Improving streamflow simulation by combining hydrological process-driven and artificial intelligence-based models

Area of Science:

• Hydrology
• Water Resources Management
• Artificial Intelligence in Environmental Science

Background:

• Accurate streamflow monitoring is vital for effective watershed management.
• Traditional hydrological models like HBV and NRECA have limitations in streamflow simulation.
• The need for advanced modeling techniques is increasing due to data scarcity in many regions.

Purpose of the Study:

• To evaluate and compare the performance of process-driven hydrological models and AI-based hybrid models for streamflow simulation.
• To assess the effectiveness of adaptive neuro-fuzzy inference system (ANFIS), support vector machine (SVM), and group method of data handling (GMDH) in streamflow prediction.
• To identify the most suitable modeling approach for Indonesian river basins, particularly under data-scarce conditions.

Main Methods:

• Employed two conceptual rainfall-runoff models: Hydrologiska Byråns Vattenbalansavdelning (HBV) and Non Recorded Catchment Areas (NRECA).
• Developed seven hybrid models integrating AI techniques (ANFIS, SVM, GMDH) with hydrological model outputs and lagged precipitation data.
• Utilized data from four Indonesian river basins for model training and testing.

Main Results:

• AI-based hybrid models generally outperformed traditional HBV and NRECA models in streamflow simulation accuracy.
• The Group Method of Data Handling (GMDH) model demonstrated the best performance across all tested stations, achieving low RMSE values.
• GMDH exhibited superior accuracy in estimating peak streamflow values, particularly in training and testing datasets.

Conclusions:

• AI-based hybrid models, especially GMDH, are highly effective alternatives for streamflow simulation, offering enhanced accuracy over conventional hydrological models.
• These AI models are particularly valuable in watersheds with limited measured hydrological and environmental data.
• Careful selection of appropriate input data is crucial for the successful application of AI hybrid models in streamflow prediction.