VANCOUVER · SO₂ · RESEARCH NOTES
From weather and emissions to a continuous 3D source scenario.
Source increments, not total ambient SO₂. Background propagation is deferred after a local particle-concentration fidelity failure. The current source calculation starts at the actual outlets, with parameterized thermal rise and 3D transport.
Explore ↗01 / METHOD
Actual equations, units and approximations.
Historical inference is separate from the new forecast.
MODEL & DISPLAY RESOLUTION
| Layer | Grid or size | Meaning |
|---|---|---|
| Weather | 9 km parent · 3 km nested · 50 levels | Prescribed wind varies with height; no resolved building wakes. |
| Native regional concentration | 151 × 91 × 16 | About 0.73 × 1.11 km horizontal cells. Surface query: 0–20 m layer mean. |
| Near-source residence fields | 3 boxes, each 96 × 96 × 96 | Each covers 8 × 8 km and 0–2 km AGL with nonuniform bins. Fine bins are not demonstrated CFD accuracy. |
| Regional Web texture | 96 × 91 × 16 | A display regrid; it does not increase atmospheric resolution. |
| Image pixels | Scientific 3840 × 2160 · native snapshot 1440 × 1000 | The interactive canvas follows the window and adapts render scale. |
Transport follows 499,920 actual source-particle trajectories through 360 hours. Its early-age integration step is 1 s, independently of the sampling grids and 15-second viewing clock. Particle-count convergence and engineering near-nozzle point accuracy are not established.
02 / CONTINUOUS FORECAST
Frozen initialization: 2026-11-04 18:00 UTC. End: 2026-11-19 18:00 UTC. All 361 weather states in both domains were validated; native transport spans 360 consecutive hours without daily resets.
GFS cycle 2026110418 supplies the entire 15-day period. The CFS extension from the former 30-day plan is not used. These are forecast conditions, not future observations.
9 / 3 km weather
50 layers · 18 / 6 s steps · Noah-MPContinuous source transport
499,920 emitted particles · 1,800 releasesInteractive display
16 output layers · native F32 surface queriesThe user-authorized criterion is per-frame RAINNC grid-RMSE≤2 mm against the same fixed CPU reference; the maximum is 1.405205 mm. It is not a per-cell maximum or validation against future rain observations. Original 1 mm failures remain preserved.
Weather uses an independent WRF-compatible JAX research implementation with documented W-boundary and cloud-freezing fixes. FLEXPART transport remains CPU-parallel. This is not an all-GPU or official NCAR GPU WRF pipeline.
| Label | Emitted | Airborne | Dry | Wet | Outflow | Residual |
|---|
The query reads the native 0–20 m layer. The current 360-hour ground chunks preserve the native F32 products; 26 independent SciPy point/time/mean/deposition controls check the query implementation. Values outside the domain remain null.
Tiny positive concentrations previously rounded to 0.000. They now use significant figures or scientific notation, such as 2.45 × 10−6 ppb. Exact model zeros are displayed as 0.
In the current selected lowland rectangle, 62.29% of area×hour source values are exactly zero. In 26.74%, the ground field is zero while a higher layer is positive. The source-only surface peak in this rectangle is 7.96435 ppb.
A zero may reflect the plume remaining aloft, removal or finite particle sampling. A nonzero 15-day mean or cumulative deposition may coexist with a zero in the selected hourly airborne field. These are native 0–20 m hourly means, not street-scale exposure measurements.
Statistics are area×hour weighted. The lowland rectangle spans longitude−123.25..−122.90, latitude 49.10..49.33, native terrain<80 m; it includes water and is not a municipal mask. Approximately 0.7×1.1 km output cells and a 0–20 m mean do not resolve breathing-height exposure. Cumulative deposition is distinct from current airborne concentration.
The dated NASA GEOS-CF model initialization is traceable, not an arbitrary random pollution field. However, representing it with 350,000 weighted particles produced local near-surface spikes around 79 ppb in a city region where the mapped starting model field is approximately 1–4.2 ppb. A deterministic 1.64-million-particle candidate improved the first hour but still produced 134 ppb within 12 hours. It was stopped and preserved, not published as a corrected field.
Background propagation is deferred at the user’s request. The current animation and surface,24 h,15-day mean and deposition queries show only the three selected sources. No missing background is silently called a computed zero. These source increments cannot establish total ambient concentration or actual exposure.
Background sampling diagnostic ↗Actual cumulative RAINC+RAINNC differences drive hourly precipitation glyphs. Within the stated city rectangle,133 hours contain at least one cell above 0.1 mm/h, mean cumulative water-equivalent precipitation is about 72 mm, and the maximum hourly value is 7.13 mm. Snow/ice/graupel diagnostics are retained.
Grey rain tint follows native source wet-removal increments; gaseous SO₂ is not represented as a falling solid. The visible source plume now comes from actual local 3D residence-concentration fields, driven by the same native particles as regional transport. It is no longer a camera-facing origin guide. The prescribed 3D wind varies with height; outlet momentum and heat are parameterized, with no feedback on the weather. Local/regional fields use a partition of unity, avoiding duplicated concentration. SO₂ is invisible: grey/custom colour and extinction gain are display aids.
03 / VALIDATION
d 02 cumulative RAINNC RMSE vs fixed CPU reference · mm
| Branch | Domain | Field | Hour | Value | Limit |
|---|
238 reports from 5 held-out stations. Overall improvement did not satisfy the original gate because regional performance deteriorated. This historical calibration is not applied to the new annual-rate source scenario.
| Stratum | Reports | Prior RMSE | Inferred RMSE | Prior MAE | Inferred MAE |
|---|
Scores are ppb; per-report held-out weights are equal. The original historical success gate remains failed.
04 / VERSION CHANGES
| First version | Current release | Remaining limit |
|---|---|---|
| Coarse terrain and industrial icons | 295,252 building footprints, coloured landmarks, ground UBC crest and outer terrain | Simplified geometry; not a 1:1 Google photogrammetry model |
| Coarse surface / isosurface display | 16 actual AGL scalar layers and freely orbiting volume rendering | Output detail does not increase the 3 km weather resolution |
| Historical 71.5 h plus a separate 120 s release | Actual continuous 360 h forecast mapped to 15 s, adjustable speed | 15-day forecast scenario, not observations |
| Source origin hard to identify | Reported outlet disks, finite-density thermal rise and actual 3D near-source concentration | Parameterized entrainment and thermal rise; no resolved building/stack CFD |
| No weather animation | Actual forecast rain/snow and wet-removal glyphs | Symbolic drops are not tracked dissolved-SO₂ trajectories |
| Fixed 3-decimal query readout | Positive values retain significant figures;24 h/15-day means and source deposition | Native Monte Carlo zeros are not evidence of clean ambient air |
| Dense overlays and repeated city passes | Settings panel, clean canvas, one city pass and adaptive volume rendering | Default 15-second playback; actual frame rate depends on the device |
05 / DATA
MIXINGRATIO numeric molar ppt /1000 → ppb. CONC ng/m³ /1000 → µg/m³. Cumulative deposition ng/m² /10¹² → kg/m²; never sum cumulative snapshots. Raw archived report text is preserved for provenance; interface and documentation are in English.