The large vertical-wind discrepancy remains, and no new CALMET configuration is adopted. Refining CALMET from 10 to 18 to 34 layers does not remove the previously identified downward excursion of about 11 m/s. Mean absolute disagreement with the upstream native model increases in all three configuration families on the same supported samples. These are differences between models sharing weather inputs; they do not measure error against observations or establish accurate vertical transport.
This historical diagnostic covers the 24 intervals of 1 June 2023 in fixed PST (UTC−8). It uses the 24 native states actually consumed, from 09:00 UTC on 1 June to 08:00 UTC on 2 June. Saved CALMET timestamps label interval starts; the native states correspond to forcing endpoints. All nine configurations retain the same 200 m horizontal grid, terrain and weather.
The reference comes from the retained British Columbia d06 WRF-derived M3D archive, with 40 mass levels on a roughly 1.333 km native grid. WRF’s W is physical vertical velocity. CALWRF averages it from adjacent vertical interfaces to mass levels and writes it at 0.01 m/s precision; its released conversion also includes QA replacement and near-zero cleanup. A retained check of one native tile reproduces 79,200 printed W and height values across these 24 times and the lower 33 levels; other tiles and the upper seven levels were not independently crossmatched. The available NetCDF tile files were truncated to the first 33 mass levels, rather than representing an unrelated 33-level grid. WRF variable definitions, released CALWRF source, BC native tile.
The comparison covers the original 101 × 101 cells, all 13 fixed reference points and the previously selected excursion column. It compares nine faces at 20, 40, 80, 160, 300, 600, 1,000, 1,500 and 2,200 m above CALMET terrain. At each target absolute-height proxy, native W is interpolated vertically in the same four native columns used by CALMET, then combined with their captured weights. All four columns must bracket the height. There is no extrapolation, replacement of missing neighbours or renormalization.
Exactly 1,421,596 of 2,203,416 cell/time/height samples (64.52%) have full support, identically for every configuration. All 781,820 exclusions are below at least one native first mass level; 758,520 are also below at least one native terrain surface. None is above the native top. Coverage is 17.00% at 20 m, 20.29% at 40 m, 28.71% at 80 m, 45.74% at 160 m, 70.53% at 300 m, 98.40% at 600 m, and complete at 1,000–2,200 m. The results describe this supported subset, with substantial limits near the ground.
CALMET’s saved W follows terrain coordinates. The primary physical estimate, B, adds terrain advection using horizontal winds interpolated to each face. On the common support, native physical W ranges from −1.98072 to +2.25456 m/s. The following differences are CALMET B minus native, in m/s. Coordinate explanation.
| Configuration | Layers | Mean absolute difference | P95 absolute difference | Maximum absolute difference |
|---|---|---|---|---|
| Adjustments off | 10 | 0.528770 | 1.829813 | 10.925514 |
| Adjustments off | 18 | 0.549300 | 1.918410 | 11.069882 |
| Adjustments off | 34 | 0.556263 | 1.946176 | 11.096438 |
| Terrain, 5 km search | 10 | 0.613157 | 2.017269 | 10.828961 |
| Terrain, 5 km search | 18 | 0.630904 | 2.088618 | 10.708030 |
| Terrain, 5 km search | 34 | 0.636897 | 2.110847 | 10.683701 |
| Terrain, 10 km search | 10 | 0.669893 | 2.159998 | 11.074185 |
| Terrain, 10 km search | 18 | 0.684770 | 2.222779 | 11.051607 |
| Terrain, 10 km search | 34 | 0.690326 | 2.244486 | 11.167936 |
The fixed excursion is at grid-cell centre UTM zone 10N 490100, 5505900 m, the 1,500 m face above model terrain, and the interval starting 20:00 fixed PST. Native W there is −0.00329 m/s, effectively near zero at its printing precision. All 216 time/height samples at this column have full support.
| Configuration | CALMET B, 10 layers | CALMET B, 18 layers | CALMET B, 34 layers |
|---|---|---|---|
| Adjustments off | +0.18724 | +0.25200 | +0.29211 |
| Terrain, 5 km search | -10.83225 | -10.71132 | -10.62140 |
| Terrain, 10 km search | -11.07747 | -11.05489 | -11.02241 |
Individual extremes can move. In the 34-layer 10 km case, the largest supported discrepancy is 11.16794 m/s at 2,200 m in the same column and interval. The adjustments-off runs also retain large discrepancies elsewhere. A smaller value at the fixed 1,500 m sample therefore does not establish improvement throughout the area.
Height matching is approximate: native heights are rounded geopotential heights from (PH+PHB)/9.81, while CALMET terrain plus AGL is a model elevation proxy. Gravity/datum differences, terrain and grid resolution, horizontal/vertical interpolation, hourly-state versus interval timing, different flow equations, and native conversion precision remain limitations. Provider conversion logs were not recovered; released-source semantics and the limited tile crossmatch do not prove the exact production executable. An alternative using horizontal wind from the layer below the face also retains the excursion; its offset shrinks as layers become thinner. WRF height documentation.
Independent arithmetic reproduces all 27,216 retained CALMET B reference-profile values exactly. This establishes calculation and alignment consistency; physical accuracy remains unresolved. Frozen protocols, hashes, support counts, alternative calculations, sign cohorts and all 13 reference summaries are retained in the machine-readable summary and checks. This wind comparison contains no pollutant-release calculation, and no new CALMET wind is adopted for viewer or plume forcing.
The next question is what produces the persistent excursion in the diagnostic flow calculation, and whether suitable independent local wind profiles support it. Controlled checks of the horizontal divergence, terrain and boundary assumptions can address its origin. Another increase in layer count alone would not establish realistic local vertical transport.