Effect of environmental covariable selection in the hydrological modeling using machine learning models to predict...

Area of Science:

• Hydrology
• Environmental Science
• Data Science

Background:

• Machine learning is increasingly used for streamflow prediction.
• Covariable selection is crucial for improving model precision and stability.

Purpose of the Study:

• To analyze the impact of Recursive Feature Elimination (RFE) and Forward Feature Selection (FFS) on machine learning model performance for daily streamflow prediction.
• To evaluate streamflow prediction accuracy using different covariable selection techniques.

Main Methods:

• Applied RFE and FFS for covariable selection in Multivariate Adaptive Regression (EARTH), Multiple Linear Regression (MLR), and Random Forest (RF) models.
• Utilized an 18-year historical dataset from the Piranga river basin, Brazil, including streamflow, precipitation, and MODIS data.
• Evaluated model performance using Nash Sutcliffe efficiency (NSE), Determination coefficient (R²), and Root Mean Square Error (RMSE) over 50 runs.

Main Results:

• All tested models demonstrated satisfactory performance with both RFE and FFS.
• Models struggled to accurately predict peak streamflow events.
• Forward Feature Selection (FFS) generally improved model performance and reduced the number of selected covariables.

Conclusions:

• Machine learning is an efficient tool for daily streamflow prediction.
• Covariable selection, particularly using FFS, enhances the efficiency of machine learning models for streamflow forecasting.