Skip to main content
ARS Home » Southeast Area » Jonesboro, Arkansas » Delta Water Management Research » Research » Publications at this Location » Publication #424314

Research Project: Optimizing the Management of Irrigated Cropping Systems in the Lower Mississippi River Basin

Location: Delta Water Management Research

Title: Machine learning models for water quality: Predicting pollutant loads and assessing conservation practice's effectiveness in agricultural fields

Author
item THAPA, ARJUN - North Carolina A&t State University
item ARYAL, NIROJ - North Carolina A&t State University
item Reba, Michele

Submitted to: Ecological Informatics
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/15/2025
Publication Date: 10/27/2025
Citation: Thapa, A., N. Aryal, and M.L. Reba. Machine learning models for water quality: Predicting pollutant loads and assessing conservation practice’s effectiveness in agricultural fields. Ecological Informatics. Volume 92. 103479. https://doi.org/10.1016/j.ecoinf.1015.103479. 2025.
DOI: https://doi.org/10.1016/j.ecoinf.2025.103479

Interpretive Summary: In recent years, Machine Learning (ML), a subset of Artificial Intelligence (AI), has emerged as a promising alternative for environmental outcomes. Various Machine Learning models have been employed to predict water quantity and quality from agricultural fields and watersheds. Commonly used models include Multiple Linear Regression (MLR), K-Nearest Neighbors (KNN), Random Forest (RF), Extreme Gradient Boosting (XGB), and Histogram Gradient Boosting. This study utilizing field-collected high-frequency data and nine regression-based machine learning models to predict daily runoff, sediment, total phosphorus, and total nitrogen loads, comparing with and without agricultural conservation practices. This study utilized real-time daily data collected from a commercial cotton field in Manila, Arkansas. The model predicted average annual runoff, sediment, TP, and TN loads from the control field were lower than those from the treatment field. The results confirmed the effectiveness of conservation practices, with cover crops and filter strips significantly reducing surface runoff, sediment, TP, and TN loads in the treatment field compared to the control field. Annual reductions in pollutant loads ranged from 35% to 51%, with the greatest reductions observed during the winter season. Generally, findings from this research help to quantify the benefit of implementing the conservation practice of a switchgrass buffer, which is useful for land managers, modelers of agricultural systems, and farmers.

Technical Abstract: Estimating real-time sources of pollutants and evaluating the effectiveness of conservation practices in agriculture are crucial for prevention of water resources contamination. Pollutant load released from agricultural fields can be estimated using process-based model or data driven models. This study utilized machine learning (ML) models to predict water pollutant loads since they need fewer input features than process-based models. Hydro-meteorological data (e.g., temperature, rainfall, runoff) were collected from treatment and control fields (2016–2022), where cover crops and filter strips were used for pollution mitigation. Pollutant loads, including sediment, total phosphorus (TP), and total nitrogen (TN), were measured and used to train nine ML models: Multiple Linear Regression (MLR), K-Nearest Neighbors (KNN), Random Forest (RF), Extreme Gradient Boosting (XGB), Histogram Gradient Boosting (HGBt), Artificial Neural Networks (ANN), Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and a hybrid CNN-LSTM model. Results showed that the hybrid model best-predicted runoff in the control field (R²=0.87) and KNN in the treatment field (R²=0.82). LSTM excelled in sediment prediction for both fields, while RF and ANN were superior for TP and TN predictions, respectively. Model performance declined from runoff to sediment to nutrient loads due to error propagation. Advanced models (e.g., LSTM, CNN, hybrid) outperformed conventional ML models, showing robustness against overfitting.Shapley Additive exPlanations (SHAP) analysis highlighted precipitation and runoff as key drivers of pollutant loads. Conservation practices significantly reduced pollutant loads, with the treatment field achieving reductions of 32.73% (runoff), 46.11% (sediment), 47.34% (TP), and 43.75% (TN). These findings demonstrate the effectiveness of ML models for real-time monitoring and the tangible benefits of conservation practices mitigating agricultural pollution.