EAGER: Exploring Machine Learning and Atmospheric Simulation to Understand the Role of Geomorphic Complexity in Enhancing Civil Infrastructure Damage during Extreme Wind Events

受2017年9月飓风玛丽亚登陆对波多黎各造成的广泛破坏的推动，这项早期概念探索研究补助金(EARGER)将研究复杂的地形如何加速风力，并最终加剧对建筑物和其他已建民用基础设施的破坏。这项研究将利用机器学习和天气预报方面的最新进展来预测山区地形和其他复杂陆地环境中的风速。该项目将利用美国国家科学基金会支持的自然危害工程研究基础设施(NHERI)佛罗里达大学的Terraform边界层风洞(BLWT)来表征波多黎各以及别克斯岛和库莱布拉市的几何模型上的地面风场。这是佛罗里达大学(为美国飓风最多发的州提供服务)之间的合作。这两所大学的研究生和本科生将积极参与实验和计算工作。预期的项目成果将包括关于地形对破坏性风行为的影响的重要新见解，将实验与先进计算方法融合在一起的新科学工具，以研究极端风对已建民用基础设施的影响，以及将提供给国家核研究所数据仓库(https://www.DesignSafe-ci.org).)其他研究人员的基准数据集该项目创造的知识可以为未来的研究和风荷载规定提供参考，以提高美国对飓风影响的韧性，从而更好地确保国家在风暴事件后的福利和繁荣。这项研究将在多个方面取得知识进步。它将调查地貌风对波多黎各的影响(即加速)，目的是促进对地貌复杂性(地形)如何增强地面风并使民用基础设施更容易受到破坏的理解。具体地说，这项研究将探索和评估机器学习和多尺度大气模拟的预测能力，即嵌套在数值天气预报(NWP)框架内的计算流体力学。为支持这项工作，将从Terraform BLWT的精确制导粒子图像测速系统中收集波多黎各以及别克斯岛和库莱布拉市按几何比例划分的模型上的高分辨率立体速度场。实验的目的将是为改进BLWT建模提供关键见解，同时产生基础数据集以评估(A)基于监督回归的机器学习在预测迎风高度变化如何修改气流方面的有效性，以及(B)用于确定在结构风荷载供应中应在哪里应用“特殊”风区的基于监督分类的机器学习方法。同时，还将应用数值预报增强大涡模拟(LES)来证明NWP-LES可以改善建筑环境中飓风风场的后向预报。如果成功，这一努力将极大地帮助工程学和大气科学领域在极端风事件期间预测地面风行为的标准化方法上达成共识。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Motivated by the extensive damage to Puerto Rico caused by Hurricane Maria's landfall in September 2017, this EArly-concept Grant for Exploratory Research (EAGER) will study how complex topography can accelerate wind and, ultimately, exacerbate damage to buildings and other constructed civil infrastructure. This research will utilize recent advancements in machine learning and weather forecasting to predict wind speed-up in mountainous terrain and other complex terrestrial environments. The project will leverage the NSF-supported Natural Hazards Engineering Research Infrastructure (NHERI) Terraformer Boundary Layer Wind Tunnel (BLWT) at the University of Florida to characterize the surface wind field over geometrically scaled models of Puerto Rico and the municipal Islands of Vieques and Culebra. This EAGER is a collaboration between the University of Florida (which serves the most hurricane prone state in the U.S.) and the University of Puerto Rico at Mayaguez (a Hispanic-serving institution still recovering from Hurricane Maria), and graduate and undergraduate students from both institutions will be actively involved in the experimental and computational work. Anticipated project outcomes will include important new insights about the influence of topography on the behavior of damaging winds, new scientific tools that fuse experimentation with advanced computing methods to study extreme wind effects on constructed civil infrastructure, and benchmark datasets that will be made available to other researchers in the NHERI Data Depot (https://www.DesignSafe-ci.org). Knowledge created by this project can inform future research studies and wind load provisions to improve the resilience of the U.S. to hurricane impacts, and thus better secure the nation's welfare and prosperity after windstorm events. This research will make knowledge advancements on multiple fronts. It will investigate topographic wind effects (i.e., speed-up) on Puerto Rico, with the goal of advancing understanding of how geomorphic complexity (topography) enhances surface winds and makes civil infrastructure more vulnerable to damage. Specifically, the research will explore and assess the predictive capability of machine learning and multi-scale atmospheric simulation, i.e., computational fluid dynamics nested within a numerical weather prediction (NWP) framework. To support this effort, high-resolution stereoscopic velocity fields over geometrically scaled models of Puerto Rico and the municipal Islands of Vieques and Culebra will be collected from a precision-guided particle image velocimetry system in the Terraformer BLWT. Experiments will be designed to yield critical insights for improving BLWT modeling, while producing foundational datasets to assess the efficacy of (a) supervised regression-based machine learning at predicting how changes in the upwind elevation modify flows and (b) supervised classification-based machine learning methods for determining where "special" wind regions should apply in structural wind load provisioning. Concurrently, NWP enhanced with large eddy simulation (LES) will be applied to demonstrate that NWP-LES can improve the hindcasting of a hurricane's wind field in the built environment. If successful, this effort can critically aid the engineering and atmospheric science fields in reaching consensus on standardizing approaches to predict the behavior of surface winds during an extreme wind event.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.