Setting the Flow Accumulation Threshold Based on Environmental and Morphologic Features to Extract River Networks from Digital Elevation Models

Round 1

Reviewer 1 Report

The manuscript by Zhang et al. titled “Setting the flow accumulation threshold based on environmental and morphological features to extract river networks from digital elevation models” proposes and compares different methods for estimating flow accumulation threshold (FAT), an essential variable for evaluating the stream network across watersheds. The first approach is regression based and estimates FAT dependency on multi-source datasets that are regressed using total river length as dependent variable to find the most significant parameters critical for estimating FAT. The second method incorporates two applications based on fractal geometry theory producing a more efficient but less accurate estimation of stream network. The performance of these methods is tested and validated across several sub-watersheds in two major provinces of China. Overall, the study proposes novel approaches for estimating an important variable for water resources applications, but the accuracy, applicability, and validity of the proposed approaches need to be compared to an existing method for making this a significant contribution.

Please see below for specific comments:

1. Why did the authors choose the D8 method for calculating flow direction instead of using the D∞ (Tarboton, 1997)? I would think this choice might have implications to the flow accumulation especially in urban areas with complex flow paths.
2. My major comment is with respect to the evaluating of the proposed methods. While the methods proposed here are useful, their applicability should also be evaluated with respect to the most common/accurate method currently used to determine FAT. The comparison with observed/surveyed/remotely sensed river networks are useful, but the results do not make the case for why the methods proposed here should be chosen over the widely accepted methods. In lines 56-70, the authors highlight several methods that have been proposed to estimate FAT in earlier studies. Please incorporate at least one of the methods in your analysis and quantify the relative improvement/similarity while also making the case for computational efficiency in the proposed approach. In summary, this manuscript can become a more significant contribution if the methods not only produce a specified level of accuracy with respect to observational networks, but also with respect to the current ‘gold standard’ in estimating FAT. The RMSE values are over 10,000 m for almost all cases, but whether these are acceptable is difficult to understand without a thorough comparison with an existing approach.
3. Optimal FAT should, in theory, depend on the intended application. The most performant stream network that gets directly affected by FAT selection may or may not be the same for different applications such as evaluating water pollution sources, monitoring disturbances on rivers caused by human activities, determining flood levels, assessing soil and water conservation measures, and enacting comprehensive management practices (these applications being the ones mentioned in this study). Even within an application, the most performant FAT can be different, for example, different FAT values might provide the most accurate streamflow estimation for different hydrologic models. Please comment.

Author Response

Dear Reviewer:

We thank you the Editor and Reviewers for insightful comments that allowed us to improve our paper. Thank you to assistant editor Ms. Anne Feng. The paper was revised to accommodate the Reviews’ comments. Please see attachment.

Best regards

Author Response File: Author Response.docx

Reviewer 2 Report

This is a review of the manuscript titled “Setting the flow accumulation threshold based on environmental and morphological features to extract river networks from digital elevation models” submitted to the International Journal of Geo-Information. The manuscript proposes two different methods – regression based, and fractal theory based – for estimating flow accumulation threshold to delineate river networks from DEMs. Overall, the manuscript does a decent job of setting up the problem statement and explaining and validating the methodologies. However, there are certain concerns regarding this research, especially the robustness of the proposed methodology, that needs to be addressed before the manuscript can be considered ready for publication. I, therefore, recommend “Major Revision”

Following are my concerns. Some minor changes are marked in the attached pdf.

• Introduction: The introduction needs to be more streamlined to clearly convey the drawbacks of current methods to estimate FAT that the authors are targeting and how their proposed methods can potentially overcome these errors. For example, Lines 63-69: what are the drawbacks of FAT determination methods proposed in these studies? Lines 100-112: The authors talk about availability of DEM resolution and then suddenly pivot to previous studies neglecting anthropogenic effects. This paragraph needs to be better phrased.
• The statistical regression method: The authors have used regression to estimate river length using reference river networks. The two provinces yield very different regression equations, which is somewhat expected given the different characteristics of the provinces. The authors should include a discussion on how the difference in regression equation reflects the difference in characteristics of these two provinces (similar to the NMF discussion).
• The authors need to quantify the degree to which the choice of training (sample) and validation (test) set is affecting the performance of the regression equations. What happens if a different set of sample and test basins are used within the same province? My suggestion would be to use the k-means validation to quantify the variation in performance due to choice of sample and test cases. Also, how is the accuracy affected if both the provinces are represented by a single equation? This will convey the robustness of these algorithms to the readers.
• Figures 3 and 4: Each subplot should be separately numbered and noted in the caption. The figures (subplots) can be reduced by showing “captured” and “missed” in a single figure instead of “extracted” and “actual” in one and “missed” in another. Another idea could be to show true positives, false positives and false negatives with three different colors in the same figure.
• Table 4 is places before tables 2 and 3. Please renumber them accordingly.
• Table 2 and 3: Please elaborate the reasons behind the high River Length Error in Area 2 compared to Areas 1 and 3.
• It is true that the Fractal method is has lower data requirements than the regression method but it seems that it also underperforms significantly compared to the regression method – with average error being more than 5 times for river length (Table 7 vs Tables 2 and 3). To somewhat extent the mitigating factor is that they are implemented at two different scales but is the trade-off in performance justified by the lower data requirements?

Author Response

Dear Reviewer:

We thank you the Editor and Reviewers for insightful comments that allowed us to improve our paper. Thank you to assistant editor Ms. Anne Feng. The paper was revised to accommodate the Reviews’ comments. Please see attachment.

Best regards

Author Response File: Author Response.pdf

Reviewer 3 Report

The document attempts to establish the flow accumulation threshold by applying two different working methodologies in the Chinese provinces of Hubei and Qinghai.

The introduction provided in the work shows a large number of authors who have worked on similar topics in different temporal spaces. Perhaps it would be advisable to order the citations according to the date of publication. Thus, the bibliographic citation of line 38 must show the author whose publication was made in 2009 before the one who carried out the work in 2015. Something similar happens in the bibliographic citations shown in lines (50, 51, 52), (73), (126, 127) and (343, 344).

The methodology follows an adequate design, establishing different action scales through a neighborhood geoprocess by the altitude levels of each established grid. The first method calculates the threshold for accumulation of flow at a reduced scale (basin) by studying parameters such as drainage density or the length of the river, these are based on different analysis factors and statistics used in works previously carried out by other authors. For its part, the second method carries out the study of a larger space (regional scale) that houses thirteen river networks and uses a multitude of statistics.

The study area is well defined and mapped, it also correctly indicates the source from which the data was obtained, the possible predictive factors, the validation and the processing performed with ArcGIS.

The results are well represented when analyzing the two mentioned spatial scales and the factors established in the methodology. It is necessary to improve the quality of the heat maps in figure 5 since the values are not well distinguished with the white color. Furthermore, in figures 3 and 4 it might be convenient to use a different colour palette in order to better visualize the areas of higher and lower altitude. 

Finally, the conclusions correctly summarize the information obtained from the research, so the results obtained with each of the methods used and the opinion of the authors are shown. Perhaps it would be useful to briefly mention how the methodology used contributes to flood risk reduction.

Section 5 (Discussion): I think it is not right. Maybe, include that paragraph in the results section and name it Results and Discussion. Or maybe develop it further and name this section after conclusions as Future lines of research or Continuity in research!

Best regards.

Author Response

Dear Reviewer:

We thank you the Editor and Reviewers for insightful comments that allowed us to improve our paper. Thank you to assistant editor Ms. Anne Feng. The paper was revised to accommodate the Reviews’ comments. Please see the attachment. Thank you.

Best Regards

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I have carefully reviewed the author's response to my comments on the previous version of the manuscript. The authors have addressed the major issue of comparing the proposed approach with other existing methods. They have also provided appropriate responses for the other comments. I therefore recommend accepting the manuscript after addressing the minor comments from the other reviewers. 

Reviewer 2 Report

The authors have satisfactorily resolved my concerns regarding the manuscript. The changes made to the manuscript have improved its readability as well as provided justifications to the issues pointed in the review. There remain some minor formatting issues - such as figures spanning multiple pages, figure subplots sometimes numbered in caps - which can be corrected during the final typesetting and publication process. As such, I recommend accepting this manuscript for publication.