Enhancing rainfall-runoff model accuracy with machine learning models by using soil water index to reflect runoff...

Area of Science:

• Hydrology
• Environmental Science
• Data Science

Background:

• Data-driven models improve rainfall-runoff estimation but struggle with soil moisture and runoff dynamics.
• Solely using rainfall data limits model performance in capturing hydrological complexities.

Purpose of the Study:

• To develop a predictor variable selection method for rainfall-runoff models using accumulated rainfall to represent soil moisture and runoff characteristics.
• To evaluate the effectiveness of rainfall products (CHIRPS, GPM) and the soil water index (SWI) for enhancing runoff estimation.
• To compare the performance of Random Forest (RF) and Artificial Neural Network (ANN) models in simulating daily runoff.

Main Methods:

• Utilized accumulated rainfall over various intervals as a proxy for soil moisture.
• Integrated remote sensing rainfall products (CHIRPS, GPM) and the soil water index (SWI).
• Employed Random Forest (RF) and Artificial Neural Network (ANN) models for daily runoff simulation.

Main Results:

• Incorporating both rainfall and SWI data significantly improved runoff estimation outcomes.
• The Random Forest model outperformed the Artificial Neural Network and conceptual models.
• Accumulated rainfall proved to be a highly effective input variable for the models.

Conclusions:

• The proposed method enhances runoff estimation accuracy by integrating soil moisture proxies and advanced rainfall data.
• Remote sensing data, particularly SWI and accumulated rainfall, are valuable for hydrological modeling in data-limited regions.
• The Random Forest model offers a robust and efficient approach for rainfall-runoff modeling without complex preprocessing steps.