Stochastic quantification of low-resolution geocoding uncertainty and its application to catastrophe modeling

Abstract

Essential to developing a robust risk-management model is to clarify first "where are the exposed properties". Geocoding is a process that binds property exposure in a built-in environment to spatiotemporal dynamics of natural catastrophes. While addresses geocoded to the finest possible spatial resolution lead to relatively precise estimation of losses from surrounding perils, incomplete/invalid input addresses must fall back to coarser spatial scales (postal code, city, and county etc.), resulting in less precise loss estimation. This work aims at providing the foundation to place the low-resolution geocoding outcomes into gridded locations at a targeted finer resolution. We take advantage of auxiliary information such as land use/cover to infer the likelihood of an exposure placement at the gridded locations. This development is showcased with a postcode in Florida to understand loss changes from low-resolution geocoding uncertainty perspectives.