NOAA hourly precipitation dataset (HPD)
Finding
Step at the 2012 digital upgrade (~+9–10% in hourly maxima)
| Verdict | Real error (instrument step) |
|---|---|
| Significance | High (Size 2 · Reach 2 · Consequence 1 = 5 of 6) |
| Outcome | Published |
| Where reported | Zenodo preprint Our write-up on Zenodo (latest version): doi:10.5281/zenodo.24327949 |
| Date | 2026-11-17 |
Summary
NCEI's COOP Hourly Precipitation Data (HPD v2) comes from Fischer & Porter weighing gauges that were converted from punched-tape recorders to digital loggers, mostly in 2012. At each station's conversion, hourly maxima jump by about 9–10% relative to the co-located daily gauge, which shows no change. The step is not flagged in the dataset, and it is as large as the trends that sub-daily studies report.
What we found
- The conversion date of each station can be read from HPD's own source flags. Of the stations converted, 123 switched in 2012; the digital share of station-hours rises from 7% (2007) to 89% (2013).
- Legacy values are mostly whole tenths of an inch; digital values are in hundredths.
- Step at the switch (±6 years, station fixed effects, against co-located GHCN-Daily totals at 114 stations):
| Series | Step at conversion | p |
|---|---|---|
| GHCN-Daily annual maximum day (independent gauge) | +0.3% | 0.90 |
| HPD annual maximum hour ÷ GHCN max day, raw 0.01 in | +9.1% | 0.014 |
| Same, with legacy-resolution emulation (cumulative 0.1 in) | +9.9% | 0.008 |
| HPD maximum daily sum ÷ GHCN max day | +5.3% | 0.006 |
The preprint also reports the step in the 15-minute data and its effect on the frequency of heavy hours.
Who is affected
Any analysis that uses HPD hourly or 15-minute data across the conversion years: sub-daily trend and scaling studies, non-stationary intensity–duration–frequency work, and NOAA Atlas 15. How much a study is affected depends on how it handles resolution: coarsening each hour to 0.1 in can hide the step, while native or cumulative resolution keeps it.
Response
The NOAA Atlas 15 team was told on 17 November 2026; no reply yet. The preprint has been revised once at an author's request (version 3 corrects how one paper's data screening is described).
What users can do
Locate each station's conversion date from the source flags (provided with the preprint code) and test for, or adjust, a step at that date. Compare with co-located daily gauges.
Reproduce
Preprint and code: doi:10.5281/zenodo.24406978 (version 3; all versions: 10.5281/zenodo.24327949). Data: HPD v2 from NCEI (doi:10.25921/p7j8-2170) and GHCN-Daily.
Discussion
Questions and discussion are welcome: ask other readers, the authors or the maintainer. Authors and developers are especially welcome to reply. Comments use GitHub Discussions (sign in with GitHub). No GitHub account? Write to us.