”Failure is success if we learn from it.” -Malcolm Forbes
We conduct research utilizing machine learning, numerical modeling, and remote sensing to study extreme hazards in coastal and ocean environments.
- Coastal hazards (storm surges, tsunamis, and compound flooding)
- Climate and coastal resilience
- Nature-based solutions in coastal and ocean environments
- Coastal vulnerability and risk assessments
- Coastal processes
(1) City Digital Twin System for Coastal Flood Risk Assessment
This system will integrate extensive built environment data with an AI-based flood model to enhance situation assessment capabilities and facilitate informed long-term urban development decision-making. This system will provide broader impacts to society by giving decision-makers a comprehensive view of the potential consequences of flood scenarios and enabling them to evaluate, predict and guide the impact of urban development that can affect housing, jobs, and public services in cities, ultimately contributing to social equity.
Sponsor: Good Systems@UT-Austin
(2) Tsunami Machine Learning Model
We extend Lee et al.’s (2021) machine-learning model to facilitate rapid prediction of tsunami inundation from heterogeneous earthquake slip distributions. This model operates without the need for additional pre-calculated databases or observation data. This new model has the potential to enhance tsunami prediction capabilities, ultimately contributing to rapid tsunami forecasting and robust tsunami hazard assessment.
(1) Storm Surge Machine Learning Model
Storm surge by hurricane is a devastating threat to coastal communities in Virginia. Significant advances in physics-based models over the last several decades allow accurate simulation of storm surge at high resolution, and thus with high accuracy. The problem, however, is that a high-fidelity model like ADCIRC is computationally intensive. In this study, we developed a machine learning model that can rapidly predict the storm surge water level based on the pre-computed high-resolution dataset of the U.S. Army Corps of Engineers’ North Atlantic Comprehensive Coastal Study (NACCS) (Lee et al., 2021).
(2) Data-Driven Tsunami Inversion Model
Tsunamis are rare but can cause serious damage to coastal communities once they occur. To better understand the tsunami generation process and its impact, tsunami inversion model has been widely developed. In this study, we developed a new TRRF-based inversion model that can directly infer a near-field tsunami source and tsunami run-up distribution from a few number of run-up records and local tectonic features (Lee et al., 2021).
(3) Data-Driven Tsunami Run-Up Prediction Model
A tsunami run-up response function (TRRF) is a data-driven model that can rapidly predict the near-field tsunami run-up distribution for any combination of fault parameters in real bathymetric and topographic conditions. The main idea of the TRRF is that the alongshore tsunami run-up distribution can be divided into source run-up (leading order contribution) and topographic run-up (residual part) terms, where the former is based on the Okal and Synolakis’ (2004) empirical formula. The results show that the TRRF is fast as an analytical approach (Computational time: < 1 sec) and as accurate as the numerical model (Lee et al., 2020).
(4) Tsunami Arrival Time Detection System
Timely detection of tsunamis with water level records is a critical but logistically challenging task because of outliers and gaps. We developed the Tsunami Arrival time Detection System (TADS), which can be applied to discontinuous time series data with outliers. TADS consists of three algorithms, outlier removal, gap filling, and tsunami detection, which are designed to update whenever new data are acquired. The results show that the overall performance of TADS is effective in detecting a tsunami signal superimposed on both outliers and gaps (Lee et al., 2016).
(5) ANN-Based Gap-Filling Algorithm
Accurate prediction of missing water levels attributable to reasons ranging from recording failure and transmission problems to mistakes made by field staff is essential in coastal and oceanic areas. We developed a new system for data recovery based on the artificial neural network. The results indicate that although the performance of the proposed system declines marginally as the gap size increases, it performs creditably in alleviating the gap-filling problem (Lee and Park, 2016).
(6) Tsunami Forerunner Simulation
The 2011 Tohoku Tsunami reached the Korean Peninsula and was recorded at numerous tide stations. In the records of the north-eastern tide stations, tsunami forerunners were found in only about a few minutes after the earthquake, which was much earlier than the expected arrival time based on a conventional numerical simulation. We investigated the tsunami forerunners observed in Korea by a numerical simulation considering the bathymetry effect (Lee et al., 2016).
(7) Global Tsunami Prediction System
We developed a global tsunami prediction system for a distant tsunami using a finite fault model and a cyclic boundary condition. We compared the numerical simulation results (tsunami height and arrival time) with different conditions (boundary condition, governing equation, grid size and fault model) and measured data (DART buoy, tide station) and showed the importance of the finite fault model and the cyclic boundary condition (Lee et al., 2015).