Machine learning can help reveal the complicated designation of which US waters are protected—or not—under various interpretations of the Clean Water Act.
The 1972 Clean Water Act protects the “waters of the United States” but does not precisely define which streams and wetlands this designation covers, leaving decisions about the interpretation of the phrase to presidential administrations, regulators, and courts. As a result, the exact coverage of Clean Water Act rules is difficult to estimate.
Our new research in the journal Science used machine learning to predict which streams and wetlands the Clean Water Act protects. A short video explains our research, and an interactive map lets users see the predictions. Our analysis finds that a rule implemented by the Trump administration in 2020, called the Navigable Waters Protection Rule, removed protections for one-fourth of all wetlands and one-fifth of all streams in the United States. These changes deregulated 30 percent of watersheds that supply drinking water to household taps.
Using machine learning to understand these rules helps clear the muddy waters and finally lets us understand what the Clean Water Act actually protects.
Prior analyses assumed that streams and wetlands sharing certain geophysical characteristics were regulated, without scrutinizing data on what waters actually were regulated according to past rulings—an approach the US Environmental Protection Agency and Army Corps of Engineers have called “highly unreliable.”
We trained a machine learning model to predict 150,000 jurisdictional decisions by the Army Corps of Engineers. Each Corps decision interprets the Clean Water Act for one site and rule. The model predicts regulation across the United States under the 2020 rule and the prior rule, which was defined in 2006 by the Supreme Court in Rapanos v. United States and previously had guided decisions by the Corps (Figure 1).
Figure 1. Probability of Regulation under the Clean Water Act According to Different Interpretations
The Clean Water Act repeatedly is debated and reinterpreted by the president, Supreme Court, and state litigation. This game of regulatory ping-pong has staggering effects on environmental protection. Our research finds that the 2020 rule deregulated 690,000 miles of stream—more than every stream in California, Florida, Illinois, New York, Ohio, Pennsylvania, and Texas combined. We estimate that the wetlands deregulated under the 2020 rule provided over $250 billion in flood-prevention benefits to nearby buildings.
Although the Clean Water Act targets water pollution, the regulations end up restricting land use. The regulations affect development involving waters that are covered under the Clean Water Act, because construction activities deposit dredge (excavated material such as soil) and fill (material such as sand that’s intended to change a body of water) into regulated waterways, which pollutes those waterways. For example, Clean Water Act regulations may restrict a residential development project located on a wetland, a solar project in the desert where an ephemeral stream flows a few days per year, or a road that crosses isolated marshes.
Our new study provides both a high-level overview of how the scope of regulation changes under different interpretations and location-specific predictions of whether a given site is regulated under each interpretation. For a subset of these sites where the accuracy of a prediction is expected to be high, the model can act as a tool to support decisions that may expedite construction permits. We estimate that this tool could save over $1 billion annually in permitting costs for regulators and developers by providing immediate estimates of the probability that a site is regulated, rather than interested parties waiting months for a permitting process to conclude.
Our study covers three recent Clean Water Act rules, but the available data ends before two changes were implemented in 2023: First, the Biden administration issued a new rule that expands Clean Water Act jurisdiction. Second, later that year, the Supreme Court ruling in Sackett v. US Environmental Protection Agency narrowed the scope of regulation once again. Once the ruling in Sackett is fully implemented, our same machine learning methodology can be deployed to clarify the scope of this newest Supreme Court decision.