Skip to main content
ARS Home » Plains Area » El Reno, Oklahoma » Oklahoma and Central Plains Agricultural Research Center » Agroclimate and Hydraulics Research Unit » Research » Publications at this Location » Publication #433829
Research Project: Development of a Monitoring Network, Engineering Tools, and Guidelines for the Design, Analysis, and Rehabilitation of Embankment Dams, Hydraulic Structures, and Channels

Location: Agroclimate and Hydraulics Research Unit

Title: A two-stage machine learning framework for integrated streamflow and water quality prediction

Author
item ABEYSINGHE, UMANDA - University Of Missouri
item HAITHCOAT, TIMOTHY - University Of Missouri
item Hunt, Sherry
item ALOYSIUS, NOEL - University Of Missouri

Submitted to: Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: 4/10/2026
Publication Date: 4/20/2026
Citation: Abeysinghe, U., Haithcoat, T., Hunt, S., Aloysius, N. 2026. A two-stage machine learning framework for integrated streamflow and water quality prediction. Meeting Abstract. University of Missouri Show Me Research Week, Columbia, MO April 20-24,2026.

Interpretive Summary:

Technical Abstract: Accurate prediction of streamflow and water quality is essential for effective water resource management and environmental protection. Here, we present a two-stage machine-learning framework to improve the forecasting of daily streamflow and water-quality variables across 157 gauged benchmark watersheds in the Mississippi River Basin. In the first stage, streamflow is predicted using data-driven models, including Random Forest, Long Short-Term Memory (LSTM), and Temporal Convolutional Networks (TCN), leveraging hydroclimatic inputs such as precipitation, evapotranspiration, temperature, and antecedent flow conditions. In the second stage, the predicted streamflow is integrated with meteorological and watershed attributes to estimate key water quality indicators, including turbidity and total phosphorus. The framework evaluates how uncertainties in streamflow prediction propagate into water quality forecasts. Model performance is assessed using several goodness-of-fit indicators, including the Kling–Gupta Efficiency (KGE), the precent bias error (PBIAS), and the root mean square error (RMSE). Our preliminary results show that incorporating predicted streamflow significantly improves water quality prediction accuracy, highlighting the importance of coupling hydrological and water quality models. The proposed approach provides a scalable and efficient tool for integrated hydro-environmental forecasting under data-driven settings.