Author(s): Maria Jose Cid Vargas; Diego Caamano Avendano; Maribet Gamboa Mendez

Linked Author(s): Diego Caamaño A.

Keywords: XGBoost; SHAP; Sediment rating curve; Megadrought; Training -window sensitivity

Abstract: Sediment transport is a key process for understanding river morphology and the hydraulic conditions that shape fluvial habitats, yet discontinuous sediment records hinder robust modeling. We evaluate the ability of Extreme Gradient Boosting (XGBoost) to predict daily suspended sediment transport rate (STR) at 37 Chilean DGA gauging stations (2000–2018), spanning diverse climatic and geological settings and including the 2010- onward megadrought. Three model configurations were tested: a base model (6 hydro -climatic variables), an extended model (19 variables, including lagged and moving averages), and an SHAP-optimized (8 predictors). Performance was quantified using Nash–Sutcliffe Efficiency (NSE) and the bias in the sediment rating -curve slope (SRC,

DOI: https://doi.org/10.64697/iahr.proc.ise2026.abs.175

Year: 2026