Bayesian analysis of population vulnerability to rainfall events in Venezuela.

Jhan Rodriguez, Universidad Simón Bolívar, Lelys Isaura Guenni, Universidad Simon Bolivar

Abstract


We use a Bayesian hierarchical model to quantify, at the district scale, the vulnerability of population to rainfall-related events, such as floods, flash-floods, and landslides. As a measure of vulnerability quantification we use the Relative Risk (RR). The RR is defined for each district and a given time span, as the ratio of the (unknown) potential proportion of people affected in the district to a pre-fixed, expected proportion of people affected. Thus, the RR is a measure of deviation from the expected behaviour of damage to population in each district. It can be used as an indicator of anomalous damage behaviour, by identifying those districts having a RR (say) significantly different from one. The model employed for the RR analysis is a log-linear model which considers the number of affected people in each district as the realization of a Poisson variable, and allows the inclusion of district-specific covariates. The model also allows to include parameters that capture any structural spatial pattern on the underlying RR surface, namely the so-called Conditionally Auto-Regressive, or CAR, effects. An important result is the RR map of Venezuela, which summarizes the posterior distribution of the RR for each district, and indicates that the most vulnerable districts form clusters in Nord-central and Western Venezuela, in addition to other districts of high RR arranged in a less structured way.

Keywords


Vulnerability; Risk; Spatial hierarchical models; Bayesian modelling

Full Text:

Download PDF