CAREER: Harnessing Heterogeneous Sources of Data and Artificial Intelligence for Informed Flood Management

该教师早期职业发展（CAREER）项目通过推进对城市和沿海洪水的科学认识以及建立新一代智能洪水预警系统和智能防洪基础设施，支持国家在气候适应方面的研究重点。该项目将通过利用异构数据源（例如交通摄像头、智能手机和无人机拍摄的地面图像和视频）来解决洪水数据可用性和当前洪水建模方面的差距，提供更快、更分散的洪水时间序列数据和视觉信息。多源异构数据的使用，加上加速的洪水建模和数据分析，将通过实时模型更新、主动测量和主动洪水管理，支持将现有洪水基础设施转变为主动防洪基础设施。该项目将研究活动与教育和外展计划相结合，以（i）培训下一代人工智能工程师和科学家，（ii）培养具有洪水意识的社区，（iii）通过开发一个由洪水建模、规划和响应、行为分析以及虚拟现实游戏和外展四个阶段组成的综合循环学习框架为决策者提供信息。 该项目将研究南卡罗来纳州的两个沿海流域，一个主要是城市化的，另一个是自然化的，有可能转移到其他城市和沿海系统。将部署新型人工智能和图像处理工具来处理洪水管理不同阶段的不同类型的输入：数据采集、洪水检测、监测、模拟和预报。该项目的研究核心围绕一个集成建模平台的详细设计和开发，该平台由用于洪水数据分析、建模和管理的多重深度学习模型（MDLM）组成。在数据分析阶段，深度学习模型将单独使用或与研究区域的 3 维重建结合使用，从地面图像和视频中提供数值数据，例如水位和淹没区域。此外，还将开发多源数据融合模块，将来自不同卫星和频段的数据馈送到多分支深度学习网络，以进行洪水检测和特征提取。在建模阶段，该职业项目将通过将分布式水文和地表水力学数学模型集成到单一建模框架中，开发模拟沿海复合洪水的全耦合模型，从而增强对复合洪水的理解。然后，将开发一套基于机器学习的替代模型，以模仿基于物理的精细洪水模型的知识，并提供及时的洪水预测。最后，该项目将提供适应性设计指南，通过实时模型更新、主动测量和主动洪水管理，将现有基础设施转变为主动防洪基础设施。该项目中创建的新技术和工具将允许利益相关者、决策者和公众就洪水前、洪水中和洪水后的直接和间接参与做出选择，并使用综合数值模拟和虚拟现实游戏来评估其影响。该职业项目由民用基础设施系统 (CIS) 和既定计划刺激竞争性研究 (EPSCoR) 项目共同资助。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力优势进行评估，被认为值得支持。以及更广泛的影响审查标准。

This Faculty Early Career Development (CAREER) project supports the nation's research priority in climate adaptation through advancing scientific understanding of urban and coastal floodings and establishing a new generation of intelligent flood early warning systems and smart flood control infrastructure. The project will address gaps in flood data availability and in present-day flood modeling by harnessing heterogeneous data sources, such as ground-based images and videos taken by traffic cameras, smartphones and drones, to provide faster and more distributed flood timeseries data and visual information. The use of multi-source, heterogeneous data, along with accelerated flood modeling and data analytics, will support the transformation of existing flood infrastructure into Active Flood Control Infrastructure through real-time model updating, active measurements, and active flood management. The project integrates research activities with educational and outreach plans to (i) train the next generation of AI-enabled engineers and scientists, (ii) foster flood-aware communities, and (iii) inform decision-makers by developing an integrated looped learning framework consisting of four phases of flood modeling, planning and response, behavioral analysis, and virtual reality gameplay and outreach. The project will study two coastal watersheds in South Carolina, one dominantly urbanized and the other natural, with potential transferability to other urban and coastal systems.Novel Artificial Intelligence and image processing tools will be deployed to process different types of inputs at different stages of flood management: data acquisition, flood detection, monitoring, simulation, and forecasting. The research core of this project revolves around the detailed design and development of an integrated modeling platform consisting of Multi-Deep Learning Models (MDLM) for flood data analysis, modeling, and management. In the data analysis phase, deep learning models, alone or in combination with the reconstruction of a 3-Dimensional of the study area, will be used to provide numerical data, such as water levels, and inundation area, from ground-based images, and videos. Moreover, a multi-source data fusion module will be developed to feed data from different satellites and bands to a multi-branch deep learning network for flood detection and feature extraction. In the modeling phase, this CAREER project will enhance the understanding of compound flooding by developing a fully coupled model for simulating coastal compound floods through the integration of distributed hydrologic and surface hydraulics mathematical models into a single modeling framework. Then, a set of machine learning-based surrogate models will be developed to mimic the knowledge of fine-scale physics-based flood models and provide timely flood predictions. Finally, this project will provide adaptive design guidelines to turn existing infrastructure into active flood control infrastructure through real-time model updating, active measurements, and active flood management. New technologies and tools created in this project will allow stakeholders, decision-makers, and the public to make choices regarding their direct and indirect involvement pre-, peri-, and post-flooding and evaluate their impacts using integrated numerical simulations and virtual reality gameplay.This CAREER project is jointly funded by the Civil Infrastructure Systems (CIS) and the Established Program to Stimulate Competitive Research (EPSCoR) programs.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.