Author Response.docx