Sediment Dynamics in the Brazos and Colorado Rivers: Insights from Satellite Imagery and Machine Learning

Accurate estimation of suspended sediment concentration (SSC) is critical for understanding sediment dynamics in the Gulf of Mexico (GOM). This study employed satellite imagery from Landsat and Sentinel to predict SSC in the lower sections of the Brazos and Colorado Rivers, utilizing machine learning (ML) techniques such as Random Forest (RF) and artificial neural networks (ANN). Ground-based SSC data and satellite-derived spectral reflectance values were integrated, with rigorous data filtering addressing key factors such as cloud cover, extreme storm events, and other critical considerations to ensure data quality and reliability. Combined flow models using four-band configurations (red, green, blue, NIR) provided the most reliable results, demonstrating superior accuracy across hydrological conditions compared to high-flow and low-flow-specific models. The ANN model consistently outperformed RF, achieving an R² of 0.89 and RMSE of 99 mg/L for the Colorado River and R² of 0.83 and RMSE of 197 mg/L for the Brazos River. This research has established a scalable SSC estimation framework suitable for diverse river systems in the GOM region, predicting sediment dynamics under changing climatic conditions.