Storm Surge Modeling in the AI ERA: Using LSTM-based Machine Learning for Enhancing Forecasting Accuracy (2403.04818v1)

Abstract: Physics simulation results of natural processes usually do not fully capture the real world. This is caused for instance by limits in what physical processes are simulated and to what accuracy. In this work we propose and analyze the use of an LSTM-based deep learning network ML architecture for capturing and predicting the behavior of the systemic error for storm surge forecast models with respect to real-world water height observations from gauge stations during hurricane events. The overall goal of this work is to predict the systemic error of the physics model and use it to improve the accuracy of the simulation results post factum. We trained our proposed ML model on a dataset of 61 historical storms in the coastal regions of the U.S. and we tested its performance in bias correcting modeled water level data predictions from hurricane Ian (2022). We show that our model can consistently improve the forecasting accuracy for hurricane Ian -- unknown to the ML model -- at all gauge station coordinates used for the initial data. Moreover, by examining the impact of using different subsets of the initial training dataset, containing a number of relatively similar or different hurricanes in terms of hurricane track, we found that we can obtain similar quality of bias correction by only using a subset of six hurricanes. This is an important result that implies the possibility to apply a pre-trained ML model to real-time hurricane forecasting results with the goal of bias correcting and improving the produced simulation accuracy. The presented work is an important first step in creating a bias correction system for real-time storm surge forecasting applicable to the full simulation area. It also presents a highly transferable and operationally applicable methodology for improving the accuracy in a wide range of physics simulation scenarios beyond storm surge forecasting.

• The paper presents an LSTM-based technique that corrects systemic biases in physics-based storm surge models using historical hurricane data.
• It employs rigorous preprocessing with a sliding window approach to effectively capture the temporal dynamics of storm surge events.
• The model demonstrates significant accuracy improvements during Hurricane Ian, showcasing its potential for operational real-time forecasting.

Enhancing Storm Surge Forecasting Accuracy Through LSTM-Based Machine Learning

Introduction

The accuracy and reliability of storm surge modeling are pivotal in predicting the potentially devastating impact of hurricanes. Despite the advancements in high fidelity physics-based models such as ADCIRC coupled with SWAN, limitations inherent in these deterministic models often introduce biases and inaccuracies, particularly in the face of the unpredictable nature of severe weather events. The paper by Stefanos Giaremis et al. on "Storm Surge Modeling in the AI Era: Using LSTM-based Machine Learning for Enhancing Forecasting Accuracy" presents an innovative approach to mitigate these issues by employing Long Short-Term Memory (LSTM) networks, a class of recurrent neural networks, in the post-processing phase of storm surge modeling to correct systemic errors and enhance forecasting accuracy.

Methodology

The cornerstone of the research methodology is the utilization of LSTM networks to model and predict the systemic errors—termed as offsets—between the observed real-world water heights and those forecasted by a physics-based model during hurricane events. The LSTM architecture was chosen for its proficiency in capturing temporal dependencies, essential for accurately modeling the sequential data of storm surges. The paper trained the LSTM model on a comprehensive dataset encompassing 61 historical storms, using offsets extracted from the comparison of modeled versus observed water levels at gauge stations. A notable aspect of the research methodology is the rigorous data pre-processing, which includes offset extraction, standardization, and reshaping through a sliding window approach, making the raw data amenable to LSTM processing. The performance of the LSTM-based model was then tested for bias correction on data from hurricane Ian, a recent and significant event, ensuring the model's real-world applicability and generalization capabilities.

Results and Discussion

The research presented compelling numerical results, with the LSTM model showing consistent improvement in forecasting accuracy for hurricane Ian across all gauge station coordinates used for the initial data. This improvement was quantified using several regression evaluation metrics, such as root mean square error (RMSE) and R-squared coefficients, showcasing strong numerical evidence of the model's efficacy in bias correction. An intriguing finding was the model's ability to achieve a similar quality of bias correction using only a subset of six hurricanes. This implies not only the model's robustness but also its operational versatility, allowing for pre-trained models to be applied in real-time hurricane forecasting, significantly reducing the computational overhead typically associated with ensemble and probabilistic forecasting approaches.

Theoretical and Practical Implications

Theoretically, this work enriches the existing literature on the integration of machine learning techniques, particularly LSTM networks, within the field of weather forecasting and storm surge modeling. Practically, the LSTM-based bias correction methodology offers a highly transferable and operationally viable solution to improve the accuracy of storm surge forecasts, potentially enhancing the preparedness and response mechanisms of affected regions. Furthermore, the paper sets the groundwork for future explorations into the application of LSTM and other advanced machine learning techniques for bias correction across a wider range of physics simulation scenarios beyond storm surge forecasting.

Future Directions

While this paper marks a significant advancement, it also opens avenues for further research, particularly in exploring alternative machine learning architectures like bidirectional LSTMs with self-attention mechanisms, which could offer further improvements in capturing the complex temporal dynamics of storm surges. Additionally, extending the LSTM-based bias correction to a geospatial domain, enabling model corrections beyond gauge stations and throughout the entire simulation area, represents a compelling future challenge.

In conclusion, the paper by Stefanos Giaremis et al. presents a pioneering approach to enhancing storm surge forecasting accuracy using LSTM-based machine learning. By accurately modeling and correcting the systemic errors of physics-based models, this research offers new possibilities for improving the resilience of coastal communities against the ever-present threat of hurricanes