University of Tehran Journal of Natural Environment 2008-7764 77 3 2024 11 21 Evaluation of machine learning methods for predicting water level fluctuations in the southern coasts of the Caspian Sea using GRACE and GRACE-FO satellites Evaluation of machine learning methods for predicting water level fluctuations in the southern coasts of the Caspian Sea using GRACE and GRACE-FO satellites 453 466 98304 10.22059/jne.2024.379065.2691 FA Mobin Eftekhari Department of Water Engineering, Faculty of Agriculture, University of Birjand, Birjand, Iran. 0000-0002-1031-1270 Mehdi Dastorani Department of Water Engineering, Faculty of Agriculture, University of Birjand, Birjand, Iran. 0000-0001-7718-9604 Ali Haji Elyasi School of Civil Engineering, College of Engineering, University of Tehran, Tehran, Iran. 0000-0002-5202-8131 Journal Article 2024 07 08 Sea level fluctuations have destructive effects on coastal cities, their environment, and climate. Therefore, identifying changes and fluctuations in sea levels and predicting them can aid in decision-making and proper management of resulting events and problems. This study focuses on modeling the time series of water level in the southern coasts of the Caspian Sea using GRACE and GRACE-FO satellite data, employing machine learning-based models such as Decision Tree (DT), Random Forest (RF), and Multivariate Adaptive Regression Splines (MARS). For this purpose, GRACE and GRACE-FO satellite data from 2003 to 2023 were used. The results obtained from the remote sensing time series were correlated with data from the Anzali Port tide gauge station. Subsequently, water levels were simulated and predicted using machine learning models. Results showed that the JPL model with R²= 0.788 indicates an appropriate relationship between satellite data and ground data. Additionally, R2 values for the three models DT, MARS, and RF were 0.545, 0.853, and 0.671, respectively, and NSE values were 0.64, 0.89, and 0.76 respectively, demonstrating the superior performance of the MARS model in simulation compared to others. Therefore, this model was used to predict water levels up to 2040, with evaluation criteria indicating the high efficiency of the MARS model. Predictions showed that in the worst-case scenario, sea level will decrease by 120 centimeters by 2040, leading to environmental damage and harm to marine industries and ports in coastal cities. The results of this study can be used as an effective tool in water resource management and long-term planning in the Caspian Sea coastal areas. Furthermore, these findings can greatly assist in assessing environmental and economic risks resulting from sea level changes and adopting appropriate strategies. Sea level fluctuations have destructive effects on coastal cities, their environment, and climate. Therefore, identifying changes and fluctuations in sea levels and predicting them can aid in decision-making and proper management of resulting events and problems. This study focuses on modeling the time series of water level in the southern coasts of the Caspian Sea using GRACE and GRACE-FO satellite data, employing machine learning-based models such as Decision Tree (DT), Random Forest (RF), and Multivariate Adaptive Regression Splines (MARS). For this purpose, GRACE and GRACE-FO satellite data from 2003 to 2023 were used. The results obtained from the remote sensing time series were correlated with data from the Anzali Port tide gauge station. Subsequently, water levels were simulated and predicted using machine learning models. Results showed that the JPL model with R²= 0.788 indicates an appropriate relationship between satellite data and ground data. Additionally, R2 values for the three models DT, MARS, and RF were 0.545, 0.853, and 0.671, respectively, and NSE values were 0.64, 0.89, and 0.76 respectively, demonstrating the superior performance of the MARS model in simulation compared to others. Therefore, this model was used to predict water levels up to 2040, with evaluation criteria indicating the high efficiency of the MARS model. Predictions showed that in the worst-case scenario, sea level will decrease by 120 centimeters by 2040, leading to environmental damage and harm to marine industries and ports in coastal cities. The results of this study can be used as an effective tool in water resource management and long-term planning in the Caspian Sea coastal areas. Furthermore, these findings can greatly assist in assessing environmental and economic risks resulting from sea level changes and adopting appropriate strategies. https://jne.ut.ac.ir/article_98304_79cef9cc5c842ee39e164009c7554da2.pdf