Improving post-disaster road network accessibility by strengthening links against failures

We study a network improvement problem to increase the resilience of a transportation network against disasters. This involves optimizing pre-disaster investment decisions to strengthen the links of the network structurally. The goal is to improve the expected post-disaster accessibility. We first propose a new dependency model for random link failures to predict the post-disaster status of the network. We show that the probability of any network realization can be computed using a Bayesian network representation of the dependency model. As the computational effort grows with the network size, we use our proposed dependency model in a network sampling algorithm. We then estimate an accessibility measure, namely, the expected weighted average distance between supply and demand points by checking pre-generated short and dissimilar paths in the sample. We minimize this measure and decide on the links that should be strengthened in a two-stage stochastic programming framework. As the failure probability of a strengthened link decreases, the discrete scenario probabilities depend on the first-stage decisions. To tackle this challenge, we develop an efficient tabu search algorithm. We apply our methods to a case study of Istanbul under the risk of an earthquake, both to illustrate the use of the methods and to derive insights for decision makers.