Скачать с ютуб Hydrologically Informed Generative AI for High-Resolution Flood Mapping в хорошем качестве

Hydrologically Informed Generative AI for High-Resolution Flood Mapping

Abstract: Physics-based models, such as hydrodynamic models, are crucial for accurate flood prediction. However, setting up and running high-resolution flood models is often computationally expensive and time-consuming. This limitation hinders the use of hydrodynamic models in real-time forecasting and probabilistic analyses, where numerous model simulations are required. Conversely, pure machine learning models, employed as surrogate models, offer both computational efficiency and prediction accuracy. Despite these advantages, they often lack explainability regarding the underlying mechanisms of their predictions. In this study, we introduce a novel physics-informed machine learning model, termed Hy-NET, designed for rapid high-resolution flood mapping. According to the Theory-Guided Data Science (TGDS) taxonomy, Hy-NET is classified as a hybrid TGDS model. Hy-NET is a U-NET-based deep learning model that leverages low-resolution hydrodynamic model predictions as initial estimates, subsequently upskilling them to align with high-resolution model outcomes. A low-resolution model operates significantly faster than a high-resolution model for the same study area, due to fewer computational cells and the ability to use larger computational time-steps. Despite these differences, the results of low-resolution models are largely correlated with those of high-resolution models. The Hy-NET model utilises low-resolution images of 512×512 pixels, along with corresponding digital elevation models, as its inputs. In the context of artificial intelligence (AI), Hy-NET falls under the umbrella of generative AI, as it generates new images that resemble the given training dataset. The proposed approach has been tested on three Australian watersheds: Wollombi, Burnett River, and Chowilla. The HEC-RAS model is employed to generate both low-resolution and high-resolution flood maps for model training and testing. While being much more computationally efficient than high-resolution models, the Hy-NET model demonstrates significant upskilling capability, achieving results that closely match high-resolution flood model outputs in terms of water depths and flood inundation extents. Dr Viraj Vidura Herath: Viraj Herath is a Senior Researcher at Macquarie University within the Faculty of Science and Engineering, a role he has held since December 2023. A civil engineer with a specialization in water resources, Viraj’s research focuses on physics-informed machine learning, flood modelling, and rainfall-runoff modelling. In 2015, Viraj completed his BSc in Engineering at the University of Peradeniya, Sri Lanka, where he was awarded the Ceylon Development Engineering Prize for Best Performance in Civil Engineering. He then worked as a temporary lecturer in the Faculty of Engineering at the same university for one year. In 2017, Viraj was awarded the President’s Graduate Fellowship from the National University of Singapore to pursue his PhD studies. He earned his PhD in 2021 with a thesis titled “Hydrologically Informed Machine Learning for Rainfall Runoff Modelling.” Prior to his current position in Australia, Viraj served as a Senior Hydraulic and Hydrological Modeller at the Hydroinformatics Institute in Singapore for three years. Professor Lucy Marshall: Lucy Marshall is Executive Dean of Faculty of Science and Engineering at Macquarie University in Sydney. Lucy received her Bachelor of Civil Engineering in 2001, Master of Engineering Science in 2002 and her PhD in Civil and Environmental Engineering in 2006, all from the University of New South Wales (UNSW) in Sydney. From 2006-2013 she was Associate Professor of Watershed Analysis at Montana State University, working at the interface of engineering and environmental science in quantifying uncertainty in hydrologic and environmental systems. Following this she was Director of the Water Research Centre in the School of Civil and Environmental Engineering and Associate Dean of Engineering (Equity and Diversity) at UNSW. Lucy’s technical expertise is in hydrologic modelling, model optimisation, and quantification of uncertainty in water resources analysis. Lucy is an expert in Monte Carlo methods, Bayesian inference, and associated methodologies aimed at improved uncertainty analysis of water resources modelling and the assessment of uncertainty in water resources.