This paper explores the applicability of neural networks (NN) for forecasting meteotsunamis affecting Ciutadella harbour (Menorca, Balearic Islands, Spain). Virtually every year, Ciutadella suffers meteotsunamis with wave heights (crest-to-trough difference in about 5-min interval) around 1 m, and at several episodes in its modern history, the waves have reached 2–4 m. A timely and skilled prediction of these phenomena could significantly help to mitigate the damages inflicted to the port facilities and the moored vessels. Once properly trained, a NN is a computationally cheap forecasting method; the approach could be easily incorporated by civil services which are responsible for issuing warnings and organizing a prompt response. We examine the relevant physical mechanisms that promote meteotsunamis in Ciutadella harbour and choose the input variables of the NN accordingly. Two different NNs are devised and tested: a dry and wet scheme. The difference between schemes resides on the input layer, while the first scheme is exclusively focused on the triggering role of atmospheric gravity waves (governed by temperature and wind profiles across the tropospheric column), the second scheme also incorporates humidity as input information with the purpose of accounting for the occasional influence of moist convection. We train both NNs using the resilient backpropagation with weight backtracking method. Their performance is tested by means of classical verification indexes. We also compare both NN results against the performance of a substantially different prognostic method that relies on a sequence of atmospheric and oceanic numerical simulations (TRAM-rissaga method). The new prediction systems work fairly well in distinguishing rissaga and non-rissaga situations, even though they tend to underestimate the amplitude of the harbour oscillation. Both NN schemes show a skill comparable to that of computationally expensive approaches based on direct numerical simulation of the physical mechanisms. The expected greater versatility of the wet scheme over the dry scheme cannot be clearly proved owing to the limited size of the training database, which lacks a sufficient number of convectively driven rissaga events. The results emphasize the potential of a NN approach and open a clear path to an operational implementation using the whole database for training, avoiding the limitations derived from splitting the available list of events into training and testing subsets.

本文探讨了神经网络（NN）在预测影响Ciutadella港口（西班牙，巴利阿里群岛的梅诺卡岛）的海啸的适用性。几乎每年，休达德亚岛都会遭受海啸，其波高（波峰与波谷之间的间隔大约为5分钟）在1 m左右，在现代历史上的几次发作中，波达到2-4 m。对这些现象进行及时而熟练的预测，可以大大减轻对港口设施和系泊船造成的损害。NN经过适当训练后，是一种计算便宜的预测方法；负责发布警告和组织迅速响应的公务员可以轻松地采用这种方法。我们研究了促进休达德亚港海啸的相关物理机制，并相应地选择了神经网络的输入变量。设计并测试了两种不同的NN：干式和湿式方案。两种方案之间的区别在于输入层，而第一种方案专门针对大气重力波的触发作用（由对流层列上的温度和风廓线控制），第二种方案也将湿度作为输入信息解释了对流的偶然影响。我们使用具有权重回溯方法的弹性反向传播训练两个神经网络。它们的性能通过经典的验证指标进行测试。我们还将两种NN结果与依赖于一系列大气和海洋数值模拟（TRAM-rissaga方法）的完全不同的预后方法的性能进行比较。新的预测系统在区分rissaga和非rissaga情况时效果很好，尽管它们往往会低估港口波动的幅度。两种NN方案都显示出与基于物理机制直接数值模拟的计算昂贵方法可比的技术。由于训练数据库的大小有限，缺乏足够数量的对流驱动的rissaga事件，因此无法清楚证明湿法方案比干法方案具有更大的通用性。