Problem
Daily max/min aggregation for tmax/tmin uses UTC-day windows instead of local-time days. For BC (UTC-7 to UTC-8) the local-afternoon tmax peak (~14-16h local, ~22-00h UTC) straddles UTC day boundaries, which produces a small systematic bias vs. a proper local-time aggregation:
- Monthly tmax biased low (afternoon peaks split across two UTC days)
- Monthly tmin biased high (overnight lows split across two UTC days)
The EDH-based Python pipeline (scripts/backfill_edh_tmax_tmin.py) flags this as a known limitation in its docstring (lines ~31-36) but has not yet been fixed.
Context
- The cd package documentation states tmax/tmin as "monthly mean of daily max/min" without specifying the day boundary — the bias is real but probably smaller than other sources of ERA5-Land error for most use cases (valley/alpine contrasts, regional trends).
- Climate departure analysis compares anomalies against a baseline, so the bias partially cancels if the same method is used for baseline and the test period. But it does not cancel for absolute-threshold questions (fire weather, growing-degree days, frost dates).
Proposed solution
Two options, in increasing fidelity and cost:
Option A: Fixed UTC offset (simplest)
Shift valid_time by -8h before .resample("1D") to approximate Pacific time. One-line change. Ignores DST and the eastern-BC switch to MST. Captures ~90% of the improvement.
hourly = hourly.assign_coords(
valid_time=hourly.valid_time - np.timedelta64(8, "h")
)
daily_max = hourly.resample(valid_time="1D").max()Option B: Per-pixel local time
Compute a UTC-offset raster from longitude (or from a time-zone polygon overlay) and apply per-pixel. Correct for eastern BC but more complex and doesn't handle DST.
Recommended path
Option A for the next backfill regeneration. It's cheap, captures most of the correctness, and matches how most Pacific-coast climate analysis handles the hourly to daily transition. Document the simplification in the cd package methodology section.
Tracking
Separate follow-up after EDH migration. CDS-based path (#33) closed obsolete; this is the only remaining tmax/tmin correctness issue.
Problem
Daily max/min aggregation for tmax/tmin uses UTC-day windows instead of local-time days. For BC (UTC-7 to UTC-8) the local-afternoon tmax peak (~14-16h local, ~22-00h UTC) straddles UTC day boundaries, which produces a small systematic bias vs. a proper local-time aggregation:
The EDH-based Python pipeline (
scripts/backfill_edh_tmax_tmin.py) flags this as a known limitation in its docstring (lines ~31-36) but has not yet been fixed.Context
Proposed solution
Two options, in increasing fidelity and cost:
Option A: Fixed UTC offset (simplest)
Shift
valid_timeby -8h before.resample("1D")to approximate Pacific time. One-line change. Ignores DST and the eastern-BC switch to MST. Captures ~90% of the improvement.Option B: Per-pixel local time
Compute a UTC-offset raster from longitude (or from a time-zone polygon overlay) and apply per-pixel. Correct for eastern BC but more complex and doesn't handle DST.
Recommended path
Option A for the next backfill regeneration. It's cheap, captures most of the correctness, and matches how most Pacific-coast climate analysis handles the hourly to daily transition. Document the simplification in the cd package methodology section.
Tracking
Separate follow-up after EDH migration. CDS-based path (#33) closed obsolete; this is the only remaining tmax/tmin correctness issue.