Finer vertical layers reduce the average change between successive calculations, but do not resolve the previously identified large vertical excursion or the airport wind-speed mismatch. The 10-, 18- and 34-layer calculations use the same day, weather inputs and horizontal terrain grid. These numerical sensitivities do not establish more accurate local winds, and they have not replaced the winds in either public viewer. No pollutant release was simulated.
All three wind configurations use the same 201 × 201 cells at 200 m spacing, the same 24 intervals on 1 June 2023, and the same native weather, terrain and surface settings. The original 20.2 km square and all 13 retained reference points receive the same comparisons. These points do not represent complete community exposure coverage.
The first two layers, 0–20 m and 20–40 m, are unchanged. Each of the eight layers above 40 m is split into two or four parts, giving 18 or 34 total layers. All original faces and the 3,000 m top remain. This also changes the heights at which upstream weather profiles are sampled; it is a coupled vertical discretization test, not just a different integration of identical winds.
The horizontal comparison averages each set of thinner layers over the corresponding original layer. The table uses every original-area cell, all 24 intervals and all ten common layers. Vector difference describes the change between calculations, not error against measured wind.
| Configuration | Layer comparison | Mean horizontal vector difference (m/s) | Largest horizontal vector difference (m/s) | Mean absolute W difference (m/s) | Largest W difference (m/s) |
|---|---|---|---|---|---|
| Interpolated regional winds | 10 → 18 | 0.060978 | 1.087224 | 0.018076 | 0.794632 |
| Interpolated regional winds | 18 → 34 | 0.014213 | 0.205631 | 0.004689 | 0.225799 |
| Interpolated regional winds | 10 → 34 | 0.070872 | 1.018023 | 0.017546 | 0.645204 |
| Terrain adjustment, 5 km search | 10 → 18 | 0.065777 | 1.764629 | 0.035478 | 2.241800 |
| Terrain adjustment, 5 km search | 18 → 34 | 0.016964 | 0.904692 | 0.013208 | 0.995821 |
| Terrain adjustment, 5 km search | 10 → 34 | 0.074691 | 1.616637 | 0.032765 | 2.054733 |
| Terrain adjustment, 10 km search | 10 → 18 | 0.069267 | 2.274691 | 0.049982 | 3.013674 |
| Terrain adjustment, 10 km search | 18 → 34 | 0.018927 | 0.904692 | 0.020319 | 1.575435 |
| Terrain adjustment, 10 km search | 10 → 34 | 0.077634 | 2.166092 | 0.045762 | 2.826024 |
W is the saved terrain-coordinate vertical component. A separate physical-vertical estimate adds terrain advection using horizontal winds interpolated to each face; it is available only at the nine bracketed faces, 20–2,200 m, and strict interior cells. Neither coordinate correction nor smaller differences alone demonstrates atmospheric accuracy.
The first-layer centre remains 10 m above model terrain. These full-area maximum vector changes show whether altering higher layers also changes this preserved near-surface layer.
| Configuration | Largest first-layer change, 10 → 18 (m/s) | Largest first-layer change, 18 → 34 (m/s) | Largest first-layer change, 10 → 34 (m/s) |
|---|---|---|---|
| Interpolated regional winds | 0.00000000 | 0.00000000 | 0.00000000 |
| Terrain adjustment, 5 km search | 1.76257829 | 0.90469155 | 1.61663694 |
| Terrain adjustment, 10 km search | 1.76254584 | 0.90469155 | 1.61663079 |
The unadjusted first-layer winds are unchanged. Small mean changes with terrain adjustment coexist with local changes exceeding 1 m/s. The source trace identifies a coupling: RDMM5 samples native temperature profiles at the altered layer centres; CGAMMA2 interpolates those temperatures to fixed ZUPT heights, which can change the stability used in terrain adjustment. FRADJ has threshold decisions and SLOPE allocates partial-layer flow. These paths explain why preserving the first-layer geometry does not guarantee identical first-layer winds; no output-only comparison isolates the share attributable to each path. The independent source review records CALMET lines 21755–21837, 3331–3375, 7383–7405 and 30569–30578. Released CALMET source.
The airport comparison reuses the same 24 dated observations and time matching as the earlier diagnostic. All three resolutions have identical first-layer airport winds within each configuration, so refinement does not change that station diagnostic. It is a secondary check at one station, not a selection criterion for all communities. See the downloadable table for bias, speed/vector error and eligible direction counts.
The following is the previously selected cell and interval, held fixed before this experiment: zero-based south-to-north row 101, west-to-east column 130, interval 20, at the 1,500 m face. It was not reselected to favour a result.
| Configuration | Layers | Terrain-coordinate W (m/s) | Physical-vertical estimate B (m/s) |
|---|---|---|---|
| Interpolated regional winds | 10 | 0.017861 | 0.187245 |
| Interpolated regional winds | 18 | 0.048878 | 0.252004 |
| Interpolated regional winds | 34 | 0.068166 | 0.292109 |
| Terrain adjustment, 5 km search | 10 | -10.921640 | -10.832249 |
| Terrain adjustment, 5 km search | 18 | -10.858052 | -10.711318 |
| Terrain adjustment, 5 km search | 34 | -10.822567 | -10.621403 |
| Terrain adjustment, 10 km search | 10 | -11.166899 | -11.077473 |
| Terrain adjustment, 10 km search | 18 | -11.153868 | -11.054895 |
| Terrain adjustment, 10 km search | 34 | -11.151501 | -11.022408 |
The complete results also retain all times and original faces at this column, every reference profile, absolute wind distributions, per-height and per-interval comparisons, and the full perimeter separately. The 3,000 m face has no baseline physical-vertical interpolation estimate.
The capacity-only executable reproduced all three original 10-layer files byte for byte. All nine final cases completed 24 intervals; the independent reader checked 541,050,192 finite wind values, headers, dates, original faces and unchanged static fields. Interior continuity checks used the existing 0.00005 m/s arithmetic tolerance. Separate implementations compared the common-height results and maps; these checks establish numerical consistency, not physical validation.
The first attempted 18-layer input stopped in the legacy input reader before any weather interval. Its controls, frozen protocol and failure logs are preserved. A recorded whitespace/delimiter amendment retained all scientific assignments; corrected cases use separate folders. Independent review clarifies that the amendment also moved one harmless group-delimiter layout per refined file. No completed baseline was rerun.
This one-day experiment does not test winter inversions, other synoptic conditions, horizontal grid refinement, independent local vertical observations, current source parameters, pollutant chemistry or community exposure. The marine neutral fallback and upstream weather limitations remain. No configuration is adopted for concentration prediction on the basis of this test.
Machine-readable results: summary, checks, independent checks, airport comparison, fixed-column profiles.