A Bayesian framework to assess and create risk maps of groundwater flooding

Groundwater flooding can cause severe damages to homes, utilities, and infrastructure and yield significant economic and social costs. Numerical models are used to understand these events and are the basis to produce imagery products for risk management and communication. However, such maps are generally produced using forward model simulations, and most of the mapping products are still deterministic. In contrast to pluvial and fluvial floods, an open issue in the analysis of susceptibility to groundwater flooding is the lack of probabilistic assessment and mapping products recognizing parametric uncertainty. Hence, we propose a Bayesian-based framework to create probabilistic risk maps and to identify the susceptibility to groundwater flood events. We aim to assess the spatial distribution and temporal variability of groundwater flooding by decomposing the uncertainty and the sensitivity of distributed groundwater numerical models. The scheme involves the use of the elementary effects method, the DiffeRential Evolution Adaptive Metropolis (DREAM) algorithm, and the exploration of the predictive posterior distributions of the groundwater heads to evaluate the susceptibility according to exceedance levels. We use the proposed Bayesian framework with a numerical model that simulates the groundwater flood event that occurred in the valley of the Alz River in 2013. This study developed two types of susceptibility maps based on the exceedance probability of certain groundwater levels and specific cellar depths. The Bayesian inference supports the parameter estimation and thereby increases the spatial confidence of the areas susceptible to inundation. This study shows that maps of susceptibility to groundwater flooding can be built over one single event models and acknowledge the inherent spatial and temporal uncertainty of such events.

Automated summary

Groundwater flooding can cause significant damages and costs. Existing mapping products lack probabilistic assessment and recognition of parametric uncertainty. Therefore, this study proposes a Bayesian-based framework to create probabilistic risk maps for groundwater flood events.