LSTM AND ANFIS MACHINE LEARNING ALGORITHMS IN ESTIMATING THE SEA WATER TEMPERATURE IN TÜRKİYE AT VARIOUS SEA LOCATIONS

Abstract

The World's temperature is experiencing a rapid increase, leading to negative consequences for aquatic ecosystems such as oceans, seas, lakes, and rivers. There are also other negative influences consisting of changing precipitation patterns, disruptions in marine current circulation, and formation of negative impacts on marine life. Ultimately, there is a compelling need for careful monitoring of sea temperatures to understand and address these interconnected environmental changes. The daily temperature of seawater (SWT) is a crucial abiotic variable that changes both the chemical composition of water and aquatic life in seas and oceans. The present study explored the capabilities of artificial intelligence techniques in one-day-ahead SWT predictions. These techniques are fuzzy c-means adaptive neuro-fuzzy inference system (ANFIS-FCM), subtractive clustering ANFIS (ANFIS-SC), grid segmentation ANFIS (ANFIS-GP), and long short-term memory (LSTM) and artificial neural network (ANN). Accordingly, daily SWT data that was collected from Alanya, Bodrum, and Akcakoca measurement stations located in Türkiye's Mediterranean, Aegean, and Black Sea locations were used in SWT predictions. Estimated results obtained by these five estimation methods were compared to the real observed values by interpreting four statistical metrics. Consequently, the most accurate estimates were obtained utilizing the fuzzy c-means (FCM) of ANFIS. Besides, it was reported that the LSTM approach closely followed the accuracy of this prediction of FCM. Both proposed models have generated superior statistical accuracy results corresponding to 0.34% MAPE, 0.0765 oC MAE, 0.1585 oC RMSE, and 0.9990 R. Those results have indicated the closest match of the predictions on the real measured data that have been acquired by ANFIS-FCM and LSTM models.