The role of livestock in nitrogen pollution of large river basins: A machine learning-based assessment

Livestock are known to be one of the leading sources of nitrogen and other organic pollutant contents in surface waters. In response to growing demand for animal-source foods, livestock production has shifted into more high-input systems, accompanied by increased fertilizer and animal manure application rates to produce feed, which has resulted in large losses of nitrogen. The effects of nitrogen contamination of water bodies on human and ecosystem health are not limited and localized but have large-scale implications and are rapidly becoming widespread. However, current approaches to estimate the changes in stream water quality often fail to explicitly account for livestock activities across large spatial scales. To overcome this challenge, we adopted a data-driven approach and developed a spatio-temporal water quality model. Our model is based on a popular supervised machine learning technique, known as random forests, that can efficiently handle large, heterogeneous geo-environmental datasets. The model was trained using several site-level measurements and a large set of gridded environmental covariates to predict monthly nitrogen concentrations across the world. We then performed variable importance analysis on the proposed model to identify influential drivers of nitrogen variability at global scale. Our results confirmed the prominent role of livestock population and nitrogen fertilizer use in pollution of the river systems. Finally, we quantified how much livestock has contributed to nitrogen pollution in 115 major river basins of the world. We found that during 1992-2010 the average increase in nitrogen concentrations due to livestock was about 15% globally. Importantly, model results also indicate that at some large basins livestock population is responsible for more than 50% of raise in the levels of nitrogen. These regions point to the global livestock ‘‘hot spots’’ where high nitrogen loading to waterways may be expected. The results and insights gained in this study can have important implications for better management of livestock faming systems and pollution control policies.