Akademik Veri Yönetim Sistemi
PAKISTAN JOURNAL OF AGRICULTURAL SCIENCES, cilt.63, sa.1, ss.23-35, 2026 (SCI-Expanded, Scopus)
In this study, unavailable inflow data for the Sar & imath;oglan Dam were supplemented using data from the neighboring Sar & imath;msakl & imath; Dam, which shares similar climate and geographical characteristics, to construct an extended dataset for long-term inflow prediction. This comprehensive dataset was analysed using advanced machine learning techniques. Initially, missing inflow data for the Sar & imath;oglan Dam were reconstructed using Sar & imath;msakl & imath; Dam records. The historically extended inflow dataset was then subjected to machine learning algorithms for the purpose of forecasting study, including Bidirectional Long Short-Term Memory (BiLSTM)-based deep neural networks, Support Vector Regression (SVR), Gaussian Process Regression (GPR), and Artificial Neural Networks (ANN). To improve prediction accuracy, data sub-band decomposition techniques were applied, including Empirical Mode Decomposition (EMD), Variational Mode Decomposition (VMD), and Maximal Overlap Discrete Wavelet Transform (MODWT). According to the obtained results, the BiLSTM model combined with EMD had the best accuracy, especially for short-and medium-term forecasts with an R2 value greater than 0.9. This study proposes a robust framework for inflow modeling and forecasting by first augmenting the dataset and subsequently applying hybrid machine learning approaches. This provides a method for developing decision support systems in dam operation and for the sustainable management of water resources.