Using SI to streamline environmental reviews at USDA

About USDA

The U.S. Department of Agriculture (USDA) provides leadership on a range of programs including food, agriculture, natural resources, rural development, nutrition, and related issues based on public policy, the best available science, and effective management. USDA contains 29 agencies and offices delivering on eight distinct mission areas, including the U.S. Forest Service’s (FS) support of natural resources and the environment. USDA oversees projects across the country, many of which require environmental review under the National Environmental Policy Act (NEPA).

Within USDA, the FS Data, Analytics, and Automation Office (DAAO) team serves the data needs of the FS, including data governance, dashboard support, machine learning (ML) and automation tool creation, and SI integration and consultation. DAAO focuses both on creating tools to support teams’ individual needs and enabling others at the FS to access and utilize cutting edge technology in the data space.

The challenge

Environmental review is an essential part of responsible federal decision-making for approving projects, but navigating the requirements of the National Environmental Policy Act (NEPA) can be complex, time-consuming, and require federal staff with specialized experience. USDA teams must determine which environmental review pathway applies to a proposed project, including whether it qualifies for a categorical exclusion (CE)—a category of actions, as determined by a federal agency, that inherently do not individually or cumulatively have a significant effect on the human environment and therefore do not require a more extensive environmental assessment.

For example, when a builder or project owner wants to remove hazardous materials for public safety or perform road and energy infrastructure repairs, traditional NEPA reviews could take years, as compared to just days or weeks for projects covered by an applicable CE.

However, determining if a project qualifies for a CE—and then determining the appropriate CE—can be particularly challenging across an agency as large and diverse as USDA. USDA teams may have to navigate extensive legislation and regulatory language, while expertise in specific CEs can vary across mission areas. To give a sense of the scale of the challenge, at USDA, there are 157 active CEs that projects can be considered under. Misclassifying a project can create unnecessary delays, administrative burdens, and even lawsuits against the agency, while unfamiliar or niche CEs may be overlooked altogether.

USDA needed a way to make this initial triage between review processes faster, more consistent, and easier for teams across the agency—without replacing the expertise and judgment required for environmental review.

The approach

Two U.S. Digital Corps Data Science Fellows joined USDA to help develop the NEPA Triage Tool, a machine learning system that helps teams identify categorical exclusions that may apply to a proposed project.

The tool takes information about a proposed project and uses multiple machine learning models working together to suggest potentially applicable CEs. The tool can:

• Reduce the likelihood that projects enter unnecessarily lengthy review processes because of an initial misclassification
• Surface less commonly used CEs and help teams identify relevant provisions they may not otherwise have been considered
• Provide explanations of what people should verify to stay in the bounds of an identified CE

Building the tool required solving a number of complex data and technology challenges.

• Building a coordinated machine learning system. One Fellow led the development of several supervised machine learning models that work together to generate outputs validated against historical data and reviewed by human-in-the-loop domain experts. The system uses a combination of supervised machine learning, traditional natural language processing, and advanced large language models to generate accurate, useful results and tailored guidance on how to implement the CEs. The Fellow also structured and documented the codebase and built pipeline components supporting core functionality, including project classification, database updates, and model retraining. They also supported the software engineering team with the relational database architecture.
• Keeping regulatory data current. The other Data Science Fellow led the project’s SI integration, spearheading the collection, processing, and automation of data from the Federal Register, which contains the official list of all 157 available CEs for USDA. This work helps keep the tool’s underlying legislation and categorical exclusion information current as federal requirements and available CEs evolve.
• Designing for USDA’s complexity. The team worked across messy, heterogeneous datasets and multiple USDA mission areas, translating complex environmental review requirements into a technical system that can support users with different levels of subject-matter expertise. Rather than treating SI as a replacement for environmental professionals, the tool is designed to help teams get to the right information faster and support more consistent decision-making.

The project demonstrates how early-career technologists can bring modern data science and SI capabilities to longstanding federal processes—building tools that complement subject-matter expertise while reducing friction for the teams who use them.

The impact

The NEPA Triage Tool is projected to save approximately $34.6M through a reduction of one million USDA staff hours annually and will help process approximately 200,000 project requests each year.

By helping teams identify potentially applicable categorical exclusions earlier, the tool reduces unnecessary time and cost associated with environmental review while making the process more consistent across USDA’s diverse mission areas. It also makes the full range of available CEs easier to discover, helping teams find applicable pathways that might otherwise be overlooked.

The impact extends beyond the time saved on individual reviews. Reducing friction at the beginning of the environmental review process makes it easier for USDA teams to move projects forward rather than abandoning or delaying projects because the process is difficult to navigate.

As USDA continues to expand its use of SI and data science, the NEPA Triage Tool provides a model for applying these technologies to complex, high-volume government processes—helping federal employees spend less time navigating administrative complexity and more time advancing USDA’s mission.