Processed data for Spatial disaggregation of asset type for hazard risk assessment usinga lattice-based Hopfield Neural Network

Characterising the spatial distribution of commercial assets is essential for assessing<br>their exposure to natural disasters. However, such information is often only available<br>at an aggregate level. We introduce a lattice-based HNN for spatial disaggregation<br>of asset data which operates by optimising a pre-specified goal function under a<br>coarse spatial resolution constraint. The method was applied to a case study in<br>which non-residential and residential properties were disaggregated from areal data<br>to the point level. For this analysis, the goal function was defined as maximising<br>the spatial dependence of the two classes with respect to the neighbourhood graph<br>distance of the properties, while the constraint was defined as the areal propor-<br>tions of non-residential properties. With this specification, the HNN was used to<br>generate multiple maps of non-residential and residential properties, validation of<br>which with ground reference data achieved a higher mean accuracy and F1 score<br>than random disaggregation. When overlaid on a flood map, the maps produced by<br>the HNN predict greater potential flood damage than random disaggregation. The<br>results demonstrate the utility of the HNN for rule-based disaggregation of spatial<br>information.