Risk-Optimal and Equitable Retrofitting Strategy Using a Siamese Graph Neural Network for Earthquake-Induced Landslide Hazards

Landslides pose a severe threat to road networks, particularly in hilly terrains, often exacerbated by triggering events such as earthquakes. Hence, it is crucial to comprehensively evaluate their risk and optimize retrofits for such hazards in road networks, taking into account their interconnected nature and the potential for widespread traffic disruptions. This study proposes an optimization framework to manage earthquake-induced landslide risk to road segments in large networks and a policy to undertake capital improvement retrofits to critical road segments to enhance a network’s overall resilience and equity outcomes of capital investment decisions. As the main contribution, a Siamese graph convolutional network (GCN) surrogate model is proposed to make the optimization process efficient and tractable. This is achieved by approximating the computation of the total system travel time increase between the undamaged and various damaged states using the GCN. Next, demographic income information is fused with the optimization framework, in terms of welfare loss, to add an equity dimension to the retrofit policy. This retrofitting policy tailored to benefit low-income commuters mitigates welfare loss gaps between income groups. The framework is applied to a large road network in the hillside area of Los Angeles, which is prone to earthquake-induced landslides. The methods developed in this study and the results provide valuable insights to enhance the overall road network resilience while achieving equity-related outcomes of capital investment decisions. This research underscores the importance of considering commuters’ economic status when formulating policies to mitigate the impact of road network disruptions, especially in earthquake-induced landslide risk contexts. The overall retrofitting framework developed in this paper has the potential to be generalized and modified to address other hazards such as floods and wildfires from a risk-optimal standpoint.