Identification of Vehicle-Pedestrian Collision Hotspots at the Micro-Level Using Network Kernel Density Estimation and Random Forests: A Case Study in Shanghai, China
Round 1
Reviewer 1 Report
The work is sufficiently structured. Some grammatical errors are present in the text. It is necessary to include more details and explanations considering the title, emphasizing better the concept of hotspots and the use of network kernel density estimation and random forests. The graphs of Figure 2 would be better represented. The results of Table 2 should be more fully described
Author Response
The authors would like to thank the reviewers for their valuable suggestions and comments. The author(s) have presented more details of the concept of hotspots and the use of network kernel density estimation and random forests (see lines highlighted in the introduction part and subsection 2.1 and 2.2 in the revised manuscript). The graphs have been reproduced and the results of Table 2 have been fully described (see Lines 238-246).
Reviewer 2 Report
This is an interesting and well written paper. Specifically, the reviewer feels that the proposed method is sound, and it is easy to follow. However, to improve the quality of this paper, I recommend the authors revise the initial version to publish. Some issues they may take into account are as follows:
1) The authors have a good job for the literature review part. However, they may add recently published articles to the Sustainability journal regarding hotspot or hazardous road location identification methods.
2) In the hotspot ID study, one of critical issues is a roadway segmentation method. This study applied an equal interval. Of course, it would not be a main part of this study, but the authors may explain the reason why they did it.
3) The authors may change the font style of variables or mathematical terms in the manuscript (not in equations) into italic font. For examples, n sub-samples in page 3 and R^2 in page 5.
4) In page 4, GPS represents global positioning system, not global positional system.
5) This study used two proxy variables for traffic volume and pedestrian flow: taxi flow and land use data. I understand why they used these proxy variables. However, in the academic perspective, this is the most critical flaw of this study. First, taxi flow could not be used as a proxy value for general traffic flow. Travel patterns and characteristics of taxi and general traffic are totally different. Second, the use of land use data as a proxy for pedestrian exposure could be a reasonable approach if they present statistical results of the relationship between land use variables and pedestrian activities (or travel).
6) In the study, taxi data for 2016 and land use data for 2014 were used. I suggest the authors apply the same period dataset (i.e., data synchronization).
7) In traffic safety research, the analysis is generally based on crash data observed for 3 to 5 year periods. However, this study used just one year period.
8) Although the approaching method presented in this study is sound, this study has various limitations. Thus, the authors may add limitations and further study items in the conclusion part.
Author Response
Reviewer 2:
This is an interesting and well written paper. Specifically, the reviewer feels that the proposed method is sound, and it is easy to follow. However, to improve the quality of this paper, I recommend the authors revise the initial version to publish. Some issues they may take into account are as follows:
Response: We would like to thank the reviewer for the valuable comments and suggestions. The detailed responses are as follows:
1) The authors have a good job for the literature review part. However, they may add recently published articles to the Sustainability journal regarding hotspot or hazardous road location identification methods.
Response: We have included two Sustainability journal papers (see reference 10 & 11)
2) In the hotspot ID study, one of critical issues is a roadway segmentation method. This study applied an equal interval. Of course, it would not be a main part of this study, but the authors may explain the reason why they did it.
Response: By analogy with standard planar KDE where the entire two-dimensional space is divided into regular grids, the NKDE method developed by Xie and Yan segments the road network with an equal interval to ensure regularly spaced locations along a network for density estimation (Lines 98-100).
3) The authors may change the font style of variables or mathematical terms in the manuscript (not in equations) into italic font. For examples, n sub-samples in page 3 and R^2 in page 5.
Response: The variables have been changed into italic font (Lines 145-161)
4) In page 4, GPS represents global positioning system, not global positional system.
Response: We would like to apologize for the oversight. It has been corrected (Line 189)
5) This study used two proxy variables for traffic volume and pedestrian flow: taxi flow and land use data. I understand why they used these proxy variables. However, in the academic perspective, this is the most critical flaw of this study. First, taxi flow could not be used as a proxy value for general traffic flow. Travel patterns and characteristics of taxi and general traffic are totally different. Second, the use of land use data as a proxy for pedestrian exposure could be a reasonable approach if they present statistical results of the relationship between land use variables and pedestrian activities (or travel).
Response: We agree that employing the proxy variables without validation is a limitation of this study, although it is not the focus of this research. One important issue is that travel patterns and characteristics of taxicabs and general traffic may differ. A typical problem is that unoccupied taxis tend to cluster in some specific types of places such as shopping malls and metro stations. Including unoccupied taxis may cause overestimation of traffic flow in these locations. As trajectories of occupied taxis are more likely to reflect travel demands and hence the variation of real traffic, only taxis with passengers were included in the sample. In the revised manuscript, the method in which taxi trajectories are selected for analysis is described in details (Lines 202-207). With this process, we believe that the problem can partly be solved. In addition, previous research has already indicated that the traffic flow can be predicted by the taxi data. We have included the reference in the revised manuscript (reference 53).
Regarding the land use variable, there were some studies investigating the relationship between pedestrian flow and land use characteristics. We have included some references (Reference 51,52).
Once again, we believe that although the focus of this research is not the validation of the three variables as proxies of vehicle and pedestrian flow, the method in which the vehicle and pedestrian exposure is measured has always been a hot topic in road safety research. With more experiments on the feasibility of proxy variables in future, more useful tools can be developed to increase the precision of the estimation, whereby the proposed method in this research can be further improved. We have included the discussion in the Result and Discussion section (see Lines 287-296) and mention this limitation in the Conclusion part (see Lines 306-308).
6) In the study, taxi data for 2016 and land use data for 2014 were used. I suggest the authors apply the same period dataset (i.e., data synchronization).
Response: Due to data availability, this study used the 2015-year vehicle-pedestrian collision data, taxi GPS data in 2016, and land use and POI datasets in 2014. Since Changning District is located in the urban area of Shanghai where the features of built environment did not vary significantly during the period from 2014 to 2016, it is reasonable to conduct analysis based on datasets collected from different years of this period (see Lines 224-228).
7) In traffic safety research, the analysis is generally based on crash data observed for 3 to 5 year periods. However, this study used just one year period.
Response: The reasons for this are twofold. First, given the length of road network in the study area, 1,200 vehicle-pedestrian collisions are able to well depict the overall picture of pedestrian safety. It is not necessary to pool 3- or 5- year datasets together to ensure the representative of the events. Second, in the recent decade, the Shanghai Police has enforced a set of safety rules year by year, which may result in significant yearly variation in safety performance.
8) Although the approaching method presented in this study is sound, this study has various limitations. Thus, the authors may add limitations and further study items in the conclusion part.
Response: The conclusion part has been revised (see Lines 306-313).
Many thanks again for your time!
Reviewer 3 Report
The paper is a well planned and conducted research. I have just one remark. When using proxies of car and pedestrian volumes, they should be somehow validated. For example, by conducting at least few traditional counts and comparing them to the proxy values.
Minor comments:
- There is section "Result and discussion", but it does not introduce any real discussion (of potential issues, limitations, what should be pursued next, comparison with other studies, etc.).
- I am not sure about the format of references (for example, listing the first names of authors). Please adapt to the journal requirements. Also please correct the surname in reference no. 34.
Author Response
The paper is a well planned and conducted research. I have just one remark. When using proxies of car and pedestrian volumes, they should be somehow validated. For example, by conducting at least few traditional counts and comparing them to the proxy values.
Response: We would like to thank the reviewer for the valuable comments and suggestions.
We highly agree that employing the proxy variables without validation is a limitation of this study, although it is not the focus of this research. One important issue is that travel patterns and characteristics of taxicabs and general traffic may differ. A typical problem is that unoccupied taxis tend to cluster in some specific types of places such as shopping malls and metro stations. Including unoccupied taxis may cause overestimation of traffic flow in these locations. As trajectories of occupied taxis are more likely to reflect travel demands and hence the variation of real traffic, only taxis with passengers were included in the sample. In the revised manuscript, the method in which taxi trajectories are selected for analysis is described in details (Lines 202-207). With this process, we believe that the problem can partly be solved. In addition, previous research has already indicated that the traffic flow can be predicted by the taxi data. We have included the reference in the revised manuscript (reference 53).
Regarding the land use variable, there were some studies investigating the relationship between pedestrian flow and land use characteristics. We have included some references (Reference 51,52).
Once again, we believe that although the focus of this research is not the validation of the three variables as proxies of vehicle and pedestrian flow, the method in which the vehicle and pedestrian exposure is measured has always been a hot topic in road safety research. With more experiments on the feasibility of proxy variables in future, more useful tools can be developed to increase the precision of the estimation, whereby the proposed method in this research can be further improved. We have included the discussion in the Result and Discussion section (see Lines 287-296) and mention this limitation in the Conclusion part (see Lines 306-308).
Minor comments:
- There is section "Result and discussion", but it does not introduce any real discussion (of potential issues, limitations, what should be pursued next, comparison with other studies, etc.).
Response: The section has been carefully revised (see lines highlighted in the section).
- I am not sure about the format of references (for example, listing the first names of authors). Please adapt to the journal requirements. Also please correct the surname in reference no. 34.
Response: The format of references has been adapted to the journal requirement. (see References)
Thank you very much again for your time!
Round 2
Reviewer 2 Report
The reviewer appreciates the effort of authors for the revision. Although the authors have revised the manuscript with respect to the comments, they may add most recently published article to the Sustainability journal regarding a new approach for identifying hazardous road locations (Chung, Y. Won, M., (2018) A novel framework for sustainable traffic safety programs using the public as sensors of hazardous road information, Sustainability, 10(11), 3892).
Reviewer 3 Report
Thank you for the revision. I accept the revised paper and recommend it for publication.
