Machine Learning Analysis of Hydrologic Exchange Flows and Transit Time Distributions in a Large Regulated River

Area of Science:

• Environmental science
• Hydrology
• Computational modeling

Background:

• Hydrologic exchange between rivers and subsurface environments is vital for water quality and river corridor ecosystems.
• Traditional high-resolution numerical models for simulating these flows are computationally expensive.

Purpose of the Study:

• To develop reduced-order models for hydrologic exchange flows and transit time distributions using machine learning.
• To identify key physical, spatial, and temporal factors influencing these processes.
• To provide computationally efficient alternatives to traditional numerical models.

Main Methods:

• Utilized Random Forest (RF) and Extreme Gradient Boosting (XGB) machine learning algorithms.
• Integrated field observations (bathymetry) and hydrodynamic simulation data (river velocity, depth).
• Incorporated hybrid clustering for hydromorphic classification of river corridors.

Main Results:

• Top predictive parameters identified: geological formation thickness, flow regime, river velocity, and river depth.
• RF and XGB models achieved 70% to 80% accuracy in predicting exchange flows and transit times.
• Evaluated optimal configurations for each machine learning model.

Conclusions:

• Machine learning models (RF, XGB) are effective and efficient alternatives to computationally demanding numerical models for simulating hydrologic exchange flows.
• The developed models enhance understanding of factors influencing river-groundwater interactions.
• Model transferability to other river systems is feasible, contingent on data availability.