Speaker
Description
Ensuring that warning sirens are audible to the population during natural disasters, technological accidents, or security crises remains a key challenge in emergency planning, while their spatial deployment is still not systematically optimized on their audibility. This study proposes a two-step approach. In France, public warning sirens are tested every first Wednesday of the month at noon. The first step relies on questionnaires distributed to residents after these monthly tests in order to derive an empirical audibility curve based on reported perception of the signal. The second step uses open-source tools for environmental noise modelling and multi-objective optimization: NoiseModelling, based on the CNOSSOS-EU propagation model, and OpenMole, implementing the NSGA-II evolutionary algorithm. Their coupling enables exploration of alternative siren configurations and identification of Pareto-optimal solutions according to two objectives: (1) the number of buildings exposed above 80 dB, and (2) the total area exposed above this threshold. A case study on Saint Barthelemy Island shows that the optimized Pareto front ranges from 7836 to 7858 dwellings and from 15.29 to 15.31 km². These results provide a set of acoustically optimal configurations and illustrate how such modelling tools can support decision-making, while complementing expert-based approaches that also consider non-acoustic constraints.