Semi-Automated Sample-Based Forest Degradation Monitoring with Photointerpretation of High-Resolution Imagery

Round 1

Reviewer 1 Report

This is an interesting methodological manuscript on the use of Plot based Data and rapid photointerpretation from VHR aerial photographs for monitoring forest degradation and assessing forest fragmentation.  Forest degradation and fragmentation is still a major environmental problem, despite the greater awareness nowadays on the benefits of forests for the environment, economy and society. This is especially true for developing countries and the tropics. Thus, developing tools and methods for accurate, inexpensive and at appropriate time intervals monitoring of forests and forest process is of particular importance. The manuscript reads well with good English and the few minor mistakes I am sure they will be corrected in the revised version.

In the introduction the authors justify the need for effective monitoring of forest processes for the implementation of programs related to forest conservation and climate change mitigating strategies. However, I believe in their effort to justify the need of implementing the approach presented in the study, they are unfair in relation to the value, importance and  the services provided by large, country or continental  scale, mapping products, such as the NLCD or perhaps the CORINE Land-cover mapping of Europe. Those products are of great value because they provide relatively accurate information for large scale analysis and monitoring of forests and other valuable ecosystems. Of course there are discrepancies in the definitions of land cover classes between those products and the needs of various monitoring programs and there will always be a need for site or region specific tools and approaches for monitoring land cover change.  Furthermore, the sensor limitations of satellite images, mentioned in line 63, is an expression that should be omitted since Satellite images are generally superior to aerial photo in terms of spatial, spectral and temporal resolution, and classification products using such images can be developed, adopting the definition and other needs of the endusers.  I believe the value of the work presented in the manuscript does not lie in the inappropriateness of existing satellite based, large scale mapping products for monitoring, but on the need for intensive and cost efficient monitoring of land cover changes at local or regional scale, and this should be pointed out in the problem statement part of the introduction. Finally, I would prefer the introduction to end with a clear statement of the aim and the specific objectives of the study, and this can be added after the short description of the work presented in the last paragraph.

The methods employed are sound, properly described, based on previous published work and scientifically justified. The fact that the analysis is performed at two different scales using real data is very interesting as well as the simulated change employed in the 3^rd study site. The metrics employed are sensible and the adoption of different scales for defining adjacency is very informative. The results achieved in this way indicate that the method can be employed for forest types with varying dispersal properties.  

The results are presented with an appropriate combination of text, figures and graphs and they are easy to follow while the discussion follows the results and does not expand to areas that are not investigated in the study.

I take also the opportunity to congratulate the authors for their great effort to build and analyze such a dataset.

Author Response

Reply to reviewer 1:

We would like to thank the reviewer for their positive feedback and kind words about our manuscript. We appreciate the reviewer’s perspective on the value of satellite remote sensing and classified imagery for large area land cover and fragmentation analysis. It has been the standard for these types of analysis for many decades. We were concerned that we were overly critical, and it sounds like we came across that way to the reviewer. For that reason, we added language to the introduction that emphasized the value of these types of products.

For example, we added:

Lines 47-51 “National- or continent-scale satellite imagery products often provide accurate information for coarse scale analysis and monitoring of forests and other valuable ecosystems, as well as serve as a base layer for many types of analyses, such as for stratification of forest inventory plots for improved estimation of forest status and trends [8].”

Starting at line 64, we also made the following change to address the reviewer’s comment that we should shift the focus from the problems with satellite imagery products to the need for local-regional scale, intensive, cost-efficient monitoring systems. We do believe that definitional mismatch is still a problem, but we feel that we de-emphasized this with our edits to the original text:

On the other hand, Fassnacht et al. [13], while acknowledging limitations of satellite remote sensing, offer commonsense strategies for effectively leveraging the improved spatial, spectral and temporal resolution of satellite images to meet users’ needs. There have been recent attempts to harmonize satellite-derived land cover maps with more complicated forest definitions using high resolution data and image segmentation [14]. However, there is still a need for intensive and cost efficient monitoring of land cover changes, forest fragmentation, and forest degradation at local or regional scales with definitions that meet specific user requirements.

Finally, we now see that we did not have a clear aim and specific objectives in our last paragraph, and we rectified that with the following addition (around line 126):

Our aim is to build on existing work in the fields of forest fragmentation analysis and sampling theory by developing and presenting new indices and analysis paradigms that allow data from intensive PI studies to be used in ways not previously foreseen. Through our presentation of case studies and distribution of computer code, our main objective is to show the value of this approach and provide tools and conceptual models that practitioners can use to improve upon their own forest monitoring systems.

Reviewer 2 Report

1- It did not point out how plots are distributed on study areas; randomly or systematically?

2- Is this true 20184 sub-plots were used for MD study and 34650 for PG?

3- In table 1 the range (max and min possible) of metrics values should be added in description so the result presented in table 2 is to be more understandable for readers. 

4- SE in table 2 are presented in absolute terms, it should be presented in relative terms. It is easier for reader to compare. 

5- It would be interesting to present the similarity of metrics values from raster maps and sampling paradigms, for instance using scatter plot.  

6- L 169- inscribes or describes?

7- How many plots were used for SC?

Author Response

Reply to Reviewer 2

We would like to thank the reviewer for their efforts and their constructive, helpful comments that made the paper better.

The sample design is described around lines 147-155 of the revised manuscript; we left this section mostly the same as the original. When we refer to spatially-balanced sampling that is an alternative to a systematic distribution of points – we describe the method briefly and provide a citation. We did add a parenthetical to help make this clearer to the reader, however:

“; this is analogous to a systematic design, but without the same degree of regularity” (lines 152-153).

Yes, that is the number of subplots interpreted. It sounds like a lot, but remember, the form and protocol was designed to minimize time per plot. In particular, for SC, the form was designed to “paint” the points with the mouse cursor, allowing for the labeling of many points almost instantaneously.

I added text and more explanation for the minimum and maximum values for the general equations in Table 1. I also corrected some small errors that other reviewers found in some of the formulas in table 1.

I edited the SE column to reflect the absolute and the percentage (relative) standard errors.

I agree, however, our concern is that the paper is very packed with information and figures. We do, however, present results using some of our metrics with raster-derived indices (see Figures 6, 7, and 8 and accompanying description). In a future study, we would like to do a more thorough comparison by comparing results under a factorial design between raster and point-based sampling of the same raster imagery, under different spatial pattern scenarios (heterogeneous to homogeneous).

Good catch, thank you. We actually changed it to circumscribes – the circle surrounds and contains the FIA plot.

We used 2099 plots (52 * 2099 points) for the SC study area, as mentioned around line 154 in the revised manuscript.