MLP-NN model-based bias correction method for significant wave height hindcast data trained based on satellite altimetry observations

Tools

Amarouche, Khalid and Akpinar, Adem and Kankal, Murat and Kamranzad, Bahareh (2023) MLP-NN model-based bias correction method for significant wave height hindcast data trained based on satellite altimetry observations. In: 10th Short Course/Conference on Applied Coastal Research, 2023-09-04 - 2023-09-06.

[thumbnail of Amarouche-etal-2023-MLP-NN-model-based-bias-correction-method-for-significant-wave-height]
Preview
 Text. Filename: Amarouche-etal-2023-MLP-NN-model-based-bias-correction-method-for-significant-wave-height.pdf
Accepted Author Manuscript
License: Strathprints license 1.0
Download (129kB)| Preview

Abstract

OVERVIEW Knowledge of wave climate has become crucial for all marine activities, e.g. coastal and offshore structure design, naval architecture and marine renewable energy exploitation. For this application, it is necessary to dispose of accurate hindcast wave data. Accurate hindcast of significant wave height (SWH) allows us to ensure sustainable and economic development of coastal and offshore structures and to ensure an accurate projection of change and trend in SWH. The performance of 3rd generation spectral wave models has been evaluated for the Black Sea through several studies (Amarouche et al., 2021a; Soran et al., 2022). the results revealed that the accuracy of the wave model for estimating SWH varies depending on the sea location and the wind climate of the area concerned by the simulation (Amarouche et al., 2021b). This variation may depend on the dominance of the swell compared to the wind sea in each location. Thus, the spatial variation in the model accuracy may depend on the precision of the wind input (Çakmak et al., 2019). The calibration of wave models often allowed an improvement in the prediction of SWHs. However, varying amounts of bias can be observed depending on the geographical area and local wind conditions. The bias variation between the different locations also depends on the swell and wind sea contribution rate. There are currently several methods proposed for wave bias correction. Those evaluated by Parker and Hill (2017) are among these methods. Recently, a deep learning-based method was proposed for ocean wave correction (Sun et al., 2022). Thus ANN models are applied for Bias Correction of Operational Storm Surge Forecasts by (Tedesco et al., 2023). STUDY OBJECTIVES This study proposes an innovative approach for SWH correction at the spatial scale and 1D wave spectra correction at the local scale. The proposed method is based on the perceptron multilayer placement ANN model and satellite altimeter data. Nothing that the estimated SWH accuracy mainly depends on the local wind accuracy, the estimated swell partition, and other geographical factors at the spatial scale, the ANN model is trained and tested based on an estimated wind sea and swell partitions, coastal distance, and bathymetry as input, and satellite altimetry observations as output data for spectral biases correction. Figure 1 shows the used MLP-NN bias correction model structure. PRIMARY RESULT The ANN model developed in this study has resulted in a considerable improvement in the accuracy of SWH hindcast data in the whole Black Sea. The methodology proposed here can be further improved by increasing the wave data measurements within the ANN model training and testing process.

ORCID iDs

Amarouche, Khalid, Akpinar, Adem, Kankal, Murat and Kamranzad, Bahareh ORCID logoORCID: https://orcid.org/0000-0002-8829-6007;
CORE (COnnecting REpositories)