Identification of flood-prone isolated areas through a probabilistic approach

Large natural disasters such as earthquakes, extensive floods or tsunamis can produce important direct impacts to roads, easily leading to a complete disruption of the transportation system. However, smaller scale processes, such as landslides or local floods, can also produce road disruptions pushing the road network functionality to a critical point. Analyzing transport networks’ vulnerability considering functional and topological properties is key to understand the indirect effects produced by a given hazard. By simulating single or multiple flood events, classic approaches use the percolation method to provide information on how well connected the road network remains after an event. In this study, we will focus on another indirect effect, the isolation problem, which is relevant for preparedness, emergency planning or post-disaster needs assessment. Roughly, many regional-scale studies deal with this issue by defining accessibility/remoteness indexes, that account for the connectivity between the locations of interest and a given set of pre-defined service centers. However, the problem of isolation has not yet been addressed from a probabilistic point of view. Hence, we aim to combine a computational probabilistic approach to simulate a set of flood scenarios that can affect a region and characterize the isolation potential of the set of vulnerable locations. Through this methodology we preserve the statistical properties of the historical events, and we simulate a large number of events, obtaining an event catalogue spanning thousands of years. Using Open-Street maps in a Python environment we analyzed Cambodia road network, identifying, as well, the most poorly connected locations (i.e., the Origin points), and the service centers (i.e., the Destination points). Combining the results of the road network analysis with the outputs of the probabilistic model we produced a map highlighting those regions where connectivity losses are expected to be high, as an indicator of where to expect direct hazard impacts to road network will host larger effects to the overall network functionality. On the other hand, we obtained the probability of disruption for each of the road segments defined. By computing which of them are associated to each Origin-Destination shortest paths, we managed to propagate the road disruptions conditioned by the hazard footprint to an indirect measure of the impact (i.e., the ability of the road network to provide services). Thus, we finally obtained a probability measure of the isolation potential for the selected locations in Cambodia.