Context-specificity of tsunami vulnerability models for roads and buildings

With a growing suite of tsunami vulnerability models for built assets, it is imperative models used in tsunami risk assessment replicate local damage processes. In this study, we assess the variables driving impacts of tsunamis on buildings and roads, using empirical data from previous studies on the 2018 Sulawesi Tsunami in Palu Bay, Indonesia. We rank variable importance and correlation, based on predictive models generated by tree-based algorithms at different aggregation levels (Overall, West, Middle and East case study clusters). Models’ predictive performance is evaluated by calculating accuracy score (relative hit rate), while also testing their transferability through model transfer analysis. For roads, “Scouring” is the dominant variable for damage, followed by “Shielding by Embankment”. For buildings, “Distance to Fault Rupture” is the dominant variable for damage, followed by “Shielding”. The model transfer analysis shows these damage datasets, are unique to their local hazard/asset/environmental characteristics, highlighting the importance of vulnerability models that consider local context for tsunami risk assessment.