The present study introduces a framework for predicting nearshore waves using two machine learning techniques of Group Method of Data Handling (GMDH) and Artificial Neural Network (ANN), trained with three global wave datasets of Japan Meteorological Agency (JMA), National Oceanic and Atmospheric Administration (NOAA), and European Centre for Medium-Range Weather Forecasts (ECMWF). Prior to our ultimate goal of forecasting nearshore waves up to one week in advance, the current study challenges to hindcast nearshore wave heights and periods for a target year at the Port of Hitachinaka, Japan using the framework compounding GMDH and ANN trained with the initially forecasted (0 h) and reanalyzed two datasets. It was found that the GMDH-based wave model, trained with NOAA and ECMWF, well predicted observed significant wave heights, while a combination of JMA and ECMWF for training gave the best performance for significant wave periods. The same tendency was found when using ANN. Since the present framework successfully transforms global waves into local nearshore waves, it can be said that the framework for the nearshore wave prediction is able to support the one week ahead wave prediction and to be implemented at a particular location, where the nearshore wave observations are available.

本研究介绍了一种使用数据处理的分组方法（GMDH）和人工神经网络（ANN）的两种机器学习技术预测近岸波浪的框架，并采用了日本气象厅（JMA），国家海洋和大气的三个全球波浪数据集进行了训练政府（NOAA）和欧洲中型天气预报中心（ECMWF）。在我们提前一个星期预测近岸海浪的最终目标之前，当前的研究面临的挑战是使用结合了初始预测的GMDH和ANN的框架将日本Hitachinaka港口的目标年份的近岸海浪高度和周期进行预测（0 h）并重新分析了两个数据集。研究发现，使用NOAA和ECMWF训练的基于GMDH的波浪模型可以很好地预测观测到的重要波浪高度，而JMA和ECMWF的组合训练在重要的波浪周期内表现最佳。使用ANN时发现了相同的趋势。由于本框架成功地将全球海浪转换为本地近海浪，因此可以说，近海浪预测框架能够支持提前一周的海浪预测，并能够在特定地点进行实施，该特定地点是近岸海浪观测所在。可用。