A Drone-Powered Deep Learning Methodology for High Precision Remote Sensing in California’s Coastal Shrubs
Round 1
Reviewer 1 Report
The authors described the background, methodology, and results of the research in great detail. They also discussed some limitations and the methods they used to improve the accuracy. It will be an interesting paper to publish.
I have a question about classifying deadwood as a single group (line 319). Deadwood of different species may have significantly different canopy sizes, heights, and slopes. Could classifying deadwood from all species in a single group reduce the accuracy of the results?
Author Response
Thank you for the question regarding the impact of combining deadwood for multiple species. We will add the following to the discussion section to clarify.
Grouping standing deadwood from multiple species into a single category may reduce the deadwood classification accuracy as different species exhibit variations in dieback patterns and decay processes, which can be overlooked when combined. However, this aggregate approach can be effective at supporting the delineation of bareground which often has similar spectral properties \citep{zielewska2020detection}. In our study, manzanitas (A. pumila and A. tomentosa), chamise (A. fasciculatum), and oak (Q. agrifolia) were the predominant species exhibiting deadwood characteristics. Creating separate classifications for each species could enhance classification accuracy and ecological relevance, providing a comprehensive understanding of dieback dynamics in the ecosystem. We attribute the high performance in deadwood detection to the abrupt variation in canopy heights and slope dynamics within manzanita and chamise dieback patches coupled with high NIR and NIR-edge absorption and low red absorption. Although our study does not delineate species-specific deadwood detection it does offer a suitable alternative to rectifying a challenge with correctly classifying bareground from standing deadwood in forest systems.
Reviewer 2 Report
The paper explores the use of drone-based imagery and deep learning techniques for mapping plant species in complex shrubland communities such as chaparral, coastal sage scrub, and oak woodland. The authors tested the effectiveness of three modeling approaches and found that the Convolutional Neural Network (CNN) coupled with object-based image analysis (OBIA) outperformed the other methods in accurately identifying tree and shrub species, vegetation gaps, and communities. The study demonstrates the potential of this approach for large-scale mapping of plant species in wildland conservation efforts, although uncertainties still exist when dealing with less common species and intermixed canopies.
I highlighted some minor suggestions in the attached pdf.
Suggested papers for citation:
Traba, J., Gómez‐Catasús, J., Barrero, A., Bustillo‐de la Rosa, D., Zurdo, J., Hervás, I., ... & Reverter, M. (2022). Comparative assessment of satellite‐and drone‐based vegetation indices to predict arthropod biomass in shrub‐steppes. Ecological Applications, 32(8), e2707.
Kabiri, K. (2020). Mapping coastal ecosystems and features using a low-cost standard drone: Case study, Nayband Bay, Persian Gulf, Iran. Journal of Coastal Conservation, 24(5), 62.
Comments for author File: 
Comments.pdf
Author Response
Dear Reviewer,
Thank you for the excellent suggestions on how to improve the manuscript.
1. Title: We have revised the title to something a bit more catchy; "A Drone-Powered Deep Learning Methodology to Identify California's Coastal Shrubs with High Precision"
2. Keywords: I've integrated the suggested keywords.
3. Maps: I've added coordinate systems and corner coordinates to all research wide maps.
4. I appreciate the suggested works to cite. I've integrated them into the introduction.
5. Figure 2 letter designation. I removed the 'A' and 'B' designation from the image and simply orient the reader to maritime chaparral on the 'left' and coastal sage scrub / oak woodland at 'right'.
I will upload the revised manuscript very soon.
~ JD
