A Hybrid Water Balance Machine Learning Model to Estimate Inter-Annual Rainfall-Runoff.

Area of Science:

• Hydrology
• Climatology
• Machine Learning

Background:

• Watershed climatic diversity complicates accurate estimation of inter-annual rainfall runoff (IARR).
• Existing non-parametric and empirical water balance models show varying reliability across different climatic zones.
• Developing robust models for IARR prediction is crucial for water resource management.

Purpose of the Study:

• To propose and evaluate a hybrid model (MR-CART) for improved IARR estimation.
• To compare the performance of different water balance models in providing input data for the hybrid model.
• To assess the reliability and dynamicity of the hybrid model across diverse climatic conditions in Northern Algeria.

Main Methods:

• A hybrid model integrating Multiple Regression (MR) and Classification and Regression Tree (CART) machine learning methods was developed.
• Input data for the hybrid model were derived from various non-parametric and empirical water balance models.
• Statistical tests (performance and distribution) were employed to evaluate model reliability and data distribution.

Main Results:

• The hybrid MR-CART model demonstrated superior performance and data distribution for IARR estimation.
• Yang, Sharif, and Zhang models proved reliable for input data across all climatic classes.
• Schreiber's model was particularly effective in humid and semi-humid regions.

Conclusions:

• The proposed MR-CART hybrid model offers a robust solution for IARR estimation in climatically diverse watersheds.
• The selection of appropriate water balance models for input data significantly influences hybrid model performance.
• The study highlights the effectiveness of machine learning approaches in hydrological modeling.