Application of Machine Learning Algorithms in Predicting the Water Area of Gorgan Bay Using Hydrological and Climatic Data

This study aimed to predict changes in the water area of Gorgan Bay using hydrological, satellite, and climatic data, including temperature, precipitation, discharge, water level, and the water area (MNDWI) of the Caspian Sea from 2000 to 2023. Advanced machine learning methods, including XGBoost, Radial Basis Function Networks (RBFN), Random Forest, and Linear Regression, were employed to model these changes. The relationships among the variables were assessed using correlation coefficients and linear regression analysis. Model evaluation metrics included RMSE, MAE, MAPE, MBE, and R², and actual data from five randomly selected years were compared with predicted values. The results revealed that changes in the water level of the Caspian Sea exhibited the highest correlation with the water area of Gorgan Bay (correlation coefficient 0.90 and significance level 0.001). Precipitation and temperature had the least impact, with correlation coefficients of 0.43 for precipitation and -0.34 for temperature. Evaluation of the predictive models showed that the XGBoost model had the best performance, with an R² of 0.93 and a mean absolute percentage error (MAPE) of 4.9%, root mean square error (RMSE) of 16.9, and mean absolute error (MAE) of 15.3. Finally, five years, including 2000, 2008, 2011, 2016, and 2018, were randomly selected to predict the water area of Gorgan Bay. The results indicated that all models adequately simulated data fluctuations, but the XGBoost model performed the best, accurately predicting severe fluctuations in the water area.