Numerical simulation of 2D real large scale floods on GPU: the Ebro River

Isabel Echeverribar,Pilar Brufau,Pilar García-Navarro,Mario Morales-Hernández,A Paquier,N Rivière

doi:10.1051/e3sconf/20184006007

Abstract

Modern flood risk management and mitigation plans incorporate the presence of numerical models that are able to assess the response of the system and to help in the decision-making processes. The shallow water system of equations (SWE) is widely used to model free surface flow evolution in river flooding. Although 1D models are usually adopted when simulating long rivers due to their computational efficiency, 2D models approximate better the behaviour in floodplains of meandering rivers using a fine mesh which implies unaffordable computations in real-world applications. However, the advances on parallelization methods accelerate computation making 2D models competitive. In particular, GPU technology offers important speed-ups which allow fast simulations of large scale scenarios. In this work, an example of the scope of this technology is presented. Several past flood events have been modelled using GPU. The physical domain (middle part of the Ebro River in Spain) has a extent of 477 km2, which gives rise to a large computational grid. The steps followed to carry out the numerical simulation are detailed, as well as the comparison between numerical results and observed flooded areas reaching coincidences up to 87.25 % and speed enhancements of 1-h of simulation time for 1-day flood event. These results lead to the feasible application of this numerical model in real-time simulation tools with accurate and fast predictions useful for flood management.

Highlights

A UN Survey [24] reveals the severity of flooding episodes regarding their consequences, in terms of material damages and concerning human losses
These models are able to simulate river flood-plain and complex geometries involving wet/dry boundaries. As they require a great level of detail in the form of number of cells to ensure a reliable accuracy, their disadvantage resides in the necessity to refine the computational grid and the large amount of computations, which turns into unaffordable simulations for real-time systems or large space and time domains
With the aim of comparing field and simulated data in the 2015 flood event, some observation points and cross sections have been distributed at strategic locations where gauging stations register water surface elevation (WSE) and discharge

Summary

Introduction

A UN Survey [24] reveals the severity of flooding episodes regarding their consequences, in terms of material damages and concerning human losses. Prediction and prevention procedures are planned in order to mitigate these effects and are essential to restraint the limits of those losses With this aim, simulation can be a competitive prediction tool if numerical models are able to provide accurate results at affordable computational time. Two-dimensional (2D) hydraulic models have been accepted as a way to overcome the 1D constraints These models are able to simulate river flood-plain and complex geometries involving wet/dry boundaries. As they require a great level of detail in the form of number of cells to ensure a reliable accuracy, their disadvantage resides in the necessity to refine the computational grid and the large amount of computations, which turns into unaffordable simulations for real-time systems or large space and time domains. The computational time required demonstrates that this type of models can be used in real time prediction systems

Governing equations and numerical scheme

Case study

Results

Conclusions