Shelter evacuation in relation to demand characteristics in Dominica.

Abstract

Reducing disaster vulnerability is one of the main objectives of government actions in countries prone to natural hazards like hurricanes. Evacuation policy planners in such areas need precise and up to date information about population location, roads, rivers etc. Moreover, they aim to employ the most efficient methods to identify vulnerable areas and suggest possible intervention solutions. This research proposes a hybrid approach using object-based image analysis, self-organising maps and network analysis to determine shelter evacuation areas. First, a method of extracting building data from multi spectral and panchromatic images is presented. Data extraction process is performed using an object image based analysis and ends up with assigning to each building attributes that are necessary for determining hurricane vulnerability. Second, obtained features are classified using self-organizing maps to identify the most vulnerable groups. Finally, extracted and classified buildings are used in network shelter analysis. The whole method is highly extendable and modifiable so it can serve not only for evacuation modelling in the study case of Dominica, but also for other areas at risk.