Application of an Artificial Neural Network for Storm Surge Forecasting

Abstract:

Accurate and timely storm surge forecasts are required during tropical cyclone events to assess the magnitude and location of the storm surge impacts. Dynamical models provide accurate measures of storm surge but are too computationally expensive to be run for real-time forecasting purposes. Therefore, real-time forecasting of storm surge impacts is usually conducted utilizing a parametric vortex model, implemented within a hydrodynamic model, which decreases computational time at the expense of forecast accuracy. Recently, data-driven artificial neural networks are being implemented as an alternative due to their combined efficiency and high accuracy. This work seeks to examine how an artificial neural network can be used to make accurate storm surge predictions and to understand how different tropical cyclone parameters contribute to storm surge. The neural network model was trained with modeled data resulting from coupling of the Hybrid WRF cyclone model (HWCM) and the Advanced Circulation Model (ADCIRC). An ensemble of synthetic, but physically plausible, cyclones was simulated using the Hybrid WRF cyclone model, and used as input for the hydrodynamic model. Tests of the artificial neural network were conducted to determine the different lead-time configurations needed to minimize storm surge forecast errors and the variables needed for successful implementation and forecasting. The artificial neural network was successful in forecasting moderate storm surge levels, while limitations were observed in the ability to accurately predict peak-value water levels. Tropical cyclone intensity and position were identified as the most important variables for accurate storm surge predictions.