Akademik Veri Yönetim Sistemi
Pure and Applied Geophysics, 2025 (SCI-Expanded, Scopus)
River sediment transport is a complex challenge in water-resource management that influences ecosystem health, water quality, and infrastructure. To accurately predict sediment discharge (Sd), the predictors should consider sediment transport processes’ nonstationary, nonlinear, and dynamic nature. These issues are more critical in data-scarce environments, highlighting the need for reliable predictive frameworks. This study focused on sediment transport prediction in the Wadi Allala Basin in Northern Algeria and determined the most suitable model. The objective was to predict sediment discharge (Sd) using univariate and multivariate long short-term memory (Bi-LSTM) models. The multivariate Bi-LSTM model outperformed the univariate by including parameters like stage (Ht), instantaneous concentration (St), and discharge (Qt). A benchmark dataset from 7247 paired observations of water discharge, sediment concentration, and sediment discharge at Sidi Akacha station in the Wadi Allala Basin, collected between 1972 and 2015, supported this analysis. The model’s results were evaluated with several statistical metrics: Spearman rank correlation, rs = 0.998, RMSE = 276.080 ton/day, MAE = 196.824 ton/day, and NSE = 0.995 for the multivariate Bi-LSTM. Compared with other methods, such as the univariate Bi-LSTM, which had R2 = 0.763 and RMSE = 1788.288 ton/day, the multivariate model generated superior results. Due to its high accuracy, the multivariate Bi-LSTM can be a dependable tool in water resource management, supporting reservoir operation optimization, sedimentation reduction in dams and structures, and long-term soil erosion and flood prevention planning. The findings indicate that Bi-LSTM models can yield superior prediction results. Selecting the discharge and sediment load for a specific day allows the multivariate Bi-LSTM to produce optimal predictions for the study area. An additional benefit is the model’s robustness to minor data errors, resulting in better estimates than traditional statistical models. Predicting suspended sediment is a valuable indicator for assessing future conditions in water resource management.