Modeling Salt Marsh Vegetation Height Using Unoccupied Aircraft Systems and Structure from Motion

Round 1

Reviewer 1 Report

Summary

This study estimates salt marsh vegetation height with unoccupied aircraft system (UAS) imagery and structure from motion (SfM) photogrammetry techniques. The authors develop a vegetation height model based on drone-derived digital terrain and surface models including a publicly available light detection and ranging (LiDAR)-derived terrain model. The proposed model is compared with field measurements of stem height, and further corrected through linear regression to predict true vegetation height of salt marshes. Moreover, the authors use predicted vegetation height as a proxy for above ground biomass (AGB) and lateral vegetation area at different stem heights. This study is technically sound since it leverages UAS technology to survey (large) coastal areas of difficult access; where field surveys are often labor intensive and conducted only at small scale. In addition, given the poor accuracy of LiDAR-derived terrain models within salt marshes, UAS and SfM offer a cost-effective alternative to real-time kinematic (RTK) surveys. This study has a broad applicability, especially in coastal flood modeling, since marsh height datasets are key inputs to correct LiDAR-derived terrain models. However, I have two major concerns and several minor comments that should be addressed/clarified in order to improve the quality of this work.

General comments 

The authors develop a vegetation height model and propose three methods to construct digital terrain models (DTMs) or ‘bare-earth’ models. However, it is not mentioned anywhere in the manuscript whether the point cloud data were collected during high or low tide conditions at the nearest tidal station or not (e.g., Beaufort Duke Marine Lab station, NOAA ID 8656483). UAS flights during high tide conditions might capture inundated marshes, and as a result, lead to an overestimation of terrain elevation and associated underestimation of true stem heights. On the other hand, point cloud data may contain high ‘water’ points within the salt marsh platform even under low tide conditions, and so lead to inaccurate DTMs too. Therefore, DTM generated with the first method should account for additional spectral indices that help mask out not only vegetation, but water points (e.g., normalized difference water index (NDWI)). In this way, the accuracy of the DTM may be even better than traditional topobathy LiDAR-derived maps.

Regarding the third method, the authors use Topobathy LiDAR data to generate a DTM of salt marsh areas. It should be noticed, however, that these data may contain a relatively large vertical bias within salt marsh areas. Consequently, comparison of the two first methods with respect to the last one may also lead to erroneous conclusions; especially if the LiDAR-derived DTM is not previously corrected for vertical bias. In fact, several studies have reported LiDAR elevation error in marsh vegetation species in the U.S. Pacific Coast [1], Gulf Coast [2,3] and Atlantic Coast [4,5] and also proposed diverse alternatives to deal with LiDAR errors.

Specific comments 

L27: Explicitly mentioning that predicted vegetation height is used as a proxy for AGB might also draw the attention of the reader.

L56: Also, LiDAR-derived DTMs are not accurate in salt marsh regions and usually lead to overestimation of marsh platform elevation.

L64: What did those authors find? Summarizing key findings of previous UAS-SfM approaches (two or three lines) will help the reader to get a better picture of the novelty of your work.

L74: Please indicate those sites in Figure 1A for a better orientation of the reader.

L89: Misspelling: altitudes.

L90: Do the authors mean a three-month period (July 2018 to September 2018)?

L97: Please define RS, RTK and GPS before abbreviating those terms.

L122 – L126: Please be consistent with the notation you use in equation (1) and (2). That is, if you use variable ‘y’ to denote dry weight in Eq. (1), make sure you use the same variable ‘y’ in Eq. (2). This also accounts for variable ‘x’.

L168: What is the spatial resolution of DSM and DTMs? This is not mentioned anywhere in the manuscript.

L171: How did the authors deal with ‘water’ points within the salt marsh platform? Was NDVI enough to discriminate bare earth from vegetation points?

L172: Please provide more details of the assessment you conducted to come out with a threshold value of 0.3. Does the threshold value represent min, mean and max value across the marsh?

L175: What’s the raster resolution of NDVI?

L182: Did the authors compare the ‘trend interpolation’ method to IDW? If so, please comment on that as you previously highlighted the benefits if IDW.

L213: While each of the models…

L214: a model that predicts…

L224: If the DTM derived from point cloud method coincides with high tide conditions, the weakest linear fit you show in Figure 4C may be partly explained due to water points (above marsh platforms) incorrectly interpreted as bare earth. Consider masking out both water and vegetation points before IDW interpolation of terrain elevation.

L234: It seems the authors are misinterpreting ‘mean’ with ‘median’ in the box plot (Figure 5).

L242: Again, it seems the authors are misinterpreting ‘mean’ with ‘median’ (Figure 6). Also, the labels below the boxplots do not coincide with your description.

L276: Do the authors mean Figure 8?

L282: and thorough the text: Correct superscript of r2

L288: 0.025-0.045

L290: Please mention those metrics.

L359: Figure 9 is missing.

L362: Perhaps adding a vertical axis next to Figure 8A would help the reader to better visualize vegetation heights.

L391: Please provide a brief summary of the three methods and the transformation you conducted before describing the comparison. This will help the reader to follow your conclusion in a logical sequence.

L392: According to the results, LiDAR method achieves the best accuracy (See L242).

References

1. Buffington, K.J.; Dugger, B.D.; Thorne, K.M.; Takekawa, J.Y. Statistical correction of lidar-derived digital elevation models with multispectral airborne imagery in tidal marshes. Remote Sensing of Environment 2016, 186, 616–625, doi:10.1016/j.rse.2016.09.020.
2. Medeiros, S.; Hagen, S.; Weishampel, J.; Angelo, J. Adjusting Lidar-Derived Digital Terrain Models in Coastal Marshes Based on Estimated Aboveground Biomass Density. Remote Sensing 2015, 7, 3507–3525, doi:10.3390/rs70403507.
3. Alizad, K.; Hagen, S.C.; Medeiros, S.C.; Bilskie, M.V.; Morris, J.T.; Balthis, L.; Buckel, C.A. Dynamic responses and implications to coastal wetlands and the surrounding regions under sea level rise. PLOS ONE 2018, 13, e0205176, doi:10.1371/journal.pone.0205176.
4. Hladik, C.; Alber, M. Accuracy assessment and correction of a LIDAR-derived salt marsh digital elevation model. Remote Sensing of Environment 2012, 121, 224–235, doi:10.1016/j.rse.2012.01.018.
5. Muñoz; Cissell, J.R.; Moftakhari, H. Adjusting Emergent Herbaceous Wetland Elevation with Object-Based Image Analysis, Random Forest and the 2016 NLCD. Remote Sensing 2019, 11, 2346, doi:10.3390/rs11202346.

Comments for author File: Comments.pdf

Author Response

Thank you for your thorough review of this manuscript. Please see the attachment for my responses! 

Author Response File: Author Response.docx

Reviewer 2 Report

Summary

 

This study used unoccupied aircraft systems to estimate plant height within coastal salt marshes.

 

Broad Comments

 

1. I think in the introduction you can also say how plant height can be used to estimate biomass with is an important input into

 

2. Some of the references seem to be incomplete and the formatting doesn’t seem to be consistent. For example citation 16 is missing the year and journal name

 

3. Alterniflora should be in italics throughout paper

 

4. It would be beneficial to know the spatial resolution of the different images. Are they all the different? Or are they all sampled to be the same as the Lidar derived digital terrain model?

 

5. lateral vegetation area should be better explained. Its not clear why its important/what it represents

 

6. Did you compare the ground elevations of the three methods with each other? I think this would be interesting or show where the elevations are different.

 

7. r2 should be r² throughout the paper

 

8. Overall this paper shows great promise since biomass measurements are important for blue carbon estimates and inputs to biogeomorphic models such as the Marsh Equilibrium model

 

Specific comments

 

1. Line 105 – Should say placed across

 

2. Line 172 – why was a threshold of 0.3 NDVI chosen?

 

3. Line 241 – Transformations should be better explained. What are the new equations used?I think it would also be nice to have a figure similar to figure 4 with an added 1:1 line, you can even add the MSE value then remove figure 6

 

4. Line 248 – It should be mentioned that for ABG calculations field collected data was used (its mentioned in the figure caption but not the text)

 

5. Line 253 (and throughout paper) you don’t have a correlation analysis. You have an r² You could do a Willmott’s index of agreement or pearson’s correlation analysis

 

6. Figure 7 – It seems like the biomass (g/m2) estimates are rather large? This paper found the max estimated biomass in a South Carolina Marsh to be something around 1,000 g/m2 https://www.mdpi.com/2072-4292/11/17/2020/htm and the paper cited for the relationship between height and AGB https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0142595#sec019found it to also be around 1,000 g/m2

 

7. Figure 8 is a little confusing. For figure 8a are the lines representative of lidar, manual and pointcloud. But for each of these three categories the predicted heights are different so the lines are at different heights, it’s not very clear.

 

8. Line 283 – “low strength of correlation.” r² isnt correlation its how well the predicted versus observed fits around the line.

 

9. Figure 2 – do the plants get pushed down from the mirror? So the only standing plants are the ones within the 25cm ?

 

10. Lines 306-307 brings up a pretty good point. I’m curious if you were able to measure the plants from the ground to the top without holding up the tip? I’m assuming it’s not possible to redo this for this paper, but it would be pretty interesting to know.

Author Response

Thank you for your edits on this manuscript. They were very helpful. Please see the attachment for my responses. 

Author Response File: Author Response.docx

Reviewer 3 Report

I am doubtful about regression models explaining no more than 30% of the total information. I wonder how effective may  be the resulting predictions.

Some comments would be welcome

Comments for author File: Comments.pdf

Author Response

Thank you for your comments. Please see the attachment for my responses. 

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Summary

This study estimates salt marsh vegetation height with unoccupied aircraft system (UAS) imagery and structure from motion (SfM) photogrammetry techniques. The authors develop a vegetation height model based on drone-derived digital terrain and surface models including a publicly available light detection and ranging (LiDAR)-derived terrain model. The proposed model is compared with field measurements of stem height, and further corrected through linear regression to predict true vegetation height of salt marshes. Moreover, the authors use predicted vegetation height as a proxy for above ground biomass (AGB) and lateral vegetation area at different stem heights.

 

General comments 

This manuscript version is much improved as it provides additional information and details (Table S1; Figures S1 and S2) that support the methodology used and key findings reported by the authors. Also, the authors did a good job clarifying the accuracy of the LiDAR-derived DEM. Please see minor corrections required below.

 

Specific comments 

L344: based on…

L371: Units in Table S1 are missing.

L377: Please provide a caption for Figure S1.

L425: Please provide a caption for Figure S2.

Author Response

Thank you for your edits. They were very helpful. I have attached my responses, and please see the revised manuscript for in-text edits. 

Author Response File: Author Response.docx

Reviewer 2 Report

Figure 4 – middle equation is missing an x. I still think it would be nice to have a 1:1 line which helps visualize what a perfect agreement between predicted and observed values would be.

 

Aboveground biomass of 5,000 - 6,000 g/m² just seems very high. But most of the datapoints fall within expected ranges. But 1.5 meter tall plants are also pretty tall. It does seem like when Spartina alterniflora is really tall the plant density also decreases which when scaling up to m² resolution might lead to an overestimation. But overall the model seems good enough.

 

Overall good improvements were made to the paper, and I think the paper helps show a useful way to estimate plant height across a landscape. Which can also be used to estimate aboveground biomass at a fine resolutions.

Author Response

Thank you for your helpful edits. Please see my attached responses as well as the revised manuscript for in-text edits. 

Author Response File: Author Response.docx