Interpretable machine learning for predicting rating curve parameters using channel geometry and hydrological...

Area of Science:

• Hydrology
• Environmental Science
• Data Science

Background:

• Climate change increases global flood events, necessitating better hydrological data for decision-making.
• Hydrological rating curves are crucial for converting river stage to streamflow, impacting flood modeling and geomorphology.
• Power law regression effectively models the non-linear relationship between stage and discharge.

Purpose of the Study:

• To develop a two-tier, data-driven approach for predicting hydrological rating curve parameters (α, β) across the contiguous United States (CONUS) stream network.
• To explore a unified solution for representing rating curves in large stream networks.
• To investigate the influence of channel geometry and hydrometeorological factors on rating curve parameters.

Main Methods:

• Utilized HYDRoacoustics in support of the Surface Water Oceanographic Topography (HYDRoSWOT), National Hydrography (NHDPlus v2.1), and STREAM-CATCHMENT (STREAMCAT) datasets.
• Compared four empirical models: Multivariate regression, eXtreme Gradient Boosting (XGBoost), Random Forest, and Support Vector Regression.
• Developed two tiers of XGBoost models: Tier-1 for gauge sites and Tier-2 for broader stream network application.

Main Results:

• Tier-1 XGBoost models achieved high prediction accuracy (R² = 0.70) but were limited to gauge sites.
• Tier-2 XGBoost models provided a good balance of accuracy (R² = 0.55) and applicability across the NHDPlus stream network in CONUS.
• The study identified key channel geometry and hydrometeorological attributes influencing rating curve parameters.

Conclusions:

• The developed two-tier approach offers a scalable solution for estimating hydrological rating curves across extensive stream networks.
• Tier-2 models provide a practical method for improving streamflow data availability, particularly in data-scarce regions.
• Understanding the drivers of rating curve parameters enhances the reliability of hydrological predictions and water resource management.