The study presented in this paper explores the impact of migration on seismic fatality risk and introduces an interesting approach to seismic fatality risk modeling through the employment of a radiation model that accounts for migration patterns. The following comments are intended to provide suggestions to increase the impact of the paper and improve its clarity for the reader.

1. Considering the authors’ selection of Sa03 as the intensity measure, is it correct to assume that the study area mainly consists of low-to-midrise buildings? If available, can the authors provide further information regarding the building typologies in the studied area, perhaps in terms of construction material and building height?
2. Do the authors think that the daily time-based fluctuation could be incorporated into their model? For example, higher occupancy in residential buildings during nighttime compared to daytime could impact the results (s. FEMA P-58/BD-3.7.8 Casualty Consequence Function and Building Population Model Development). Please comment on this possibility.
3. The authors define the incidence rate ratio as the factor by which the fatality rate multiplies when that variable increases by one unit, assuming all other variables remain constant. For some of the variables, it is not clear how this increase is calculated. For instance, assuming the fault-impacted area is 0.02, does an increase by one unit mean the area becomes 0.03? Please provide details.
4. Further, the reviewer questions the rationale behind the “increase by one unit” approach, as many variables have different units. It might be more effective to increase these variables by the same percentage rather than by one unit. For example, regarding spectral acceleration Sa03, increasing it by 1g would likely cause many buildings to collapse.
   
   
   
   Consider the following scenario: On site A, assume a low-rise building with a spectral acceleration Sa03 at collapse of 0.6g (capacity). This means any Sa03 larger than 0.6g would cause the building to collapse (ignoring record-to-record variability). If the observed Sa03 on site A is 0.1g and we increase it by one unit, making it 1.1g, the expected fatalities due to collapse should not differ between 0.6g and 1.1g, as the fatality rate saturates at collapse, which occurs at 0.6g. Therefore, it may be more beneficial to increase the variables by the same percentage rather than by one unit for ease of comparison. Please comment.
5. Does the number of fatalities saturate at the spectral acceleration level associated with collapse? If no, it is suggested to mention this as a limitation of the study.
6. The seismic hazard clearly has a strong influence on the number of fatalities. However, regarding the incidence rate ratios in Table 2, the high ratio associated with Sa03 may be due to the effect discussed in Comment 4. As a result, a direct comparison between variables based solely on their incidence rate ratios (computed by the “increase by one unit” approach) may be challenging.

This paper investigates the 1999 Chi-Chi earthquake in Taiwan across 4,052 neighborhoods, employing Poisson regression and maximum likelihood estimation to predict incidence rate ratios and determine the significance of various covariates related to hazard, exposure, vulnerability, and migration patterns. The authors used the radiation model to estimate migration patterns and examine their effect on seismic risks and fatalities. The topic is highly relevant, the study is well-constructed, and the methodology is both innovative and thoroughly explained. However, I have a few comments for the authors to consider:

Introduction:

• Line 31: Replace “natural disaster” with “natural hazard” for accuracy.
• Line 139: Consider replacing “resource scarcity (vulnerability)” with just “vulnerability” to avoid conflating different concepts, as vulnerability refers to the potential for loss, whereas resource scarcity is a distinct term.

Methods:

• Line 158: Update to “In this study, we adopted the neighborhood geographic unit to estimate the socio-spatial effects of …” to improve readability and relevance. Additionally, I would suggest incorporating the risk formula into either the Data or Models section for clarity.
• Consider adding a map to illustrate your study area at the neighborhood level. You could include a boundary in Figure 2 to enhance understanding.
• Line 180: Add "n=4502" in Table 1 for clarity.
• Reorganize section “2.2.4 on Independent Variables: Vulnerability” to follow the order presented in Table 1. For example, in Line 214, specify that “data was collected from…” and again in Line 216 for the household income part. Merge the “proportion of indigenous population” information with Line 236 to improve readability and flow.
• Line 259: Equation 1 does not include “r_ij” as indicated in lines 258-261.
• For Equations 2 and 3, please include the left-hand side of the equations for completeness.

Models:

• Some variables have positive effects (e.g., high income), while others have negative effects (e.g., income disparity). Consider adding a more detailed explanation of how these effects are demonstrated within your models.
• Regarding the migration pattern calculated using the radiation model, it would be beneficial to discuss the model's precision and accuracy. This will help clarify the model's strengths and limitations in predicting real-world migration flows and identifying areas where the model may over- or underestimate movements. Such a discussion could guide future refinements or improvements in its application.

Results:

1. Page 14: Please specify the exact number of fatalities in suburban or urban fringe areas.
2. Include the boundaries of the study areas in Figure 3. The text on vulnerability in Figure 3 is confusing.
3. Lines 355-356: The discrepancy in the effects of population over age 64 could be due to higher resolution data, but another potential explanation is that older adults may have more experience and unique perceptions in earthquake situations, which could influence the results. Consider discussing this in more depth.
4. Ensure consistency throughout the paper when reporting significance (e.g., use “P” versus “p”).