The 1972 Clean Water Act has played a crucial role in protecting the “waters of the United States,” but the specific streams and wetlands covered by this legislation have often been subject to interpretation. The lack of precise definition has resulted in the responsibility of determining coverage falling on presidential administrations, regulators, and courts. Consequently, it has been challenging to estimate the exact reach of Clean Water Act rules. However, a recent research study conducted by the University of California, Berkeley, utilizing machine learning techniques, has shed light on this issue.
In this groundbreaking analysis published in Science, the UC Berkeley team employed machine learning to improve the accuracy of predicting which waterways are protected by the Clean Water Act. The research unveiled that a 2020 rule implemented by the Trump administration led to the removal of protections for a significant portion of U.S. wetlands and streams. Notably, approximately 25% of U.S. wetlands and 20% of U.S. streams lost their Clean Water Act coverage due to this ruling. Moreover, the 2020 rule also resulted in the deregulation of 30% of watersheds responsible for supplying drinking water to households.
Author Joseph Shapiro, an associate professor of Agricultural and Resource Economics at UC Berkeley, acknowledges the importance of utilizing machine learning in understanding environmental policy. He states that this approach allows policymakers to gain a comprehensive understanding of the protection provided by the Clean Water Act. Previous analyses focused on geophysical characteristics to assume regulation; however, this study brings forth more reliable information by assessing the actual data on regulation.
The machine learning model developed by the researchers predicted 150,000 jurisdictional decisions made by the Army Corps, providing valuable insights into the 2020 rule and the Supreme Court’s “Rapanos” ruling, which previously guided Corps decisions. The analysis revealed that the 2020 rule resulted in the deregulation of an astounding 690,000 stream miles, surpassing the combined length of streams in California, Florida, Illinois, New York, Ohio, Pennsylvania, and Texas. Additionally, the wetlands no longer protected under this rule were estimated to provide over $250 billion worth of flood prevention benefits to nearby buildings.
Author Simon Greenhill, a Ph.D. candidate at UC Berkeley, emphasizes that the back-and-forth nature of regulatory changes has massive impacts on environmental protection. The study suggests that adopting the machine learning model’s predictions could save regulators and developers over $1 billion annually in permitting costs. This estimation results from the ability of machine learning to provide immediate estimates regarding the probability of a site falling under regulation, streamlining the often uncertain permitting process.
The study’s findings also consider subsequent events, such as the 2023 Biden White House rule that expanded Clean Water Act jurisdiction and the Supreme Court’s 2023 Sackett decision, which limited it. As the implementation of the Sackett decision progresses, the researchers believe that their machine learning methodology can provide clarity on its scope.
This innovative research employing machine learning has provided valuable insights into the impact of the Clean Water Act on U.S. waterways. By improving the understanding of regulations and their coverage, policymakers can make informed decisions to ensure the protection of vital ecosystems and the well-being of local communities.