Woodfibre LNG: weather and background-air data roadmap

Public-use priority — 11 September 2026. The immediate deliverable is a practical community scenario tool: documented flare cases, consistent place comparisons, matching-period air-quality benchmarks and a short explanation of potential health relevance. Additional internal wind-model tracing and temperature-precision investigations are optional research, not prerequisites for publishing that tool. Current release/heat records, representative local weather and suitable transport, pollutant-specific chemistry and background remain the most useful improvements for quantitative local exposure claims. The existing map remains illustrative; its physics have not been relabelled as a forecast.

Updated 11 September 2026 · Living data roadmap

The highest-value next input is time-resolved, quality-controlled weather at the Woodfibre site, together with vertical wind and temperature measurements. Airport wind is useful evidence, but it cannot determine whether a plume stays in a shallow valley layer, passes above the city, or meets a return flow. Background air is a separate input: it must be measured and treated consistently with the pollutant and averaging period.

The accompanying machine-readable catalogue records providers, variables, resolution, dates, quality handling, licences, access routes and what was actually acquired. “Available” never means that a dataset has already been downloaded, validated or incorporated into the model.

What was acquired in this research pass

The vertical-wind discrepancy now has a traced mechanism. Captured, unchanged-output replays show that terrain-tangent redirection creates much of the selected large vertical increment, with smoothing partly cancelling it. Independent checks cover the original area, time/height distributions and reference profiles. Stage attribution and data.

Temperature preparation and convective history are now separated. Two unchanged replays reproduce the actual gradient calculation exactly. The complete common-height comparison supports 38,935 of 244,824 original-area cell-hours; none of the 13 reference points has all required near-ground brackets. Preparation/sampling and the convective override dominate this descriptive comparison. Separately, unrounded temperature, pressure and humidity were recovered for a preselected 100-column native tile. The tile printing-sensitivity check is now complete and independently reproduced; it does not resolve the wind-performance or near-ground coverage limits. Further internal thermal work is parked while the public-use release is completed. Higher precision does not fill the near-ground support gap. Thermal calculation and figure, higher-precision archive, earlier vertical-stability comparison.

Three historical cases are now selectable. The Explorer adds the 14-hour LPCF tank-shutdown reconstruction, full-day time controls and explicit scheduled-versus-emitted mass. A source comparison shows rates and event totals for PM₂.₅, NOx, CO and total VOC. The calculation still uses illustrative fixed weather and prescribed injection height. Scenario inputs, timing and checks.

Vertical refinement does not resolve the transport discrepancy. Nine matched 10-, 18- and 34-layer cases retain the original study area and all 13 references. The largest preselected vertical excursion persists around −11 m/s, while the airport comparison is unchanged. A separate comparison with native model W uses identical support across all nine cases; it is a model comparison, not observation validation. Refinement results, native comparison and limits.

The marine thermal path is now identified. The current wind-only configuration takes the neutral overwater fallback. Populated native skin temperature and heat-flux fields are not presently used through that path. A prognostic thermal sensitivity would change donor selection, low-level temperature and surface exchange together; it requires explicit assumptions and captured diagnostics before interpretation. Source trace and next experiment.

Saved winds can now be explored across the study area. A new interactive wind study shows all 24 hourly intervals and three settings from the expanded 1 June 2023 calculations, with a common wind-speed scale and consistent reference-point sampling. It presents actual saved diagnostic winds, separately from the illustrative plume. Display data and methods.

The low-level processing trace narrows the airport mismatch. Using the same four native points and interpolation weights, mean airport speed is 4.06 m/s before profile processing, 4.02 m/s after that processing and 4.01 m/s in the final adjustments-off field, against 1.71 m/s observed. Most of this day's station speed mismatch is already present in the regional forcing. All original study cells and 13 reference points were compared; no location was tuned. Trace and independent reproduction.

Native water and thermal fields are now checked across the forcing grid. All 20 archive tiles link to the retained 2,000 native cells and 25 states. Water code 16 includes 39 lake-marked cells; the airport-nearest native point is among them. All 50,000 SST values are zero upstream, while simulated skin temperature and surface heat flux are populated. These fields have not been substituted into the model. Native-surface evidence and data.

Coordinate conversion does not explain away the vertical excursions. Two explicit collocation methods reconstruct model-implied physical vertical velocity from saved terrain-following W. The persistent terrain-case excursion remains about −11 m/s under either method. Their differences are retained across every interior cell, not just the extreme. Physical validation remains necessary. Vertical-coordinate reconstruction.

Calculated winds now checked against airport observations. For the complete 1 June 2023 model day, all six CALMET cases overestimate mean airport speed: 4.01–4.28 m/s against 1.71 m/s observed. Added surrounding terrain changes that station result only slightly. An independent reconstruction reproduces all 168 pairs, seven summaries and 42 direction/cohort rows. Source inspection establishes the interval-end wind alignment; the subsequent low-level trace quantifies the fixed extrapolation assumption’s modest effect for this day. One station/day and unresolved upstream assimilation do not establish airshed validation. Full-day comparison and data.

Large vertical components persist in the expanded domain. The separate source-and-equation audit checks 28,512,720 interior values within the existing arithmetic tolerance. The large original terrain-following W excursion remains well inside the enlarged boundary. Equation consistency and added terrain alone do not establish physically realistic vertical transport. Vertical-wind findings and methods.

Expanded-domain winds now compared across the whole study area. Three additional full-day CALMET runs add 10 km of terrain on each side of the original grid. The comparison covers all original cells and all ten existing place references under common criteria. Every original inner terrain and surface input reproduces exactly; 87,266,160 new wind values passed structural and finite-value checks. The largest near-surface domain change is 4.79 m/s, while area-wide averages are much smaller. Radius sensitivity remains after the expansion; neither setting is validated by this test. A new community-coverage audit identifies reserve polygons missed by regular grid centres. Full-area findings and maps, community coverage. Earlier five-run results remain available in the first study and wind-driver diagnostic.

Background candidates and station context. Fixed values by averaging period, annual/monthly hour-of-day NO₂ tables and a 2025 ozone preparation copy now have independent native-data checks. The ozone copy contains 8,120 observed hours and 640 explicitly flagged substitutions; it is numerically complete but still needs a site-representativeness decision. NAPS provides a nominal 4 m inlet and neighbourhood exposure classification in its Stations2024 metadata. Dated inlet/instrument history and simultaneous town/onsite comparison are the next selection gates. Background results and data, station records and targeted gaps.

Complete 2023–2025 town-air record. All 157,824 native hourly entries for PM₂.₅, NO, NO₂, NOx, ozone and SO₂ are now acquired, with 149,871 numeric values and 7,953 retained NA readings. Every expected timestamp is present, including leap day. Independent calendar and objective-metric checks passed. The new report provides current B.C. comparisons, full daily timelines, methods and data. Spatial background selection remains separate work. Three-year findings and reproduction archive.

Earlier Squamish Elementary sample, preserved. The corrected calendar-day CSV contains 432 native hour-ending rows covering all of 1–3 May 2025: 72 each for NO, NO₂, NOx, O₃, PM₂.₅ and SO₂. There are 417 numeric readings. Each gas has three provider-marked NA hours; PM₂.₅ has 72 valid hours. Its hour-ending timestamps run from 1 May 01:00 through 4 May 00:00 PST, representing intervals from 1 May 00:00 through 4 May 00:00. Original values, units, instrument fields and missing values are preserved. The corrected retrieval provenance documents 426 original rows reconfirmed exactly and six additional final boundary rows. These are ambient concentrations, with no attribution to flaring.

The original seed CSV and original provenance are preserved separately. That extraction selected normalized hour-end timestamps from 1 May 00:00 through 3 May 23:00, so it included the last hour of 30 April and omitted the last hour of 3 May. Native DATE/TIME 24:00 means the end of that calendar day; normalized DATE_PST places it at the following midnight. Use the corrected calendar-day file or explicit UTC intervals for interval-based analysis. Weather timestamps do not imply the same averaging interval. Observed-case findings and timing audit, BC hour-ending convention, §3.1

The files came from the BC provider's verified 2025 archive by anonymous FTP. Its README identifies the annual folders as QA/QC validated except folders explicitly marked _Unverified. The downloaded CSV does not contain a per-row QA flag column, so no additional QA labels were invented. Current catalogue information identifies verified pollutant data through 2025 and meteorology through 2024; a 2025 folder does not establish that its meteorology is finalized. The earlier sample established the access route; the full-year acquisition above now supports station statistics. Spatial representativeness and model-background selection remain separate questions. BC verified-data catalogue

Paired surface weather. The regional weather CSV and provenance cover the same three local-standard-time days. They retain all returned original fields and flags:

The missing airport hours are 1 May 13:00 and 3 May 03:00 PST; Pam Rocks is missing 1 May 04:00 PST. This supplements the existing complete-year airport acquisition of 8,741 rows for 2025, with 19 absent hours. Original wind direction is in tens of degrees from true north, speed in km/h, pressure in kPa and temperature in °C. LOCAL_DATE uses PST, UTC−8, including summer; UTC_DATE is retained separately. Individual wind observations are not necessarily averages over the entire clock hour. Pemberton's wind and pressure are from distinct stations. ECCC hourly collection, ECCC station inventory

Annual two-station wind benchmark. The full 2025 Pam Rocks archive has now been acquired and joined to the retained airport year, producing 8,730 usable simultaneous pairs. The airport reported calm while Pam Rocks reported at least 5 m/s in 914 pairs. Twelve observation-selected seasonal cases are preserved from 356 qualifying runs, with threshold sensitivity, missing timestamps and fixed-PST/UTC checks. Matching numerical weather is now partially acquired for two of the twelve cases: six hourly forecast times in each, with 24 matched station-hours. Neither selected run is fully covered, the other ten cases have no acquired matching forecast fields, and no annual forecast dataset has been acquired. Matching source-site winds remain unavailable. Annual findings and episode inventory, partial HRRR comparisons

Two archived HRRR forecast windows acquired and tested. The cycles were prescribed before error inspection: 2 February 2025 at 18 UTC, valid 19 UTC through 3 February 00 UTC, and 25 July 2025 at 12 UTC, valid 13–18 UTC. These are only six-hour portions of observed runs containing 103 and 22 consecutive samples respectively. NOAA's anonymous AWS archive supplied 56 original indexed GRIB messages, 88,001,957 bytes, including 10 m U/V, 2 m temperature, terrain and two pressure levels at lead six. Original indices/listings, Lambert rotation checks, UTC joins and ground masks are preserved. All three locations and the display square are inside the actual 3 km grid. HRRR findings, NOAA archive

The winter result reveals a concrete failure: Pam Rocks observations remain northerly while HRRR briefly predicts southerly winds. Wind-vector RMSE is 2.636 m/s at the airport and 10.118 m/s at Pam Rocks in the winter window; the summer values are 1.776 and 2.285 m/s. These observation-selected short tests do not establish seasonal skill or a ranking against September 2026 HRDPS results from different dates. The Woodfibre proxy's nearest HRRR cell is 1.128 km away at 38.26 m elevation, with neighbouring model terrain exceeding 600 m; a small terrain-height discrepancy at one point does not establish representative source winds. No source forcing was accepted. Exact values and limitations

The new frozen HRRR source pack is 72,161,945 bytes. Its 102 members, all 56 original field hashes and ZIP integrity were verified. The manifest and archive check preserve provenance and attribution; prior observation and forecast archives remain unchanged. The full-year observation inputs remain in their separate annual evidence pack rather than being duplicated here.

One archived operational forecast acquired and compared. We subsequently acquired the actual HRDPS 2.5 km cycle initialized 2 September 2026 at 00:00 UTC, leads 1–24, valid 2 September 01:00 through 3 September 00:00 UTC. The 92 original GRIB files total 219,461,500 bytes and include hourly 10 m U/V winds and 2 m temperature, independent wind checks and three low-level pressure profiles at one time. All 48 station pairs have temperature and wind values; UTC joins, units and true-earth wind rotation were audited. This is one operational forecast test, not a reanalysis, multi-episode hindcast or validation of local atmospheric/dispersion prediction. Forecast comparison, raw-file hash/URL manifest, processing audit

The result makes the remaining gaps more concrete: wind-direction mean absolute error was 30.8° at Squamish Airport and 104.6° at Pam Rocks, for 19 and 23 hours respectively with both speeds at least 1 m/s. The nearest Woodfibre proxy grid point is 1.15 km away with model orography 568.4 m, far above the shoreline. Its 985 and 950 hPa levels are below the model surface and are rejected as atmospheric samples there. This cell must not be used directly as measured or representative source wind. Detailed representativeness and timing limits

A separate matched resolution test. The experimental West 1 km and operational 2.5 km 5 September 00 UTC cycles, leads 1–6, were acquired and compared against the same 12 observed station records. This adds 68 raw fields, 112,406,800 bytes, and 24 model–observation pairs. At the Woodfibre proxy, 1 km sampling reduces the nearest-point distance to 170 m and model orography to 135 m, still far above the roughly 9 m shoreline reference. West has worse airport wind errors and slightly smaller but still large Pam Rocks errors in these six hours. At 06 UTC, Pam Rocks observed 10.83 m/s, versus 4.55 m/s at 1 km and 2.04 m/s at 2.5 km. This is a same-hours diagnostic; it does not compare skill across dates or accept either model as source forcing. Resolution findings, exact manifest

Additional road-weather records checked. The June 2023 Tantalus and Eagle Ridge files each contain 720 unique calendar labels and numeric wind fields. All 27,360 native field tokens agree with a separate parser. Eagle Ridge retains 139 zero-speed records with numeric bearings; the applicable calm, clock, bearing, exposure and QA definitions are still required. Road observations and qualification needs.

B.C. WRF native weather acquired. Original 1.333 km d06 files with 40 native levels have been acquired for January and June 2023. The source/Elementary June 2 profile sample and fixed-cell Airport/Pam monthly surface-wind comparisons are complete and reproducible. The two months show station-dependent errors; assimilation independence and source-site/vertical observational validation remain unresolved. These are archived model data, not measured profiles. June profile sample, January comparison, June comparison.

Station-history follow-through. Retrospective WMO records list nominal 10.0 m Airport/Pam wind deployments beginning April 29, 2016, with no recorded end date and no reference-surface type. Actual January/June 2023 mounting, exposure and station-specific averaging remain unresolved; no height correction or relocated comparison point follows. January methods and station records.

Station and terrain inventories. We acquired nearby ECCC and BC station metadata, NOAA upper-air station inventory, original 5 September HRDPS/RDPS file listings, and local NRCan terrain catalogue responses. The original listing snapshots remain historical evidence; HRDPS 2.5 km and experimental West 1 km now have separate actual acquisition records. RDPS remains inventory-only, and no radiosonde series was downloaded. Two small published lidar project boundaries were acquired and clipped in this GeoJSON. They are actual project extents, not inferred coverage.

Highest-priority work, in order

Onsite daily records and comparison: the full 2025 onsite daily-table acquisition and onsite–town comparison are complete; all 30 June 2026 daily rows are also retained. These products do not supply native hourly wind direction or establish meteorological timing, heights or final QC. Full-year paired comparison. The Level 0 statement concerns air quality, with separate meteorological QC unestablished. Three inspected monthly PDFs contain no embedded data files. Acquired tables and precise hourly-data route

Vertical-archive follow-through: the public SNOW-V10 archive was recovered and audited; its 160 members contain surface/scalar observations, not Squamish radar/RASS profiles. The 2016 lidar and later institutional searches did not recover measured local profiles within their bounded coverage. Exact existing-record drafts are prepared, not sent. The known 2010 and 2016–2018 campaigns precede the acquired B.C. 2019–2023 model period; historical profiles could inform process evaluation and campaign design, but would not directly verify 2023 forcing.

Additional vertical-data lead: a primary study documents a separate Squamish Halo lidar deployment beginning in May 2016 and identifies ECCC as data producer. Its former archive hostname failed DNS resolution; current operation and downloadable Squamish profiles remain unverified. Do not confuse it with the 2010 radar/RASS dataset or the public Iqaluit/Whitehorse lidar datasets. Source and archive-routing evidence

Immediate model work: the expanded-domain experiment, initial airport comparison, low-level reader trace and vertical-coordinate reconstruction are complete. Next, qualify native surface representation and thermal inputs, then predeclare a comparison using independently documented local observations and weather regimes. Recover production configuration and actual assimilation records before claiming independence. Avoid tuning a setting to one station/day; retain common criteria across all original cells and community references. Input evidence, low-level trace, vertical reconstruction.

The priority order reflects the information each action adds, rather than the size of the download. An exact source release history is a parallel prerequisite handled by the emissions workstream. A better DEM cannot compensate for an unknown release or an unmeasured inversion.

For a measurement campaign, investigate a Doppler wind lidar or comparable profiler, temperature-profile capability and well-exposed surface instruments. A useful design would span the source shoreline, the estuary/city and an up-valley location, with seasonal coverage and event sampling. Sensor type, placement, achievable height range and campaign length require a documented site-specific design; a fixed specification has not been assumed here. Historical profiles can help design the campaign but do not replace contemporary observations.

Local observations and background air

The May 2025 Woodfibre report confirms an operating site meteorology station at NAD83 UTM zone 10N 481610 E, 5501369 N, next to the AQMS. It lists wind, temperature, rainfall, humidity and pressure. The AQMS continuously measures PM₂.₅, PM₁₀, TSP and NO₂; SO₂ and VOC sampling is passive over much longer periods. The report describes Level 0 validation for the presented air-quality data. Its figures and daily tables establish that observations exist; they are not a substitute for a raw, quality-controlled time series. Request the complete machine-readable archive, including uncertainty and final QA. A wind rose cannot be converted into the chronological winds needed for an event simulation. Woodfibre May 2025 AQMS report, sections 1–4 and appendices

BC's station metadata identifies Squamish Elementary, E304570, at 49.70516, −123.15133, elevation 6 m, opened 17 December 2015. Langdale Elementary E222778 and Port Mellon E226338 are also listed as operational in the acquired 2025 metadata. Multiple instrument/averaging records can share one station name and identifier, so preserve the full station name and instrument fields. Metadata opening dates do not guarantee all pollutants were measured continuously since opening. BC station metadata archive

NO and O₃ alongside NO₂ are valuable for evaluating chemical context. PM background can include smoke and other regional/local sources; Port Mellon is industrially influenced and must not be labelled clean background by default. Use event and wind information to choose an appropriate method and check sensitivity to the background selection. Avoid adding a measured concentration that already includes the modelled source without accounting for that overlap. A single receptor cannot identify source emissions or validate the entire 400 km² concentration field.

The BC current raw NO₂ feed was also reachable, with station coordinates and IDs in its header. It is preliminary, not interchangeable with the verified archive. Exact rolling retention was not established.

Vertical structure: useful evidence, a substantial remaining gap

An Environment Canada report documents a 915 MHz wind profiler and RASS at Squamish Airport installed in 2006, with low-altitude wind profiles to roughly 3 km and virtual-temperature profiling to roughly 1.5 km. The Vancouver 2010 network and subsequent Howe Sound research demonstrate that such local observations were made. Their existence does not establish that the instrument is operating today or that its raw archive is openly downloadable. A library entry for a research paper is not a data service. Environment Canada Vancouver 2010 report

The Howe Sound literature describes important changes along the channel and with height, including shallow outflow and different flow aloft in particular events. These findings justify testing vertical and spatial wind structure; they do not justify assigning the depth from one published case to every scenario. Request profile gates, measurement averaging, quality/SNR filters and time stamps with any historical data. Along-Channel Winds in Howe Sound: Climatological Analysis and Case Studies, Gap Winds in a Fjord, Part I

NOAA's IGRA 2.2 provides an accessible upper-air archive with pressure, temperature, geopotential height, humidity and winds, together with station history and quality processing. The acquired inventory lists ongoing Port Hardy and Quillayute records; their distances from the Woodfibre site proxy are approximately 315 km and 215 km respectively. Vancouver's listed upper-air record ends in 1996 and Kelowna's in 2017. These stations can constrain regional conditions but cannot measure a shallow Squamish inversion. Exact sounding availability and level flags must be checked before use; no sounding values were acquired here. NOAA IGRA documentation, station inventory

Numerical weather fields: verified access, different levels of suitability

HRDPS documentation, HRDPS West documentation, NOAA HRRR archive and terms, NCEP HRRR product inventory, RDPS documentation, ERA5 provider overview

The catalogue preserves the original file inventory snapshots from the 5 September 2026 00 UTC cycles and their example URLs. Separate 2 September HRDPS acquisition and 5 September matched resolution-comparison records link actual retained files, hashes, values and diagnostics. They supersede inventory-only status for operational HRDPS and experimental West; RDPS still has no acquired values. The model's internal vertical levels are not all distributed in the pressure-level product. For example, HRDPS West describes 62 internal levels, but a much smaller set of public pressure levels. Pressure surfaces below terrain are not atmospheric samples, and masks must be evaluated for the particular model and time: at the operational Woodfibre proxy point, 985/950 hPa were below ground on 2 September; on 5 September, 985 hPa was below and 950 hPa was only barely above model terrain.

The two 2025 HRRR acquisitions are separate archived operational forecasts, not reanalysis or full coverage of the annual episode inventory. All twelve acquired 975/950 hPa point samples are above their HRRR model surface, but two levels at one time per case do not validate an inversion or full vertical wind field. The only listed 10 m WIND diagnostic is an interval maximum; it was not substituted for instantaneous vector speed. Lambert wind rotation was checked against NOAA's formula and independent projection geometry. Do not apply the HRDPS rotated-pole formula to these Lambert fields. HRRR processing audit, NOAA rotation documentation

The main Datamart's current 30-day retention is capacity-dependent, with dated directories and a /today route. It cannot provide a May 2025 forecast merely by changing today's URL. ECCC documents an archive-retrieval service with cost recovery; request a quote and confirm the exact historical fields, levels, model versions and years. No request, payment or recurring downloader was created. Datamart retention, archived-data retrieval service

ERA5 is a reanalysis combining observations and a model, not a local observation. Its long record is useful when historical regional forcing is needed. Preliminary ERA5T is normally about five days behind real time and can change before the final release roughly two to three months later. A 0.25° cell is too coarse to resolve this valley; nesting or diagnostic downscaling still requires local validation. ECMWF ERA5 overview

Terrain and surface properties

Canadian terrain processing is now numerically checked. A pinned horizontal transformation carries native NAD83(CSRS) CDEM into AERMAP 24142 while retaining CGVD28 heights. The independent check passes 1,516 checks with zero discrepancies; all 18 point elevations agree within 0.0041 m arithmetic error. This is not survey accuracy. A single unchanged terrain-domain extension does not establish final hill-search adequacy, and it does not select AERMOD as the final fjord model. Terrain methods, results and scope.

The visualization uses an NRCan CDEM terrain asset aggregated to a 100 × 100 grid at 200 m spacing. The current browser calculation resamples that asset to 64 × 64 cells, about 313 m across; the stored terrain spacing is not the solver resolution. Its source is legacy and uses CGVD28 heights. The existing terrain evidence documents the source, sampling, missing-value checks and point checks. It represents major mountain/valley geometry; it is not a current construction survey.

The live NRCan catalogue now exposes a June 2026 MRDEM 30 m mosaic and local 1 m lidar projects, including Squamish 2016, Sunshine Coast 2020, and other overlapping projects. MRDEM's update date is not the acquisition date of every source pixel. The HRDEM lidar collection provides DTM/DSM and detailed coverage assets in EPSG:3979, referenced to CGVD2013; it notes that separate projects are not edge-matched. The catalogue includes exact asset URLs. NRCan lidar collection, MRDEM collection

Checking the actual published project extents, the Squamish 2016 project intersects about 48.27 km² of this domain and includes both the Woodfibre proxy point and Squamish Elementary. The Sunshine Coast 2020 project intersects about 51.73 km² and includes neither point. These are boundary intersections, not proof every enclosed raster pixel is valid. The next terrain step is a small raster window and detailed coverage/nodata audit, followed by a common vertical datum. This is more defensible than assuming an entire large catalogue bounding box has 1 m coverage. Squamish project extent, Sunshine Coast project extent

LidarBC offers provincial point clouds, DEMs and DSMs under the BC open licence; local tile downloads were not undertaken. ESA WorldCover 2021 v200 supplies 10 m land-cover classes and quality information under CC BY 4.0. Either resource requires acquisition-vintage checks. Land-cover classes are only a starting point for assigning roughness, albedo and heat/moisture exchange; they are not measurements of those parameters. Construction clearing and present structures need current mapping. LidarBC portal guide, ESA WorldCover access

Gates before treating a result as prediction

1. Establish the actual release and source geometry, time-varying meteorology, usable profiles and background method for the chosen episode.
2. Validate the meteorological field independently across location, height and wind regime. Report low-wind and transition failures rather than hiding them in an annual average.
3. Run the full changing episode with suitable initialization, boundaries, chemistry and deposition treatment. Preserve the assumptions that still remain and test them explicitly.
4. Compare concentrations only at compatible species, units and averaging times. Keep measured ambient concentration, modelled source increment and combined estimate distinguishable. A brief plume snapshot cannot establish a 24-hour or multi-year objective statistic.
5. Evaluate against independent receptor measurements and report an uncertainty range. Sparse monitors may support an episode check without establishing skill everywhere on the map.

The September HRDPS diagnostics, two short 2025 HRRR windows, June 2023 native WRF profile sample, fixed January/June 2023 station comparisons and Canadian terrain interpolation check are complete. No candidate has been accepted as Woodfibre source-site forcing. The HRRR work supplies partial matching weather for two of twelve selected 2025 episodes, with no full selected run or annual coverage. A representative multi-episode meteorological evaluation, raw onsite weather archive, contemporary local vertical observations, updated terrain raster and validated predictive meteorological field remain unacquired or unfinished. Their access status and next steps are preserved in the catalogue rather than replaced with assumed values.