Forecasting Lake Baikal Water Levels: A Novel Hybrid ANN-PySOA Model and Its Comparison with ANN and Random Forest

Accurate forecasting of lake water levels is essential for sustainable water resource management, ecosystem protection, and climate change adaptation. This study focuses on Lake Baikal, the deepest and oldest lake in the world, which contains approximately one-fifth of the Earth's surface freshwater. A novel hybrid model, ANN-PySOA, is developed by integrating an Artificial Neural Network (ANN) with the Python Snake Optimization Algorithm (PySOA)—a bio-inspired metaheuristic that mimics python hunting behavior through three phases: searching, scanning, and attacking. Monthly water level data from Lake Baikal (۱۹۹۲-۲۰۲۵, ۸۴۱ records) were used. A ۵-lag input window was selected via autocorrelation analysis, and the dataset was split into ۷۰% training and ۳۰% testing. Model performance was evaluated using correlation coefficient (R), root mean square error (RMSE), and mean absolute error (MAE), and compared against standalone ANN and Random Forest (RF) models. The ANN-PySOA hybrid model significantly outperformed both benchmarks, achieving R = ۰.۹۴۱, RMSE = ۰.۱۲۶ m, and MAE = ۰.۱۰۰m for Lake Baikal. The standalone ANN model achieved R = ۰.۸۴۲, RMSE = ۰.۲۵۷ m, and MAE = ۰.۲۰۸ m. The Random Forest model achieved R = ۰.۹۲۱, RMSE = ۰.۱۸۱ m, and MAE = ۰.۱۴۶m. Compared to standalone ANN and RF, the proposed hybrid model improved prediction accuracy by approximately ۱۱.۸% and ۲.۲% in terms of R, respectively. This superior performance is attributed to the PySOA optimizer's ability to balance exploration and exploitation, avoiding local optima and enhancing ANN learning capability. The findings confirm that the hybrid ANN-PySOA model provides a reliable and accurate approach for water level forecasting in large lake systems.

کلیدواژه ها:

Lake Baikal ، Water level forecasting ، ANN-PySOA ، Python Snake Optimization Algorithm ، Hybrid model ، Random Forest