Prepared 5 September 2026. This is a proposed research and modelling programme, not a regulatory submission or a finding about present air quality. Sources were checked against the documents themselves. Page references below are printed page numbers unless stated otherwise.
The next useful improvement is to replace assumed winds, source height and mixing rates with independently checked inputs in an established modelling system. Adding detail to the current browser solver would not, by itself, increase confidence in its concentration maps. The most defensible public product will distinguish an engineering scenario, an actual-event reconstruction and a forecast, because each needs different evidence.
The existing terrain model demonstrates conservation of a released tracer and lets people explore imposed inflow, outflow and restricted vertical mixing. It does not calculate the thermal processes that create those regimes. Its grid-refinement check did not converge: changing from 48 to 64 columns changed the maximum near-surface concentration by approximately +54% and the value at the Squamish marker by approximately −45% for the examined case. A wind field satisfying continuity is not necessarily a realistic wind field.
Consequently, retain this version as an educational display and input/output prototype. Do not tune its diffusivity, artificial inversion restriction or effective release height until its map resembles an expected plume, and then call that calibration. The test results and limits remain in the version 2 method note.
The earlier isolated AERMOD 26135 exercise reproduced EPA’s Martin’s Creek sample using supplied preprocessor outputs. The subsequent complete reference chain built and ran AERMAP 24142 and AERMET 26135 on the sample’s raw terrain and weather, compared their generated outputs with EPA’s references, then passed those new files to AERMOD. All 1,559,741 printed numeric tokens across the supplied preprocessing and dispersion reference files matched, with normalized text also matching. Original model source and sample inputs remain preserved. Preprocessing checks, complete-chain AERMOD comparison.
This establishes working software for that official input pathway. No Woodfibre source or weather was used, and the result does not validate local concentrations or demonstrate that AERMOD resolves changing fjord flow. Canadian terrain/current weather conversion, surface-characteristic preparation and CALMET/CALPUFF remain unverified. Living readiness and scope.
A separate AERSCREEN 21112 / MAKEMET 16216 flat-terrain flare example now reproduces the official reference with AERMOD 26135: 7,095 printed numeric fields and normalized text match. This checks a sustained-source screening pathway, not changing fjord winds or a finite flare event. The terrain, downwash, NO₂-conversion and fumigation branches were not exercised. Flare-screening verification
That verified pathway has now run 36 historical, normalized sensitivities, three search-distance extensions and one fixed-heat mass-rate scaling check. The source geometry and explicit 1.5 m flat-ground receptors were checked in all 40 generated inputs. The three selected 40 km runs retained their 20 km peaks at printed precision. All 13 executions for the historical CDGF source carry AERMOD's equivalent-diameter warning; the remaining 27 have no model warnings. Neither status establishes physical applicability. The separate AERSCREEN rise diagnostics reach implausibly large heights and must not be used as trajectories, effective-height defaults or proof that emissions avoid the community. Further review of highly buoyant plume and inversion-penetration treatment is required. Historical screening results, independent applicability review
The expanded hot-temperature sensitivity is traceable to ECCC's reported 43°C on 28 June 2021 at Squamish Airport. Both the original 250–310 K and expanded 250–320 K input sets are retained because MAKEMET samples the midpoint as well as endpoints; changing the interval also changes its sampled weather. Their union is not a continuous bound, and neither temperature set is a recovered Woodfibre event. Temperature evidence
Supported MODEL/METEOR diagnostic output now reproduces the selected peak from each historical source, using the exact original synthetic record and unchanged AERMOD binary. The 906 receptor calculations pass component-sum and coherent/meander-blend checks at printed precision. At the three selected peaks, more than 99.99% of the modelled ground contribution comes from the penetrated component, despite different mass penetration fractions. This identifies a controlling formulation issue; it does not validate that treatment. The separate component-height parameters also differ from AERSCREEN's final-rise report. A failed STARTEND attempt showed why nonchronological matrix labels cannot be treated as event dates. The original attempt is retained and no HBP option was enabled. Component evidence and next applicability question
A chronological Baldwin 1982–1983 field benchmark now compares default AERMOD 26135, an ALPHA control and experimental HBP using one common AERMET 26135 reconstruction. The old processed weather was incompatible with the current reader; the original failure and a frozen preprocessing amendment are retained. Default and ALPHA agree exactly at printed precision. Across 86,024 common hourly pairs, HBP reduces MAE from 14.506 to 12.737 µg/m³ and RMSE from 71.427 to 55.451 µg/m³, while its mean falls further below the processed observed mean (7.178 versus 10.215 µg/m³; default 9.333). Both treatments retain a substantial stable-period shortfall. Separate AI-assisted audits verify raw data, chronology, diagnostics and 3,035 scoring fields. This is one previously used tall-stack dataset, not flare validation, a qualified independent review or an exact reproduction of EPA’s evaluation. HBP remains experimental and no community contours change. Field results, figure and reproducibility package
A separate first test compared one actual 2 September 2026 HRDPS forecast with 48 local station records. It exposed a large Pam Rocks direction error and a roughly 559 m source-proxy/model-terrain mismatch. That forecast cannot simply replace the browser’s assumed winds as validated source-site forcing. Retain this episode when testing proposed weather improvements and expand to independent regimes and seasons. Forecast comparison and processing checks
A matched 5 September 2026 test compared operational 2.5 km and experimental West 1 km forecasts initialized at 00:00 UTC against the same six hours at each of two stations. The 1 km product placed its nearest Woodfibre-proxy grid point closer to the shoreline elevation, but still about 126 m too high. Wind errors were mixed: worse at Squamish Airport and somewhat smaller at Pam Rocks, where both forecasts substantially underestimated the strong observed winds. This six-hour diagnostic cannot rank general forecast skill or establish source-site forcing; the models also differ in configuration, so the change cannot be attributed solely to grid spacing. The frozen 2 September test remains separate. Matched resolution findings and limitations.
Archived HRRR forecasts for two preselected 2025 episodes add a winter and summer stress test: six forecast hours at two stations per case. Winter Pam Rocks winds temporarily reverse in the model while observations stay northerly. The smaller source-area terrain mismatch in HRRR does not resolve that wind failure, and the pressure-level samples do not validate an inversion. Do not pool different-date HRRR/HRDPS scores into a model ranking. Archived weather findings
The July 2026 B.C. dispersion guideline selects models by flow complexity, distance and assessment purpose. A straight-line model can still answer some maximum-concentration questions; curved transport to receptors around a valley bend supports CALPUFF when representative meteorology is available. The guide recommends at least three years where public concern is significant, with 90% completeness by month and required meteorological variable. Its flare treatment links pseudo-source parameters to gas heat release and distinguishes transient emission profiles; a fixed arbitrary release height is insufficient. These are guideline recommendations, not independently established permit conditions. B.C. guideline, §§1.1, 2.3, 5.4–5.5, 10.1, pp. 1, 10–11, 41–42, 82–84.
The guidance also calls for examination of modelled winds at several heights and in different seasons, and prohibits applying an arbitrary correction factor to concentration output to match observations. That makes input checking and independent evaluation central to this project. B.C. guideline, §§9.1–9.2, pp. 78–81.
The separately published July 2022 NO₂ guidance, still the document retrieved from B.C.’s current official URL, requires presenting a total-conversion result in model reports. ARM2, OLM and PVMRM are options chosen for the application and supporting data, not a ladder guaranteeing better accuracy. OLM/PVMRM need a justified source NO₂:NOx ratio and ozone input; background NO₂ is added after conversion. Where corresponding historical emissions and ambient measurements exist, Level 3 applicants are encouraged to evaluate model performance. The document describes both distribution-based comparisons and time-matched comparisons; these assess different kinds of skill. B.C. NO₂ guidance, §§3.2–3.3 and 4.2–4.3, pp. 8–21.
The following choices are recommendations for this project, subject to a qualified dispersion modeller’s documented assessment and the relevant agency’s modelling-plan review if the results are to support a regulatory decision.
| Product | Proposed model and input approach | What it can establish |
|---|---|---|
| Source screening | AERSCREEN with verified flare geometry, gas composition, heat and event segments; inspect an AERMOD comparison where appropriate | A check on possible maxima and unusual refined-model results. It cannot describe a plume travelling around Howe Sound’s bends. |
| Community-scale engineering assessment | A CALMET/CALPUFF candidate configuration, with representative numerical weather output and local observations where available, tested against independent local weather | Concentrations conditional on specified operating cases and meteorological years, including changing trajectories. This is the leading candidate to evaluate, not a declaration that CALPUFF is always superior. |
| Recorded flare reconstruction | The evaluated configuration driven by the same event’s time-resolved source and weather records | Whether observed plume timing, direction and concentrations are consistent with the release. Source and weather timestamps must actually overlap. |
| Demonstrated flow failure | Compare a better-resolved prognostic weather configuration or another independently evaluated transport approach | A targeted remedy if the initial configuration misses observed reversals, vertical decoupling or terrain blocking. More resolution is useful only if it improves withheld observations. |
| Future operational forecast | A separately evaluated forecast weather feed, current source-state information and forecast-error ensemble | Probabilistic future conditions only after retrospective forecast testing; historical scenario frequencies alone are insufficient. |
A current access check found that the standard CALPUFF Version 7 installation package requires a provider request, with licence terms affecting use, source modifications and redistribution. No package has been obtained or terms accepted. An older publicly listed demonstration would verify only its own version. The access-readiness record and unsent provider inquiry make this dependency explicit. CALWRF is a WRF netCDF converter; a direct HRDPS-to-CALMET pathway has not been established.
Keep the public map at approximately 20 × 20 km if that is the useful community view. Determine the meteorological and transport domains independently: upstream conditions, fjord entrances, plume departure and possible return can require a larger domain. Demonstrate adequate extent by expanding it and checking the results. A coastline cutting through coarse weather cells must be inspected rather than assumed accurate because a finer terrain raster exists underneath it.
Distinguish terrain resolution, weather resolution and receptor spacing. A 20 m receptor grid samples a meteorological field; it does not make the field resolve 20 m flow. Place explicit receptors at publicly accessible shoreline locations, residential areas, schools and care facilities, and use measured elevations. Define the receptor height and model quantity consistently. A modelled concentration at a specified receptor height is different from the prototype’s average over a bottom layer tens or hundreds of metres deep.
LNG Canada, October 2025 — an operating-flare assessment. The assessment used CALPUFF with five years of 2011–2015, 1 km WRF meteorology; its domains were 55 × 110 km and a nested 35 × 35 km. It updated NO₂ and SO₂, while PM₂.₅ and CO were not remodelled in that exercise. NO₂ used OLM. A separate review examined September 2024–August 2025 ambient data at three stations. LNG Canada Flaring Assessment, 27 October 2025, §§1 and 6–7, pp. 2, 10–37.
Transfer the documented source-update workflow, multi-source context and monitor comparisons. Do not transfer its emissions, backgrounds, objective table or air-quality conclusions to Woodfibre. In particular, historical 2011–2015 model weather and 2024–2025 monitoring do not form an event-by-event hindcast validation. Our proposed reconstruction should explicitly close that time mismatch.
Ksi Lisims, June 2024 — a coastal LNG assessment. Its documented chain used CALMET 6.5.0, CALPUFF 7.2.1 and CALPOST 7.0.0. It used 2011–2015 WRF at 4 km, a 50 × 50 km CALMET domain at 250 m, twelve vertical layers, and nested receptors. Emergency blowdown and thermal-oxidizer-trip flaring were separate cases. Local meteorological observations were not assimilated; WRF was checked against Prince Rupert Airport, approximately 85 km away. The plan explicitly reported that a standard concentration-versus-monitor performance evaluation was unavailable. Ksi Lisims Air Quality TDR, June 2024, §§3.6.9 and 3.7, pp. 31–38; Appendix A §12.3; Appendix B pp. B.1–B.2.
Transfer the explicit model versions, separate abnormal-operation cases, nested domains and disclosed limitations. Do not import its grid, vertical levels, remote-station representativeness judgement or chemistry switches as Woodfibre defaults. Its detailed documentation does not establish that an unevaluated Howe Sound configuration will predict local winds or exposure correctly.
These gates are proposed for this research programme. They are not additional B.C. legal requirements.
| Gate | Evidence to assemble and check | Claim permitted after passing |
|---|---|---|
| 1. Reproducible source case | Current source identity and coordinates; tip and flare geometry; event-specific flow, composition, heating value and duration; pollutant factors and their provenance; consistent mass basis and timestamps | A stated, reproducible engineering release scenario. Historical designs remain clearly dated. |
| 2. Representative weather | Raw and processed observations, station heights and exposure, flags and missing periods; numerical-weather configuration; local seasonal and vertical checks | A weather reconstruction supported for the tested regimes. This alone does not validate concentrations. |
| 3. Verified dispersion setup | Input/output archive; terrain and shoreline checks; source-height and buoyancy checks; receptors; boundary, resolution and time-step sensitivity; independently reproduced processing | Numerically checked conditional concentrations, with unresolved sensitivity reported. |
| 4. Independent evaluation | Withheld weather periods and stations; then actual-event concentrations where the source signal is measurable; preregistered metrics and uncertainty treatment | Evidence of demonstrated skill, restricted to observed regimes, pollutants and averaging periods. |
| 5. Decision-ready interpretation | Complete averaging periods; total ambient treatment; criteria version; sensitivity envelope; qualified independent review and any required agency acceptance | A defensible assessment for the stated decision, with limits attached. |
Passing one gate does not substitute for another. A pre-operational project can pass meteorological evaluation without having post-startup flare events available. In that situation, keep concentration validation explicitly pending and present engineering predictions with their source and transport uncertainty.
Set the protocol before looking at the final concentration maps. Freeze input versions and reserve complete weather episodes or seasons for testing; randomly withholding individual hours allows neighbouring hours from the same event to leak into calibration. Use leave-one-station-out checks if the observing network supports them. Report the number of independent episodes and confidence intervals, not only the number of hourly rows.
Weather. Publish a table for each station and season containing valid coverage, wind-speed bias and error, vector wind error, calm frequency and circular direction error above the instrument’s meaningful low-wind threshold. Separately score inflow, outflow, transitions and weak-wind periods. Count observed and predicted regime transitions, false transitions, and timing errors. For measured vertical profiles, compare the height, speed and direction of low-level jets and decoupled layers; assess inversion base and top only when temperature observations actually support them. A ceilometer’s aerosol-layer height alone is not a temperature inversion measurement.
Evaluate the raw numerical-weather field and the adjusted field against the same withheld observations. A downscaling step should not be retained just because its arrows follow the valley more neatly. Require a documented improvement or a transparent trade-off across regimes. The relevant score is performance at the source, transport corridor and community, not only at an airport outside the fjord.
Numerical and configuration sensitivity. For each important operating and weather case, increase domain size, refine the controlling spatial scales and alter the calculation time step. Compare community receptor values, domain maxima, peak locations and arrival times. As a proposed research target, seek less than 10% change in concentrations of decision relevance over two successive refinements. This is a project target, not a provincial acceptance criterion. At values near zero, use an absolute tolerance tied to measurement capability and decision relevance; a relative percentage is misleading. If the target cannot be met, carry the demonstrated spread into the result and avoid presenting contour positions as precise boundaries.
Actual-event concentrations. Align source, meteorological and ambient records to one explicit time standard, correcting documented logger offsets and daylight-saving conventions. Calculate matched hourly and daily averages from the underlying records with declared completeness rules. Compare absolute ambient concentration and the separately estimated flare increment. Use time-series plots, bias/error, upper-tail distributions, peak timing and the frequency of material increments. Compare at the actual monitor coordinates and inlet heights, not the nearest convenient map pixel.
An attribution analysis should include non-flaring control periods matched for wind regime and season, and relevant traffic, shipping, construction, wildfire and other background activity. Report sensitivity to control selection. A pollution rose alone cannot establish causation. If the expected flare increment is smaller than background variability or instrument uncertainty, classify that event as unable to discriminate between model cases rather than as a successful validation of negligible effects.
Publish failed regimes and monitor misses. Separate source uncertainty from weather and chemistry uncertainty with a small, physically coherent ensemble: do not independently combine maximum pollutant rate with minimum heat if the process cannot produce that pair. Label unweighted alternatives as a sensitivity range. They become probabilities only when their weights are supported by operating frequencies and meteorological evidence.
Use separate source inventories and treatment choices for CO, NOx/NO₂, SO₂, primary PM₂.₅ and individual VOCs. An undifferentiated VOC mass cannot predict benzene concentration. A PM emission factor needs an explicit account of particle size and whether condensable material is included. Secondary PM and deposition need additional chemistry and particle evidence; an assumed constant settling rate is a sensitivity case.
Report a total-conversion NO₂ screen alongside any supported refinement. Keep the NOx mass convention explicit, retain the chosen NO₂ fraction and ozone series in the input archive, and publish the chemistry sensitivity separately from transport sensitivity. A refined conversion must not conceal uncertainty in the emissions inventory.
Distinguish the source contribution from total ambient concentration. Match each objective’s averaging and statistical form exactly, using its current authoritative definition. A single 24-hour scenario cannot determine an annual percentile or multi-year attainment statistic. A map of each receptor’s maximum over a year is not one simultaneous plume. Do not interpret crossing a concentration contour as an individual health outcome or interpret a below-objective source increment as proof of safe total exposure.
For the community interface, attach four short facts to every result: source case and date; weather period or scenario; pollutant and averaging period; and evidence status. Show uncertainty as a range or alternative contours only when the meaning of that range is explained. Give users the underlying numeric receptor values and retain unrounded results in downloadable files.
| Priority | Obtain or resolve | Why it changes prediction | If it remains unavailable |
|---|---|---|---|
| P0 | Current Woodfibre source specification and event time histories, including heat as well as pollutant mass | Determines buoyant rise, release duration and the amount of each pollutant | Retain dated design cases and physically linked sensitivity bounds; do not label them current measured emissions. |
| P0 | Raw local meteorology with flags, exposure metadata and overlapping numerical-weather fields | Tests where the plume travels and whether airport or coarse-grid weather represents the fjord | Publish a conditional ensemble and explicitly untested regimes. |
| P0 | Existing ambient raw records, instrument/QA metadata, and verified event timestamps | Enables baseline estimation and eventual time-matched evaluation | Weather validation can proceed; source attribution remains unresolved. |
| P1 | Coordinated surface winds along the source–community corridor and vertical wind/temperature observations during representative outflow, inflow and reversal episodes | Resolves the largest uncertainty introduced by vertical separation and fjord flow | Avoid assigning observed probability or confidence to an assumed inversion cap. |
| P1 | Representative ozone and NO/NO₂ measurements; information supporting the emitted NO₂:NOx ratio | Narrows the conversion from NOx emissions to NO₂ exposure | Keep total-conversion and justified alternative chemistry cases visible. |
| P1 | Complete current modelling files already submitted for Woodfibre, including preprocessing, flare parameter calculations and QA | May reproduce or improve an existing site assessment faster than building a new one | Record the missing files explicitly; a PDF alone is not a reproducible model. |
| P2 | Source-area survey, building geometry, current land cover and sufficiently resolved shoreline/terrain | Refines source elevation, downwash, surface exchange and local receptors | Quantify sensitivity and keep precision consistent with available inputs. |
| P2 | PM size/speciation, condensable fraction, combustion-performance evidence and relevant VOC speciation | Supports pollutant-specific treatment and source uncertainty estimates | Do not turn generic PM/VOC factors into chemical-specific predictions. |
| P2 | Independent review and a reproducible model/output archive | Identifies configuration and interpretation errors before public reliance | Continue as a documented research product, with no claim of regulatory acceptance. |
Prioritize retrieval and auditing of existing records before proposing new equipment. If a field campaign is necessary, design its locations and timing after the first weather evaluation reveals which regimes and parts of the corridor are poorly constrained. The defensible end point is a model with a traceable source case, demonstrated weather skill, appropriate pollutant treatment and explicitly bounded remaining uncertainty—not the elimination of every assumption.