ℹ
Context — 12Z time-definition matters: ASOS/COOP daily tmax uses a midnight-to-midnight window; PRISM uses 24h ending at 12Z UTC. This mismatch inflates ASOS-vs-PRISM tmax MAE to ~5°F. PEMN stations rebuilt with a 12Z-aligned daily file confirm this: tmax MAE drops to 0.67°F with proper alignment. The pairwise PRISM-to-PRISM statistics below are unaffected by this issue. See PEMN comparison dashboard for the full 12Z analysis.
Resolution effect · Tmax MAD
0.38°F
800m vs 4km · 6-yr network mean
Interpolation effect · Tmax MAD
0.06°F
ON vs OFF at 800m · negligible
Group 1 coverage
Complete
4 combos × 6 years · 24/24 files
Max station elev mismatch
475 ft
STVP1 · 800m=1194 vs 4km=1669 ft
Full 6-Year Pairwise Differences · 513,571 station-days
| Comparison | Isolates |
ppt (in) | tmin (°F) | tmean (°F) | tmax (°F) | tdmean (°F) | Verdict |
Elevation Assignment · 241 Stations (2020-01-01 snapshot)
| Comparison | Mean (ft) | MAD (ft) | Max (ft) |
|---|
| 800m ON vs 800m OFF | +4.5 | 14 | 110 |
| 800m ON vs 4km ON | −35.2 | 90 | 475 |
| 4km ON vs 4km OFF | +3.7 | 37 | 201 |
Interpolation toggle barely shifts assigned elevation (±4–5 ft mean). Resolution is the driver — 90 ft MAD across 241 stations for 800m vs 4km. 149/241 stations are warmer at 800m (DEM assigns lower elevation); 89 are cooler. Network-wide temperature is +0.12°F warmer at 800m than 4km.
Per-Year Resolution Effect (800m ON vs 4km ON)
| Year | Status | Tmin MAD | Bias | Tmax MAD | Bias | N |
|---|
| 2020 | stable | 0.321 | +0.087 | 0.356 | +0.115 | 88,206 |
| 2021 | stable | 0.425 | +0.127 | 0.421 | +0.156 | 87,965 |
| 2022 | stable | 0.365 | +0.113 | 0.361 | +0.119 | 87,965 |
| 2023 | stable | 0.345 | +0.105 | 0.367 | +0.121 | 87,965 |
| 2024 | stable | 0.366 | +0.098 | 0.403 | +0.104 | 88,206 |
| 2025 | provisional | 0.349 | +0.108 | 0.395 | +0.089 | 73,264 |
2021 is the outlier year — highest resolution MAD (0.42–0.43°F) vs 0.32–0.37°F other years. No trend across 6 years. Precipitation MAD flat at 0.004–0.005 in every year regardless of combo.
Recommended Combo by Use Case
| Use Case | Recommended | Reason |
|---|
| RFSI training feature (all temperature vars) | 800m ON | Best or tied for tmax, tdmean; DEM-aware blend most physically appropriate |
| Precipitation | Any — all identical | Zero practical difference across all 4 combos and both resolutions |
| Resolution bias / lapse-rate correction | 800m ON − 4km ON | Per-station delta captures DEM mismatch — use as RFSI covariate feature |
| Strict grid-cell audit | 800m OFF | No blending ambiguity — exact cell value, cleanest for lookup tables |
| ERA5 background alignment | 4km ON | 4km matches ERA5 downscaling resolution; marginally best tmean |
| 12Z-aligned obs comparison | 800m ON | PRISM uses 12Z definition — with 12Z PEMN obs, tmax MAE = 0.67°F vs 5.05°F (midnight) |
⚠
ASOS/COOP tmax ~5°F MAE is a time-definition artifact, not a PRISM error. ASOS reports tmax over the midnight-to-midnight local calendar day. PRISM computes daily extremes over the 24h ending at 12Z UTC (≈7–8 AM EDT). A warm afternoon on Day N falls in PRISM's Day N record, but in ASOS's Day N record only if it occurs before midnight. This affects tmax most severely, tmin less so. The fix: use 12Z-aligned obs (like the corrected PEMN file) — tmax MAE drops from 5.05°F to 0.67°F. All ASOS tmax vs PRISM comparisons should carry this caveat.