Predictive capabilities of novel deep neural networks learning for long-term streamflow prediction: Insights from the Barandouz Chay River

Dams and reservoirs play a crucial role for public health, food security, economic growth, and flood protection, especially as climate change challenges sustainable water management. This study explored the application of regression-based deep learning models for long-term prediction of Barandouz River flow. In the present study, four models were used, namely, Continuous Weighted Residual Method (CWERM), Physics-Informed Koopman Networks (PI-Koopman), Convolutional Neural Network (CNN) and Implicit Neural Representations (INRs). Forty years (1980–2020) of daily time-series data were adopted , including temperature, precipitation, relative humidity, wind speed, and discharge. To improve prediction accuracy, lagged discharge (up to three days) was incorporated. Data pre-processing consisted of missing value reconstruction (KNN), outlier removal, and normalization to mitigate overfitting risks. Feature selection and scenario construction were guided by the Minimum Redundancy Maximum Relevance (MRMR) method, resulted in five predictive scenarios. Model evaluation was based on multiple criteria (R 2 , MAE, RMSE, MAD, NSE, and KGE). Results showed that scenario 4 excluding historical flow variables exhibited severe predictive failure across all models in both training and testing phases, with testing R 2 ranging from 0.087 (CWERM) to 0.189 (PI-Koopman) and NSE remaining below 0.23, highlighting the critical role of antecedent discharge in modeling hydrological memory. While, CNN consistently outperformed all other models in both phases. During testing, R 2 values of 0.991 (SN1), 0.992 (SN3), and 0.981 (SN5), with corresponding NSE > 0.910 and RMSE as low as 4.157 m 3 /s (SN5) were observed for CNN. Its average testing, R 2 (0.977), was 14.0–18.3% higher than those of ANN-OA (0.857), INRs (0.855), and PI-Koopman (0.826), while its average RMSE (7.25 m 3 /s) was 34.5–44.1% lower. PI-Koopman, despite moderate performance (best testing R 2 = 0.858 in SN3), maintained predictions within physically reasonable bounds, particularly in degraded scenarios like SN4. In conclusion, the results demonstrated CNN's exceptional ability to capture complex, nonlinear streamflow dynamics under historical conditions, while also underscored the limitations of purely data-driven approaches when hydrological stationarity is violated providing crucial insights into deploying deep learning models in semi-arid, human-influenced basins under climate uncertainty.