|
|
Publications
Latas, M., Ali, A., Silva, D. and Guedes Soares, C. (2024), Improving the accuracy of significant wave height hindcast data with long short-term memory models, Advances in Maritime Technology and Engineering, Guedes Soares, C. & Santos T.A. (Eds.), Taylor and Francis, London, UK, pp. 651-658.
To determine the capabilities of long short-term memory (LSTM) neural networks to extend the skill of wave forecasts, the ERA5 hindcast database is used as a test case. Data from a buoy located in the North Atlantic Ocean near Portugal continental coast is used as the reference value. The error between the ERA5 data set and the wave buoy records is used to train a LSTM model, which is then used to correct other samples of the ERA5 database. ERA5 feature selection is performed by computing the variance inflation factor and by applying sequential feature selection with forward selection, which resulted in the selection of 5 features for training the models. After extensive testing, an optimal number of 50 neurons and a filtering window size of 10 is selected, yielding LSTM models that improve the ERA5 estimates. The accuracy of the data sets adjusted with the LSTM model is compared with the one of the initial ERA5 data using four metrics: the normalized bias, the scatter index, the normalized root mean squared error, and the correlation coefficient.
If you did not manage to obtain a copy of this paper: Request a copy of this article
For information about all CENTEC publications you can download: Download the Complete List of CENTEC Publications
|