Machine learning model identifies waters protected under different interpretations of the U.S. Clean Water Act

(Press-News.org) The U.S. Clean Water Act is a critically important part of federal water quality regulation, but the act does not define the exact waters that fall under its jurisdiction. Now, Simon Greenhill and colleagues have developed a machine learning model that helps to clarify which waters are protected from pollution under the United States’ Clean Water Act, and how recent rule changes affect protection. The model demonstrates that the waters protected under the act differ substantially depending on whether the act’s regulations follow a 2006 U.S. Supreme Court ruling or a 2020 White House rule. Under the 2006 Rapanos Supreme Court ruling, the model suggests that the Clean Water Act protects two-thirds of U.S. streams and more than half of its wetlands. Under the 2020 White House rule, protection declines to under one-half of streams and a fourth of wetlands, “implying deregulation of 690,000 stream miles, 35 million wetland acres and 30% of drinking water sources,” the researchers write. Waters of the United States-Machine Learning (WOTUS-ML) offers the first national, geographically resolved estimate of legally binding Clean Water Act regulation, and predicts jurisdictional determinations under different rules and court cases, Greenhill et al. conclude. As such, the model could be helpful in some cases as a decision-support tool for permitting and policy design, they note. Greenhill et al. trained the model using aerial imagery of waters, hydrological data and wetland coverage from national inventories, soil, groundwater and elevation data and regulatory district and state boundaries, testing the model against more than 150,000 Army Corps of Engineers cases to determine waters jurisdiction.

END