Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Stefanos Giaremis, Noujoud Nader, Clint Dawson, Hartmut Kaiser, Carola Kaiser, Efstratios Nikidis

TL;DR

This work proposes and analyzes the use of an LSTM-based deep learning network machine learning (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.

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 machine learning (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.

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

TL;DR

This work proposes and analyzes the use of an LSTM-based deep learning network machine learning (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.

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 machine learning (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.
Paper Structure (19 sections, 7 equations, 11 figures, 5 tables)

This paper contains 19 sections, 7 equations, 11 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Data and Methodology
  3. Overview
  4. Data
  5. Machine Learning Model
  6. Long Short-Term Memory
  7. Convolution Layer
  8. Fully Connected Layer
  9. Activation Functions
  10. Structure
  11. Computational resources and model parameters
  12. Windowing and Data Preprocessing
  13. Evaluation Metrics
  14. Results and Discussion
  15. Scenario 1: ML bias correction using training data from a single hurricane
  16. ...and 4 more sections

Figures (11)

  • Figure 1: Overview of the methodology framework: (A) Data Pre-processing phase includes offset extraction using Eq. \ref{['eq:offsets']}, data cleaning, and standardization. (B) Modeling phase includes training/test data preparation with the sliding window approach for the offsets time series at each station. Different scenarios were analyzed and evaluated with several commonly used regression evaluation metrics. A single ML model is used in each scenario for all the considered stations.
  • Figure 2: (a) Visualization of hurricane Ian water levels in contour, along with the National Hurricane Center best track, available gauge stations data and the selection of three stations used for the demonstration of the performance of the ML model, within the CERA framework (CERA, CERA2023). (b) Visualization of the observed and modeled water level, along with the corresponding offset time series (as defined by eq. \ref{['eq:offsets']}, from an example gauge station (Springmaid Pier) during hurricane Ian, as obtained from the interactive CERA framework (CERA, CERA2023).
  • Figure 3: Architecture of the proposed ML-model for offset prediction. The input samples are the offset time series windows. The architecture consist of a Convolution layer followed by multiple LSTM layers. The output of the model is the predicted offset window.
  • Figure 4: The sliding window approach: Data samples processed from a time series training sample by creating input-output pairs.
  • Figure 5: (a) Performance of the ML model in predicting offset data, as evaluated with the R$^2$ metric, as a function of input window size for different sizes of the prediction window. (b) Distribution of R$^2$ obtained in the considered range of input window sizes, as a function of the prediction window size. (c) Runtime required for the training of the model as a function of the input window, for different sizes of the prediction window. In all cases, the model was trained on the earliest 75% of offsets data from hurricane Ian (2022), while the rest 25% of the data of the same hurricane were used for testing its performance. 4 Intel® Xeon® Platinum 8260 processor cores were used for the training of the ML models.
  • ...and 6 more figures