Investigating the Potential of Using POI and Nighttime Light Data to Map Urban Road Safety at the Micro-Level: A Case in Shanghai, China
, Yufan Liu 1,2, Jinzi Wang 1,2, Xingjian Qian 1,2, Xizhi Zhao 3, Jianping Wu 1,2, Bin Wu 1,2, Shenjun Yao 1,2,* and Lei Fang 4,*Round 1
Reviewer 1 Report
The paper has good quality and a good method is applied but need the following comments to be adddressed before publication.
1- Line#20 in the abstract section using two present participle tenses one after each other may not be very correct, plus changing spells are not written correctly.
2- Have you considered different type of vehicles in your study such as trucks, SUVs, and etc., since you have both urban and expressway type of roadways?
3- In the introduction section, it is recommended to add some results out of speed limit changes paper, which is included in "evaluating the safety impacts of increased speed limits on freeways in Kansas using before-and-after study approach" paper, and it is already published in sustainability journal. Since, the CMF is estimated in that journal, is there any way to estimate CMF in your regression model like that journal named above? Also, in another paper entitled "impact of advertising signs on freeway crashes within a certain distance in Michigan", the regression model was used and it is recommended to add the results of that paper in your introduction as well to support other types of regression models such as NB model.
4- In line 85, in the introduction section it is written that you have only considered property damaged only and injured crashes. Wasn't it possible to consider fatal crashes in your model too, which is more important than injury crashes?
5- The method named in your paper does need some more clarifications on how the method separated the results of urban crashes versus expressway crashes. How?
Author Response
The paper has good quality and a good method is applied but need the following comments to be addressed before publication.
Response: We would like to thank the reviewer for acknowledging the contribution of the method. The responses to the reviewer’s valuable comments and suggestions are as follows:
1- Line#20 in the abstract section using two present participle tenses one after each other may not be very correct, plus changing spells are not written correctly.
Response: We feel sorry for the confusion. For clarification, the sentence has been revised as “By using a district of Shanghai as the study area, this research employed the two types of urban sensing data to map vehicle-pedestrian and vehicle-vehicle collision risks at the micro-level by road type with random forest regression (RFR) models.”(see lines 20-23)
2- Have you considered different type of vehicles in your study such as trucks, SUVs, and etc., since you have both urban and expressway type of roadways?
Response: We would like to thank the reviewer for the valuable suggestions. Although we are not able to categorize the vehicle types from our database, we do believe that the issue is quite worthy of being investigated if detailed data on vehicle type is available. We have added some lines in the final section as follows:
“In addition, this study roughly categorised the traffic collisions into vehicle-vehicle and vehicle-pedestrian collisions because of the data availability. However, the influences of POI and NTL factors on traffic collisions involving different types of vehicles may differ significantly. If detailed traffic collision data on vehicle types are available, more research efforts can be focused on the extent to which the models are sensitive to the types of vehicle-vehicle collisions.” (see lines 342-346)
3- In the introduction section, it is recommended to add some results out of speed limit changes paper, which is included in "evaluating the safety impacts of increased speed limits on freeways in Kansas using before-and-after study approach" paper, and it is already published in sustainability journal. Since, the CMF is estimated in that journal, is there any way to estimate CMF in your regression model like that journal named above? Also, in another paper entitled "impact of advertising signs on freeway crashes within a certain distance in Michigan", the regression model was used and it is recommended to add the results of that paper in your introduction as well to support other types of regression models such as NB model.
Response: The two papers are very interesting and well-written. We have included them into our manuscript (see lines 51-53, 55-57, reference 9 & 19).
As the purpose of this research is to explore the ability of predicting vehicle-pedestrian crashes, it concentrates on the accuracy of the prediction. That’s why we employed a machine learning method. Since we did not consider much on the exact relationship between variables and collisions, it is difficult to estimate CMF in this research. Next step, we will make more efforts on the CMF by investigating detailed influence caused by each feature in the model. We believe that the non-linearity relationship captured by the random forest models may tell more interesting stories. We have elaborate a bit more on this in the final section as “It should be pointed out that the purpose of this research is to explore the potential of using POI and NTL data to map traffic collisions. It placed emphasis on the prediction of traffic collisions on urban roads and did not consider the detailed influence caused by each feature in the model. Hence, it is difficult to obtain rules such as crash modification factors mentioned in previous studies. As investigating impacts of explanatory variables on traffic collisions can help policy-makers to conduct safety improvement programs, future research works could be dedicated to the association of the POI and NTL features with traffic collisions.”(see lines 333-339)
4- In line 85, in the introduction section it is written that you have only considered property damaged only and injured crashes. Wasn't it possible to consider fatal crashes in your model too, which is more important than injury crashes?
Response: We would like to apologize for the oversight. The traffic collision database includes property damaged only, injured and fatal crashes. We have already made the revision in the manuscript. (see lines 99-100)
5- The method named in your paper does need some more clarifications on how the method separated the results of urban crashes versus expressway crashes. How?
Response: The fitness of the models are evaluated by using OOB scores. All the models of collisions on expressways have very low OOB scores (no more than 0.2) while the models of collisions on other types of urban roads have relatively high OOB scores. For more clarification, we have added some lines on OOB scores (see lines 193-198, lines 247, 249,256).
Reviewer 2 Report
The article is interesting and it describes a method including two promising variables for predicting urban road crashes, by considering road types.
The article is well written and structured. However, I have some comments for improving it:
I think that, given the methods used in the study, differences between expressways, arterial, secondary and branch roads should be defined in detail. Do they belong to number of lanes, territorial function, access control, speed limits...? In my opinion, there are some aspects of methods which should be better clarified. For example, variables in Eq. 1 could be better explained through a figure which may help their meaning. Terms in Fig. 3 should be better explained as well. OOB scores, which are also used as performance measure in Table 5, should be better defined and explained. The Getis Ord Statistic should be defined. The first part of the Methods section could be difficult to follow, if the RFR are not defined yet. Moreover, the first part of the Results and discussion section includes threshold used for assessing the correlation coefficient. This part could be moved in the methods section. I think that the difference between the results belonging to the two periods (6-18, 18-6) inquired should be more thoroughly discussed, given that they are mentioned in the research questions.Author Response
The article is interesting and it describes a method including two promising variables for predicting urban road crashes, by considering road types.
The article is well written and structured. However, I have some comments for improving it:
Response: We would like to thank the reviewer for acknowledging the contribution of the method. The responses to the reviewer’s valuable comments and suggestions are as follows:
1- I think that, given the methods used in the study, differences between expressways, arterial, secondary and branch roads should be defined in detail. Do they belong to number of lanes, territorial function, access control, speed limits...?
Response: Thank you very much for the valuable suggestions. We have inserted Table 1 that describes the differences between expressways, arterial, secondary trunk and branch roads.(see Lines 87-88, Table 1)
2- In my opinion, there are some aspects of methods which should be better clarified. For example, variables in Eq. 1 could be better explained through a figure which may help their meaning. Terms in Fig. 3 should be better explained as well. OOB scores, which are also used as performance measure in Table 5, should be better defined and explained. The Getis Ord Statistic should be defined.
Response: We have taken the following actions：
1) inserted a figure to explain the NKDE. (see Figure3, lines 136-137)
2) added some lines on the correlation coefficient matrix . (see lines 221-223)
3) presented definitions of OOB score and OOB error in the RFR section. (see lines 193-198)
4) added formula (7)(8)(9) to explain the Getis Ord Statistic (see lines 309-313).
3- The first part of the Methods section could be difficult to follow, if the RFR are not defined yet.
Response: We have revised the paragraph (see lines 128-133).
4- Moreover, the first part of the Results and discussion section includes threshold used for assessing the correlation coefficient. This part could be moved in the methods section.
Response: The description of the thresholds on both correlation coefficient and VIF have modified and moved to subsection 3.2 .(see lines 176-178)
5- I think that the difference between the results belonging to the two periods (6-18, 18-6) inquired should be more thoroughly discussed, given that they are mentioned in the research questions.
Response: We have discussed the result in the revised manuscript as “When comparing models of daytime and nighttime, one may observe that vehicle-pedestrian collisions happening on daytime were better modelled than those occurring on nighttime, while there was no significant difference between models of vehicle-vehicle collisions. It implies that there might be more complex risk factors influencing pedestrian safety at night.” (see lines 286-289)
”
