PNNL Predicts River Health Using Machine Learning Data

Researchers from Pacific Northwest National Laboratory and Parallel Works, Inc. utilised machine learning to predict oxygen and nutrient use by microorganisms in river sediments, analysing a dataset of 367 samples with 133 features. Their models, incorporating over 100 environmental factors including stream order and soil characteristics, determined that sediment organic matter chemistry is crucial for predicting respiration rates across the Columbia River Basin. This research, leveraging data from the Worldwide Hydrobiogeochemistry Observation Network, aims to improve river health management and support larger-scale Earth system modelling by explaining factors controlling carbon dioxide fluxes.

Modelling Sediment Respiration with Machine Learning

Researchers at Pacific Northwest National Laboratory and Parallel Works, Inc., employed machine learning methods to predict oxygen and nutrient consumption by microorganisms within river sediments, a process central to understanding river health and ecosystem services. Their modelling considered over 100 environmental factors, encompassing variables such as stream order and soil characteristics, and utilised data gathered across the contiguous United States. The team’s approach leveraged publicly available databases alongside direct oxygen use measurements obtained from samples crowdsourced via the Worldwide Hydrobiogeochemistry Observation Network for Dynamic River Systems (WHONDRS) consortium.

A dataset comprising 367 samples and 133 features was compiled by the multi-institutional team, acknowledging that complete feature data was not available for all samples. The researchers developed a two-tiered machine learning approach, utilising a stacked ensemble of models to optimise hyperparameters and employing feature permutation importance analysis to identify significant predictive variables. This analysis revealed that the chemistry of organic matter in the sediment is crucial for predicting oxygen use, a key component of river sediment respiration.

Spatial maps predicting oxygen consumption were generated for the Columbia River Basin, demonstrating the influence of sediment organic matter chemistry – as determined by high-resolution mass spectrometry – on sediment respiration rates. Larger-scale variables, including stream order, geography, and population density, were also identified as significant factors influencing these rates. By predicting these factors over large areas, this research supports the development of regional and global Earth system models, and aids in the parameterisation of watershed models.

Key Environmental Factors Influencing River Health

The research elucidates which environmental factors influence oxygen and nutrient use by microorganisms in river sediments, ultimately controlling carbon dioxide fluxes. Machine learning techniques were used to predict these factors across extensive geographical areas, potentially supporting both regional and global Earth system models. Furthermore, the methodology aids in the parameterisation of watershed models, enhancing their predictive capabilities.

The team’s analysis identified sediment organic matter chemistry, specifically as determined by high-resolution mass spectrometry, as a critical factor in predicting sediment respiration rates. Larger-scale variables, such as stream order, geography, and population density, also play significant roles in determining these rates, demonstrating a multi-faceted influence on microbial activity within river sediments. A dataset of 367 samples and 133 features was compiled, although not all samples contained complete data for every feature.

Implications for Earth System and Watershed Models

This research supports regional and global Earth system models, and aids in parameterising watershed models. The team generated spatial maps predicting oxygen consumption across the Columbia River Basin, demonstrating the influence of sediment organic matter chemistry on sediment respiration rates. Larger-scale variables, such as stream order, geography, and population density, also play significant roles in determining respiration rates, contributing to a comprehensive understanding of microbial activity within river sediments.

The multi-institutional team compiled a dataset with 367 samples and 133 features, acknowledging that not all samples contained all features. Their approach utilised a stacked ensemble of models to optimise hyperparameters, and a feature permutation importance analysis to detect significant features, enabling the identification of key predictive variables.

This research helps explain which environmental factors affect oxygen and nutrient use by microorganisms in river sediments, which in turn controls fluxes of carbon dioxide. Using machine learning, researchers can predict these factors over large areas, which could support regional and global models of the Earth system, and aids in parameterising watershed models.