A Hybrid ARIMA–LGBM Framework with Optuna Optimization for Long-Term Dissolved Oxygen Prediction in Aquatic Systems

Accurate prediction of dissolved oxygen (DO) is essential for effective water quality monitoring and aquatic ecosystem management. However, DO dynamics are governed by complex temporal patterns and nonlinear interactions among physicochemical and meteorological variables, which limit the performance of standalone statistical or machine learning models. To address this challenge, this study proposes a hybrid modeling framework that integrates AutoRegressive Integrated Moving Average (ARIMA), Light Gradient Boosting Machine (LGBM), and Optuna-based hyperparameter optimization. The framework was applied to a 30-year water quality dataset (1989–2019), incorporating key variables such as water temperature, air temperature, depth, salinity, pH, and Secchi depth. The proposed hybrid model achieved excellent predictive performance, yielding a coefficient of determination (R2) of 0.98 and a Root Mean Squared Error (RMSE) of 0.043 mg/L in dissolved oxygen prediction, outperforming the standalone LGBM model (R2 = 0.94, RMSE = 0.081 mg/L). The ARIMA component effectively captured linear temporal dependencies and seasonality, while LGBM modeled complex nonlinear relationships, and Optuna optimization significantly enhanced model generalization and stability. Comparative analysis with existing studies further confirmed the superior accuracy and robustness of the proposed approach. Overall, the results demonstrate that the hybrid ARIMA–LGBM framework provides a reliable and efficient tool for long-term dissolved oxygen forecasting, supporting advanced water quality monitoring and informed environmental management.