Faster, More Informed Environmental Permitting with AI-Guided Support

RICHLAND, Wash.—It began with a simple question. Can we use artificial intelligence (AI) tools and techniques to speed up the arduous federal environmental permitting process for proposed energy-related projects? 

The Department of Energy was tasked to make it happen in an October 30, 2023, White House executive order. By April 2024, DOE had announced a $13 million VoltAIc initiative to streamline siting and permitting of critical clean energy infrastructure. DOE’s Pacific Northwest National Laboratory data scientists and subject matter experts received $300,000 to fund a pilot “PolicyAI” prototype to assist federal regulators in speeding the environmental review process. 

With unprecedented speed and efficiency, PNNL data scientists collected and extracted 28,212 documents across 2,917 different National Environmental Policy Act reviews into an AI-ready searchable database. The current version contains 4.8 million pages and over 3.6 billion bits of information related to federal environmental impact statements. Until now, these records were not available or searchable in one location. The entire database is now publicly available. 

“There’s a lot of variation in the structure and format of NEPA documents,” said Davie Nguyen, the deputy director for state, local, tribal, and territorial policy in DOE’s Office of Policy. “Utilizing AI to help us quickly search, interpret, and synthesize information from thousands of federal projects is a gamechanger—we can use AI tools to build capacity in federal agencies and drive more effective decision-making.”

But efficient searching was just the start. Now, the PolicyAI team is ready to deploy the AI toolkit as part of its collaboration with the DOE Office of Policy. A new funding infusion of $10 million will permit the team to expand its scope and reach, with the goal of having publicly searchable, safe, secure, and reliable AI tools and applications available over the next two years. 

“The environmental review and permitting process is a very tedious, expensive and time-consuming decision-making process that is a shared problem across almost all government agencies that do these national environmental reviews,” said PNNL data scientist Sameera Horawalavithana, the principal investigator of PolicyAI. “Right now, to conduct an environment impact statement usually take around two to four years and costs from $100,000 to $1 million.” 

Reducing the cost and time to conduct environmental reviews will require a uniform “playbook.” PolicyAI researchers are currently working to add value to collected data by annotating data from previous environmental reviews to make it compatible with AI large language models. This crucial step is expected to  deliver the value-added necessary to deliver time and cost savings.

“Like starting from scratch each time”

It began with the best of intentions. The 1969 National Environmental Policy Act created a uniform process to assess the potential for negative environmental effects of proposed federal projects. When signed into law in 1970, nearly all federal agencies were charged with preparing environmental impact statements for major federal actions having a significant effect on the environment.

Each year, federal agencies evaluate thousands of NEPA documents, including environmental impact statements (EIS), environmental assessments (EA), and categorical exclusions (CX), among others. These documents promote transparency in the decision-making process and provide the public an opportunity to better understand the environmental impacts of proposed projects in their community. Now more than 50 years old, the NEPA documentation process has evolved at the federal agencies charged with implementing the law.

While fundamental principles remain the same, agency mandates, missions, expertise, resources, guidance, regulations, and the types of projects evaluated create variability in the NEPA documents and can lead to inconsistencies, added Nguyen. As a result, the overall time and cost of conducting reviews has become weighty for everyone involved. 

When he started the PolicyAI project, Horawalavithana spoke with NEPA reviewers to learn how they implement NEPA regulations.

“When they are starting these reviews, we learned that there is lack of access to previous reviews, which made it like starting from scratch each time,” said Horawalavithana. “We are trying to identify these pain points and developing several applications that basically automate certain parts of this documentation processes. Of course, we are not ever going to automate the entire process with one single model, and NEPA evaluators will always be driving the process.”

Instead, the research team is using its data analysis skills to work with NEPA practitioners to identify critical data needs that are often referenced while preparing environmental reviews. The team is helping tag such data objects so that AI can easily find and synthesize information found in those data.

Looking ahead

The PolicyAI team is currently beta testing SearchNEPA, an interactive AI-driven toolkit with an interface designed for federal NEPA reviewers. About 30 DOE evaluators are putting the AI component to the test, and several other government agencies plan to test the AI applications.

For example, the DOE Office of NEPA Policy and Compliance is testing SearchNEPA to identify patterns and best practices in the analysis of potential environmental effects in proposed energy transmission projects. This is part of a larger DOE initiative called VoltAIc, which is aimed at improving siting and permitting of energy infrastructure. PolicyAI is the cornerstone of the VoltAIc initiative. 

“We aim to offer a one-stop comprehensive NEPA database by aggregating and organizing data from several agencies, with versions that are available to industry and the public coming within one to two years,” said Sai Munikoti, a PNNL data scientist and co-principal investigator of PolicyAI.

Initiatives underway include: 

• Developing shared data standards and a shared data architecture to help federal agencies collaborate in performing environmental reviews.
• Creating standard benchmarks for the use of AI models in NEPA decision-making, thereby reducing costs while increasing accuracy.
• Deploying applications and tools to facilitate NEPA decision-making workflows. These future deployments include:
  • ChatNEPA, an AI-powered assistant that is being trained to return reliable and relevant answers to questioners' queries about prior NEPA decisions.
  • EngageNEPA, which automatically extracts public engagement comments published in the Federal Register and integrates a conversational AI-powered assistant for searches and summaries. 

“Our aim is simple,” said Munikoti. “We want to make these reviews go faster and be less cumbersome through safe, secure use of AI tools in a trustworthy environment.”