Enhancing Travel Demand Forecasting Using CDR Data: A Stay-Based Integration with the Four-Step Model

Comments 1:

Line 117:
Please specify the exact year and month of the data used for the analysis.

Response 1: Thank you for your comment. We have revised the manuscript to specify the exact year of the data. The sentence now reads: (line 167)
“The dataset covers 400,000 callers over the month June 2013”

Comments 2:

Lines 119–122:
Please provide more details about the household travel survey data. Indicate the survey year, survey method, and response rate. Also, discuss how the differences between CDR-based data and the household travel survey were addressed.

Response 2: Thank you for this valuable suggestion. We have updated the manuscript to include additional details about the Household Visit Survey (HVS) (line 175-179) .Specifically, we now note that the survey was conducted in 2013 as part of the CoMTrans study, using a face-to-face interview method. The survey covered 44,000 households.

To address the differences between the CDR-based data and the household survey data, we introduced spatial aggregation techniques that map cell towers to administrative boundaries and used statistical validation at both district and DSD levels. Moreover, adjustments were made to account for discrepancies in temporal coverage and data granularity. These methodological steps are now described in the revised “Study Area and Data” and “Results and Validation” sections.

Comments 3:

Lines 150–160:
How were students treated in the analysis? For students, the most frequent location during office hours would likely be schools rather than workplaces. Please clarify this point.

Response 3: Thank you for this insightful observation. We agree that students' travel patterns differ from typical commuters, and that their frequent daytime locations are more likely to be schools rather than workplaces. In our analysis, we did not explicitly differentiate students from working individuals due to the anonymized nature of the CDR data, which lacks demographic labels such as age or occupation. However, frequent appearances at non-home locations during core daytime hours (10:00 am–4:00 pm) were interpreted as Work/Study locations. Therefore, for students, school locations would have been classified under this category. We have clarified this point in the manuscript.(line 233-237)

Comments 4:

Line 22:
The phrase “a promising alternative” should clearly specify: an alternative to what?

Response 4: Thank you for pointing this out. The whole introduction section is adjusted to cover more background.

Comments 5:

Line 42:
The statement “informed by en-route cell data” is unclear. Please clarify the intended meaning.

Response 5: Thank you for highlighting this. We agree that the phrase lacked clarity. We have revised the sentence to explain that route assignment is guided by identifying the sequence of cell tower connections (en-route cells) made during actual commuting trips, which reflect the likely travel path of users. (line 81-83)

Comments 6:

Figure 5:
The figure only provides a comparison based on relative values. A comparison using absolute values is also necessary to properly assess the differences

Response 6: Thank you for your valuable suggestion. In response, we have added a comparison of absolute Home-Based Work (HBW) trip counts at the district level, alongside the existing relative validation. (Table 6 and line 339-403)

Comments 7:

Figure 5:
Figures 5–9:
These figures are not appropriately referred to or discussed in the main text. Please ensure that all figures are adequately cited and interpreted in the manuscript.

Response 7: Thank you for bringing this to our attention. We have revised the manuscript to explicitly reference and interpret Figures 5–9 within the relevant sections..

Comments 8:

References:
Several journal names are missing in the reference list. Please ensure that all citations are complete and follow the appropriate journal style.

Response 8: Thank you for pointing this out. We have thoroughly reviewed the reference list and updated it to ensure that all citations include complete information, including journal names, volume, issue, page numbers, and DOI (where available). All entries have also been formatted to comply with the journal’s reference style.

Comments 1:

The literature review (2C) would benefit from more specific comparisons and concrete details, rather than general references, particularly regarding prior attempts at such integration. To strengthen the contribution claim, the authors could explicitly highlight the novel methodological elements in contrast to the closest existing studies.

Response 1: We appreciate the reviewer’s observation regarding the need for clearer differentiation from prior studies. In response, we have revised Section 2C to include more specific comparisons with closely related works, such as those by Gundlegård et al. (2016), Iqbal et al. (2014), and Fekih et al. (2020), focusing on their methodologies for OD matrix estimation and route assignment. We then explicitly contrast these with our study’s novel aspects, including:

·       The individual-based regularity model used to detect significant locations,

·       The caller-level OD estimation framework, and

·       The integration of en-route cell paths into route choice modeling using user-specific alignment probabilities.

These elements are now discussed in a dedicated subsection at the end of the literature review to highlight their distinctiveness. (line 131-162)

Comments 2:
Line 3 – 'Amal S. Kumarage' should be used instead of 'Prof. Amal S Kumarage

Response 2: "Prof. Amal S Kumarage" has been changed to "Amal S. Kumarage" in both the author line and correspondence section.

Comments 3:

Line 72 – the section header appears oddly at the bottom of the page

Response 3:

The awkward page break before the section header has been corrected to ensure logical section placement.

Comments 4:

line 94 – This subsection not marked as 'D.' like the previous ones

Response 4:

The missing subsection label ('D.') has been added for consistency with prior subheadings (A–C).

Comments 1:

The paper would benefit from a more explicit discussion on the limitations of the CDR-based approach in areas with low mobile phone penetration or data sparsity.

Response 1: We agree with the reviewer that mobile phone penetration rates can significantly influence the robustness of CDR-based models. In response, we have added a discussion in the Conclusion section acknowledging this limitation .(line 465-472)

Comments 2:
Although the methodology is clearly presented, a more detailed explanation of the assumptions behind the load-sharing correction method would enhance the reproducibility of the results..

Response 2: Thank you for this important suggestion. We have expanded the description of the load-sharing correction method in the Methodology section (line 197-210)

Comments 3:

Figures 5 and 6 demonstrate a promising correlation; however, the text could offer a clearer explanation of the implications of the outliers and how they might affect the interpretation of model accuracy.

Response 3:

We appreciate this observation. In the Results and Validation section, we have expanded the discussion following Figures 5 and 6. (line 392-400)

Comments 4:

The authors may wish to consider briefly discussing the potential privacy concerns related to CDR usage, even if anonymized, in order to address the ethical dimensions of big data usage.

Response 4:

We fully agree. A brief paragraph has been added to the Conclusion section to address the ethical and privacy considerations. .(line 475-479)

Comments 5:

t may be useful to provide a brief comparative table or schematic summarizing key differences and advantages between the traditional and CDR-based models.

Response 5:

Thank you for the helpful suggestion. We have added a comparative table (Table 2) summarizing the main distinctions between the traditional four-step model and the CDR-integrated approach.