Development of Machine Learning Models to Predict Daily Gas Ebullition Flux in Waterways from Sediment Characteristics

Gas ebullition plays a critical role in facilitating contaminant transport from the sediment to the water column. Predicting gas ebullition accurately is necessary for managing sediments and assessing sediment/water flux, and many researchers have published models to predict gas ebullition in lakes, ponds, and rivers. The need for site-specific data and a strong emphasis on water column parameters in published models (rather than sediment data where gas ebullition occurs), coupled with the narrow range of systems studied, limits the application of current published methods to predict ebullition. This study develops machine learning (ML) generalized models to predict gas ebullition accurately across a diverse range of waterways under various ecohydrological conditions. Input data are common sediment parameters, depth and temperature, and gas ebullition flux rates acquired through various methods. We trained and evaluated Multivariate Linear Regression (MLR), Random Forest (RF), eXtreme Gradient Boosting (XGB), and Neural Network (NN) models to predict gas flux using data from over 40 sites across all seasons. ML approaches significantly enhance prediction accuracy compared to results predicted by ten published regression models, most of which had low/no predictive capability for the waterway data set. Among the ML models, the results show that RF and XGB performed significantly better (<i>r</i> ² = 0.79 and 0.80, respectively), and temperature, COD/TOC ratio, and water depth are the most influential parameters, consistent with known mechanisms of methane production and sediment fracture. The results further show that currently published regression models for ebullition based on lacustrine systems have little predictive capability for waterways. The ability to predict gas ebullition accurately using commonly measured parameters suggests immense potential to enhance ebullition modeling, site assessment, and remediation design.