Enhancing overland flow tsunami modelling across urban topography with novel statistical emulation

现有的海啸模拟模型往往依赖于裸地地形来模拟淹没和陆地流。实际上，淹没沿海地区的水流要复杂得多。建筑物、构筑物、植被和其他障碍物的存在都会影响海啸淹没期间的水流。然而，许多海啸模型为了提高计算速度而忽略了地面障碍。平滑粒子流体动力学（SPH）模型能够模拟如此复杂的流动，但即使在局部尺度上，计算要求也很高，并且需要对稳定性和精度进行调整。此外，SPH模型不能快速有效地模拟其产生和越洋传播。因此，我们将把传统的海啸模型与SPH求解器dualspphysics结合起来，使用最先进的GPU集群，生成逼真的沿海海啸影响的端到端模拟。这将是全球海啸模拟的首次创新，因为最近的工作只是在简化的环境下完成的。该研究项目将侧重于特定的案例研究（迄今为止，新西兰的Whakatane和印度尼西亚的Cilacap）。在第二步中，我们将探讨使用统计模拟来产生概率海啸危害评估。仿真器是复杂计算机模型的近似值，与基于仿真的建模相比，计算量减少了几个数量级。我们将在耦合模拟器上安装一个模拟器。它将提供模拟输出的近似值，包括淹没范围、速度和高度，并考虑到城市地形。它将至关重要地使不确定性从海啸源传播到可能的影响，这是这种规模的第一次。然后，我们将通过将仿真过程应用于一系列应用程序来验证仿真过程，以帮助评估它们在灾难管理中的实际用途。迄今为止，此类应用的目标是量化城市结构的压力和应力，并了解港口内的流体速度。该研究还将通过模拟过去和潜在的未来事件来测试这些模拟器的准确性和精度，在此过程中，比较仿真模型、模拟器和现实产生的输出。后来，使用搅拌机使用先进的图形可视化技术，并展望这些拟合模拟器的未来可能性将进行调查和讨论。

Existing tsunami simulation models often rely on bare-earth topography to model inundation and overland flow. In reality, the flow of water inundating coastal regions is far more complex. The presence of buildings, structures, vegetation, and other obstacles, all influence the flow of water during tsunami inundation. However, many tsunami models disregard terrestrial obstacles in favour of increased computation speeds. Smoothed particle hydrodynamics (SPH) models are able to model such complex flows but are computationally demanding, even at a local scale, and require tuning for stability and precision. Furthermore, SPH models cannot quickly or efficiently model the generation and transoceanic propagation. So, we will couple a traditional tsunami model with an SPH solver, DualSPHysics, to generate realistic end-to-end simulations of coastal tsunami impacts, using state-of-the-art GPU clusters. This would constitute a first innovation in tsunami modelling worldwide, as recent work has been done only on simplified settings. This research project will focus on particular case studies (to date, Whakatane, New Zealand, and Cilacap, Indonesia).In a second step we will explore the use of statistical emulation to produce probabilistic tsunami hazard assessments. Emulators are approximations of complex computer models with orders of magnitude reduction in computation, compared to the simulation-based modelling. We will fit an emulator to the coupled simulators. It will deliver approximations of simulation outputs, including inundation extent, velocity and height, accounting for the urban topography. It will crucially enable the propagation of uncertainties from tsunami sources to possible impacts, a first at this scale.We will then validate the emulation process by applying it to a range of applications, looking to aid the evaluation of their practical use in disaster management. Such applications have, to date, looked at aiming to quantify pressures and stresses on urban structures and understand fluid velocities within harbours. The research will also test the accuracy and precision of these emulators by modelling past and potential future events, in doing so, comparing outputs produced by simulation models, emulators, and reality. Later, the use of advanced graphical visualisation techniques using Blender, and looking into the future possibilities for these fitted emulators will be investigated and discussed.