A Comparison of Four Methods for Automatic Delineation of Tree Stands from Grids of LiDAR Metrics

Round 1

Reviewer 1 Report

This study is focused on the detailed comparison of various methods using LiDAR metrics calculated for grids of 5 m by 5 m raster cells as the data. The tested segmentation methods were region growing (RG), cellular automaton (CA), self-organizing map (SOM) and simulated annealing (SA).

Both the methods used and the results presented have scientific value and are beneficial to the scientific community in the fields of Remote Sensing and Forestry.

I recommend subsequent corrections to the article:

-         -   Fig. 1 The numerical and graphic scale of the map showing the research area is missing;

-          authors use inconsistent terms and abbreviations, e.g. the abstract states: unmanned aerial vehicle laser scanning (ULS), but on line 103: Laser scanning from an unmanned aerial vehicle (UAVLS). It should be uniform in the article;

 -          180 between line 180 and Fig. 3 put an empty line (space);

 -          252 – 253 equation No. 7 is written in a different font size than the other equations in the article;

 -          277 between line 277 and equation (8) put an empty line (space);

 -          304 between line 304 and equation (11) put an empty line (space);

-          336 between line 336 and Figure 5 put an empty line (space);

-          495 between table 2 and line 495 put an empty line (space).

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper compared the performance of region growing, cellular automaton, self-organizing map, and simulating annealing in automated segmentation of forest into stands using UAVLS data. All four methods tested in this study are capable of stand delineation in the context of forest management. With the highest proportion of small segments and most heterogeneous segments, RG is outperformed by the other three methods.

This paper describes four methods adequately and presents the results clearly; however, the introduction may need improvements. The authors should discuss the importance of automatic segmentation methods in delineating forest stands and add more citations for applying different methods.

For the results section, I have two questions about selecting segmentations in Figures 6 and7,

1.       The authors did not clearly explain how they selected the segmentations, although they mentioned, “Considering that small within-segment variation in the LiDAR metrics is the most 397 important measure of a good stand delineation, the segmentations shown with large 398 markers in Figure 6 and Figure 7 were selected for further analyses.”

2.       In Figure 7, why does SA have 10 segmentations? I think they all should have 9 instead of 10.

This is a complete paper with good overall quality, but the authors should check the writings carefully before resubmitting; for instance,

Line 69, “The third method that was recently adapted to stand delineation is the self-organizing map developed by Kohonen (1982) [19]. ”  Is this the fourth method?

Line 79 “According to the articles in which these methods were presented [3,4,9]. All of…” Rewrite these sentences.

 

…

Author Response

please see the attachment. Thanks.

Author Response File: Author Response.docx

Reviewer 3 Report

Dear authors, thank you for your interresting contribution.

Keywords:

unmanned aerial vehicle laser scanning (ULS) - this is not atypical keywords, UAV or UAS (better - it is whole system with terrestrial and pilot segment) or RPAS (remote piloted aerila system) and ALS (aerial laser scanning); ULS - you created a new abbreviation, is it necessay or more sexy?

I recommend a few sentences in the introduction about the reason for the inventory and other methods, it will be good for the non-specialist reader.

Figure 1. for cartographic reasons add bar scale and orientation to north.

row 130 Laser scanning from an unmanned aerial vehicle (UAVLS) ... you use  in keywords ULS and now UAVLS ? Join it and find better abbreviation.

Tab. 1 Accuracy/Precision (mm) 10/5...define it, what is the difference from your point of view?

row 110 The average point density was 136 pulses/m2.  OK, in row 118 you write The average density of ground points was 17 pulses/m2.  ???

The test area was 1x1,8 km2. How many fligts? Joining precision?

How it is with LiDAR - is it witch one echo or more echos (full waveform lidar)?

row 119 Then, the digital terrain model (DTM) was constructed using the ground points and the Kriging interpolation method with a 1-m spatial resolution [23].

It is wrong, you interpolated the DEM (digital elevation model); DTM has different meanings for different disciplines. DTM for geographers or soldiers is everything that is on the surface, soldiers fight in the terrain, geographers work in the terrain. Surveyors think that DTM is pure relief. :-) So better DRM. Digital surface model is clear - DSM (digital surafe model) cerated from photogrammetry for example.
row 191 2.3.2. Cellular automaton ?

row 512 discussion

This study compared the performance of four different segmentation methods to seg ment the forest into stands using UAVLS data.

OK, write 1-2 sentences, why you do it, what is the final aim. Number of thees for example?

Conclusion

Overall, SA was evaluated to be the best method for automatic stand delineation. The results suggest that methods based on cellular automata, self-organizing maps and combinatorial optimization should be used more  in automated delineation of forest stands.

It means, that from four method three are ok and the best, based on testing, is SA?

References:

Join style :

16. STRANGE, N.; MEILBY, H.; BOGETOFT, P. Land use...

Capital letters?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Thank the authors for their response and revision. I am OK with the current version.