A spatial regression and clustering method for developing place-specific social vulnerability indices using census and social media data

Abstract

This paper presents a new social vulnerability index construction approach that utilizes geographically weighted local regression modeling and spatial clustering to determine location-specific weights of vulnerability indicators. Sub-regions within the study region with homogeneous vulnerability profiles are identified by clustering the local regression coefficients with spatial contiguity constraints. The weights of the indicators are determined using an independent data set showing the disaster impact and fitting a spatial regression model. The approach is applied to Hurricane Harvey as a case study in which block-group level census data were used to obtain the indicators by factor analysis and geocoded rescue request calls posted on social media was used as independent data to compute the indicator weights.