NEPA Decarbonization Technology Analysis: Deliverable 6
Technology-Specific Analyses: Transmission Lines, Geothermal, and Pipelines
Executive Summary
This deliverable examines three technology-specific questions from the NEPATEC 2.0 corpus, each addressing a dimension of how NEPA reviews operate for major decarbonization technology infrastructure types. This report delivers the following:
Technology-specific inquiries
- Transmission lines: Length of lines from project summary correlated with timelines, location, etc.
- Geothermal: Timelines of environmental reviews for different phases of the same project.
- Carbon and hydrogen pipelines (if available): Length of pipelines from project summary correlated with timelines, location, etc.; compare to natural gas pipelines.
Transmission Lines
Key Findings
- 263 electricity transmission projects were identified with extractable line lengths; of these, 141 have both initiation and decision dates and contribute to NEPA duration analyses.
- Most projects involve short line segments: 44% are under 10 miles; the median extracted length is 11.1 miles. 25 projects involve lines of 100 miles or more.
- NEPA reviews are fast overall, with a median duration of 107 days. However, reviews for the longest projects (100+ miles) take a median of 477 days — nearly 4.5× longer.
- Upgrades and replacements dominate electricity transmission activity: 94 projects involve equipment upgrades or reconductoring, while only 12 are classified as purely new-build construction.
- Geography matters: Western states have both the highest project counts and the longest median line lengths. Multi-state projects (33 in total) are concentrated in the West and South.
Transmission Line Identification
Transmission projects are identified through a four-gate classification applied to all decarbonization technology projects in NEPATEC 2.0. A project must pass all four gates to enter the analysis:
Electricity Transmission type tag — the NEPATEC
project_typefield must includeElectricity Transmission. This is the primary narrowing step: of 20,725 decarbonization technology projects, 7,697 carry this type tag.Build-related text — the project title and description must match explicit construction language: phrases such as “new transmission line,” “construct/install/rebuild [kV] transmission line,” “double-circuit transmission line,” or “right-of-way … transmission line.” ROW renewals, fiber-optic installs, substation transactions, and DOE research grants that carry the transmission type tag but lack construction language are excluded here. This step reduces the candidate pool to 1,741 projects.
Not maintenance-flagged — projects whose titles contain explicit maintenance language (vegetation management, herbicide treatment, line inspection, and related terms) are excluded before length extraction. This removes 322 projects, leaving 1,419 candidates.
Extracted line length ≥ 1 mile — a numeric mileage value must be recoverable from the project text. Length is first extracted from the project title and description metadata; if absent, the NEPA document page text is searched (up to 10 pages of main documents for EA/EIS; all pages for CE). Of the 1,419 non-maintenance candidates, 151 projects are recovered from metadata and an additional 136 from document pages, yielding a final set of 263 projects.
Figure 1 shows the project count at each filtering stage.
Length figures come from project_transmission_length_final. For projects with two or more plausible candidate values, an LLM adjudication step selects the most likely total line length; all others use rule-based selection. The rule-based selector applies a priority cascade — preferring explicit total-length language (“X miles long,” “X miles in length”), build-verb context, and highest hint-score — before falling back to the largest candidate value.
Final Transmission Line Sample
The 263 projects that pass all four gates form the analysis universe for length-based findings. However, NEPA duration analyses require both an initiation date and a decision date to be extractable — only 141 of the 263 projects meet this requirement. All duration statistics in this report (median review times, length–duration relationships, regional and action-type comparisons) are based on this smaller sample of 141 projects. Length-only findings (distributions, geographic patterns, action-type shares) use the full 263.
Figure 2 shows the breakdown of the 263 projects by date availability.
Length Distribution
Figure 3 shows the full distribution of extracted transmission line lengths. The distribution is heavily right-skewed: the majority of projects involve short line segments of 25 miles or less, while a small number of large-scale projects extend well beyond 100 miles. The dashed line marks the median.
Length vs. NEPA Duration
Figure 4 examines how project count and median NEPA duration vary across length bands. The left panel shows that most projects are short-segment (<10 miles), while the right panel reveals that review duration rises with line length — projects in the 50–100 mile band take a median 133 days, and 100+ mile projects take 221 days, compared to just 25–41 days for projects under 50 miles.
| Length band | Projects | Median length (mi) | Median duration (days) | P90 duration (days) |
|---|---|---|---|---|
| <10 mi | 113 | 3.2 | 82.0 | 684.0 |
| 10–50 mi | 112 | 20.0 | 110.0 | 1,146.8 |
| 50–100 mi | 18 | 65.0 | 292.5 | 746.7 |
| 100+ mi | 19 | 220.0 | 704.5 | 2,480.0 |
Figure 5 plots each project’s extracted line length against its NEPA duration. The dark blue trend line shows the overall linear relationship. There is a positive association — longer lines tend to have longer reviews — but the correlation is modest, and many short projects still take substantial time.
Geographic Patterns
Figure 6 and Figure 7 decompose the transmission sample across states. The first figure shows which states have the most projects; the second shows which states have the longest median line lengths. The two pictures are related but distinct — states with many projects are not necessarily those with the longest lines.
Project Action Type
The project_transmission_action field classifies each project’s primary activity using keyword patterns applied to the project title and description. Five categories appear in the figures:
- New Build: explicit construction language — e.g., “construction of a new 230 kV transmission line” or “new substation.”
- Upgrade: reconductoring, uprating, or capacity expansion of an existing line — e.g., “reconductor existing 69 kV line” or “voltage uprate to 138 kV.”
- Acquisition: land or right-of-way purchase without construction — e.g., “right-of-way acquisition” or “land purchase for corridor.”
- Mixed: the document matched keywords for more than one action type. These are typically multi-component projects involving both new infrastructure and modifications to existing facilities.
- Unknown / None: no action-type keyword was detected. The project passed the transmission build-text gate but its description lacks sufficient language to assign a specific activity.
Figure 8 shows project counts by action type. Figure 9 and Figure 10 show how line length and NEPA duration vary across categories; both figures top-code at 300 mi and 1,200 days respectively so that the box shapes are visible — a small number of outliers beyond those thresholds are hidden but included in the underlying statistics.
Upgrades are the most common classified activity, reflecting the active reconductoring and maintenance of the existing grid. Mixed projects are also frequent, suggesting many NEPA actions cover both new construction and modifications to existing facilities. New builds are relatively rare. The Unknown and None categories together account for a substantial share, consistent with many short project descriptions in the dataset that do not include explicit build-type language.
Geothermal Energy
Key Findings
- Geothermal NEPA actions in the dataset span exploration, drilling, plant construction, operations, and multi-phase projects; phase is reported at the project level by collapsing multiple NEPA filings for the same physical development into a single phase label.
- Drilling is the most common identifiable phase, followed by multi-phase projects where a single project spans multiple distinct development stages across its NEPA history.
- Phase classification uses a two-stage pipeline: regex pattern matching on project text for the majority of projects, followed by a fine-tuned SciBERT ML classifier for actions where no regex cues are detected (median classifier confidence ~0.90).
- Drilling reviews are often resolved very quickly (frequently categorical exclusions), while exploration and multi-phase projects show longer median durations.
- Within-project sequencing shows that some geothermal developments span multi-year permitting sequences, with individual NEPA actions separated by years as projects progress from exploration wells to full plant construction.
Geothermal Project Identification
Geothermal projects are identified using a keyword match on the NEPATEC project_type field — analogous to the Electricity Transmission type-tag gate used for transmission. The pattern geothermal | enhanced geothermal | EGS is matched against project_type; projects where this tag is absent are not included even if geothermal language appears in the title or description. This produces a precise sample of NEPA actions explicitly categorized as geothermal by NEPATEC.
Figure 11 shows the two-stage identification path.
Phase Classification Method
Geothermal projects are unique in that a single physical development typically triggers multiple sequential NEPA actions — one for exploration drilling, one or more for production well programs, and a final review for the power plant. This analysis groups geothermal projects by phase and examines timeline patterns by phase.
Phase classification runs in two stages.
Stage 1 — Regex classifier. Each NEPA action is scanned against five phase-specific keyword sets applied to concatenated project text (title + description + project type + NOI title + document titles). Assignment logic:
- If one phase set matches → that phase is assigned.
- If two or more phase sets match → the action is classified as
multi_phase. - If no phase set matches → the action is classified as
unknown.
Stage 2 — ML classifier. Actions classified as unknown by the regex stage are reclassified using a fine-tuned allenai/scibert_scivocab_uncased model. The model is trained on the regex-labeled rows with class-weighted loss (to correct for the drilling-heavy imbalance) and a self-training step that expands the effective training set using high-confidence pseudo-labels (threshold 0.70). Input text is title + project type + first 100 words of description, supplemented by up to 300 words of page-level document text where available. At median, the classifier assigns phase labels with approximately 90% softmax confidence.
Project-level rollup. After per-action phase assignment, results are collapsed to the project level. A normalized title key is constructed by stripping common geothermal vocabulary (“geothermal”, “well”, “drilling”, “plant”, “project”, etc.) from the project title; actions whose normalized titles match are treated as filings for the same physical development. The 873 NEPA actions collapse to 753 inferred projects — a reduction of 120 rows driven by well fields and multi-well programs that generated several sequential CEs or EAs over time. Of these, 753 projects have a resolved phase (shown in the bar chart below); the remaining 0 could not be classified and are excluded from phase figures.
The table below illustrates the merge with two simplified examples.
| NEPA action (row in data) | Process | Normalized key | Action phase | → Project phase |
|---|---|---|---|---|
| Blue Mtn Geothermal Exploration Well Drilling Program | CE | blue mtn | Exploration | Multi-phase |
| Blue Mtn Geothermal Production Well Program | CE | blue mtn | Drilling | ↑ (merged) |
| — | — | — | — | — |
| Ormat Nevada Geothermal Production Well Program Phase I | CE | ormat nevada | Drilling | Drilling |
| Ormat Nevada Geothermal Production Well Program Phase II | CE | ormat nevada | Drilling | ↑ (merged) |
A project is assigned a single phase if all its actions agree; if actions span two or more distinct phases the project is classified as multi_phase. 319 projects are classified as multi-phase. git Table 3 shows representative text cues for each phase.
| Phase | Sample cues |
|---|---|
| Exploration | exploration, exploratory, resource assessment, seismic survey, temperature gradient, feasibility study, slim hole, core hole |
| Drilling | drilling, production well, injection well, well pad, wellfield, well program, well stimulation, wellhead, geothermal well |
| Plant | power plant, binary plant, flash plant, turbine, generating station, power generation, substation, interconnection, steam gathering |
| Operations | steam supply, reinjection, make-up well, makeup well, fluid management, working fluid, geothermal resource utilization |
| Multi-phase | (two or more phase cue sets detected in the same document) |
Multi-Phase Combinations
Figure 13 shows phase combinations for 241 geothermal projects that span multiple sequential development stages. Each bar shows the count of projects with exactly that combination; filled dots indicate which phases are present. Combinations like Exploration + Drilling reflect the typical progression from resource assessment to well development, while three- and four-phase bars represent projects that advanced through plant construction and operations.
The remaining 78 multi-phase projects (out of 319 total) are excluded. These were classified as multi-phase by the Stage 2 ML classifier: the Stage 1 regex found no strong phase-specific vocabulary match (returning unknown), and the ML model inferred from the overall document text that the project spans multiple development stages — but as a single-label classifier, it cannot identify which phases are present. Without that information, these projects cannot be placed in a specific combination bar. The 241 shown represent projects where phase assignments were confirmed by the regex pipeline and can be unambiguously attributed to specific development stages.
NEPA Duration by Phase
Figure 14 shows the distribution of NEPA review duration for each geothermal phase among projects with calculable timelines. The exploration phase is excluded from the main plot due to its small, spread-out sample; its distribution is summarized in the figure footnote. Drilling-phase projects tend to be reviewed very quickly — consistent with the widespread use of categorical exclusions for well drilling on established geothermal fields. Multi-phase projects, which span multiple development stages, show the widest interquartile ranges.
Within-Project Sequencing
Figure 15 visualizes the temporal sequencing of NEPA actions within individual geothermal development projects. Each horizontal segment represents one NEPA review (initiation to decision date); rows are grouped by inferred project identity (normalized title key). Only a random sample of projects with at least two dated actions spanning at least two distinct development phases is shown, making the phase progression visible — exploration wells, production well programs, and plant construction can each require separate reviews separated by years.
Note
Data requirement: This figure requires both an initiation date and a decision date for each NEPA action. Many geothermal actions in NEPATEC 2.0 are missing one or both timeline dates, which limits the number of projects that can be shown. If the figure contains very few or no projects, the BERT timeline extraction coverage for this subset likely needs improvement before this analysis is meaningful.
Carbon and Hydrogen Pipelines
Key Findings
- 5,324 pipeline NEPA projects are identified across six technology groups: carbon capture, hydrogen, natural gas, oil/petroleum, water/irrigation, and other. Oil/petroleum and natural gas dominate by count; carbon (4 + 183) and hydrogen are smaller but policy-relevant.
- A new-build filter isolates 899 likely construction reviews (17% of all pipeline-tagged projects). Carbon and hydrogen are exempted from the build-text requirement because their entire infrastructure base is nascent; natural gas and oil/petroleum require explicit construction language.
- Duration comparisons use the new-build subset to avoid conflating genuine construction reviews with routine operational filings. Sample sizes for carbon and hydrogen remain small and results should be treated as descriptive.
- Length data is moderately sparse across groups; pipeline length is extracted from metadata only (no document page search, no LLM adjudication), unlike transmission.
Pipeline Identification
Pipeline projects are identified using the controlled-vocabulary project_type field: a project receives the pipeline tag (project_is_pipeline) if its project_type contains Pipeline. This mirrors the geothermal identification approach — type-tag-only, with no free-text keyword sweep of project titles or descriptions. The result is a universe of 5,324 pipeline-tagged NEPA projects across all technology groups.
A new-build filter further narrows the pipeline universe for duration analyses. Because the broad keyword gate captures the full lifecycle of pipeline-related NEPA activity — new construction, expansions, operational modifications, safety certifications, right-of-way renewals, and administrative approvals — mixing all of these into duration comparisons would conflate genuinely new infrastructure reviews with routine regulatory filings. A two-gate new-build filter (project_is_pipeline_new_build) is applied to isolate likely construction reviews for that purpose:
- Build-text gate (applied to project title + description): the project contains explicit construction language — e.g., new pipeline, construct/build/install/lay pipeline, pipeline project/route/corridor/expansion/segment/lateral.
- Maintenance exclusion (applied to title only): the project title does not contain maintenance language — e.g., pipeline inspection, cathodic protection, pigging, pipeline repair/maintenance, routine inspection, leak detection.
No length threshold is applied (unlike transmission’s ≥ 1 mile gate) because pipeline length coverage is substantially sparser; requiring a length would discard many genuine new-build projects that lack an extractable length value.
Carbon and hydrogen pipelines are exempted from the build-text gate. These represent nascent infrastructure: nearly all of their NEPA reviews are for genuinely new projects (CCS corridors, CCUS facilities, liquefaction terminals, hydrogen conveyance systems) rather than operational or maintenance filings. Applying the build-text gate to them would incorrectly exclude large integrated energy projects whose NEPA title describes the broader facility rather than the pipeline component specifically. The maintenance exclusion still applies. As a result, 98% of carbon and hydrogen pipeline projects (183 of 187) pass the new-build filter. For natural gas and oil/petroleum pipelines — which have large existing infrastructure bases generating routine operational NEPA filings — the build-text gate is required and only 17% of all pipeline projects (899 of 5,324) are classified as new-build.
Note
The new-build filter is applied specifically to duration analyses (length vs. duration scatter, length-band duration violin). All other figures — length distributions, technology group breakdowns, and the pipeline identification funnel — use the full 5,324 pipeline-tagged universe.
Figure 16 shows the identification path from the full NEPATEC 2.0 corpus through the pipeline tag, the new-build filter, and finally the decarbonization technology subset.
Pipeline Length
Pipeline length is extracted from the same metadata fields used for tagging (project title, description, and project type). Unlike transmission length extraction — which searches NEPA document pages as a fallback when metadata lacks a length value — pipeline length is derived from metadata alone. Pattern-based rules identify numeric mileage candidates in the text; when multiple plausible values are present, a priority cascade selects the best one, preferring explicit phrasing such as “X miles long” or “X-mile pipeline” over incidental numeric mentions.
There is no LLM adjudication step for pipelines. Transmission length adjudication uses a Claude-based model to resolve ambiguous multi-candidate cases (e.g., distinguishing a segment length from a total corridor length); no equivalent step was run for pipelines. This, combined with the metadata-only search scope, means pipeline length coverage is substantially sparser than for transmission.
Figure 17 shows what fraction of the 5,324 pipeline-tagged projects have an extractable length value.
Of the 3,014 projects with an extractable length, a further subset also have calculable NEPA durations (both an initiation and a decision date); length–duration analyses draw only on this smaller intersection.
Technology Groups and Decarbonization Status
The 5,324 pipeline NEPA projects span six technology groups:
- Carbon pipeline: pipeline tag + keyword match for carbon, CO₂, carbon dioxide, CCS, carbon capture, or carbon sequestration.
- Hydrogen pipeline: pipeline tag + keyword match for hydrogen.
- Natural gas pipeline: pipeline tag + keyword match for natural gas, gas pipeline, gas gathering, or gas line.
- Oil/petroleum pipeline: pipeline tag + controlled-vocabulary
project_typeincludes “Oil & Gas,” “Oil and Gas,” or “Petroleum.” - Water/irrigation pipeline: pipeline tag + controlled-vocabulary
project_typeincludes “Water Resources” or “Irrigation” (and not Oil & Gas). - Other pipeline: all remaining projects carrying the pipeline tag but unmatched by any of the above classifiers.
For this analysis, carbon and hydrogen pipelines are treated as decarbonization technologies — overriding the NEPATEC energy type tag, which classifies both as “Fossil fuel” in the source data. This reflects that CCS infrastructure and low-carbon hydrogen are policy-relevant decarbonization technology investments regardless of their association with fossil fuel systems. Natural gas and oil/petroleum pipelines are classified as fossil fuel; water/irrigation and other pipelines are classified as Other.
Figure 18 shows the project counts by group, colored by this classification.
The “Other pipeline” category (313 projects) captures actions where the pipeline tag fired on incidental or non-specific text — projects where pipeline appears in the title or description but none of the substance-specific classifiers apply. This group includes: propane and LPG distribution lines; steam distribution networks (district heating systems); ammonia pipelines for industrial or agricultural use; produced-water disposal lines associated with oil and gas operations (which often lack explicit natural gas keywords); generic pipeline right-of-way actions where no commodity is named; slurry and multi-product pipelines; and projects where pipeline is used in a procedural sense (e.g., references to a regulatory or permitting pipeline process rather than physical infrastructure). Some of these may be fossil fuel–related but fall outside the current classification vocabulary.
Pipeline Length Distribution
Figure 19 shows the overall distribution of extracted pipeline lengths across all groups combined, topcoded at 50 miles. The distribution is heavily right-skewed: the great majority of projects with a recoverable length are short — under 5 miles — with a long upper tail of interstate corridors and large-scale transmission systems. The dashed line marks the overall median.
Figure 20 breaks the same data out by technology group (violin + box + jitter, topcoded at 25 miles). Carbon capture pipelines have the widest spread relative to their count — their interquartile range extends further than natural gas gathering lines, reflecting that CCS infrastructure ranges from short connector lines at industrial facilities to long-distance CO₂ transport corridors. Natural gas pipelines concentrate heavily near zero (most are short gathering and flowlines for well sites) but have a thick upper tail. Oil/petroleum pipelines show a similarly short median but a slightly tighter distribution. Hydrogen pipelines are too few to draw distributional conclusions. Water/irrigation pipelines skew short — most serve single field or ranch systems — but a handful extend into the tens of miles.
Pipeline Length vs. NEPA Duration
Duration analyses in this section use the new-build subset (899 projects) rather than the full pipeline-tagged universe. This avoids conflating construction reviews with operational/maintenance filings in the duration comparisons.
Figure 21 examines how project count and NEPA duration vary across length bands. The left panel uses all pipeline-tagged projects with an extractable length (broad universe); the right panel is limited to new-build projects with both length and duration data. Most pipeline projects with an extractable length fall in the sub-5-mile range; the duration panel suggests modest positive association with length, though small sample sizes in the upper bins limit firm conclusions.
Figure 22 plots pipeline length against NEPA duration for new-build projects, with separate trend lines by group. The relationship is noisy across all groups. Note that hydrogen pipelines are absent from this plot (no projects with both length and duration data in the new-build subset), and carbon pipeline coverage is minimal. The natural gas trend line, based on the most projects, provides the most informative reference.
Timeline Duration by Pipeline Type
Figure 23 compares NEPA review durations for decarbonization technology pipeline projects with calculable timelines (n ≈ 228), topcoded at 500 days. Carbon pipelines show the widest distribution and highest median, consistent with the greater regulatory complexity and novelty of CCS infrastructure — though the very small sample prevents firm conclusions. Natural gas pipelines provide the most stable baseline. Groups with very few projects will show minimal violin density; the n labels indicate sample size for each group.
Caveats and Limitations
Important Considerations
Transmission:
- Length extraction: Lengths are extracted by regex from project descriptions and adjudicated by LLM for ambiguous cases. Some lengths may reflect segments, rights-of-way widths, or other non-linear measures misidentified as line lengths.
- Classification: Projects are only counted if they have both a type tag and explicit build-related language plus a non-trivial length. This likely undercounts purely administrative or renewal-only actions.
- Action type completeness: ~15% of transmission projects could not be assigned an action type; these are excluded from action-type figures.
- Short median duration: The 38.5-day median reflects that many short-segment projects are processed as categorical exclusions. This does not reflect the full range of transmission permitting complexity.
Geothermal:
- Phase detection: Phase classification relies on regex pattern matching against project title and description text and will miss projects that use non-standard terminology. Projects with no text cues are assigned a phase based on process type (CE → drilling, EA/EIS → exploration) as a weak prior rather than direct evidence. Multi-phase classification occurs whenever two or more phase cue sets co-fire in the same document, which can overcount genuinely multi-phase projects if broad terms appear incidentally.
- Project key matching: Within-project grouping uses normalized title keys, which may merge unrelated projects with similar names or split a single project whose titles vary across actions.
Pipelines:
- New-build filter recall: The build-text gate has imperfect recall for natural gas and oil/petroleum pipelines — projects that describe construction in terms not covered by the regex (e.g., “leg-off”, “tie-in”, or project-level titles without explicit pipeline construction language) may be incorrectly excluded. The filter is best understood as a conservative lower bound on the new-build population for those groups.
- Carbon/hydrogen exemption: Carbon and hydrogen pipelines are exempted from the build-text gate on the basis that their infrastructure is nascent. If future data include operational or maintenance reviews for these technologies, the exemption should be revisited.
- Small duration sample: With only 183 carbon and hydrogen new-build projects that also have timeline data, all duration comparisons are highly sensitive to individual project characteristics and should be treated as exploratory.
- Length coverage: Most pipeline projects lack extractable length figures (metadata-only extraction, no document page search, no LLM adjudication), making length-based comparisons substantially sparser than for transmission.
Report generated 2026-03-18 | NEPA Decarbonization Technology Analysis