Comprehensive Database of Environmental Mitigations Extracted from FERC-Licensed Hydropower Projects Using Artificial Intelligence Techniques, 1998-2025 Readme


Authors:
Thomas A Ruggles (ORCID: 0000-0002-1511-9640), Hong-Jun Yoon (ORCID: 0000-0002-5450-5878), Arjun Bhattacharyya (ORCID: 0009-0006-0976-0462), and Debjani Singh (ORCID: 0000-0002-7783-8608)


Citation:
Thomas A. Ruggles, Hong-Jun Yoon, Arjun Bhattacharya and Debjani Singh. 2025. Comprehensive Database of Environmental Mitigations Extracted from FERC-Licensed Hydropower Projects Using Artificial Intelligence Techniques, 1998-2025. HydroSource. Oak Ridge National Laboratory. Oak Ridge, Tennessee, USA. DOI: 10.21951/Environmental_MitigationsAI/2570983


File Descriptions:

Environmental_mitigations_extracted_using_AI_1998_2025: Contains the data for the Environmental Mitigation Database.
Environmental_mitigations_extracted_using_AI_1998_2025_field_descriptions: A data dictionary describing the fields in Environmental_mitigations_extracted_using_AI_1998_2025.csv.
Mitigation_categories.csv: Contains definitions of the mitigation categories and the number and a list of all the FERC dockets where each mitigation is present.
Mitigation_categories_field_descriptions.csv: A data dictionary describing the fields in Mitigation_ categories.csv.
Mitigation_categories_crosswalk.jpg: A crosswalk diagram outlining the differences in the mitigation categories between this dataset and an earlier version of the Environmental Mitigation Database developed Bevelhimer et al. (2015).



Abstract:

This dataset provides a comprehensive inventory of environmental mitigation measures required by Federal Energy Regulatory Commission (FERC) licensed hydropower facilities from 497 licenses that were issued from 1998 to 2025. These licenses constitute 478 of the 1006 FERC projects that were active in September 2026 (when this dataset was last updated). 19,113 mentions of environmental mitigations were identified and categorized in 128 unique categories. Mitigations were identified using a Natural Language Processing (NLP) approach, specifically with a Bidirectional Encoder Representations from Transformer (BERT) model. Model-derived results were then reviewed and updated by a subject matter expert as needed. This dataset introduces important enhancements to previous efforts to inventory environmental mitigations, such as including associated license text for each mitigation, tracking the number of instances a mitigation was identified within a license, and providing improved location information. These enhancements significantly expand the dataset's utility, offering greater analytical capabilities and ensuring reproducibility. The dataset is downloadable as a zip file containing the metadata and dataset files.
In the September 2026 update, mitigations from 37 licenses issued in 2024 and 2025 were appended to this dataset. 


Version history:
Initial release - June 30, 2025: Initial release of dataset
Second release – September 30, 2026: Mitigations from 36 licenses issued in 2024 and 2025 were appended to this dataset. Duplicated rows for P-2230 and P-12715 were removed. Mitigation_categories.csv and the readme file were updated.

Methodology:

In this dataset, 497 Federal Energy Regulatory Commission (FERC) licenses issued from 1998-2025 were analyzed. As of September 2026, 478 of these licensees corresponded to active FERC projects. The other 19 licenses were inactive, with the following project statuses: retired (P-2897, P-10809, and P-10810); revoked (P-10808); surrendered (P-2601, P-2660, P-12737, P-13011, P-13404, P-13405, P-13406, P-13407, P-13408, P-13411, P-14327, P-14677); terminated (P-12478, P-12569, and P-12740). This dataset also contains 291 of the 308 licenses that were analyzed in a previous analysis of environmental mitigations (Bevelhimer et al, 2015), which covered 1998-2013. For the remaining 17 licenses (P-13, P-420, P-460, P-1417, P-1835, P-2004, P-2032, P-2077, P-2416, P-2487, P-2566, P-2588, P-2651, P-11480, P-11910 ,P-12429, P-12455, P-12470, and P-12646), a version with digitized text was unavailable.

To identify environmental mitigation measures, an approach leveraging Natural Language Processing (NLP) approach was developed. The NLP method employed a Bidirectional Encoder Representations from Transformer (BERT) model, chosen for its effectiveness in understanding the context of words in relation to each other within paragraphs. The BERT model was used to analyze each paragraph of the FERC licenses and assign one or more environmental mitigations along with a corresponding confidence score.

A training dataset was created by a subject-matter expert who reviewed FERC licenses from 2014-2017. Sentences and paragraphs that specifically outlined environmental mitigations were identified and marked with their respective mitigation IDs. In total, 1,869 text segments were identified, covering at least one segment for 93 out of 132 mitigation categories. This annotation process ensured that the model would be trained on high-quality, accurately labeled data. Note that the mitigation categories were later revised, as detailed in Mitigation_categories_crosswalk.png. Licenses from 2014-2016 were used for training, while licenses from 2017 were set aside for model validation.

The model was made up of two sub-models. The first was a detection model that identified whether a paragraph contained language indicating required actions, such as phrases like "this license requires," "is required," and "must." The second assigned mitigation categories to each paragraph based on its content. Hyperparameter tuning and performance evaluation were conducted using the 2017 validation set, with metrics such as F1-score used to assess the model's accuracy. This dataset was used to assess licenses from 2014-2023.

To ensure the quality and reliability of the model's outputs, a decision-support tool was developed. This tool presents the model's results to a subject-matter expert, enabling them to quickly perform quality assurance by confirming, rejecting, or editing the model-derived outcomes. The decision-support system integrates an intuitive user interface that displays the model’s predictions alongside the relevant text segments, streamlining the review process.

This process was then repeated for licenses issued from 1998-2013, and 2024-2025. The results from 2014-2023 were used to retrain the model, which included 6893 text segments across 115 mitigations. 

Environmental mitigations across the 1998-2025 period were concatenated into a single file and additional spatial location information was added. 

The result is a reproducible dataset containing an inventory of environmental mitigations, accompanied by the associated license text that provides validation and additional context. This demonstrates an improvement from previous work to inventory environmental mitigations, as it facilitates a standardized approach to documenting mitigation measures across FERC licenses, supporting further research and analysis.


Changes in Mitigation Categories between Bevelhimer et al (2015) and this dataset:

In this dataset, changes were made to the mitigation categories presented in Bevelhimer et al. (2015) to better capture and categorize mitigation measures. These changes are outlined in Mitigation_categories_crosswalk.jpg. Eelway and Adult Fishway were both split into upstream and downstream mitigations (Downstream Eel Passage and Upstream Eel Passage, and Downstream fishway and Upstream fishway, respectively) to better align with their respective Tier 2 categories which indicates the direction of fish passage. Benthic macroinvertebrate monitoring and Macroinvertebrate monitoring were combined into a single mitigation, Macroinvertebrate monitoring, since adequate information was often lacking in the license documents to differentiate between the original two categories. To better capture mitigations that were not collected in Bevelhimer et al (2015), two new mitigations were added: Aquatic threatened and endangered species protection plan and Revegetation Plan. A handful of mitigation categories were removed, as there were not identified in this analysis. These categories had very low numbers of occurrences in Bevelhimer et al (2015) and are likely captured as another mitigation in this dataset (e.g. the fish passage mitigations are likely captures as Downstream fishway or Upstream fishway). Lastly, three mitigations were renamed to improve clarity. 

Applications:

This dataset provides a synthesis of environmental mitigation measures required by FERC for licensed hydropower facilities. Users can explore spatial and temporal trends in mitigation requirements. These trends might inform hydropower developers and managers of measures they may be required to employ as a result of the initial licensing or relicensing process. Researchers and policymakers might use this data to explore the relationship between mitigation measures and various environmental indicators to explore their efficacy. 

Example Use Case:

A user might be interested in the mitigation measures required at a nearby hydropower facility. A project of interest can located on the HydroSource Data Explorer (https://hydrosourcedataexplorer.ornl.gov/). If the project was relicensed between 1998 and 2025 (see ‘FERC License Issuance’ column) it will be in the environmental mitigation database and the FERC docket number (‘FERC License Docket #’) can be recorded. 

From within the Environmental_mitigations_extracted_using_AI_1998_2025.csv file, mitigations can be subset by the FERC docket number (‘FERC_Docket’) and then the mitigations associated with the facility can be explored. Some additional context surrounding the nature of a mitigation can be found in the ‘License_text’ column, which quotes text from the license document. 

If the user wants to explore the mitigation in further detail, the accession number (FERC_accession) can be used to locate the full license from the FERC eLibrary (https://elibrary.ferc.gov/). Use the FERC docket number to locate other documents in the FERC eLibrary that might be associated with the mitigation, such as studies or plans. 


Caveats and Limitations:

The authors recognize that there are some caveats surrounding the use of data provided in this dataset. This dataset included the number of presences of a given mitigation and the associated license text from those presences. This information serves first and foremost as validation of the finding, and secondly to provide additional context to users about the specific requirements of the mitigation. Due to the nature of how mitigations were determined via a paragraph-by-paragraph analysis, the provided instances and licenses text may not encompass all the text within the license associated with the mitigation. Users seeking a complete understanding of a given mitigation should review the FERC license document and other FERC documents that may be associated with a mitigation. 

This dataset uses the FERC docket number as the resolution for analysis since licenses are issued by docket number. A given FERC docket may be associated with multiple hydropower facilities and dams. Different projects, dams, stream reaches, and reservoirs covered under a FERC docket may have unique mitigation measures or mitigation measures that do not apply to them. Thus, users seeking to apply the mitigations to a single feature covered under the FERC docket must look at each on a case-by-case basis rather than a blanket application of all mitigation measures to all project features. Additionally, hydropower projects can be cross listed under multiple FERC docket numbers. This dataset only displays a single FERC docket number for each license that was analyzed (the first docket listed in the license document). FERC docket numbers for a given piece of infrastructure also may change over time through license transfers.

A similar caveat applies to the spatial data presented in this dataset. Most spatial data were obtained from HILARRI (Hansen et al., 2024) and EHA (Johnson et al., 2024) datasets. These datasets are at hydropower dam and facility resolution and contain cases where multiple records are associated with a given FERC docket. The spatial information used in this dataset was collected from the first instance of the FERC docket in those datasets. 

Lastly, the determination of whether a mitigation was present inherently involved a level of nuance and ambiguity in many cases. This stems from both the nature of how mitigation categories were defined and the level of detail available in the license text. For example, a license may mention a requirement to ‘stock fish species’, making it hard to determine if this should be categorized as ‘Stocking fish species of concern: 414103’ or ‘Stocking recreational fish species: 619129’. Contexts such as the location in the document and surrounding information helped, but cases arose where it was not possible to decide with absolute certainty, and a ‘best guess’ was selected. For this reason, it was an important improvement in this dataset to record the license text from areas where mitigations were identified as vouchers and provide users with the ability to arrive at their own conclusions. 


Related Datasets:

A previous analysis of environmental mitigations from 1998-2014:
Bevelhimer, M.S., M.P. Schramm, C.R. DeRolph. 2015. US Hydropower Mitigation Database. HydroSource. Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA. Available at: https://hydrosource.ornl.gov/dataset/us-hydropower-mitigation-database 


References:

Bevelhimer, M.S., M.P. Schramm, C.R. DeRolph. 2015. US Hydropower Mitigation Database. HydroSource. Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA. Available at: https://hydrosource.ornl.gov/dataset/us-hydropower-mitigation-database 

Megan M. Johnson, Shih-Chieh Kao, and Rocio Uria-Martinez. 2024. Existing Hydropower Assets (EHA) Plant Database, 2024. HydroSource. Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA. https://doi.org/10.21951/EHA_FY2024/2344934

Hansen, C. H., Matson, P. G, Bozeman, B. B., Turner, S. W. D. 2024. Hydropower Infrastructure – LAkes, Reservoirs, and RIvers (HILARRI), Version 3. HydroSource. Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA. Available at: https://hydrosource.ornl.gov/data/datasets/hilarri-v3/ 


Dataset contacts:
Thomas Ruggles: rugglesta@ornl.gov
Debjani Singh: debd@ornl.gov  

Acknowledgement:
Funding and support were provided by the U.S. Department of Energy's (DOE) Water Power Technology Office through Oak Ridge National Laboratory, which is managed by UT-Battelle, LLC
