Abstract An accurate estimation of the sea surface temperature (SST) is of great importance. Therefore, the objective of this work was to develop an adaptive neuro-fuzzy inference system (ANFIS) model to predict SST in the Çanakkale Strait. The observed monthly air temperature, evaporation and precipitation data from the Çanakkale meteorological observation station were used as input data. The Takagi–Sugeno fuzzy inference system was applied. The grid partition method (ANFIS-GP) and the subtractive clustering partitioning method (ANFIS-SC) were used with Gaussian membership functions to generate the fuzzy inference system. Six performance evaluation criteria were used to evaluate the developed SST prediction models, including mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), Nash-Sutcliffe efficiency (NSE) and correlation of determination (R2). The dataset was randomly divided into training and testing datasets for the machine learning process. Training data accounted for 75% of the dataset, while 25% of the dataset was allocated for testing in ANFIS. The hybrid algorithm was selected as a training algorithm for the ANFIS. Simulation results revealed that the ANFIS-SC4 model provided a higher correlation coefficient of 0.96 between the observed and predicted SST values. The results of this study suggest that the developed ANFIS model can be applied for predicting sea surface temperature around the world.

中文翻译：

---

开发自适应神经模糊推理系统 (ANFIS) 模型来预测海面温度 (SST)

摘要 准确估计海面温度（SST）非常重要。因此，这项工作的目标是开发一种自适应神经模糊推理系统 (ANFIS) 模型来预测恰纳卡莱海峡的海温。从恰纳卡莱气象观测站观测到的每月气温、蒸发和降水数据被用作输入数据。应用了 Takagi-Sugeno 模糊推理系统。网格划分方法（ANFIS-GP）和减法聚类划分方法（ANFIS-SC）与高斯隶属函数一起用于生成模糊推理系统。六个性能评估标准用于评估开发的 SST 预测模型，包括均方误差 (MSE)、均方根误差 (RMSE)、平均绝对误差 (MAE)、平均绝对百分比误差 (MAPE)、Nash-Sutcliffe 效率 (NSE) 和测定相关性 (R2)。数据集被随机分为机器学习过程的训练和测试数据集。训练数据占数据集的 75%，而 25% 的数据集分配给 ANFIS 中的测试。选择混合算法作为 ANFIS 的训练算法。模拟结果表明，ANFIS-SC4 模型在观测到的和预测的 SST 值之间提供了更高的相关系数 0.96。这项研究的结果表明，开发的 ANFIS 模型可用于预测世界各地的海面温度。而 25% 的数据集被分配用于在 ANFIS 中进行测试。选择混合算法作为 ANFIS 的训练算法。模拟结果表明，ANFIS-SC4 模型在观测到的和预测的 SST 值之间提供了更高的相关系数 0.96。这项研究的结果表明，开发的 ANFIS 模型可用于预测世界各地的海面温度。而 25% 的数据集被分配用于在 ANFIS 中进行测试。选择混合算法作为 ANFIS 的训练算法。模拟结果表明，ANFIS-SC4 模型在观测到的和预测的 SST 值之间提供了更高的相关系数 0.96。这项研究的结果表明，开发的 ANFIS 模型可用于预测世界各地的海面温度。