Urban Vulnerability Assessment for Pandemic surveillance: The COVID-19 case in Bogotá, Colombia

: A Pandemic devastates the life of global citizens and causes signiﬁcant economic, social, 1 and political disruption. Evidence suggests that Pandemic’s likelihood has increased over the past 2 century because of increased global travel and integration, urbanization, and changes in land use. 3 Further, evidence concerning the urban character of the Pandemic has underlined the role of cities in 4 disease transmission. An early assessment of the severity of infection and transmissibility can help 5 quantify the Pandemic potential and prioritize surveillance to control of urban areas in Pandemics. 6 In this paper, an Urban Vulnerability Assessment (UVA) methodology is proposed. UVA investigates 7 the possible vulnerable factors related to Pandemics to assess the vulnerability in urban areas. A 8 vulnerability index is constructed by the aggregation of multiple vulnerability factors computed on 9 each urban area (i.e., urban density, poverty index, informal labor, transmission routes). UVA provides 10 insights into early vulnerability assessment using publicly available data. The applicability of UVA is 11 shown by the identiﬁcation of high-vulnerable areas where surveillance should be prioritized in the 12 COVID-19 Pandemic in Bogotá, Colombia.

those models. Different methodologies allow to transform the experts' knowledge into a mathematical 48 language (i.e., Analytic Hierarchy Process), but these methodologies have some limitations such as 49 a-priori knowledge, expert bias, or hierarchical criteria [19,20]. 50 In this paper, a conceptual framework for Urban Vulnerability Assessment (UVA) for Pandemics 51 is proposed. UVA conducted a comprehensive review of relevant literature to identify vulnerable 52 factors influencing Pandemics. These factors are used to generate an index that allows us to identify 53 and rank potentially vulnerable urban areas. The rank is built using Borda's count aggregation method, 54 which does not need experts knowledge nor additional parameters for the construction of the ranking. 55 Then, the vulnerability rank is associated with a vulnerability index, i.e., a higher rank indicates 56 higher vulnerability. UVA is tested in the current COVID-19 Pandemic in Bogotá, the crowdest city

64
This paper is divided into four sections. Section 2 develops the methodology of Urban (a) Literature review (b) Find distribution (c) Group by similar

Borda Count (d) Aggregation method (e) Vulnerability index
Vulnerable factors V={V 1 , . . . ,V M } Probability distribution of V f V (x)=( f V 1 (x), . . . , f V M (x)) T Similar groups C C={C 1 , . . . ,C L } Unique rank for C R(C 1 )=2, . . . ,R(C L )=1  The search retrieved studies for which the study's title, abstract, or keywords indicated the study 82 examined a type of vulnerability in Pandemics. Then, a manual assessment is made for every study 83 against eligibility criteria: The study provided a quantitative or conceptual analysis of a type ( is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted December 30, 2020. ;https://doi.org/10.1101https://doi.org/10. /2020   is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted December 30, 2020. ; https://doi.org/10.1101/2020.11.13.20231282 doi: medRxiv preprint range 0 (best) to 1 (worst), see Figure 1(b). Different normalization methods exists in the literature [29].
anomalies, orthogonality, linear dependency). One solution is to build an estimation of the Probability 104 Density Function (PDF) of the data, and then transform it via its Cumulative Density Function (CDF), 105 so intervals with higher likelihood of containing data are assigned to higher portion of the normalized 106 interval [0,1]. This is call probability integral transform [30]. We estimate the PDF f V k at specific spatial 107 unit x using the Kernel Density Estimation (KDE) method.
where K is the kernel (a non-negative function) and λ is the smoothing parameter called the bandwith.

109
Then, to normalize the raw data at spatial unit x over the range 0 (best) to 1 (worst) in the k-th 110 vulnerable factor, the probability integral transform is applied.
where F V k is the CDF of the k-th vulnerable factor. of the Cluster C i in the rank R k , and w R k the weight assigned for the rank R k . A new aggregated value 131 of ranking or the i-th Cluster is defined as: The rank built using Borda's count aggregation method does not need experts knowledge nor 133 additional parameters for the construction of the ranking (i.e., the weights could be equal for each 134 rank). However, if the rank has to be weighted in some way, the method allows assigning this 135 weight for each rank. For the vulnerability level assignation problem, a set of ordered lists or ranks 136 is calculated using the centroid of each cluster C i . Therefore, Borda's count aggregation method is 137 used, that is, vulnerability factor ranks R k were made sorted the values of each centroid for the M 138 vulnerability factors. Next, these M ranks were combined using Borda's method to construct the 139 aggregated vulnerability rank, see Figure 1(d).

140
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted December 30, 2020. ; https://doi.org/10.1101/2020.11.13.20231282 doi: medRxiv preprint Version December 30, 2020 submitted to Sustainability 6 of 14 Finally, the vulnerability rank is associated with a vulnerability index, i.e., higher rank indicates 141 higher vulnerability, see Figure 1(e).    is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The proposed domains are used for the convenience of the reader and could change depending on the data analysis made in the geographic area. It helps the reader to associate vulnerability factors related. These domains do not influence the process of assigning vulnerability to a spatial unit.
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted December 30, 2020. ; https://doi.org/10.1101/2020.11.13.20231282 doi: medRxiv preprint

Vulnerability analysis 199
To understand the distribution of the vulnerability factors over the Urban sectors, the raw data for is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted December 30, 2020. ;https://doi.org/10.1101https://doi.org/10. /2020

Vulnerability index 209
To provide better response for vulnerability assessment, UVA generates three different 210 vulnerability index to assess vulnerability in different ways (depending on the k partitions).

211
Vulnerability index I have three different clusters (k = 3) to get a vulnerability index from low 212 to high (i.e., low, medium, high). Vulnerability index II has five different clusters (k = 5) to get a 213 vulnerability index from lowest to highest (i.e., lowest, low, medium, high, highest). And, Vulnerability 214 index III has ten clusters (k = 10) to get a vulnerability index from 1 to 10. Figure 4 shows the different 215 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted December 30, 2020.
Class R(C i ) C 1 10 C 5 9 C 7 8 C 9 7 C 10 6 Class R(C i ) C 4 5 C 8 4 C 3 3 C 2 2 C 6 1 (c) Vulnerability index III Figure 4. Three different proposed vulnerability indexes assess the urban Vulnerability. Vulnerability I with k = 3 (left panel), Vulnerability II with k = 5 (midst panel), Vulnerability III with k = 10 (right panel). For each Vulnerability index: clusters generated using the k-means method (top), its corresponding centroid values for each vulnerable factor (middle), and the unique rank generated using the Borda's count method (bottom) 6 . vulnerability index (i.e., Vulnerability index I with k = 3 (3 clusters), Vulnerability index II with k = 5 216 (5 clusters), Vulnerability index III with k = 10 (10 clusters)).

217
After getting the clusters for each vulnerability index (showed in Figure 4 -top), the centroids of 218 the clusters (showed in Figure 4 -midst) are getting and used to sort from higher to lower values the 219 each vulnerability factor. These vulnerability factors sorted (by the centroids) are assumed as vulnerable 220 ranks that would be used for the analysis. Then, to aggregate the 14 ranks (one for each vulnerable 221 factor in Table 2) the Borda's count aggregation method build a unique vulnerability ranking for each 222 cluster (showed in Figure 4 -bottom).

223
A vulnerability index is assigned for each cluster based on the ranking (i.e., higher rank indicates 224 higher vulnerability). Figure 5 shows the final vulnerability index for the three different vulnerability The class identifier 1, . . . , k for the clusters of the vulnerability indexes with different k partitions (k = 3 left, k = 5 medium, k = 10 right) does not be the same between models (i.e., the class identifier variate from index to index).
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted December 30, 2020. ;https://doi.org/10.1101https://doi.org/10. /2020  (c) Vulnerability index III Figure 5. Vulnerability indices generated using UVA for the current COVID-19 Pandemic in Bogotá, Colombia. Vulnerability index I has 3 levels from low to high (left); Vulnerability index II has 5 levels from lowest to highest (middle); and Vulnerability index III has 10 levels from 1 to 10 (right).   lowest-highest, and 1-10) to Pandemic surveillance. Surveillance is of primary importance to monitor 245 the burden of disease and will give both local authorities and the global community a chance for a 246 quick response to public health threats.

247
Our work has demonstrated how high-vulnerable level contexts contribute to increasing the 248 impact and spread of the disease at different geographic levels. This approach also enables the design 249 of comprehensive plans, implemented at the city scale, for addressing urban vulnerability at national, 250 regional, and provincial scales. Further, the results allow to build evidence for planning, modeling, 251 and epidemiological studies to better inform the public, policymakers, and international organizations 252 to where and how to improve surveillance, response efforts, and delivery of resources, which are 253 crucial factors in containing the COVID-19 Pandemic. It must concern the spatial inequality problems 254 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted December 30, 2020. ; https://doi.org/10.1101/2020.11.13.20231282 doi: medRxiv preprint  The real COVID-19 cases are available in the publicly available dataset provided by Observatorio de Salud de Bogotá. The map shows the concentrations of the cases in 1000 meters on December 30, 2020.
. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted December 30, 2020. ; https://doi.org/10.1101/2020.11.13.20231282 doi: medRxiv preprint report of the Harvard-LSHTM Independent Panel on the Global Response to Ebola. The Lancet 2015, 274 386, 2204-2221. 275