Ontology and LLM-based Extraction of Wastewater Unit Process Data from Permit Documents for Improved Sector-Level Cost and Emissions Analysis

Wastewater treatment plant (WWTP) process configuration data help researchers and practitioners understand infrastructure investment needs, estimate environmental impacts, and monitor technology deployment. The U.S. Environmental Protection Agency’s Clean Watersheds Needs Survey (CWNS) provides an inventory of installed unit processes, but it relies on voluntary reporting with well-documented data gaps and inconsistencies. This study develops and validates a large language model (LLM)-based method for extracting the same unit process data from a new source: the National Pollutant Discharge Elimination System (NPDES) permits for California’s municipal WWTPs. We begin by manually constructing and verifying a reference dataset of installed processes at 15 California facilities. We establish baseline accuracy metrics for both the CWNS and manually extracted process data from NPDES permit text, finding that the NPDES permit text has a lower F 1 error on identified unit processes (13% vs. 50%). We then compare the accuracy of three distinct methods for accelerating the extraction of unit process data from NPDES permits: keyword-based extraction, LLM-based extraction to a predefined unit process list, and LLM-based extraction using the Water Treatment ontology. We find that the ontology-based LLM approach reliably returns a structured output and achieves a unit process F 1 score of 0.89 (11% F 1 error) using GPT-5 mini. We apply this method to compile a dataset of unit processes for 609 California facilities with active permits and demonstrate the utility of that dataset in estimating facility-level infrastructure investment needs and greenhouse gas emissions. We estimate that infrastructure investment needs rose 6% from 2021 to 2026. We also find that improved resolution of nutrient removal processes at CA WWTPs yields N 2 O emissions estimates that are 81% higher than estimates from CWNS. We envision ontology-assisted, LLM-based extraction of permit documents as a tractable approach for enhancing high-reliability data availability across the wastewater sector.

Authors

Institutions

Publication Details

Journal
Environmental Engineering Science
Published
2026-11-14
DOI
https://doi.org/10.1177/15579018261504213
Primary Topic
Wastewater Treatment and Reuse
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Ontology and LLM-based Extraction of Wastewater Unit Process Data from Permit Documents for Improved Sector-Level Cost and Emissions Analysis

Fletcher Chapin, Meagan S. Mauter, Daly Wettermark, Constance Rouffet et al.
Environmental Engineering Science
Wastewater Treatment and Reuse
article

Ontology and LLM-based Extraction of Wastewater Unit Process Data from Permit Documents for Improved Sector-Level Cost and Emissions Analysis

Fletcher Chapin, Meagan S. Mauter, Daly Wettermark, Constance Rouffet, Ashley C. Ramirez
article en

Abstract

Wastewater treatment plant (WWTP) process configuration data help researchers and practitioners understand infrastructure investment needs, estimate environmental impacts, and monitor technology deployment. The U.S. Environmental Protection Agency’s Clean Watersheds Needs Survey (CWNS) provides an inventory of installed unit processes, but it relies on voluntary reporting with well-documented data gaps and inconsistencies. This study develops and validates a large language model (LLM)-based method for extracting the same unit process data from a new source: the National Pollutant Discharge Elimination System (NPDES) permits for California’s municipal WWTPs. We begin by manually constructing and verifying a reference dataset of installed processes at 15 California facilities. We establish baseline accuracy metrics for both the CWNS and manually extracted process data from NPDES permit text, finding that the NPDES permit text has a lower F 1 error on identified unit processes (13% vs. 50%). We then compare the accuracy of three distinct methods for accelerating the extraction of unit process data from NPDES permits: keyword-based extraction, LLM-based extraction to a predefined unit process list, and LLM-based extraction using the Water Treatment ontology. We find that the ontology-based LLM approach reliably returns a structured output and achieves a unit process F 1 score of 0.89 (11% F 1 error) using GPT-5 mini. We apply this method to compile a dataset of unit processes for 609 California facilities with active permits and demonstrate the utility of that dataset in estimating facility-level infrastructure investment needs and greenhouse gas emissions. We estimate that infrastructure investment needs rose 6% from 2021 to 2026. We also find that improved resolution of nutrient removal processes at CA WWTPs yields N 2 O emissions estimates that are 81% higher than estimates from CWNS. We envision ontology-assisted, LLM-based extraction of permit documents as a tractable approach for enhancing high-reliability data availability across the wastewater sector.

Environmental Engineering Science
Cerritos College (US), Cardiovascular Institute of the South (US), SLAC National Accelerator Laboratory (US), Stanford University (US)
Wastewater Treatment and Reuse
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon