Modeling the Dynamics of Social Preparedness with Regional Interactions to Realize Impact-based Flood Early Warning System

In designing a flood early warning system, it is necessary to consider how people respond to forecasts and warnings—i.e., impact-based forecasting and warnings are required. The capacities to anticipate and respond to an imminent or ongoing disaster are called "social preparedness," and the following three factors are considered important in its formation: 1. direct experiences with hazard events, 2. indirect experiences, which include media and hazard witnessing, 3. trust in warnings. However, no socio-hydrological models have considered all of these factors. This study aims to model the dynamics of social preparedness affected by three key factors and propose socially optimal warning strategies. We extend the existing socio-hydrological model (i.e., a model in which the social preparedness of a region is shaped by the region's direct disaster experience and trust in warnings) and formulate a network model with each region as a node. The model describes the interactions among regions, in which the social preparedness for a region is determined by the disaster experiences and warning performances of other regions. In other words, such interactions allow the model to include the effect of indirect experiences on social preparedness. The analysis shows that when there is heterogeneity among regions, i.e., different infrastructure levels (e.g., levee heights) and different flood frequencies according to the region, the cry wolf effect is more pronounced due to regional interactions. That is, many false alarms undermine the credibility of the early warning systems and make it difficult to induce preparedness actions. The finding suggests the importance of considering the risk of disaster occurrence in surrounding areas and interactions with them when designing flood warning systems. Furthermore, the analysis shows that the single warning criterion optimized for all regions may not lead to optimizing individual regions. This result suggests that it may be appropriate to set warning criteria according to the region.