Overall Description

This article presents seismic risk results from 18 earthquake scenarios in the Sabana Centro province, an intermediate hazard zone in Colombia. The 18 scenario events were chosen based on hazard disaggregation results on a site within the region of interest, for a 475 years return period hazard level. The epicenters of the scenario events were located within the region. The exposure model was gathered from previous studies, complemented with census data and remote surveys. Finally, the set of fragility curves for the considered typologies was gathered from previous studies. Results show that, on average, the occurrence of one of the scenario earthquakes might result in about $800 million USD of economic losses (about $1000 million USD when adjusted by social vulnerability) and about 20% of all the buildings collapsing.

The study is well performed and written. In terms of scientific merit, although the article does not provide new methods for seismic risk assessment, it provides novel results on the seismic risk faced by the Sabana Centro province. I have some comments that may help to improve the quality of the article (I have marked the most important issues that I comment). In particular, compared to the damage produced by previous earthquakes around the World, the numbers presented in this article seem to be too high, especially considering that: (1) these are mean values, not low-probability values (thus they are not even the “worst-case scenario”); and (2) the authors state that the risk results should be considered as a lower bound. Therefore, I strongly suggest a careful revision and discussion of the results.

Specific Comments (Individual scientific questions/issues)

1. Line 130: The authors state that the Bucaramanga’s seismic nest earthquakes occur at depths between 140 and 200km. However, in Figure 3 it is possible to observe several events at depths between 70 and 150km. I suggest the authors comment something about this inconsistency.

2. Line 144: The description of Figure 3 does not match what is observed in the figure. The authors state that Figure 3 shows: (1) close (distance < 75km) and shallow (depth < 70km) events; and (2) far (150km < distance < 200km) and deep (depth > 70km) events with magnitudes greater than 6.5. However, the figure presents many other events. For example, there are shallow events at distances larger than 75km with magnitudes lower than 6.5, deep events with magnitudes lower than 6.5, etc. The description and the figure should be consistent.

3. Line 155: The authors state that the population centroid is located within Tenjo. Moreover, they use this municipality throughout the article as a reference (e.g., Figures 2 and 3). However, given the map provided in Figure 6c and the population of each municipality provided in Table 1, it seems like the centroid should be located somewhere between Zipaquirá and Chía (probably within Cajica), which are the two municipalities with the largest populations.

4. Figure 4: What is the distance type of the disaggregation? Rupture distance? Epicentral distance? Joyner-Boore distance? Given the depth of the deep sources, the difference between the distance types might be important.

5. Table 2: The authors provide a list of the selected scenarios. A map showing these events might be useful for visualizing the epicenters with respect to the municipalities.

6. Table 3 and its description: Although the logic tree was proposed by Arcila et al. (2020), I suggest the authors provide a justification for the different weights to the Cauzzi et al. (2014) and Abrahamson et al. (2014) ground-motion models.

7. *Important comment* In Line 105, the authors state that “The majority of the building stock of the region is comprised of one- and two-story houses.” However, as shown in Table 4, the considered building typologies include buildings with either 1 or 4 stories. The question is then, how are two-story houses classified into this system? This question is especially important because it has been previously demonstrated that one- and two-story houses present a significantly different seismic behavior and therefore levels of damage and losses (see, for instance, Heresi and Miranda 2022). In particular, classifying two-story houses as one-story structures may result in a significant underestimation of the seismic risk of these two-story structures.

Heresi, P., & Miranda, E. (2022). Evaluation of relative seismic performance between one-and two-story houses. Journal of Earthquake Engineering, 26(2), 857-886.

8. Table 5 presents the main parameters of the considered fragility curves. As stated by the authors, these fragility curves were selected from different studies after a thorough literature review. Although this is perfectly fine, it has an important drawback that should be commented on: the final set of fragility curves comprise curves developed with very different methods (e.g., analytical vs empirical) which have very different reliabilities (e.g., generally speaking, empirical fragility curves developed after earthquakes have higher uncertainties both in the probability of damage and in the ground-shaking intensity). The authors are encouraged to discuss about the limitations and reliability of the considered fragility curves, taking into account the methods, the data, and the assumptions used to develop them. They address some of these issues in the Caveats and Limitations section, specifically the issue of fragility curves not being developed directly for Colombian structures and not having a uniform description of the damage states, but there are other issues that are missing in this section, as those previously stated in my comment.

9. *Important comment* Results show that a Mw5.95 event at Chía is expected to cause the collapse of more than 17% of the buildings in the region, and some level of damage in about half of the building portfolio. In particular, 6722 out of 14959 (about 45%) of houses made out of non-ductile unreinforced masonry with adobe block walls (1-story) are expected to collapse, according to the authors. Moreover, in Chía, more than 44% of the buildings are expected to collapse due to this Mw5.95 scenario. These numbers seem incredibly high for a Mw5.95 event at a first glance (even more when the authors state, in Line 441, that these estimates should be considered as a lower bound). Note that these are mean (i.e., expected) values, not low-probability values that might represent a somewhat “worst-case scenario” (or, in other words, somehow answer the question “how big may be the consequence if this earthquake occurs tomorrow?”). To put these numbers in perspective, we can compare them with the damage produced by the 2010 Haiti earthquake, Mw7.0:

• According to DesRoches et al. (2011), the 2010 Haiti earthquake damaged nearly half of the structures in the epicentral region.
• Eberhard et al. (2013) performed two field surveys of: (1) 107 structures in Port-au-Prince, where 30 (28%) of them collapsed and other 35 (33%) had enough damage to require repairs; and (2) 52 structures in Léogâne (closest population center to the epicenter), where 32 (62%) of them collapsed and other 16 (31%) had enough damage to require repairs.
• Rathje et al. (2011) performed a field survey of over 400 structures in Port-au-Prince. Of the 414 surveyed structures, 157 (38%) had significant damage (i.e., collapse or very heavy damage, EMS Grade 4).

Considering that the Haiti earthquake was not only 32 times larger in terms of magnitude, but also affected a more socially vulnerable country, it is expected that a Mw5.95 event in the region of interest would result in considerably less damage and losses, especially if we talk about mean values.

In terms of losses, in Figure 12 we can observe that some of the earthquake scenarios have a 20% probability of producing more than 50% of the total replacement cost as economic losses (about 40% of the GDP of the region!). Considering that these curves were computed neglecting the spatial correlation of ground motion intensities (comment about this below), this probability for such a high loss is extremely large. For perspective, the 2010 Chile earthquake, Mw8.8, produced an economic loss of about 14% of the GDP of the country at the moment of the event.

The previous remarks highlight the importance of comparing risk results from scenario events with previous events to put the numbers in perspective. I suggest the authors include comparisons like the ones proposed above, but also include other events, such as, for example, the 2020 Puerto Rico earthquake, Mw6.4. Moreover, in the Introduction, the authors mention two historical earthquakes that affected the region of interest, which may also be used to evaluate the reliability of the resulting damage produced by the considered scenario earthquakes. These comparisons would further support the risk results of the article.

DesRoches, R., Comerio, M., Eberhard, M., Mooney, W., & Rix, G. J. (2011). Overview of the 2010 Haiti earthquake. Earthquake Spectra, 27(S1), S1-S21.

Eberhard, M. O., Baldridge, S., Marshall, J., Mooney, W., & Rix, G. J. (2010). The Mw 7.0 Haiti earthquake of January 12, 2010: USGS/EERI advance reconnaissance team report. US Geological Survey Open-File Report, 1048(2013), 64.

Rathje, E. M., Bachhuber, J., Dulberg, R., Cox, B. R., Kottke, A., Wood, C., ... & Rix, G. (2011). Damage patterns in Port-au-Prince during the 2010 Haiti earthquake. Earthquake Spectra, 27(S1), S117-S136.

10. Table 10 presents the resulting SVI for the 11 municipalities of the region. Although the authors previously explain the variables involved in this index (Table 6), I have two comments about this:

• I suggest the authors provide more detailed information about how the index of each category is obtained. This explanation would improve the reproducibility of the reported results.
• There are many variables used for the SVI that are strongly correlated. For example, in the “Population” category, there are 7 variables, where, for instance, “Female population” and “Total population” are expected to be strongly correlated, unless the percentage of women varies significantly from one municipality to another for some reason. As the authors did not provide too much detail on how the index is computed, I’m not sure if they tested for collinearity between these variables, for example. We can even expect some correlation between different categories. For instance, municipalities with a high index in Economy will probably have also a high index in Infrastructure. These correlations might result in biased SVI’s when all the variables are considered.

11. *Important comment* As one of the limitations, the authors state that they did not consider the spatial cross-correlation when modelling the ground motion fields. However, they do not justify this arbitrary exclusion. For example, the OQ-Engine has models of spatial correlation already implemented, and therefore I do not see a good reason for neglecting it. As the authors correctly state, the inclusion of a spatial correlation model would increase the dispersion of the curves presented in Figure 12, making them more “realistic”. Thus, I suggest either including a spatial correlation model, or giving a strong justification for its arbitrary exclusion.

Technical corrections

1. Line 43: Change “7248 injured” for “7248 injured people” or “7248 injuries”.
2. There is an inconsistency in the use of thousand separators. For example, in Line 45 the authors state “… and 35000 buildings that collapsed…”, but then in Line 86, they write “resulted in 200,000 deaths”. In Table 1, the authors use thousand separators again.
3. Line 118: Review the word “gro”.
4. Line 158: The authors use the Quetame earthquake for defining the rupture geometry of the scenario events. I suggest adding an annotation in Figure 3, showing which one is the Quetame earthquake, for those of us who are not familiar with the historic seismicity of Colombia.
5. Line 249: There is an incomplete phrase.

The article addresses a relevant issue for a seismic country as Colombia; it gives relevant information about the seismic risk of the Sabana Centro providence. Nonetheless, there is no novelty on the article and key information is not given in the paper. Results from the selected scenarios indicate consequences of concern not well supported. My main concern regards the selected fragility curves: I find it complex to perform a risk analysis using fragility functions developed by different methodologies and as stated by the authors, with different limit states definitions. I believe this is one issue that requires additional explanation. For example, comparison between fragility functions is not presented. Do the set of curves behave as expected? Information given in Table 5 is not enough. The article should present figures that allows for a visually appreciation of the curves. A brief explanation of the methodologies uses for the curves’ development should also be included, as well as an opinion about how reliable the curves are. How can the authors explain that the number of collapse buildings is almost three times the number of buildings with extensive damage? Furthermore, the number of collapse buildings exceeds the number of buildings in any other damage state.

Additional comments:

• Why do all the scenarios are crustal shallow events? In the article it is mentioned that for SA (1.0s) there is an important contribution of subduction events. As the number of scenarios is important (18) some of them should be subduction events.
• The authors mentioned the use of population census data to infer the number of buildings added to the original exposure model of SARA. It is not clear why the authors did not use the census data to directly obtain the number of buildings. The 2018 Census provides relevant information that can be used to have a more precise number of buildings.
• How was the building typology assigned to the added buildings to the original exposure model?
• Which replacement cost did the authors use? The authors only mention that the cost is assigned according to the socio-economic levels, but it is not clear which cost was used and how was it computed: cost per area? Cost per building? It is suggested to include the replacement cost in Colombian pesos as the exchange currency fluctuates.
• It is not clear how the information of the base exposure model (SARA) was complemented with the information of the 6249 surveys. Furthermore, all these buildings belong to the same municipality. A description of the buildings characteristics of each municipality should be included.
• The authors mention that 8.24% of the stock are wood buildings. How does this information compare to the Census data? (The Census provides information about building’s wall material). In addition, the authors assigned a fragility function for wood buildings developed for Chilean buildings. Although the reference of the fragility functions used for wood has not yet been published, it is not clear that Colombian wood buildings have the same seismic behavior as Chilean wood buildings. A support for the use of Chilean wood fragility functions is needed.
• It is not clear why the authors use only two building heights: 1 and 4. Does the exposure model only comprise building with 1 and 4 stories? Or does the exposure model have buildings of several number of stories, but the authors decided to group them in just to building heights? Whatever the option, for a region where most of the buildings are low-rise buildings (as stated in the paper) a differentiation of number of stories is very important.
• Results should include the uncertainty as 1000 ground motion fields were generated and two GMPEs were used.
• The taxonomy MCF/DNO/H:1 is not correct as it is missing the lateral load resisting system.
• The taxonomy CR/LFINF/DUM/H:4 is used for buildings constructed using thin RC walls. This is not the original definition in the GEM taxonomy. It is suggested to use a different taxonomy.
• The taxonomy W/H1 is missing the information about the lateral load resisting system and the ductility level.
• It is not clear why the authors present mean values for the 18 seismic events. As each scenario has a different epicenter and different consequences mean values are not representative (results for each scenario should be presented by themselves). See Table 19 and Figures 10 and 13.
• The sentence of line 410 “One out of four buildings will experience extensive damage or collapse” is a strong conclusion that requires a big certainty to be written. I suggest the authors to revise the fragility curves of the masonry buildings (as most of the buildings are of this typology) and to compare the ground motion fields with the building damages to be sure that results are correct. Furthermore, as all the buildings form this typology are one-story buildings results should not be as bad as shown in the article.
• Line 440. The authors mention “the damage and losses estimates presented in this study should be considered as lower bound”. See the previous comments.
• Figure 4. Add a color scale. It is difficult to read the percentage associate to each bin.
• Figure 9. It is suggested to include the earthquake epicenter as well as a figure with the ground motion field generated by the event.
• Figure 11. Expected losses including SVI should be greater than the expected losses without SVI. This is not shown at Chía and Sopó. For the ease of understanding it is suggested to use the same color scale in both maps of the figure.  
• Figure 12. It is suggested to include the uncertainty in the figure.
• Table 2. Add the distance for the epicenter to the study area. Complement the information with a figure in which the epicenters are shown. As the events have an associated municipality, is the epicenter located at each municipality? How feasible is this? Results indicate important consequences that can be misinterpreted if the article does not mention the possibility that such events occur with epicenters in each municipality.
• Line 179. How does the “significant number of low-rise stiff buildings” relate to the selected crustal events?

The study deals with the analyses of various seismic risk scenarios for the Sabana Centro region in Colombia, located in the northern region of the capital city Bogotá, which concentrates important industrial facilities, educational facilities. It is an interesting study that follows the state-of-the-art procedures of scenario risk analyses (at least up to the computation of direct losses), for a region that hasn’t been studied before, and thus is a good contribution to the scientific literature that communicates the seismic risk in the country. However many important issues should be addressed before publication.

Here are some comments about important issues in the study, that hopefully will help improve its clarity, coherence, and thoroughness. It must be said that after doing the review, the reviewer saw that many of the comments and limitations of the study were included in the discussion section as further developments, however there are many that should be included in this study to make the results sound and representative, otherwise many of the presented results could be very misleading.

Section 2:

It would be interesting to include why the Sabana Centro region is of particular interest. In previous studies of the major cities, generated GDP or % of the population in comparison with the whole country were presented as reasons for the study of a particular city or region. Additionally, it is mentioned that it concentrates many economic and industrial activities, but at the end the analysis only deals with the residential building stock.

Major comment: Line 105: “The majority of the building stock of the region is comprised of one- and two-story houses” It would be good to show a reference with the numbers based on the 2018 Census for this. It is interesting that this is mentioned and still no two-story houses are considered in the analyses.

Section 3.1 Seismic hazard:

Line 171: “eighteen crustal events were selected from the catalogue to be used in this study” Does this mean that only ‘historical’ events included in the catalogue were included? No new possible events from the event-based tables from the PSHA model?

Table 2: Include column with the distance to the population centroid taken as reference point in the disaggregation to be able to compare this selection of scenarios with the disaggregation graphs presented in Figure 4 and their representativeness in the overall 475-years return period hazard in the region. It would also be important to mention the contribution of each tectonic regime in the overall 475-years return period hazard to be able to discard the long subduction events.

Given the proximity of the events based on the disaggregation, directivity effects should be considered for some of these events. As further seen in the discussion this was not considered but it would be good to mention something about how to account for it and not leaving it till the discussion as a further development.

Section 3.2 Exposure model for the residential building stock:

Was the replacement cost updated to 2021? 2022? In which way was this done if indeed it has been updated? If not, it should be done and explained.

Major comment: Were the new inflated exposure building numbers (based on population as proxy) in any way compared to the dwellings or building numbers reported in the 2018 Census for these regions?

Major comment: Table 4 only considers unreinforced masonry of 1 storey, which is known to be less vulnerable than the unreinforced masonry of 2 stories, which is actually more common in many urban areas. This typology should be included (assuming something probably based on census data or the surveys), as in the region it is very common to find 2-storey, in some cases more than single storey houses (as previously mentioned in the study also). In the current version, the study may be underestimating the losses in this sense.

Section 3.3 Physical vulnerability of residential building stock to seismic ground shaking

Major comment: Chilean wood structures are known to be in better shape than those in Colombia, and they consider a different type of construction technique. The same goes to the curves used in HAZUS, which are not as representative of the local conditions and may be underestimating the risk. If different vulnerability curves are going to be used a more thorough explanation of the limitation of using them should be included and some kind of calibration or validation would be needed.

Major comment: Table 5 needs a clarification of what each curve considers in each of the damage states. If the vulnerability model is considering a unique consequence model, there may be incompatibility between the loss ratios of the derived vulnerabilities, as each one considers each damage state in a specific different way. This is one of the main issues when combining vulnerability functions from different sources and is particularly true given the damage results of the studies are shown considering these categories of the damage states. Additionally there are no validations or calibrations on the reliability of the selection of the curves. The reviewer saw this mentioned in the further developments of the discussion, but it is something thaat should be included in the computation of the vulnerability curves here in some way, for the results to be coherent.

Major comment: Given the exposure is not considering separately the 2-storey housing, there are no vulnerabilities for 2-storeys considered, even when it is more common in the urban environment than the single storey houses. This typology should be included.

Section 3.4. Social Vulnerability (SV)

Major comment: One of the main criticisms of the paper is in the consideration of the social vulnerability index as a percentage increase using this expression (1+SVI). As stated in the study “The min-max normalization was used to standardize the SV indicators from zero to one to estimate the SVI per municipality. Higher scores indicate more socially vulnerable municipalities, and lower scores reflect less vulnerable ones. Then, the indicators were integrated by summing them with equal weight, as followed in Contreras et al. (2020c). The resulting SVI index is therefore used to adjust the percentage of economic losses with respect to the costs presented by the building inventory, i.e., multiplying them by (1+SVI) (Carreño et al., 2007).” The problem with this is that there is no analysis done on the significance of the variables included within the study and no way to know if there are variables that shouldn’t be included and if anything is counted double. Additionally, considering this index as a “percentage increase” is extremely misleading. If there was a way to correlate the SVI of each variable in economic terms to the direct economic loss, then this could be done. But this is not done and there is no parametric study or anything else to validate any of the assumptions. This SVI cannot be considered a percentage unless there is backup data validating this. This has been done also in fatality modelling where the models that are presented in any publication are previously calibrated and validated with data from historic events. Moreover, considering previous events reporting post-loss amplification that include costs from the response and recovery stages in some disasters, it has been shown that numbers over 30-40% are almost non-existent (What is demand surge? Olsen and Porter 2010), while this study mention cases with increases of up to 60%. There may be a problem with the explanation of the methodology, but as it is right now it is very difficult understand how it can relate to economic losses, especially direct physical losses after an event.

(These limitations are also afterwards mentioned by the authors in the discussion, but it is a MAJOR limitation of the inclusion of the SVI methodology in the results in this study, as there is no validation or calibration of any kind for the methodology)

Table 9 numbers are misleading as a direct non-weighted average of the 18 scenarios is not probabilistically and statistically sound. It should consider the contribution of each event, otherwise the less probable events are counted in the same way as the more probable ones. In this way, as when computing AAL from a probabilistic analysis, the contributions should consider the probability of occurrence of each scenario. After saying this, it is advised not to present this table and instead present one with the analysis of each scenario done separately as in a deterministic approach, unless it is possible to demonstrate that the 18 scenarios included account for the 100% of the 475 years return period loss and a weighted average is calculated based on the contribution of each. It is advised to present each scenario on its own reporting the probability of occurrence of this event. As a good way to do a calibration it would be interesting to check the return period of the loss that is being simulated and compare what similar events in the PSHA are also reporting (from the event loss table).

Section 4.4:

As stated previously, to present these analyses, it would be important to show the contribution of the 18 scenarios to the total hazard in the region (based on the disaggregation results). If not Figure 13 is misleading, considering that it says: “Within the municipalities, the mean percentage of losses is presented with respect to the total expected losses in the region”.

“The economic losses experienced by a province due to an earthquake depends on the event’s epicenter as it is depicted in Figure 12.” Figure 12 does not show in any way anything regarding the epicenter. It may not be only the epicenter but also the Mw for each event the cause of the differences, so this statement is not provable from the Figure. Delete it.

Major comment: Presenting only the mean or median results in this kind of analyses is not recommended, given the amount of uncertainties that are included within the whole process (selection of GMPE's, weightings, exposure, taxonomy assignment, selection of vulnerability curves, assignement of these vulnerability curves, great number of ground motion fields). This is a major limitation and concern for the paper given there are no validations with other studies or numbers to establish if the assumptions are reasonable. Additionally results seem to be in the high side when compared with what has been reported in literature. 

Discussion:

Just until this section this is stated: “The simulations of eighteen seismic scenarios with a return period of 475 years show that half of the building stock will experience some degree of damage”.  How was this ‘475 years’ return period calculated? Even when the disaggregation was done for the ‘475 years’ return period, how can it be confirmed that the 18 scenarios add up to the 100% contribution for the hazard for this return period? Either way this statement should be included in some way in previous sections and not only until the discussion.

Effects of SV:

Major comment: The main criticism for this approach is also stated by the authors in this sentence: “First, as not all social aspects exert equal effect after an earthquake, it is necessary to develop a weighted approach to best estimate a more realistic SVI for earthquake events. A second improvement required is to devise a better way of estimating the economic impact of social vulnerability. One potential approach is to generate a database of past earthquakes with different consequences that include the economic costs.” These are not needed future improvements but major limitations of the proposed approach. In this kind of analyses, as when performing a linear regression, it is important to avoid counting double and establish the significance of each variable within the analysis, if not this could be overestimating the vulnerability and losses in the region considerably. Also, using the min and max approach is very subjective as many variables as unemployment and poverty are tempered with by local organisms. This kind of indicators are good to compare and prioritize actions within regions but cannot be used in the way they are presented in this study to increase direct physical losses.

Minor modifications:

Line 118: Typo: This GROWTH counts for 64% of the total population of the region

Line 248: Incomplete sentence: “The number of masonry buildings represents the 88.61% of the total buildings in Sabana Centro, whereas those of concrete and wood represent 3.16% and 8.24%, respectively, with.”

Line 261: Repeated to “In the absence of specific curves locally developed for the Sabana Centro province, fragility curves were selected to to represent these structures”