An Improved Forest Structure Data Set for Europe
Round 1
Reviewer 1 Report
I appreciate vey much the effort to complete and update an already successfully published method to use RS for improving and spatialize NFI field data. It could help and fill several gaps for a uniform knowledge of forest resources. Moreover, it could trespass the strong limitations of different methods and standards in data processing at continental level.
The paper could benefit of some minor improvements and coherent discussion of the perspectives from the adoption of this proposal as operational tools for reporting and comparing forest data, namely at least:
- it is necessary to discuss reasons and choice of truncated thresholds for some indicators (volume e.g.)
- an introductory discussion on differences in definitions among NFIs and with the adopted method
- details on Land Cover information for countries with missing data are needed, since they could also explain some over or under estimations
- a better discussion on suggestions to improve the approach and make it operational could be very effective, such as the adoption of a common European forest mask able to include both forests and other wooded lands, extremely important in southern west and east Europe
Author Response
We would like to thank the reviewer for the helpful and supportive review comments. Based on the suggestions we revised the draft and refer to each comment in turn:
Comment: it is necessary to discuss reasons and choice of truncated thresholds for some indicators (volume e.g.)
Response: Thanks, we used truncated values only for display reasons in Figures 2, 3, 4, 7 and 8 and clarified this the figure captions.
Comment: an introductory discussion on differences in definitions among NFIs and with the adopted method
Response: We added additional text on differences in definitions among NFIs and the difficulty of harmonization in the “Discussion” section in L421-428 and “4.1 Comparison with FAO statistics” L469-507 and in section “2.1 National Forest Inventory Data” L127-133.
Comment: details on Land Cover information for countries with missing data are needed, since they could also explain some over or under estimations
Response: We now discuss this topic in the discussion section in L482-504, focusing on France and Spain, two countries where we have inventory data. Additionally, the Appendix covers the potential use of higher-resolution CORINE land cover data as input. We further discuss how different definitions of forest and forest available for wood supply (e.g. land use vs. land cover based, different tree cover thresholds) can lead to different estimates.
Comment: a better discussion on suggestions to improve the approach and make it operational could be very effective, such as the adoption of a common European forest mask able to include both forests and other wooded lands, extremely important in southern west and east Europe
Response: Thanks for this important issue, which we discussed in “4.3 Room for Improvement”.
Reviewer 2 Report
The goal of this paper is to provide an improved version of gridded forest structure data using the methodology provided in “Moreno et al. 2017”. Current paper is providing nice concept in utilizing NFI-plot information for pan-European forest mapping. However, the authors are often referring to former articles and the description of methodology is very general. It would be good to add some methodological details to ensure proper documentation of used methodology.
Authors indicated that data harmonized using method developed in Neumann et al (2016). In that article harmozation is applied for estimating Terrestrial NPP. They use temperature and precipitation to capture important small-scale regional effects such as elevation or topography. I don’t know used models very well, but “European level” reliability of those models is maybe not very widely known. I would appreciate some comments related to those models (at least in discussion) as well as it would be also good to know other variables harmonized ?
Data was aggregated to fixed 8 x 8 km, which is much better than previous geographical coordinate based resolution (0.133 degree ?? unit not dscribed). The mapping probability/resolution should same in entire area, if someone is using data for reporting.
The authors use the MODIS land cover information product and spatial aggregation from 500 m to the 8 km resolution. Value of MODIS data is not very high in land use mapping. I would like to see some discussion about value of Modis data in this context. I would imagine that only NFI plot locations would provide similar reliability of estimates (even without Modis data). At least some kind of justification should be given in “Introduction” and “Discussion”.
Authors use k-Nearest Neighbor-method to fill empty cells in their mapping work. It is definitely good way, but it would be advisable to describe method properly. The parameters used in Euclidean distance are part of research report. The variables used co-variate space should reported (not only given that those can be defined by the user)
Authors has provided excellent result tables related to reliability. Those table could be analyzed detailed way in “Results” section. We can see that in some countries very large differences were identified. The reasons could be discussed and possible sources for unreliability could be identified
There are long methodological descriptions in “Result” sections 3.2-3.4. I would move those under the method section. Also, discussion contains some results, which is not fully good practice.
Discussion part don’t have methodology related comparison to other alternative methodologies and global raster products (eg. Hansen et al.). One such type of paragraph would improve the quality of paper significantly.
As a conclusion: paper has high importance and superior quality field information for this type of exercise. However, there are some structural and methodological gaps which need to be corrected. After major revision, paper would provide solid value for scientific community!
Author Response
We would like to thank the reviewer for the helpful and supportive review comments. Based on the suggestions we revised the draft and refer to each comment in turn:
Comment: However, the authors are often referring to former articles and the description of methodology is very general. It would be good to add some methodological details to ensure proper documentation of used methodology.
Response: The “Materials and Methods” section has been carefully revised. We added methodological details and new sub-sections. The methodological descriptions in the “Results” section were moved to the “Materials and Methods” section.
Comment: Authors indicated that data harmonized using method developed in Neumann et al (2016). In that article harmonization is applied for estimating Terrestrial NPP. They use temperature and precipitation to capture important small-scale regional effects such as elevation or topography. I don’t know the used models very well, but “European level” reliability of those models is maybe not very widely known. I would appreciate some comments related to those models (at least in discussion) as well as it would be also good to know other variables harmonized?
Response: More details regarding the harmonization of the data is now given in “2.1 National Forest Inventory Data”. In the “Discussion” section we now also discuss the differences in definitions among NFIs and the resulting difficulties in harmonizing European forest data.
Comment: Data was aggregated to fixed 8 x 8 km, which is much better than previous geographical coordinate-based resolution (0.133 degree unit not described). The mapping probability/resolution should same in entire area, if someone is using data for reporting.
Response: The missing unit “degree” was now added to the text. The reviewer is correct that the mapping resolution of 8 x 8 km is the same in the entire study area. We emphasized this important improvement throughout the text.
Comment: The authors use the MODIS land cover information product and spatial aggregation from 500 m to the 8 km resolution. Value of MODIS data is not very high in land use mapping. I would like to see some discussion about value of Modis data in this context. I would imagine that only NFI plot locations would provide similar reliability of estimates (even without Modis data). At least some kind of justification should be given in “Introduction” and “Discussion”.
Response: Using the described ‘leave-one-out’ and ‘country-wise’ cross validation we tested which combination of bioregion map, land cover map and number of k-mean clusters provides the best result, i.e. the least bias and error (the approach is described in more detail in Moreno et al, 2017). Our results showed including a land cover map leads to more accurate results compared to not including a land cover map. Landcover is important for capturing differences in forest type and we mention this in L482-507.
Comment: Authors use k-Nearest Neighbor-method to fill empty cells in their mapping work. It is definitely good way, but it would be advisable to describe method properly. The parameters used in Euclidean distance are part of research report. The variables used co-variate space should reported (not only given that those can be defined by the user)
Response: Thanks, we introduced a new section entitled “2.4 K-means clustering and k-Nearest Neighbor gap-filling”. The co-variates are already listed in the section “2.3 Co-Variables for gap-filling” and are listed again in the revised draft to improve the readability.
Comment: Authors has provided excellent result tables related to reliability. Those table could be analyzed detailed way in “Results” section. We can see that in some countries very large differences were identified. The reasons could be discussed and possible sources for unreliability could be identified
Response: The results of the cross-validation and the comparison with FAO statistics is discussed in more detail in the “Discussion” section. More emphasis was given to the possible reasons for the occurring differences.
Comment: There are long methodological descriptions in “Result” sections 3.2-3.4. I would move those under the method section. Also, discussion contains some results, which is not fully good practice.
Response: Thanks, we revised the paper accordingly – the methodological descriptions in the “Results” section was moved to the “Materials and Methods”, and the results in the “Discussion” section have been moved to the “Results”.
Comment: Discussion part don’t have methodology related comparison to other alternative methodologies and global raster products (eg. Hansen et al.). One such type of paragraph would improve the quality of paper significantly.
Response: The Hansen et al. (2010, 2013) papers were included in the discussion by adding some text related to this topic (see L493-497).
Reviewer 3 Report
The manuscript provides a welcome data set, (with open access I understand?), to characterize forests in Europe. Such a dataset would be very useful background information for many studies on the role of forests in climate change mitigation. I have some concerns, however, on the methodology that I hope the authors would address before the manuscript is accepted for publication.
I only list the major issues here as they are crucial to the correct understanding of the authors' results:
- I did not find any mention on the possible use (or lack thereof) of a forest mask to estimate the forest area under each 8 by 8 km grid cell. So are the forest parameters reported for each cell just an example of the kind of forest that appears in this cell, or is some forest inventory data (aggregated for each grid cell?) implied? The latter would require a forest area mask.
- This same issue is also reflected in the significant visual difference between the 8 by 8 km and the 500 by 500 m mean volume and carbon content maps in Figs 3 and 4 vs. Figs 7 and 8. The maps are in very coarse scale, yet the distribution of forest volume is dramatically different between them! So it should be stressed that the dataset is not about forest inventory, but about forest structure type, and that it should not be used for forest inventory or quantitative carbon accounting over any land area. Unless I have misunderstood something?
- Because of this, Figures 10 and 11 are also baffling. The authors conclude that their estimates are unbiased, but in reality it seems that massive overestimation in Romania and Sweden is compensated with a similarly massive underestimation in France and partly in Germany. The authors do tell as much in section 4.1 but they still use their dataset to aggregate national forest inventory volumes that does not seem justified because of such huge national discrepancies.
- One thing that would help readers in assessing how to interpret the dataset correctly would be scattergrams at grid cell level between NFI plots and Forest structure estimates for timber volume. I strongly recommend that the authors present these for Europe as a whole and at least for the most forested countries in the continent,
Author Response
We would like to thank the reviewer for the helpful and supportive review comments. Based on the suggestions we revised the draft and refer to each comment in turn:
Regarding the major issues raised we hope that the following comments can clarify some of the, as we understand, misunderstandings of the methodology and aim of the study.
Comment: I did not find any mention on the possible use (or lack thereof) of a forest mask to estimate the forest area under each 8 by 8 km grid cell. So are the forest parameters reported for each cell just an example of the kind of forest that appears in this cell, or is some forest inventory data (aggregated for each grid cell?) implied? The latter would require a forest area mask.
Response: The gridded datasets describe the average forest conditions within an 8 x 8 km cell, e.g. the mean height of the trees or the mean volume in m³ per hectare. For cells where we had National Inventory data (see Figure 2) these average values directly come from that data. Cells where we did not have NFI data were gap-filled with data from cells with inventory data. A forest area mask was not necessary for the gap-filling, as we used a land cover map to capture differences in forest cover.
Comment: This same issue is also reflected in the significant visual difference between the 8 by 8 km and the 500 by 500 m mean volume and carbon content maps in Figs 3 and 4 vs. Figs 7 and 8. The maps are in very coarse scale, yet the distribution of forest volume is dramatically different between them! So it should be stressed that the dataset is not about forest inventory, but about forest structure type, and that it should not be used for forest inventory or quantitative carbon accounting over any land area. Unless I have misunderstood something?
Response: It is true that we do not produce a gridded forest inventory dataset. We do produce a gridded forest structure dataset. We also agree that looking at these 8 x 8 km gap-filled maps can be misleading as also cells with only little forest cover not commonly associated with forest (e.g. areas in Spain) do have forest structure data. This is why we use land cover data at 500 m resolution to generate a forest area mask and account for differences in forest cover. The creation of the forest area mask is described in section “2.5 Forest area mask”. The forest area mask is also used for converting per hectare values (e.g. m³/ha) to absolute values (e.g. m³). Many land cover types to not contribute any forest area at all (e.g. Grasslands, Savannas, Urban) in our study and consequently the mean values of their corresponding 8 x 8 km cell are irrelevant. This is also the reason why way more gaps are present in Figures 7-9 when compared to Figures 3-6. These maps at the finer 500 m resolution now better reflect the actual forest area in Europe. This data cannot readily and is not intendent to be used as NFI data or for official reporting. This is now also discussed in section “4.3 Room for improvement”. An Appendix was added which also discusses the use of CORINE land cover for creating a forest area mask.
Comment: Because of this, Figures 10 and 11 are also baffling. The authors conclude that their estimates are unbiased, but in reality, it seems that massive overestimation in Romania and Sweden is compensated with a similarly massive underestimation in France and partly in Germany. The authors do tell as much in section 4.1 but they still use their dataset to aggregate national forest inventory volumes that does not seem justified because of such huge national discrepancies.
Response: While there is over- or underestimation for some countries, our results are unbiased at the European scale. This was our objective. We changed the text to make it clearer. In the “Discussion” section we give more emphasis to the possible reasons for over- or underestimation and highlight shortcomings when analyzing particular regions or countries.
Comment: One thing that would help readers in assessing how to interpret the dataset correctly would be scatter grams at grid cell level between NFI plots and Forest structure estimates for timber volume. I strongly recommend that the authors present these for Europe as a whole and at least for the most forested countries in the continent
Response: Although this suggestion seems interesting, a comparison of aggregated NFI plots with the forest structure estimates seems to provide little insight, as they match each other.
Reviewer 4 Report
The paper entitled “An improved Forest Structure Data set for Europe” reflects the development of applied research, the topic is interesting and the manuscript has an approach innovative. However, the methods, results and discussion need to be improved. Thus, major changes are recommended.
Comments
1) Lines 36-37 – If e.g. is used there is no need to use etc.
2) Lines 44-51 – References should be added.
3) Line 115 – Either the names of the countries should be included between brackets or a reference should be made to table 1.
4) Lines 137 – 0.133x0.133 resolution – units should be added.
5) Materials and methods – The statistical methods to compare volume are not described.
6) Materials and methods – The software used should be included.
7) Along the text – Age class and dominant age class should be defined in the text.
8) Lines 258-271 – It is not clear why the authors used factors. Please clarify and provide references.
9) Lines 313-323 – The text is not clear. Please clarify.
10) Lines 350-355 – References should be included.
11) Lines 381-389 – Please revise English.
12) Some references are incomplete, e.g., 2, 3 25, 26, 28, 39
Author Response
We would like to thank the reviewer for the helpful and supportive review comments. Based on the suggestions we revised the draft and refer to each comment in turn:
We very much appreciated the encouraging feedback and points raised in the comments. The methods, results and discussion sections have been revised accordingly.
Comments:
1) Lines 36-37 – If e.g. is used there is no need to use etc.
3) Line 115 – Either the names of the countries should be included between brackets or a reference should be made to table 1.
4) Lines 137 – 0.133x0.133 resolution – units should be added.
Response: Thanks, all are corrected.
Comment: 2) Lines 44-51 – References should be added.
Response: Six references were added.
Comments:
5) Materials and methods – The statistical methods to compare volume are not described.
6) Materials and methods – The software used should be included.
Response: The statistical methods used to compare volume are now described in the “Materials and Methods” section. The software used is now also included.
Comment: 7) Along the text – Age class and dominant age class should be defined in the text.
Response: We now use the term “most frequent age class” instead of “dominant age class”. The definition is given in “2.1 National Forest Inventory Data”.
Comment: 8) Lines 258-271 – It is not clear why the authors used factors. Please clarify and provide references.
Response: The creation of the forest area mask is covered in the section “2.5 Forest area mask”. Reasons for using the factors are provided in the “Discussion” section.
Comments:
9) Lines 313-323 – The text is not clear. Please clarify.
10) Lines 350-355 – References should be included.
11) Lines 381-389 – Please revise English.
Response: These “Discussion” sections have been carefully revised.
Comment: 12) Some references are incomplete, e.g., 2, 3 25, 26, 28, 39
Response: We carefully revised the references, and could not spot what is missing. We ask for more specific advice in this regard.
Round 2
Reviewer 3 Report
The authors have addressed most of my concerns.
Reviewer 4 Report
The paper entitled “An improved Forest Structure Data set for Europe” has improved in the second version of the manuscript. Thus it is recommended to be accepted in the current form.
