A Comprehensive Vulnerability Assessment of Rincon’s Neigh

EXPOSURE AND ADAPTIVE CAPACITY TO FLOODS: A Comprehensive Vulnerability Assessment of Rincon’s Neighborhoods

Saylisse Dávila, PhD

Associate Professor Department of Industrial Engineering, University of Puerto Rico at Mayagüez

Lourdes Medina, PhD

Associate Professor Department of Industrial Engineering, University of Puerto Rico at Mayagüez

OBJECTIVES

Table 1. Status of Project Objectives

1- Advance the understanding of the vulnerability to floods of Rincón’s neighborhoods.

Accomplished: 95%

Outcomes: Conceptual frameworks for the vulnerability dimensions of exposure and adaptive capacity were developed: exposure, sensitivity, and adaptive capacity. Feature selection methods and analyses on internal consistency were used to identify redundant vulnerability attributes in order to further simplify the vulnerability assessment and make it more appealing for emergency responders.

2- Characterize Rincón’s flood-prone zones in terms of exposure, sensitivity, and adaptive capacity.

Accomplished: 85%

Outcomes: Numerous data collection activities and interviews have been carried out in order to explain Rincón’s vulnerability to floods and use this data to fit the overall vulnerability index based on: exposure, sensitivity, and adaptive capacity. Data collection, data processing on the vulnerability attributes selected for the model, and model fitting for the adaptive capacity index is currently taking place.

3- Support emergency responders and strategic planners in the development of mitigation, response, and recovery strategies for floods.

Accomplished: 85%

Outcomes: Multiple meetings were conducted with emergency responders, while encouraging their engagement in the project. In addition, a tsunami pedestrian evacuation model was implemented in QGIS using the r.walk tool. QGIS model was augmented with custom models that address: evacuation response, population distribution, reaction time, and fatigue factor.

METHODS

The focus of the work in Year 1 was to generate tools and methodologies that would make vulnerability indexes more appealing to practitioners. These initiatives includes: (1) a new methodology for removing the number of redundant attributes from a vulnerability index, (2) an Excel-based application that automatically can estimate a vulnerability index, (3) a mobile application to capture and document geo-referenced data of the infrastructure of Rincón, (4) the development of a conceptual framework for a pedestrian evacuation model, and (5) the development, execution, and analysis of questionnaires used to generate inputs for the vulnerability index as well as the pedestrian evacuation model.

The work for Year 2 has focused on the development of the vulnerability index and the implementation of the tsunami pedestrian evacuation model. The initiatives for the remaining time include: (1) implementation of an adaptive capacity index, (2) estimation of flood height as a proxy for sensitivity, and (3) data collection and processing for the vulnerability index.

RESULTS AND FINDINGS

Adaptive Capacity Survey

During the summer period of the academic year 2014-2015, a collaborative research experience between Texas A&M University and UPRM funded two research students for the summer period. These students focused on developing an instrument to quantify the vulnerability dimension of adaptive capacity in terms of its three main constructs: mitigation (before the emergency), response (during the emergency), and recovery (after the emergency). The purpose of this survey was twofold. First, it is intended to unveil the structural patterns defining the response of individuals to a potential tsunami threat. Second, it was intended to gather data to measure the dimension of adaptive capacity. Overall, the resulting instrument was a survey with a series of questions that capture the pedestrian’s demographical information as well as a pedestrian’s mitigation, response, and recovery attributes. The document was submitted to the Institutional Review Board on July 7, 2015 and was approved on August 20, 2015.

During the fall semester of 2015, substantial changes were made and a revised version of the questionnaire was resubmitted and approved by the IRB on September 1, 2015. The data collection process started immediately after the revised version was approved. Most of the respondents correspond to individuals that were interviewed by the project’s staff in the data collection visits to Rincón and surrounding territories that took place between September 2015 and May 2016. Other respondents correspond to individuals that were either interviewed while on the phone with one of the members of the project’s staff, and an even smaller portion of the sample corresponds to individuals that completed the survey online. To date, 192 respondents have completed the survey. This sample was obtained with stratified sampling where age group, residence status, and gender were used as strata.

Adaptive Capacity Index

The conceptual framework for the flood adaptive capacity index is described in Figure 4. The breakdown of variables corresponds to the three dimensions of adaptive capacity: mitigation, response, and recovery. Mitigation corresponds to what individuals do to prepare before the onset of the natural hazard. Response refers to how individuals deal with the emergency as it is unveiling. Lastly, Recovery relates to how individuals overcome from the onset of the natural hazard. The variables proposed in this index correspond to an extensive literature review on vulnerability to natural hazards. The data in the adaptive capacity survey was used to validate the conceptual framework for the model shown in Figure 1.

In the 2015 & 2016 ISERC Presentations and Papers folder, please refer to manuscript titled Adaptive Capacity to Floods: A Case Study of Rincón, PR for a more in depth discussion of the methodology used to validate the original conceptual framework using Cronbach’s alpha based on the smaller sample (��= 139) that was available by the end of January 2016. Since then, the conceptual framework has been modified and the most recent version of the proposed conceptual framework is described in Figure 1

Cronbach’s Analysis

For validating the instrument, the Cronbach’s alpha analysis was used. This analysis measures how closely related a set of items are as a group, and it is considered a measure of scale reliability. The idea behind this analysis was to take the conceptual framework described in Figure 1 and evaluate individually each sub-dimension of adaptive capacity (i.e. Mitigation, Response, Recovery). If the attributes measure the same construct, then, the sub-dimension would show internal consistency with a Cronbach’s alpha greater than or equal to 0.70 the rule of thumb suggested in the literature. The vulnerability attributes

Mitigation

Self-preservation

Measures

Perceived Preparedness

Response

Perceived Likelihood

ThreatAppraisal

ThreatExperience

that were identified using Cronbach’s alpha are described in Table 2. In order to reduce complexity, redundant attributes were removed from the Cronbach’s alpha analysis (i.e. attributes that triggered a minimal or no increase in ��). In the data analysis, variables that did not appear to measure the subdimension of adaptive capacity to which they were originally assigned to in the conceptual framework were further evaluated in other dimensions of vulnerability. Results in Table 2 show that between 2 and 9 attributes can be used to assess each of the sub-dimensions of adaptive capacity. Results also show that attributes that might have been considered to belong to one sub-dimension in the conceptual framework (e.g. scholarity, age), actually belonged to another sub-dimension. In the case of age, also the nature of the relationship was different between the conceptual framework and the findings from the Cronbach’s analysis.

HouseholdStructure

Perceived Severity

RelianceonPublic FloodProtection

SignalAwareness

PerceivedSignal Awareness

ObservedSignal Awareness

Knowledgeof NativeLanguage

Recovery

PerceivedCoping Appraisal

PerceivedAreaAwareness

Awarenessof SafeZones

Familiarity withtheZone

Table 2. Results for Cronbach analysis on proposed conceptual framework for adaptive capacity index using survey data. All analyses selected indicate that as each of the following attributes increases, so does the sub-dimension of adaptive capacity under consideration, except for age which shows an inverse relationship with mitigation.

Construct (Cronbach’s ��)

Mitigation (�� =0.7037)

Variables Detected

Personal experience with floods, experience with floods through family or friends, perceived likelihood of a flood, perceived likelihood of a tsunami, perceived severeness of a flood, perceived severeness of a tsunami, reliance on public flood protection, level of preparedness for natural hazards, age*

Response (�� =07674) Awareness of safe zones, familiarity with the zone Recovery (�� =08097) Household income, personal savings, scholarity *inverse relationship

Pedestrian Evacuation Model

Population Allocation

The initial scenario that will be evaluated with the pedestrian evacuation model is a night between Monday and Friday where all the residents of Rincón are strictly distributed in households. This distribution can be done in a completely randomized manner and completely disregard an individual’s attributes as it is currently done in most models in the hazards literature. Nevertheless, the goal of this work is to simulate more accurately how an individual’s set of attributes and family configurations influence evacuation response. This initialization relies on four data sources. The first data source is the adaptive capacity survey previously described as well as Rincón’s digital cadaster, the US Census Bureau, and the American Community Survey (U. S. Census Bureau, 2016). Refer to Table 4 for further details on how the external sources of data were used.

The statistical software R was used to implement the pedestrian evacuation initialization. The total count of houses in Rincón, PR were extracted from the American Fact Finder and linear programming is used to estimate the number of houses of size 1, 2, 3, and 4 for each sector. Data from the adaptive capacity survey, in turn, is used to estimate the number of houses with dependents in each household size. First, all dependent population (up to 19 years old) is allocated. Then, adults are added to the house to complete the household size. For example, if a dependent is allocated to a household of size 2, then this house can no longer receive a dependent since the maximum number of people in the house is 2. Therefore, any adult at least 20 years older and no more than 40 years older than the dependent will be added to the household. In case there is no sample left to fit that description, the code will place any adult 20 years or older. In houses we there is room for two parents, families are configured assuming mothers are between 20 to 40 years older than the oldest dependent in the household. A male adult 10 years younger to 10 years older than the female adult is then added. In cases where there are no sample available to fit the age restrictions, any random adult will be added. If all household sizes were created in previous section, the code will randomly sample the remaining adults in household sizes 2, 3, and 4. If a household size is missing, the code will distribute the remaining adults to that household size up to the maximum number of houses necessary. If there are still remaining adults and houses, the code will randomly distribute them between all the household sizes.

Data Source Uses

Table B11016 American Community Survey

Table DP1 Census 2010

Contains information on household type by household size. The information from this table helped in the estimation of the number of houses per household size.

Contains information about the population distribution for all the Rincón municipalities divided by gender and age. This estimate was used to distribute the population based on their age and household size.

This table containes unique ID’s for all the parcels in Rincón. The information on this table was used to create and identify the different houses to which distribute the population.

Throughout the spring and fall semester of 2016, the team evaluated different alternatives for the tsunami evacuation model (TEM). The intent of this model is to relax some of the restrictive assumptions found in most pedestrian evacuation models in the literature, such as a constant speed for all individuals regardless

of the fact that the person belongs to an elderly or disabled population, a constant speed throughout the entire evacuation route (i.e. no fatigue), individuals do not take anyone else into account (i.e. families do not evacuate together), and an identical response for all population (i.e. evacuate immediately). In order to overcome these limitations, the model should be able to address an individual’s attributes (i.e. age group, gender).

For such reason, an approach that blends features of agent-based simulation (ABS) and anisotropic least cost distance (ALCD) models was chosen. The first alternative that was evaluated was the ArcGIS evacuation analyst tool created by the USGS. Unfortunately, this tool is not customizable enough to allow for the implementation of an ABS model. The second alternative considered was to code an in-house ABS in C, nevertheless, once again, the need to process and output georeferenced data made it extremely difficult to implement. Finally, a good compromise between the two evaluated alternative was found in r.walk in QuantumGIS (QGIS) an open source geographical information system.

The framework for the proposed pedestrian evacuation model is described in Figure 2. Between the fall semester of 2016 and the spring semester of 2017, the model was implemented in QGIS and R. The r.walk algorithm in QGIS allows to implement an ALCD and with custom code in R additional features were added Some of the features coded in R to improve upon the r.walk algorithm include: improved population distribution, detail distribution fits for running speed as well as fatigue factor and reaction time penalties. More details on these additional features are described in the following sections.

Figure 3 shows the study area for the proposed tsunami PEM, which includes Rincón and neighboring territories in Añasco and Aguada. Figures 4 – 6 show the road layer, the land use land cover (LULC), and the digital elevation model (DEM), which are essential inputs for the ALCD model. The road layer and the land use land cover provide the areas that represent either pathways or blockages in the evacuation routes of individuals. Friction costs in the LULC depict how more difficult is walking or running on any given terrain when compared against paved roads. The higher the friction costs, the more difficult it becomes to go through any given area. For instance, a body of water represents a complete blockage for individuals, hence, its friction times is roughly 1,000 times higher than any other type of land use. Finally, the ALCD aspect of the r.walk model relies on the knowledge of slopes, which is based on the DEM. Uphill is always penalized because it is expected to increase evacuation times. In this implementation of r.walk, seldomly, downhill slopes are also penalized. Negative slopes with a magnitude greater than 12 degrees are also penalized, as they are also expected to increase evacuation time. Figure 7, on the other hand, shows the evacuation routes for each of the groups in the model, which allow individual evacuation as well as groups that range anywhere from two to ten individuals.

Figure 2. Framework of the proposed pedestrian evacuation model

Figure 3. Area of study in proposed pedestrian evacuation model includes all of Rincón’s subcounties as well as two as areas from neighboring subcounties in Añasco and Aguada.

Evacuation response

The adaptive capacity survey was used to fit a model to better understand the evacuation response of individuals in the event of a tsunami. Three standalone supervised learners were evaluated along with three meta learners. A parallel ensemble of decision trees (i.e. random forest) was the top performer in a thorough evaluation considering parameter tuning on prior probabilities and three complementary performance measures. The proposed model is able to predict one of six responses: contact relatives first, do not evacuate, evacuate immediately, gather dependents first, help others first, and others In the previous response categories, the term first is used to depict an event where an individual would immediately evacuate after having completed the task described in the category. The proposed prediction model estimates the probability of a pedestrian assuming each response and, then, assigns a prediction based on the category with the largest probability. The attributes used to determine a pedestrian’s response include gender, age group, resident status (tourist or resident), number in household, and an indicator variable stating whether the individual is in the proximity of dependent population (i.e. children, older parents) when the tsunami warning signals are recognized (near dependent population). Further details on the proposed evacuation response model are available in Dávila et al. (2017).

Running Speeds and Fatigue Factors

To determine the running speeds that will be used in the pedestrian evacuation model, data from a fivekilometer race from four consecutive years was used (2013-2016). The race chosen was “El Corazón de la Montaña” since the topography of the race included mountains, which in the event of a tsunami is where the pedestrians will be heading to. Also, individuals of all types, regardless of their physical condition, participate in this type of race. That is, individuals who trained for the event and even individuals who are not able to run and choose to walk. This event, nevertheless, does not take place in Rincón, PR. It was still more suitable than other events that take place in Rincón because of the nature of the event. Sports-event data for events taking place in Rincón, PR included paddle board races and triathlons, for which the physical condition of the individuals participating cannot be expected to be representative of the average resident or tourist. Initially, 22 different groups based on age group and gender combinations were used, and, in order to reduce complexity, non-parametric inferential statistics were carried out to assess differences in the medians and variances of these groups. The rationale was that groups with no statistically significant differences in their medians or variances can be grouped together to reduce the number of parameters that need to be handled within the pedestrian evacuation model. With this non-parametric assessment, the number of groups was reduced from 22 to 8 groups and a distribution was fitted to each of the groups. The idea behind the distribution is to be able to estimate different running speeds for the traditional evacuation scenarios depicted in the literature: 15th percentile of running speeds for slow run and 85th percentile of running speeds for fast run. Also, the distribution fit has the added advantage that random percentiles can be drawn and further relax the assumption of the same evacuation running speed even within groups. Further details on the running speed and fatigue factor estimation are available in Hernández et al. (2017).

Table 4. Summary of results for speed data. Each row represents one of the proposed groups and ks and ad refer to the p-values for the Kolmogorov-Smirnov and Anderson-Darling goodness-of-fit tests, respectively. Group n Female Age Groups Present Male Age Groups Present Kruskal Wallis PValue Fligner Killeen P-Value

and Parameters P-Value

1 98 <19, 20-24, 25-29 - 0.772 0.199 Gamma (min=0, α=12.83, β=0.23)

0.485

2 168 30-34, 35-39, 40-44, 45-49, 50-54, 55-59, 60-64, 65+ - 0.734

3 12925-29 ,40-44, 60-64

4 23830-34, 35-39, 45-49, 50-54

SB (min=0, λ=8.01, γ=2.09, δ=2.58) KS 0.693 AD 0.718

Weibull (min=0, α=4.61, β=3.59)

(min=0, α=5.13, β=3.84)

5 71 - 20-24 -Weibull (min=0, α=4.72, β=4.10)

6 109 - <19 -Weibull (min=0, α=6.52, β=3.97)

7 30 - 65+ -Weibull (min=0, α=5.26, β=3.06)

8 37 - 55-59 - -

Weibull (min=0, p=10.72, β=3.47)

0.974

0.783 AD 0.886

0.961

0.642 AD 0.667

0.601

The estimated groups were also used to determine the fatigue factor, that is, the rate at which the speed decreases as the distance traveled increases. The work by Riegel (1981) was used to add a time penalty to the evacuation times. The time was estimated assuming no fatigue and assuming the fatigue factors described in Table 2. The difference between these two times was treated as a time penalty that was added to the evacuation time estimated with the r.walk model in QGIS.

Table 5. Estimated percentiles for multiple case scenarios in the pem. In terms of evacuation speed, the percentiles shown portray the following scenarios: slow (p15), average (p50), and fast run (p85).

Reaction Time

A literature search was carried out to estimate a reaction time for each of the response categories described in the Evacuation Response section. In this case, a literature review was carried out and the commonly used approach of applying a Rayleigh distribution to estimate reaction times was chosen. In the proposed approach, the scale parameter was chosen to be the mean reaction time for the six different responses considered in the model, which were extrapolated based on previous works on reaction time. The mean reaction times considered and the rationale being them is described in Table 6.

Table 6 Estimated mean reaction time for different response categories.

R. F. (1997)

Gather

G., & Fahy, R. F. (1997) Help

(1997)

(1997)

Panic and evacuate

Seek Evacuation Assistance

Wait for Official Notification

Vulnerability Index

63

Office reaction with cool weather Proulx, G., & Fahy, R. F. (1997)

169 Residential with good alarm alarm Proulx, G., & Fahy, R. F. (1997)

480

Unsupervised Variable Selection

Institutional decision and notification times Yuzal, H., Kim, K., Pant, P., & Yamashita, E. (2015)

The hazards literature suggests a large number of attributes that can be used to assess the construct of vulnerability. Nevertheless, often times, many of these attributes convey the same type of information and not at of them are needed simultaneously to assess vulnerability. Given there is a gap of unsupervised feature selection methods in the literature, three different methods have been developed and are currently under evaluation. The first method uses principal component analysis and applies three different voting schemes to select the most relevant vulnerability attributes. The other two methods are a unsupervised adaptation of the artificial contrast ensemble (ACE) feature selection scheme by Tuv et al (2009), which was developed in the context of supervised learning scenarios. Details on the PCA method are available in the 2015 & 2016 ISERC Presentations and Papers folder. Partial results on the unsupervised adaptation of ACE are available in the QUEST poster presentation.

Variable Weights

An analytic hierarchy process (AHP) model will be used to model individually the exposure and sensitivity indexes. Since this model needs a weight for each attribute in the model, a survey instrument was developed to determine these weights. Subject-matters experts were approached, and the weights will be determined using the Delphi method. The subject-matter experts targeted include emergency responders and researchers with knowledge on one or more of the following topics: vulnerability, floods, and tsunamis. It must be acknowledged that the instrument underwent IRB approval and 23 surveys have been completed to date.

Sensitivity

Sensitivity is used as a weight to exacerbate or to reduce the impact of exposure in the construct of vulnerability. It is often treated as a weight between 0 and 1, and the higher the sensitivity, the more impact exposure will have within the vulnerability assessment. In the context of this work, sensitivity will be estimated using the wave height polygons in the hazard zone, where the hazard zone is defined as the maximum of maximum using FEMA inundation, tsunami evacuation zone, and category 5 storm surge. As a preliminary assessment, the tsunami wave height polygons were processed using ArcGIS.

Table 7 Tsunami wave height polygons

Wave Height

Adaptive Capacity Index Implementation

The adaptive capacity index will be implemented using an analytical hierarchical process. Code was written in R to automate the pre-processing of survey data (i.e. adaptive capacity survey, variable weights) and generating a script for running the AHP model using the ahp package. Model implementation is expected to be completed by the end of the spring semester of the year 2017.

Georeferenced Inventory

The georeferenced inventory of infrastructure in Rincón is more than 95% complete. The data was collected using the ArcGIS Collector application developed by one of the undergraduate students funded by the project. Additional data used consisted of publicly available (e.g. Portal Geodatos, Google maps) or extracted from the registry of businesses in Rincón, PR provided by the Rincón Planning Department. Regardless the origin of the data, numerous visits to Rincón were used to validate the data that was not in the ArcGIS Collector application. Duplicates have been removed and, to date, 84 points have not been validated. The goal is to perform additional visits to Rincón to validate these observations and generate the final map product for the georeferenced inventory dividing the data points as critical, necessary, and public facilities. Critical facilities refer to those needed during the event of a flood or tsunami. Essential facilities are those needed to continue operations after the emergency has passed, and public facilities are all those that are open to the public, whether they are owned by the government or private entities. Figure 8 shows the currently available infrastructure information for Rincón with 553 businesses, including those that need to be validated. The georeferenced data set is expected to be completed by the end of the spring semester of 2017.

Figure 8 Map of the most recent version of the georeferenced inventory

Table 8. Project Participants

LIST OF STUDENTS SUPPORTED

Table 9. Students Supported

María del Mar Cruz Mendoza

Carlos J. Sánchez Bonet

B.S. in Industrial Engineering (787)-448-1803

maria.cruz30@upr.edu

B.S. in Industrial Engineering (787)-217-4120

Norbert Franqui

Spring 2016

$ 552.18 Spring 2015

carlos.sanchez5@upr.edu $ 552.18 Spring 2015 $ 640.53 Spring 2016

norbert.franqui1@upr.edu $ 340.51 Fall 2015 $ 347.88 Fall 2016

B.S. in Industrial Engineering (787)-604-2729

Beatriz Torres

Xaimarie Hernández

B.S. in Industrial Engineering (787)-300-8366 beatriz.torres2@upr.edu

$ 73.63 (until April 3, 2017) Spring 2017

$ 147.25 Spring 2016

$ 283.45 (Estimated) Summer 2016

Gabriel González

Nolgie O. Oquendo

B.S. in Industrial Engineering (787)-546-3828

xaimarie.hernandez@upr.edu $ 51.54 Spring 2016 $ 206.15 (Estimated) Summer 2016 $ 346.04 Fall 2016

B.S. in Industrial Engineering (787)-467-3369 gabriel.gonzalez1@upr.edu $ 125.15 Spring 2016

B.S. in Industrial Engineering (787)-604-2729 nolgie.oquendo@upr.edu

Mayra R. Rios

B.S. in Industrial (787)-420-6422

$ 187.74 Fall 2016

$ 195.11 (until April 3, 2017) Spring 2017

$ 169.33 Fall 2016

Engineering mayra.rios2@upr.edu

LIST OF PRESENTATIONS

$ 64.43 (until April 3, 2017) Spring 2017

▪ Hernández, X., Dávila, S., and Franqui, N. (2017) “Relaxing Assumptions in Evacuation Models using Sports Event Data.” To be presented at the 2017 Puerto Rico Interdisciplinary Meeting (PRISM), Humacao, PR.

▪ Faucher, J.E., and Dávila, S. (April 2017). “A Hybrid Approach to Pedestrian Evacuation Models.” To be presented at the 2017 Puerto Rico Interdisciplinary Meeting (PRISM), Humacao, PR.

▪ Castiel, N., Sánchez, C., Macías, Jorge, and Dávila, S. (2017) “Tsunami Exposure Indexes: Vulnerability Attribute Selection and Substitution.” Poster Presentation QUEST, Ponce, PR.

▪ Hernández, X., Dávila, S., and Franqui, N. (2017) “Relaxing Assumptions in Evacuation Models using Sports Event Data.” Emerging Researchers National Conference in STEM (ERN), Washington, DC.

▪ Dávila, S., Franqui, N., and Hernández, X. (2017) “Modeling Pedestrian Evacuation Response for Tsunami Events ” To be presented at the 2017 Industrial Systems Engineering Research Conference, Pittsburgh, PA.

▪ Hernández, X., Dávila, S., and Franqui, N. (2017) “Relaxing Assumptions in Evacuation Models using Sports Event Data ” To be presented at the 2017 Industrial Systems Engineering Research Conference, Pittsburgh, PA.

▪ Dávila, S., Melgar, S., Power, T. (June 2016). “Adaptive Capacitive to Floods.” Poster Presentations of Undergraduate Research for Texas A&M Study Abroad Program, University of Puerto Rico, Mayagüez, PR.

▪ Faucher, J.E., and Dávila, S. (May 2016). “A Hybrid Approach to Pedestrian Evacuation Models.” Industrial and Systems Engineering Research Conference, Anaheim, CA.

▪ Dávila, S., Franqui, N., and Medina, L.A. (May 2016). “Tsunami Exposure Indexes: Vulnerability Attribute Selection.” Industrial and Systems Engineering Research Conference, Anaheim, CA.

▪ Dávila, S., Castiel, N., Rodriguez, C., and Medina, L.A. (May 2016). “Adaptive Capacity to Floods: A Case Study of Rincón, PR.” Industrial and Systems Engineering Research Conference, Anaheim, CA.

▪ Castiel-Camacho, N.G., Sánchez-Bonet, C.J., and Dávila, S. (March 2016). “Vulnerability Attribute Selection.” 31st Annual Symposium of “Geología del Caribe, ” University of Puerto Rico, Mayagüez, PR

▪ Franqui, N., and Davila, S., (March 2016). “Tsunami Exposure Indexes: Vulnerability Attribute Selection.” 2016 Junior Technical Meeting (JTM) and the Puerto Rico Interdisciplinary Meeting (PRISM), Ponce, PR.

▪ Dávila, S. (November 2015). “A Decision Theory Approach to Tsunami Exposure Indexes.” GIS Day, Mayagüez, PR

▪ Dávila, S., Vasquez, B., and Church, R. (June 2015). “Adaptive Capacitive to Floods.” Poster Presentations of Undergraduate Research for Texas A&M Study Abroad Program, University of Puerto Rico, Mayagüez, PR.

▪ Cruz, M., García, T.M., Bonet, S.A., and Dávila, S. (May 2015). “Global Ranks in High-Dimensional Tsunami Exposure Indexes.” XXII Applied Social Science Symposium (CISA), University of Puerto Rico, Mayagüez, PR.

▪ Castiel-Camacho, N.G., Sánchez-Bonet, C.J., and Dávila, S. (May 2015). “Tsunami Exposure Indexes: Vulnerability Attribute Selection.” XXII Applied Social Science Symposium (CISA), University of Puerto Rico, Mayagüez, PR

▪ Dávila, S., Ruiz, Roy, Medina, L. (May 2015). “Exposure and Adaptive Capacity to Floods: A Comprehensive Vulnerability Assessment of Rincón’s Neighborhoods.” Community-based Climate Adaptation Plan for Rincón, Rincón, PR.

▪ Dávila, S., Castiel-Camacho, N.G., and Sánchez-Bonet, C.J. (May 2015). “Tsunami Exposure Indexes: Vulnerability Attribute Selection.” Simposio “La investigación y la innovación como aceleradores de nuevas oportunidades en Puerto Rico, University of Puerto Rico, Mayagüez, PR.

▪ Dávila, S., (May 2015). “Exposure and Adaptive Capacity to Floods: A Comprehensive Vulnerability Assessment of Rincón’s Neighborhoods.” Community-Based Climate Adaptation Plan for Rincón, Rincón, PR.

▪ Dávila, S., Cruz, M., García, T.M., Bonet, S.A., Ruiz-Vélez, R. (May 2015). “Global Ranks in High-Dimensional Tsunami Exposure Indexes.” Proceedings of the Industrial and Systems Engineering Research Conference, Nashville, TN.

▪ Castiel-Camacho, N.G., Sánchez-Bonet, C.J., and Dávila, S. (April 2015). “Tsunami Exposure Indexes: Vulnerability Attribute Selection.” XV Sigma Xi Student Poster Day, University of Puerto Rico, Mayagüez, PR.

LIST OF PEER REVIEWED PUBLICATIONS

▪ Dávila, S., Franqui, N., and Hernández, X. (2017) Modeling Pedestrian Evacuation Response for Tsunami Events. To appear in the Proceedings of the 2017 Industrial Systems Engineering Research Conference, Pittsburgh, PA.

▪ Hernández, X., Dávila, S., and Franqui, N. (2017) Relaxing Assumptions in Evacuation Models using Sports Event Data. To appear in the Proceedings of the 2017 Industrial Systems Engineering Research Conference, Pittsburgh, PA.

▪ Dávila, S., Franqui, N., and Medina, L.A. (2016). Tsunami Exposure Indexes: Vulnerability Attribute Selection. Proceedings of the Industrial and Systems Engineering Research Conference, Anaheim, CA.

▪ Dávila, S., Castiel, N., Rodriguez, C., and Medina, L.A. (2016). Adaptive Capacity to Floods: A Case Study of Rincón, PR. Proceedings of the Industrial and Systems Engineering Research Conference, Anaheim, CA.

▪ Dávila, S., Cruz, M., García, T.M., Bonet, S.A., and Ruiz-Vélez, R. (2015). Global Ranks in High-Dimensional Tsunami Exposure Indexes. Proceedings of the Industrial and Systems Engineering Research Conference, Nashville, Tennessee.

LIST OF PRINCIPAL INVESTIGATORS SUPPORTED

Dr. Saylisse Dávila (PI)

Dr. Lourdes Medina (Co-PI)

14.6% (2 credits/semester –0.75 SG, 1.25 Match)

7.3% (1 credit/semester –1 Match)

$5,173.52 M Spring 2015

$2,165.79 / $3,609.65 M Fall 2015

$2,471.25 / $3,954 M Spring 2016

$2,471.25 / $3,954 M

Fall 2016

$2,821.92 M Spring 2015

$2,821.92 M Fall 2015

$2,821.92 M Spring 2016

$ 3,150.24 M Fall 2016

RELEVANCE, RESPONSE, AND RESULTS

Title: UPR Sea Grant characterizes flood vulnerability of Rincón Puerto Rico coastal communities in terms of exposure, sensitivity, and adaptive capacity.

Recap: A tsunami pedestrian evacuation model and a flood vulnerability assessment that includes FEMA, storm surge, and tsunami flood are currently under development for Rincón, PR. Custom conceptual frameworks and partial implementation of the vulnerability assessment has been carried out as two separate analytic hierarchy process (AHP) models aimed at quantifying the dimensions of flood exposure and adaptive capacity. A custom geo-referenced inventory of facilities was developed for the purposes of better understanding the exposure of Rincón’s subcounties in terms of commercial infrastructure. Data collection and data preprocessing has been carried out to fuse data from multiple sources using geographical information systems (GIS). Codes have been developed in the R statistical software to automate the process of extracting and pre-processing survey data for the purposes of fitting the AHP models in the ahp package.

The proposed tsunami pedestrian evacuation model is a hybrid model that merges anisotropic least code distance with agent-based approaches and relaxes several restrictive assumptions found in the PEM literature. An analysis of sports-event data was used to determine groups that behave similarly in terms of their running speed and fit probability distributions that can be used in sensitivity analyses of the amount of population that reaches safety according to different case scenarios (e.g. slow run, fast run). A fatigue factor was introduced to further relax the assumption that the evacuation speed is constant through the entire evacuation route. Lastly, an optimization algorithm was used to more accurately distribute individuals using Rincón’s digital cadaster and data from the US Census Bureau.

Relevance: As coastal areas become increasingly populated, what were once hazard-prone areas become potential sites for disaster. The level of disaster varies with the type of flooding, needs of the neighborhood, and infrastructure value. A GIS database and vulnerability assessments at the county subdivision level that integrate this information allow emergency responders to identify geographical hotspots for floods and potential strategies to mitigate their effect.

Response: The GIS inventory of infrastructure will provide emergency responders with an estimate of the infrastructure at risk in the event of a flood. The tsunami pedestrian evacuation model will allow to estimate the time to safety and, more importantly, pinpoint those geographical areas where individuals will not be able to reach safety in the event of a tsunami. Both data sources combined will allow emergency responders to identify which and where mitigation strategies (e.g. vertical evacuation) are most needed.

Results: Conceptual frameworks were created to measure exposure and adaptive capacity to floods, and the implementation of adaptive capacity index is currently under way. Feature selection and internal consistency statistical analyses have been carried out to reduce the complexity of the models. A total of 192 surveys on a sample of tourists and residents of Rincón were used to gather data for the vulnerability model and to understand the evacuation response of individuals in the event of a tsunami. A total of 23 surveys were carried out on subject matter experts to determine the weights of the vulnerability attributes within the exposure and adaptive capacity indexes. The PEM model has been implemented using r.walk in QGIS and complemented with custom code in R to relax several restrictive assumptions found in the literature: all individuals evacuate immediately and separately, the evacuate speed is constant through the entire evacuation route, all individuals evacuate at the same speed.