Multidimensional Vulnerability Index

Data:

• Results of the Socio-Economic Impact Assessment (SEIA) conducted by UNDP Accelerator Labs Lebanon after the August 4th Beirut blast

Analysis Objectives:

• Identify where the most vulnerable groups exist within the city’s neighborhoods 
• Develop and use a multidimensional vulnerability index (MVI) to understand intersectional vulnerabilities
• Conduct analysis at the household or business level, and then aggregate to the neighborhood level. 

Method:

• Based on a literature review of existing multidimensional vulnerability indices and their methodologies, consultation with the UNDP Accelerator Lab team, and availability of relevant inndicators from the SEIA dataset, two MVI frameworks were designed, one for households and the other for micro, small, and medium sized enterprises (MSME). 
• The methodology for calculating the MVI is largely based on the UNDP 2020 Multidimensional Poverty Index. The MVI frameworks are composed of three dimensions each of equal weights, with multiple indicators under each dimension. 
• The Households MVI three dimensions are: 1) Health, 2) Standard of Living, and 3) Employment & Social Security. 
• The MSME MVI three dimensions are: 1) Exposure, 2) Sensitivity, and 3) Adaptive Capacity. 
• Each household is assigned a deprivation score based on the household’s deprivations across the different indicators. The maximum deprivation score is 100 percent, therefore the maximum deprivation for each dimension is 33.3 percent. The weightings for each dimension is then further distributed across the indicators. For example, there are eight indicators under the health dimension so each indicator has a weight of 1 / 24.

Households MVI

MSMEs MVI

• Similar to the Human Development Report’s Multidimensional Poverty Index, to identify multidimensionally deprived households or businesses, each household and business is assessed across the indicators, then the sum of the weightings of all deprived indicators is calculated. A cut off of 1 / 3 is used to categorize households and businesses as deprived or not. For example, if a business is deprived in all three of the exposure indicators and one of the sensitivity indicators, the deprivation score would be 0.44 which is higher than the cut off so the business is considered multidimensionally vulnerable.
• Percentage of vulnerable households or MSMEs - is the proportion of multidimensionally vulnerable households or businesses in the population, in this the total respondents. This can be calculated by neighborhood, for example, enabling the following statement: “The [xx]% of households / businesses in [nieghborhood] are multidimensionally vulnerable.”
• Intensity of vulnerability - for households and MSMEs that are considered multidimensinally vulnerable, the average of the deprivation sums of all households or MSMEs is calculated. This can be calculated by neighborhood, for example, enabling the following statement: “The average multidimensionally deprived household / business in [neighborhood] is deprived in [xx]% of the indicators.”
• MVI Score - is the product of the percentage of vulnerable households / businesses and the intensity of vulnerability. The score ranges from 0 to 1. Higher values indicate higher vulnerability. 

Population Analysis 

Data:

• Facebook World Population Data
• Lebanon Administrative Boundaries Level 3
• OSM Road Network

Analysis Objectives:

• The main objective of this analysis is to understand how populations are spatially distributed within the wider Beirut municipal area.

• Short of official census data that are spatially linked, the present analysis looked to compare how ancillary population data (i.e. Facebook World Population Data) could be aggregated to both the lowest administrative boundary in Beirut and possible neighbourhood clusters.

• Neighbourhood clusters in this case are a structural component to Beirut. They have been derived from clustering algorithms that take into account connectivity within Beirut's road network. The primary data source for this analysis was obtained from the crowdsource OpenStreetMap platform.

Method:

Facebook World Population Data

• The data was obtained in a GeoTIFF format, with cell values corresponding to population densities across the whole of Beirut. The base coordinate reference was noted as EPSG4326. No other reference transformations were performed.
• The obtained GeoTIFF was first clipped to the Beirut Municipal boundary. Null cell values were assigned a population value of 0. This standardisation was used to ensure all missing cell data errors were applied equally across all of municipal Beirut.
• The subsequent dataset was vectorised into point data across all of Beirut (Figure below).

• The population points (N.B: one point representing a population density value) were spatially linked with both the Beirut municipal boundary and the neighbourhood clusters boundaries derived from the below urban morphology analysis.

• The corresponding population was then aggregated to obtain two variable population distributions.

Urban Morphology Analysis

Data

• OSM Street Network

Analysis Objectives

• This analysis takes note of the disproportionate sizes between Beirut’s municipal boundaries.

• The present dataset shows that municipal areas in Beirut range between 0.42 km2 54(Saifeh) to 4.1 km2 (Moussaytbeh).
• As indicated above, this has downstream implications on population distribution and possibly, facility allocation and resource management within the wider Beirut area. 

Method

OSM Road Network Processing and Clustering

• In lieu of an available transport network from the Lebanese Government, open source data was chosen. The OpenStreetMap platform provides crowdsourced data on all urban, topological, and network features for Lebanon - and, this was chosen as the primary datasource.

• The OSM data download is dated November 2020.

• Given the nature of the data, several preprocessing steps were required to ensure functionality of the data.

• The road network was first isolated by attributes (where fclass = highway). In the OSM nomenclature, highway classes refer to all roadways for all transport modes (cars, bridleways, cycleways, and pedestrian ways).

• From the above subset, cycleways, bridleways, raceways, and tracks were omitted. The derived transport network with each remaining class was assumed to be traversable by foot; and, therefore, left for analysis

• Subsequent preprocessing steps included the rectification of topological errors within the transport network. This preprocessing step involves addressing issues of disconnections, overlaps, and isolated islands within the road network. A reiterative selection algorithm was lastly employed from all motorways and primary roads to ensure that a single, fully-connected road network was obtained as the main dataset.

• An unsupervised road network clustering algorithm was then employed to determine road networks clusters within the Beirut Municipal area.
• The above algorithm iteratively identifies the optimum number of road network clusters based on the degree of connectivity of each road intersection.
• Each cluster was bounded with a convex hull. Topological overlaps arising from the delineation process were rectified by allocating all overlapping zones to the larger cluster.

Accessibility Analysis

Data

• OSM Street Network
• OSM Facilities Data
• UNDP Hospital and School Data

• Facebook World Population Data

Analysis Objectives

• This analysis aims to understand how the spatial distribution of essential facilities in Beirut may contribute to issues of urban equity
• It characterises Beirut’s numerous neighbourhoods with respect to the approximate ease of reach to schools and hospitals
• This has been done in two steps. First, with walking as the primary mode of movement, access to these above facilities is determined by their distance to both derived neighbourhood clusters and the municipal Beirut boundaries. Second, a synthetic flow of potential population movement to each facility. This is created through a spatial interaction model to understand how each facility may affect the interaction of population in space to each facility.

Methodology

Isochrone Mapping

• The cleaned road network used in the previous analysis appended with a walking speed of 4.2 km/h, providing a movement time for each road vertice.
• 6 time thresholds, between 5 to 30 minutes at 5 minute increments, was used to understand the reach of each facility to municipal Beirut’s population
• Separate isochrone maps at the above intervals were created for each facility type.
• Each isochrone threshold was then used to aggregate Beirut’s population to obtain the population reach of facility type at each time threshold.
• The resulting data was grouped by neighbourhoods to provide an indication of accessibility at the same level

Spatial Interaction Modelling

•  The spatial interaction model conducted here was only done for hospitals given the current data limitations. The aim of this model was to only derive potential flows of population within the wider Beirut metropolitan.
• A production-constrained gravity model was used given the availability of neighbourhood population obtained from Facebook.
• It should be noted that given the limited availability of actual flow data available for Beirut, the model could not be calibrated with respect to its fit. Nevertheless, parameters chosen for the spatial interaction model have been selected from existing literature. A beta value of 0.2 was used in this case to calibrate the model’s distance-decay parameter.