Datasets:

PNNL
/

NEPATEC3.0

license: cc0-1.0
task_categories:
  - text-generation
  - text-classification
  - image-classification
language:
  - en
tags:
  - environment
  - permitting
  - nepa
  - gis
  - geoai
size_categories:
  - 100K<n<1M

National Environmental Policy Act Text Corpus (NEPATEC3.0)

Dataset Description

The National Environmental Policy Act of 1969, as amended (NEPA), is a major environmental law in the United States, requiring Federal agencies to consider and document potential environmental impacts before deciding on a proposed action. Modernization of NEPA and permitting processes faces significant challenges due to the lack of standardized formats and interoperable systems for organizing and sharing NEPA-related information across agencies. Much of the information gathered during NEPA reviews is written into documents such as categorical exclusions, environmental assessments, and environmental impact statements, then stored in independent agency repositories with varying formats, metadata standards, and access mechanisms. The application of metadata and data standards, such as those recommended by the Council on Environmental Quality (CEQ), provides a shared vocabulary and structure for key entities such as projects, processes, and documents, helping streamline information exchange and collaboration across systems.

In this work, we publicly release NEPATEC3.0, an expanded corpus of public Federal environmental review documents with associated metadata, full text, and image information. Developed by Pacific Northwest National Laboratory (PNNL), the National Environmental Policy Act Text Corpus (NEPATEC) is an AI-ready dataset and standardized metadata repository designed to compile, organize, and enrich environmental review documents. The NEPATEC dataset is the primary focus and contribution of the PermitAI Data Thrust. NEPATEC3.0 substantially increases the number of metadata attributes extracted or generated from source documents and expands the diversity of contributing agencies. The dataset includes 188,330 files from 81,347 projects, 42 metadata attributes, and text and image modalities. Documents were collected from EPA, DOE, USDA, BLM, BOEM, USACE, and DHS using public URL endpoints, agency-facilitated downloads, agency-provided data files, and the ERDC API.

Curated by: Pacific Northwest National Laboratory
Funded by: Office of Policy, Department of Energy
Language(s) (NLP): English
License: CC0

Usage

The dataset repository is organized into three top-level folders:

Folder	Contents	Size
`nepatec3_metadata`	The full NEPATEC3.0 metadata (all 42 attributes described in Dataset Structure), split by agency.	1.83 GB
`nepatec3_text_full`	The same metadata as `nepatec3_metadata`, plus the full extracted text for every file (each page's `page_text`).	25.59 GB
`nepatec3_gis`	All extracted map images (PNG) and GeoTIFFs referenced by the metadata, arranged in subfolders that follow each map image's `image_document_path` and `image_name` (and, for GeoTIFFs, `geotiff`) metadata fields.	2277.72 GB (~2.278 TB)

If you only need the metadata (no full text or images), use the code in Load the Metadata Version below to stream it directly or save an exact local copy of just nepatec3_metadata -- no separate download step needed. If you'd rather not use Python, see the CLI/Git alternatives in Alternative: Download via CLI or Git (No Python).

Install the required package:

pip install datasets

Load the Metadata Version

The snippet below covers both ways to get the metadata: dataset_metadata streams records directly into memory (nothing written to disk), and the block right after it is an optional extra step that instead saves an exact local copy of the nepatec3_metadata folder -- use whichever one fits your use case, or both.

import httpx
from datasets import load_dataset
from huggingface_hub import HfApi, hf_hub_download
from huggingface_hub.hf_api import RepoFile
from huggingface_hub.utils import set_client_factory
from huggingface_hub.utils._http import hf_request_event_hook

set_client_factory(lambda: httpx.Client(
    event_hooks={"request": [hf_request_event_hook]},
    follow_redirects=True,
    timeout=httpx.Timeout(120.0, write=60.0),
))

dataset_metadata = load_dataset(
    "json",
    # To load only one agency, add its folder name to the path, e.g.:
    # "hf://datasets/PNNL/NEPATEC3.0/nepatec3_metadata/epa/**/*.jsonl"
    data_files={"train": "hf://datasets/PNNL/NEPATEC3.0/nepatec3_metadata/**/*.jsonl"},
    split="train",
    streaming=True,
)

# Optional: instead of the flattened records dataset_metadata yields above,
# mirror the exact on-Hub folder structure (nepatec3_metadata/<agency>/<batch>/
# merged/<filename>.jsonl, one file per file) onto local disk.
#
# Note: snapshot_download() (and the `hf download`/huggingface-cli command)
# lists the ENTIRE repo tree first -- all ~188K files across nepatec3_metadata
# + nepatec3_text_full + nepatec3_gis (~2.3 TB) -- then filters by allow_patterns
# afterwards, so it's slow/prone to timing out on a repo this size even though
# only ~1.83 GB actually matches. Listing this folder's subtree directly
# instead (like the glob above resolves internally) skips that full-repo walk.
api = HfApi()
folder = "nepatec3_metadata"  # or e.g. "nepatec3_metadata/epa" for one agency

paths = [
    item.path
    for item in api.list_repo_tree(repo_id="PNNL/NEPATEC3.0", path_in_repo=folder,
                                    repo_type="dataset", recursive=True)
    if isinstance(item, RepoFile)
]
for path in paths:
    hf_hub_download(repo_id="PNNL/NEPATEC3.0", repo_type="dataset", filename=path, local_dir=".")

Load the Metadata + Full Text Version

Same two options as above, scoped to nepatec3_text_full instead: stream via dataset_metadata, or use the optional block after it to save an exact local copy of the folder.

import httpx
from datasets import load_dataset
from huggingface_hub import HfApi, hf_hub_download
from huggingface_hub.hf_api import RepoFile
from huggingface_hub.utils import set_client_factory
from huggingface_hub.utils._http import hf_request_event_hook

set_client_factory(lambda: httpx.Client(
    event_hooks={"request": [hf_request_event_hook]},
    follow_redirects=True,
    timeout=httpx.Timeout(120.0, write=60.0),
))

dataset_metadata = load_dataset(
    "json",
    # To load only one agency, add its folder name to the path, e.g.:
    # "hf://datasets/PNNL/NEPATEC3.0/nepatec3_text_full/epa/**/*.jsonl"
    data_files={"train": "hf://datasets/PNNL/NEPATEC3.0/nepatec3_text_full/**/*.jsonl"},
    split="train",
    streaming=True,
)

# Optional: instead of the flattened records dataset_metadata yields above,
# mirror the exact on-Hub folder structure (nepatec3_text_full/<agency>/<batch>/
# merged/<filename>.jsonl, one file per file) onto local disk.
#
# Note: snapshot_download() (and the `hf download`/huggingface-cli command)
# lists the ENTIRE repo tree first -- all ~188K files across nepatec3_metadata
# + nepatec3_text_full + nepatec3_gis (~2.3 TB) -- then filters by allow_patterns
# afterwards, so it's slow/prone to timing out on a repo this size even though
# only ~25.59 GB actually matches. Listing this folder's subtree directly
# instead (like the glob above resolves internally) skips that full-repo walk.
api = HfApi()
folder = "nepatec3_text_full"  # or e.g. "nepatec3_text_full/epa" for one agency

paths = [
    item.path
    for item in api.list_repo_tree(repo_id="PNNL/NEPATEC3.0", path_in_repo=folder,
                                    repo_type="dataset", recursive=True)
    if isinstance(item, RepoFile)
]
for path in paths:
    hf_hub_download(repo_id="PNNL/NEPATEC3.0", repo_type="dataset", filename=path, local_dir=".")

Alternative: Download via CLI or Git (No Python)

The Python snippets in Load the Metadata Version and Load the Metadata + Full Text Version above already cover streaming and saving a local copy -- use this section only if you'd rather not use Python at all.

Hugging Face CLI

hf download (formerly huggingface-cli download) uses snapshot_download() under the hood, which lists the entire repo tree -- all 188k files across nepatec3_metadata + nepatec3_text_full + nepatec3_gis (2.3 TB) -- before filtering by --include, so it can be slow or time out on a repo this large. For a faster, scoped alternative, see the Python snippets above instead.

Download only the metadata folder:

hf download PNNL/NEPATEC3.0 \
  --repo-type dataset \
  --include "nepatec3_metadata/**" \
  --local-dir ./nepatec3_metadata

Download the metadata-plus-full-text folder:

hf download PNNL/NEPATEC3.0 \
  --repo-type dataset \
  --include "nepatec3_text_full/**" \
  --local-dir ./nepatec3_text_full

Selective Git LFS Clone

Unlike the CLI download above, this doesn't go through the same repo-tree listing endpoint, so it isn't subject to the same slowness on a repo this large -- clone the (lightweight, pointer-only) structure once, then pull just the LFS content you need.

First, clone only the repository structure:

GIT_LFS_SKIP_SMUDGE=1 git clone \
  https://huggingface.co/datasets/PNNL/NEPATEC3.0

cd NEPATEC3.0

Download only the metadata folder:

git lfs pull --include="nepatec3_metadata/**"

Download only the metadata-plus-full-text folder:

git lfs pull --include="nepatec3_text_full/**"

Dataset Structure

Each JSONL record represents a project and its associated NEPA process, documents, files, pages, geographic metadata, and map-image metadata.

The following structure is abbreviated:

{
  "project": {
    "project_id": "UNIQUE PROJECT ID FOR PUBLIC VERSION",
    "project_title": {
      "value": ""
    },
    "project_description": {
      "value": ""
    },
    "project_sector": {
      "value": []
    },
    "project_type": {
      "value": []
    },
    "project_sponsor": {
      "value": []
    },
    "geojson": {
      "type": "FeatureCollection",
      "bbox": [],
      "metadata": {
        "location_description": {
          "value": ""
        },
        "location_short_name": {
          "value": ""
        }
      },
      "features": [
        {
          "type": "Feature",
          "geometry": {
            "type": "Point",
            "coordinates": [0.0, 0.0]
          },
          "properties": {
            "administrative_areas": {
              "country": {
                "value": ""
              },
              "region_name": {
                "value": ""
              },
              "subregion_name": {
                "value": ""
              },
              "state_province": {
                "value": ""
              },
              "county": {
                "value": ""
              },
              "municipality": {
                "value": ""
              },
              "postcode": {
                "value": ""
              }
            }
          }
        }
      ]
    }
  },
  "process": {
    "process_id": "UNIQUE PROCESS ID",
    "process_type": {
      "value": ""
    },
    "process_family": {
      "value": ""
    },
    "federal_agencies": [
      {
        "agency": "",
        "bureau": "",
        "agency_detail": "",
        "role": ""
      }
    ],
    "tribes": {
      "value": []
    },
    "nonfederal_agencies": {
      "value": []
    },
    "documents": [
      {
        "document_metadata": {
          "document_id": {
            "value": "UNIQUE DOCUMENT ID"
          },
          "document_type": {
            "value": []
          },
          "document_title": {
            "value": ""
          },
          "supplement": {
            "value": false
          },
          "programmatic": {
            "value": false
          },
          "publish_date": {
            "value": {
              "iso": "",
              "year": -1,
              "month": -1,
              "day": -1
            }
          },
          "prepared_by": {
            "value": []
          },
          "main_document": {
            "value": false
          },
          "ce_summary": {
            "ce_description": {
              "value": "",
              "source_file_uuids": []
            },
            "proposed_action": {
              "value": "",
              "source_file_uuids": []
            },
            "ce_selection": {
              "value": "",
              "source_file_uuids": []
            },
            "supporting_documentation": {
              "value": "",
              "source_file_uuids": []
            },
            "issues": {
              "value": "",
              "source_file_uuids": []
            },
            "public_engagement": {
              "value": "",
              "source_file_uuids": []
            },
            "consultation": {
              "value": "",
              "source_file_uuids": []
            }
          },
          "ea_eis_summary": {
            "proposed_action": {
              "value": "",
              "source_file_uuids": []
            },
            "purpose_need": {
              "value": "",
              "source_file_uuids": []
            },
            "alternatives": {
              "value": "",
              "source_file_uuids": []
            },
            "affected_environment": {
              "value": "",
              "source_file_uuids": []
            },
            "environmental_consequences": {
              "value": "",
              "source_file_uuids": []
            },
            "mitigation": {
              "value": "",
              "source_file_uuids": []
            },
            "public_engagement": {
              "value": "",
              "source_file_uuids": []
            },
            "consultation": {
              "value": "",
              "source_file_uuids": []
            }
          }
        },
        "files": [
          {
            "file_metadata": {
              "file_id": {
                "value": "UNIQUE FILE ID"
              },
              "file_hash": {
                "value": ""
              },
              "file_name": {
                "value": ""
              },
              "file_group_id": {
                "value": 0
              },
              "file_type": {
                "value": []
              },
              "file_date": {
                "value": {
                  "iso": "",
                  "year": -1,
                  "month": -1,
                  "day": -1
                }
              },
              "section_or_volume_title": {
                "value": ""
              },
              "sort_order": {
                "value": 0
              },
              "file_provider": {
                "value": ""
              },
              "total_pages": {
                "value": ""
              }
            },
            "pages": [
              {
                "page_number": 1,
                "page_text": "PAGE 1 TEXT",
                "map_object": [
                  {
                    "image_id": "UNIQUE IMAGE ID",
                    "image_document_path": "",
                    "image_name": "",
                    "geotiff": {},
                    "bbox": {
                      "crs": "EPSG:4326",
                      "min_lon": 0.0,
                      "min_lat": 0.0,
                      "max_lon": 0.0,
                      "max_lat": 0.0
                    },
                    "legend_units": {
                      "<legend item name>": {
                        "geometry": "Polygon"
                      }
                    },
                    "cartographic_elements": {
                      "Legend": false,
                      "InsetMap": false,
                      "NorthArrow": false,
                      "ScaleBar": false,
                      "basemap": {
                        "value": ""
                      }
                    },
                    "figure_caption": {}
                  }
                ]
              }
            ]
          }
        ]
      }
    ]
  }
}

Metadata attributes are grouped by entity.

Entity	Metadata Attribute	Description	Datatype
Project	project_id	PNNL-generated identifier for the project.	text
Project	project_title	Descriptive name of the project.	text
Project	project_description	Description or summary of the project or proposed action.	text
Project	project_sector	High-level project sector. Selection should be the best fit of available options.	text/list
Project	project_type	Type or types of project. A subtype of project sector.	text/list
Project	project_sponsor	Name of the responsible entity, organization, or person for the project.	text/list
Location	country	Country associated with the project location.	text
Location	region_name	Region associated with the project location.	text
Location	subregion_name	Subregion associated with the project location.	text
Location	state_province	State, province, or equivalent administrative area.	text
Location	county	County or equivalent administrative area.	text
Location	municipality	Municipality associated with the project location.	text
Location	postcode	Postal code associated with the project location.	text
Location	location_description	Detailed description of the project location.	text
Location	location_short_name	Short description of the project location.	text
Location	geojson	Geographic representation of the project location.	object
Process	process_id	Identifier for the review or permitting process.	text
Process	process_family	Major category to which the process belongs.	text
Process	process_type	Type of review or permitting process.	text
Process	federal_agencies	Federal agencies and organizational components associated with the process.	list[object]
Process	tribes	Tribal governments or Tribal Nations associated with the process.	list[text]
Process	nonfederal_agencies	Nonfederal agencies associated with the process.	list[text]
Document	document_id	Identifier for a logical document.	text
Document	document_type	Type or types of document.	list[text]
Document	document_title	Title of the document, reflecting the actual title rather than the file name.	text
Document	supplement	Indicates whether the document is a supplement.	boolean
Document	programmatic	Indicates whether the document is programmatic.	boolean
Document	publish_date	Normalized publication date.	object
Document	prepared_by	Agency or entity responsible for preparation.	list[text]
Document	main_document	Indicates whether the document is a main review document.	boolean
Document	ce_summary.ce_description	Description of the categorical exclusion applied (populated for CE documents).	text
Document	ce_summary.proposed_action	Summary of the proposed action, from the CE summary.	text
Document	ce_summary.ce_selection	Rationale for selecting the categorical exclusion.	text
Document	ce_summary.supporting_documentation	Supporting documentation referenced for the exclusion.	text
Document	ce_summary.issues	Issues or extraordinary circumstances considered.	text
Document	ce_summary.public_engagement	Description of public engagement activities, from the CE summary.	text
Document	ce_summary.consultation	Description of consultation activities, from the CE summary.	text
Document	ea_eis_summary.proposed_action	Summary of the proposed action (populated for EA/EIS documents).	text
Document	ea_eis_summary.purpose_need	Purpose and need for the proposed action.	text
Document	ea_eis_summary.alternatives	Alternatives considered.	text
Document	ea_eis_summary.affected_environment	Description of the affected environment.	text
Document	ea_eis_summary.environmental_consequences	Summary of environmental consequences.	text
Document	ea_eis_summary.mitigation	Mitigation measures described.	text
Document	ea_eis_summary.public_engagement	Description of public engagement activities, from the EA/EIS summary.	text
Document	ea_eis_summary.consultation	Description of consultation activities, from the EA/EIS summary.	text
File	file_id	PNNL-generated identifier for an individual file.	text
File	file_hash	Hash associated with the file.	text
File	file_name	Name of the file.	text
File	file_group_id	Identifier used to group related files.	integer
File	file_type	Type or types assigned to the file.	list[text]
File	file_date	Normalized date associated with the file.	object
File	sort_order	Inferred sort order for files belonging to a multi-file document.	integer
File	section_or_volume_title	Title of a specific document section or volume.	text
File	file_provider	Agency or repository that provided the file.	text
File	total_pages	Number of pages in the file.	integer/text
Page	page_number	Page number within the file.	integer
Page	page_text	Text extracted from the page.	text
Map Image	image_id	Identifier for an extracted image or map.	text
Map Image	image_document_path	Path associated with the extracted image.	text
Map Image	image_name	Name of the extracted image.	text
Map Image	geotiff	GeoTIFF-related information, when available.	object
Map Image	bbox	Bounding-box information, when available.	object
Map Image	legend_units	Units associated with the map legend.	object/text
Map Image	cartographic_elements	Presence of a legend, inset map, north arrow, or scale bar.	object
Map Image	cartographic_elements.basemap	Type of basemap underlying the map image (e.g. topographic, satellite imagery, street map), when identifiable.	text
Map Image	figure_caption	Extracted figure caption, when available.	object/text
Map Image	figure_references	References to the figure elsewhere in the document.	object/list

The table above lists 67 rows. Some rows document the individual components of a single attribute rather than a separate attribute in their own right: the 7 administrative_areas.* rows are components of one administrative_areas attribute, the 7 ce_summary.* rows are components of one ce_summary attribute, the 8 ea_eis_summary.* rows are components of one ea_eis_summary attribute, and cartographic_elements.basemap is a component of cartographic_elements. Rolling these up, and excluding the 5 per-entity identifiers (project_id, process_id, document_id, file_id, image_id) that are corpus-generated/pass-through record keys rather than metadata extracted or generated about the document, NEPATEC3.0 contains 42 metadata attributes.

Data Sources

Documents were collected with the support of agency stakeholders in coordination with DOE using public URL endpoints, web scraping, agency-facilitated downloads, agency-provided data files, and the ERDC API.

Data Source	Download Source	Terms of Use
EPA	https://cdxapps.epa.gov/cdx-enepa-II/public/action/nepa/search	https://edg.epa.gov/epa_data_license.html
DOE	Agency-facilitated downloads from https://www.energy.gov/nepa/nepa-documents	https://www.energy.gov/web-policies
BLM	Web scraping and agency-facilitated downloads from https://eplanning.blm.gov/eplanning-ui/home	https://www.doi.gov/copyright
USDA	Agency-provided data ZIP file	Public-domain concurrence from agency stakeholder
BOEM	Agency-provided data ZIP file	No public terms of use identified
USACE	ERDC API: https://erdc-library.erdc.dren.mil/home	No public terms of use identified
DHS	Web scraping from https://www.dhs.gov/ocrso/eed/epb/nepa/archive	No public terms of use identified

Users should review the applicable terms of use and policies for the original source documents.

NEPATEC3.0 Dataset Statistics

The following statistics are preliminary.

Metric	Total Count
Projects	81,347
Metadata attributes	42
Files	188,330
Pages	10,237,240
Map images	637,477
Agencies	7

Breakdown by NEPA Process

Process	Projects	Files	Pages	Maps	Geotiffs
Categorical Exclusion (CE)	45,772	57,093	263,119	22,472	2,400
Environmental Assessment (EA)	15,308	52,621	1,782,725	146,247	23,465
Environmental Impact Statement (EIS)	12,555	66,160	7,546,485	409,573	100,217
Other	7,712	12,456	644,911	59,185	11,038
Total	81,347	188,330	10,237,240	637,477	137,120

Version History

Version	Release	Data Source	Projects	Files	Metadata Attributes	Modality
NEPATEC1.0	June 2024	EPA	2,917	Approximately 26,000	5	Text
NEPATEC2.0	August 2025	EPA, DOE, USDA, BLM	61,811	143,886	21	Text
NEPATEC3.0	August 2026	EPA, DOE, USDA, BLM, BOEM, USACE, DHS	81,347	188,330	42	Text, Images

NEPATEC1.0 is available at:

https://huggingface.co/datasets/PNNL/NEPATEC1.0

NEPATEC2.0 is available at:

https://huggingface.co/datasets/PNNL/NEPATEC2.0

NEPATEC3.0 repository:

https://huggingface.co/datasets/PNNL/NEPATEC3.0

Notice

Released under the Creative Commons 0 Public Domain Dedication:

https://creativecommons.org/publicdomain/zero/1.0/

This material is free to use, and attribution is always appreciated.

Please cite the following in your work:

@misc{NEPATECv3,
  author       = {Sai D Koneru, Kaustav Bhattacharjee, Matthew E Raffel, Johnny L Chen, Siddhartha Shankar Das, Daniel M Nally, Anastasia Bernat, Heng Wan, Alexander C Buchko, Kathy Nwe, Aaron Moreno, Timothy J Vega, Sridevi N Wagle, Leah R Hare, Micah S Taylor, Derek B Lilienthal, Scott T Spare, Michael J Parker, Reilly P Raab, Sai Munikoti, and Yasanka S Horawalavithana},
  title        = {NEPATEC v3.0: NEPA Text and Image Corpus v3.0},
  howpublished = {\url{https://huggingface.co/datasets/PNNL/NEPATEC3.0}},
  year         = {2026},
  note         = {PNNL-SA-225916}
}

We welcome your feedback and suggestions to help improve this dataset.

If you have any comments or questions, please email us at permitai@pnnl.gov.

DISCLAIMER

This material was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor the United States Department of Energy, nor the Contractor, nor any of their employees, nor any jurisdiction or organization that has cooperated in the development of these materials, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, software, or process disclosed, or represents that its use would not infringe privately owned rights.

Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof, or Battelle Memorial Institute. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.

PACIFIC NORTHWEST NATIONAL LABORATORY
operated by
BATTELLE
for the
UNITED STATES DEPARTMENT OF ENERGY
under Contract DE-AC05-76RL01830

Acknowledgement

This work was supported by the Office of Policy, U.S. Department of Energy, and Pacific Northwest National Laboratory, which is operated by Battelle Memorial Institute for the U.S. Department of Energy under Contract DE-AC05–76RL01830.