# Actual NOAA daily station observations: Central Park, 2024

This is an intentionally small, actual government-data example, **not simulated
weather**. It contains NOAA NCEI GHCN-Daily station `USW00094728`, named
`NY CITY CNTRL PARK` in the preserved station metadata, for January 1 through
December 31, 2024. All 366 leap-year calendar days are represented.

Source: [NOAA NCEI GHCN-Daily](https://www.ncei.noaa.gov/products/land-based-station/global-historical-climatology-network-daily).
Credit Menne, M. J. and colleagues, *Global Historical Climatology Network—Daily,
Version 3*, NOAA National Climatic Data Center, [dataset DOI](https://doi.org/10.7289/V5D21VHZ),
accessed 2026-10-23. The accompanying version marker reports GHCN-Daily 3.35,
update `3.35-upd-2026102117`, full reprocessing `3.35-por-2026101623`.
The methods overview is [Menne et al. (2012)](https://doi.org/10.1175/JTECH-D-11-00103.1).
NOAA does not endorse Aner or this analysis.

`manifest.json` records the exact original download URLs, download time, byte
counts, SHA-256 hashes, HTTP last-modified values, source selection, conversion
rules and independently computed checks. Upstream files and the version marker
were retrieved separately, not as an atomic archive transaction; their individual
identities are retained. The full original station file is available from
[NOAA](https://www.ncei.noaa.gov/pub/data/ghcn/daily/all/USW00094728.dly).
NCEI can revise historical values; this local snapshot deliberately stays fixed.

Preserved inputs are `source-months.dly` (36 original fixed-width monthly lines,
three elements across twelve months), the original `station.txt` line and the
retrieved `ghcnd-version.txt`. The normalized `observations.csv` was generated by
[the offline converter](../../../tools/import_climate_noaa.py), not manually
transcribed. Its SHA-256 is
`96d02e4aa0e083e50e5283ba6b06fc95d15992c666e78496e3aff45351a211e7`.

The [official format documentation](https://www.ncei.noaa.gov/pub/data/ghcn/daily/readme.txt)
defines TMAX/TMIN in tenths of degrees Celsius, PRCP in tenths of millimetres,
missing value -9999 and the accompanying measurement/quality/source flags.
The converter scales by ten, maps missing values to `NA`, preserves flags,
excludes noncalendar padding slots and derives date/observation-ID/day-of-year
labels. No quality filtering, missing-value filling or timezone interpretation is
performed during conversion. Blank flag characters become present empty Strings.
Dates remain calendar labels, rather than invented midnight UTC instants.

Verify the normalized snapshot entirely offline from the project root:

```sh
python3 tools/import_climate_noaa.py \
    examples/fixtures/climate_noaa_2024/source-months.dly \
    examples/fixtures/climate_noaa_2024/observations.csv --check
```

This snapshot has no missing TMAX/TMIN/PRCP values and no nonblank quality flags.
It has 44 trace-precipitation flags, all with a recorded numeric total of zero;
a trace is not silently reclassified as a dry day. It has 125 days with positive
recorded precipitation and 39 days whose recorded daily maximum is at least
30 degrees Celsius. This threshold is illustrative, not a heatwave definition.

The data are a deliberate single-station/single-year/three-element subset of the
upstream archive. `origin_complete` records whether Aner consumed its supplied
source ancestry, not whether an analyst acquired every NOAA station/year/element.
One year is insufficient to establish a climate trend, a climatological normal,
causation or representative city-wide exposure. Preserve station history,
observation-window semantics and appropriate long records for those questions.
NOAA's [open-data policy](https://www.ncei.noaa.gov/sites/default/files/2023-12/NCEI%20PD-10-2-02%20-%20Open%20Data%20Policy%20Signed.pdf)
describes public access to NOAA-produced environmental observations. Retain
source credit and identify this normalized example as an Aner-derived subset.