Harnessing Novel Data-Driven Techniques for Precise Rainfall–Runoff Modeling

Rainfall and runoff are considered the main components of the hydrological cycle, and their forecasting is of great significance in water resource management, particularly for reservoir operation. Developing an accurate model to capture the dynamic connection between rainfall and runoff remains problematic and challenging in water resource management due to the nonstationary characteristics of hydrologic processes and the effects of noise. In this study, data-driven techniques, such as the group method of data handling (GMDH), extreme learning machine (ELM), and two hybrids of artificial neural network (ANN) with Cuckoo search algorithm (ANN + Cuckoo) and genetic algorithm (ANN + GA) were used to model the rainfall–runoff relationship. For a comprehensive analysis, four scenarios were examined based on the different input combinations to test and select the best scenario and best model performance. The results indicated that the performance of ELM and GMDH in predicting runoff was more accurate than that of ANN + Cuckoo and ANN + GA. Although the GMDH predicts runoff with higher accuracy, ELM provides reliable performance in simulating both low and high values. The models' performance can be ranked based on the testing data in the following order: GMDH, ELM, ANN + GA, and ANN + CUKOO. The root mean squared error (RMSE) was recorded as 56.7 and 69.7 m³/s for the GMDH and ELM models, respectively. These low RMSE values highlight the potential of these models in effectively addressing the challenges associated with the complexity of rainfall–runoff simulations. Moreover, the results demonstrate that the machine learning models could be used as a simple, rapid, and inexpensive approach for timely and reliable runoff prediction that is expected to benefit reservoir management.