Clustering methods for decision making: application to flood risks and radiological emergencies

In case of natural and / or technological disaster, decision making relies on predictions based on available information, monitoring data and model-based forecasts. Uncertainties are particularly high in emergency situations, with scarce information and strong time constraints [1].

Uncertainty quantification and propagation methods are well established and used in numerous applications such as meteorological forecasting and risk evaluation in various domains (seismic hazard, flooding, environmental consequences of radioactive or chemical releases…). However, there are still challenges in taking these uncertainties into account for decision making, particularly in case of emergency. These challenges are of different natures, shared among different domains and types of risks: (1) how to properly account for all sources of uncertainties, including deep uncertainties that cannot be quantified, inherent to crisis situations? (2) how to fit this uncertainty evaluation within the time constraints of emergency response? (3) how to present and communicate these evaluations in an understandable and practical way for decision makers, accounting for interpretation biases?

We propose a scenario-based approach that combines meta-modelling, to generate many simulations in a short time, with a clustering method that allows to select a few situations or “scenarios”, described by their probability of occurrence and associated impact. This approach is illustrated on two applications: flooding risk [2] and nuclear emergency [3]. This method will be applied in the Natech project within the France 2030 Risks-IRIMA program, to a marine submersion in the Gironde estuary combined with nuclear and industrial accidents. The aims will be (1) to include decision-oriented parameters (such as population or critical infrastructures) in the clustering process, (2) to involve stakeholder panels in the design of evaluation products, (3) to better understand how cognitive biases will affect the decision-making process for different kinds of risks and evaluation products.

[1]          P. Bedwell et al., ‘Operationalising an ensemble approach in the description of uncertainty in atmospheric dispersion modelling and an emergency response’, Radioprotection, vol. 55, no. HS1, Art. no. HS1, 2020, doi: 10.1051/radiopro/2020015.

[2]          C. Sire, R. Le Riche, D. Rullière, J. Rohmer, L. Pheulpin, and Y. Richet, ‘Quantizing Rare Random Maps: Application to Flooding Visualization’, J. Comput. Graph. Stat., pp. 1–16, Apr. 2023, doi: 10.1080/10618600.2023.2203764.

[3]          Y. El-Ouartassy, I. Korsakissok, M. Plu, O. Connan, L. Descamps, and L. Raynaud, ‘Combining short-range dispersion simulations with fine-scale meteorological ensembles: probabilistic indicators and evaluation during a 85Kr field campaign’, EGUsphere, vol. 2022, pp. 1–35, Aug. 2022, doi: 10.5194/egusphere-2022-646.