Seawater Temperature Prediction that Adapts to Changes in Water Depth
Seawater Temperature Prediction that Adapts to Changes in Water Depth
复制标题
DOI：
10.1145/3605423.3605426
复制
发表时间：
2023-05
期刊：
影响因子：
--
通讯作者：
Shuhei Aoyama;Takuma Miwa;T. Otsuka
中科院分区：
文献类型：
--
作者：
Shuhei Aoyama;Takuma Miwa;T. Otsuka
海水温度的变化影响养殖户的日常决策。为了做出最佳决策，农民需要高度准确地预测农场特定深度的海水温度。然而，很少有研究集中在预测海水温度的深度所需的农民，相比之下，海表面温度（SST）的预测，这一直是许多研究的重点。本文的目的是建立这些海下预报作为一个分支的海水温度预报除了SST预测。我们专注于在五个具有不同特征的点上由系泊浮标测量的海底温度数据，以处理多个深度，并详细描述了进行预测所涉及的广泛分析和预处理。然后，我们使用这些数据分别评估24小时和7天的短期和长期预测。作为预测模型，我们提出了一种具有自适应水深的门控递归单元（GRU）的深度学习模型，并将其与LightGBM和CatBoost等标准数学模型进行了比较。其中一个结果是，我们的模型比数学模型预测7天的准确率高出10%以上。此外，为了支持我们的深度自适应模型的优越性，我们通过消除对模型输入和输出的多个深度的考虑来测试预测精度的降低。我们广泛分析了水深变化对温度预测结果的影响，并提出了相应的预测模型，为满足水产养殖中真正需要的深度海水温度预测的需求提供了基础。
Changes in seawater temperature affect the daily decisions of aquaculture farmers. To make optimal decisions, farmers need highly accurate predictions of seawater temperature at specific depths in their farms. Nevertheless, few studies have focused on the prediction of seawater temperature at the depths required by farmers, in contrast to the prediction of sea surface temperature (SST), which has been the focus of much research. The purpose of this paper is to establish these undersea predictions as a branch of seawater temperature prediction alongside SST prediction. We focus on undersea temperature data measured by moored buoys at five points with different characteristics to handle multiple depths and provide a careful description of the extensive analysis and pre-processing involved in making predictions. We then use these data to evaluate short-term and long-term predictions for 24 hours and seven days, respectively. As a predicting model, we proposed a deep learning model with gated recurrent units (GRUs) adaptive to water depth and compared it with standard mathematical models such as LightGBM and CatBoost. One of the results is that our model is more than 10% more accurate than the mathematical model for 7-day-ahead predictions. In addition, to support the superiority of our depth-adaptive model, we tested the reduction in prediction accuracy by eliminating the consideration of multiple depths for the inputs and outputs of the model. Our extensive analysis of the impact of changes in water depth on temperature prediction results and our corresponding proposed prediction model provides the foundation for meeting the demand for seawater temperature prediction at the depth that is truly needed in the aquaculture.