B0596
Title: Linearization approach for aggregated landslides data
Authors: Man Ho Suen - University of Edinburgh (United Kingdom) [presenting]
Mark Naylor - University of Edinburgh (United Kingdom)
Finn Lindgren - University of Edinburgh (United Kingdom)
Abstract: In spatial statistics, it is not uncommon to have spatial misalignment in observed responses at point locations and covariates data at various resolutions and shapes. One of the common approaches is to aggregate the point observations into count data with respect to the area polygon. One of the popular approaches in landslide literature is to aggregate based on slope units that cluster landslide observations beneath the surface. This takes away the point location information and introduces both bias and uncertainty. Starting with a Poisson point process, the domain is discretised into subspaces. The definition of these subspaces can be flexible based on various scenarios. Assuming the intensity of the process is log-linear, an implementation trick is used and the first-order Taylor linearization in the INLA and inlabru R packages. The approximation bias is computed with the help of the omitted second-order terms. This turns out to provide insights into improving the modelling of aggregated data. This approach is illustrated in earthquake-induced landslide data.
