A Machine Learning Approach for Prediction of Surface Water Quality Vulnerabilities of American Rivers under Climate Change

Abstract:

Water public utilities operate water storage and treatment facilities together with distribution and transport networks to deliver high-quality water to their customers. With changing climate and land use/cover underway, water public utilities must plan for potential future alterations of the natural systems supplying the continental surface water feeding their systems. While a significant amount of work in the past decades has been dedicated to the prediction of water quantity for the decades to come, less attention was paid regarding the evolution of water quality. This research intends to address this gap. Physically-based approaches have been used to model the transport of contaminants along with the river networks. However, their application for long-term prediction of water quality is complicated by their computational and data requirement costs (data that are often location-specific and not easy to obtain). As such, the overarching goal of this research is to assess the use of machine-learning methods as a generalizable prediction tool to predict the change in surface water quality. The considered prediction tool relies on a combination of regression, neural network, and classification methods. The predictors include hydrometeorological variables (precipitation, temperature, and streamflow), topologies and land cover/use description of the simulated catchments. The list of predictands includes six major water quality indicators from which the Water Quality Index is estimated (i.e., turbidity, dissolved oxygen, water temperature, total nitrogen, total phosphorus and total coliform). Water quality data is obtained from the Water Quality Portal, EPA (the Environmental Protection Agency). The machine-learning prediction tool is developed and tested across nearly 250 catchments from USGS (the United States Geological Survey) database, sampling the range of hydroclimatic and land cover conditions in the continental United States. Results show that the deterioration of water quality is magnified under climate change unless mitigation and adaptation are conducted.

Keywords: Machine Learning, Surface Water Quality, Vulnerability, Climate Change, Water Quality Index