Accounting for Spatio-Temporal Dependencies in Flood Hazard Assessment at the Basin Scale

Flood hazard assessment and mapping often rely on event-based approaches that assume uniform return periods for peak flows across an entire watershed. However, this simplification neglects the spatial and temporal heterogeneity intrinsic to flood events, potentially leading to inaccuracies in hazard estimation and, consequently, risk assessment. While many recent studies applying multivariate extreme value models focus on large-scale systems, this research applies the Heffernan and Tawn (HT) multivariate conditional exceedance model at the basin scale, using the Lambro River in Northern Italy as a test case.

The used hydrometric data required careful preprocessing to address gaps due to gauge malfunctioning or the lack of an appropriate rating curve to convert measured depths into flow rates. Missing data were handled using the Multiple Imputation by Chained Equations (MICE) method. This approach iteratively models missing values by leveraging relationships among variables, ensuring that the imputed data preserves the underlying structure and variability of the original dataset.

The Heffernan and Tawn (HT) multivariate conditional exceedance model was used to analyze the spatio-temporal dependencies of extreme flow rate values. The HT model characterizes the joint behavior of variables by conditioning the distribution of one variable on the exceedance of a high threshold by another, allowing the realistic modeling of flood scenarios. After the dependence structure was determined, Monte Carlo simulations were employed to generate synthetic events based on the estimated model’s parameters, producing a comprehensive set of scenarios that account for the spatial heterogeneity and temporal variability in extreme flows. The synthetic event generation thus captured the intricate dependencies between peak flows across locations, enabling the synthetic events to reflect realistic flood scenarios. By focusing on a small-scale rather than a regional or continental one referred to in prior applications of this method, this work aims at improving hazard assessment tools at the basin level.

In order to exploit the generated events in hazard assessment, one needs to (i) develop an approach to obtain the multivariate probabilities of occurrence for the generated scenarios, which remains a challenging and unresolved task, (ii) execute multiple hydrodynamic simulations across the range of generated scenarios, and (iii) statistically synthesize the simulation results.