Recognizing Building Group Patterns in Topographic Maps by Integrating Building Functional and Geometric Information
Round 1
Reviewer 1 Report
See my comments attached.
Comments for author File: 
Comments.pdf
Author Response
Thank you for your comments. Please check the attachment which includes all responses to your comments.
Author Response File: Author Response.pdf
Reviewer 2 Report
Major?
The paper proposes a method of clustering of building features into groups based on their characteristics. The novelty of the presented work is the use of functional and geometric attributes of said building features, in addition to their commonly used spatial characteristics.
Because the primary application of detecting building clusters is map generalization, building groups are usually identified on the basis of their proximity to one another, as well as similar characteristics such as orientation and alignment, while attributes such as building function are rarely considered.
The proposed methodology involves inferring building functions from user density data obtained from citizen mobility information. using selected locations (residential, office and commercial buildings) as reference samples. The established building functions are then used to create a constrained Delaunay triangulation of building blocks. Finally, the graph is segmented on the basis of the spatial continuity index and the azimuth angle of adjacent buildings. The proposed method has been applied to a dataset representing the Chinese city of Chengdu.
While the general flow of the paper is fine, I have several issues with the proposed methodology. First of all, the authors propose a method of building classification based on citizen density data, which yields questionable results. The identification accuracy of the proposed method for residential, commercial and office buildings has been shown to be 91.70%, 96.77% and 47.82%, respectively. The impact of those inaccuracies on the remainder of the grouping process is not discussed. Moreover, the citizen density data used in the study was collected in June 2020, during the COVID-19 pandemic. The possible influence of workplace limitations, quarantines as well as selective lockdowns on the accuracy of dataset classification also has not been discussed in the paper. In the above context it is not clear why the authors did not simply use a pre-classified building dataset from a verified source such as data.gov.uk.
More importantly, however, my main issue is with the proposed verification methodology. The authors compare the results obtained by their method with an unnamed "method without building function recognition". Since there are many building grouping methods described in literature, and all of them produce somewhat different results, the choice of a reference method should be well explained and argumented. Ideally, the authors should present a comparison of their approach with the output of several popular building grouping methods. Moreover, I have also identified several issues with presentation (see detailed recommendations below). In my opinion both the methodology as well as results should be improved before the paper may be considered fit for publication.
Detailed recommendations:
Lines 228-330: The "method without building function recognition", which is used as reference should be described in more detail. Is this one of the methods described in the literature review? If so, this should be clearly stated in the text.
Lines 313-327: Please explain why the generalization results sometimes intersect with the road network, despite the fact that the initial building groups are segmented using the road graph. Moreover, the meaning of blue circles in Figure 9 should be explained in this section.
Figure 4: The applied method of marking incorrectly recognized building groups by using a red outline is not sufficient. Consider applying a hatched pattern over the area of these building groups, or at the very least make the red outline 2-3 times bolder.
Figure 5: It would be a good idea to remind the reader (eg. in the figure caption) that June 05 was a working day, and June 06 was a non-working day.
Figure 8: The applied method of marking incorrectly recognized building groups by using a red outline is not sufficient. Consider applying a hatched pattern over the area of these building groups, or at the very least make the red outline 2-3 times bolder.
Figure 9: The meaning of colored circles should be explained in the figure caption.
Author Response
Thank you for your comments. Please check the attachment which includes all responses to your comments.
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear authors,
congrats on your work and the well structured manuscript. My main concern is the presentation of the results, as I was having a hard time to read and understand the figures. My detailed comments:
- Line 92: "Road net" -> "network"?
- Figure 1: Would help if you drew arrows from (a) to (b) and from (b) to (c). Still one can understand but it is not intuitive and takes some time to decode the relationship (even for a cartographer like myself). Also, the numbers of the building blocks can be increased in size (and/or use halo around the labels).
- Line 111: What is "abnormal data"? Please provide examples and methods to remove them (automatically, with visual inspection).
- Line 146: I suggest modifying the name "original triangles" (e.g. to "adjacent triangles"). One may ask: "Why original? The other triangles are not original?"
- You are saying using constraint triangulation, however this looks more of a conforming triangulation. How come the building edges are triangulated in regular intervals? Have you densified the edges? For the constraint TIN one would expect to see the corners of the buildings to be triangulated but not the entire edge at regular distances. Please clarify / elaborate / modify.
- Line 188-192: Understood that the method is described in other works, however it would help readability if authors included a figure showing the concept.
- Line 205: "connecting" - > "connected"
- Line 209: "restarted"-> "recalculated"
- Line 239: "dotted" - > "dashed"
- Figure 4: Figure needs to be improved designed. Using inconsistent color makes the comparison between (a) and (b) to the reference (c) impossible. I can't see any similarities / shared colors that would explain the "81% accuracy" three lines after the figure (line 253). I trust that random colors have been used, when consistent colors should have been selected for the same categories in all three sections of this figure. Also, the dashed lines around buildings is impossible to see and even when zooming in is still difficult. A better way to present this would be with two figures: in one figure the authors can show the group recognition results (again, using the same color for each category) while in a second figure authors can show the correct and wrong groups (e.g., in blue all buildings that are correct and in red those wrongly classified)
Author Response
Thank you for your comments. Please check the attachment which includes all responses to your comments.
Author Response File: Author Response.pdf
Reviewer 4 Report
The authors attempted to recognise groups of buildings based on topographical and functional criteria. The innovation is to consider not only the building geometric formation but also the building functionality. The study was very interesting.
The proposed methodology is presented in the form of Table 1 withe main steps and detailed description. Such an approach greatly organizes the actions taken.
I have some technical and graphical remarks:
Figure 1: a & b – please, provide scale or scale bars, c – the scale bar is not properly done (350 m divided into three blocks results in 116 m as the base of scale bar), please provide the 100 m section as a base to this scale bar. The North direction arrow should be placed on the map, not in the legend. Could it be the simplest arrow with letter N?
Figure 4: there is a mistake in the title of part B: should be without instead of whtout, the incorrect groups are difficult to identify (red dotted lines), the scale bar is not properly done (see remarks to Figure 1). It is also confusing to give three colors in the legend, while more are used on the map. i propose to give one pattern in gray without outline and with a red line.
Figure 5: very interesting maps, please provide correct scale bar.
Please check the axis scaling in Fig. 6. It is not entirely clear what the horizontal axis captions mean. Please check the axis scaling in Fig. 6. It is not entirely clear what the horizontal axis captions mean. I assume that 05/ is day and /07 is hour, but it is not clear. The dates do not match the descriptions in Fig. 5 (06/07 and 06/11 are missing, 09/07 is incorrect).
Fig. 7-9 – please provide properly done scale bars.
Author Response
Thank you for your comments. Please check the attachment which includes all responses to your comments.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Most of my comments have been addressed, and I believe that the paper has been significantly improved as a result. This being said, certain parts of the text could be further improved. In particular, it may still be unclear to the reader what the "method without building function detection" actually is. I would suggest to instead write "standard CTD (without building function detection)" or something similar.
Author Response
Thank you for your advice. A description about the compared method was added to Step 7 (Line 237-Line 244). We also replaced the corresponding subtitles of pictures with standard CTD.
Author Response File: 
Author Response.docx
