Data products, code, and reproducibility

Per-format usage

• Unzip the archive — every zip extracts to a single nps-open-climate-data-v1.0.0/ folder:

  unzip nps-open-climate-data-v1.0.0-summary.zip

• Daily CSVs are gzipped — read directly with pandas (auto-detects the .gz extension) or decompress first:

  # in Python
  import pandas as pd
  df = pd.read_csv("daily/yellowstone/yellowstone.csv.gz")
  
  # from the shell
  gunzip daily/yellowstone/yellowstone.csv.gz

• Summary JSONs are plain JSON; load with any standard parser. Schema lives on the Methodology page.

  import json
  summary = json.load(open("summary/yellowstone.json"))
  print(summary["headline_trends"]["tmean_c"])

• GeoJSON boundaries open natively in QGIS, geopandas, Mapbox, Leaflet, etc.

  import geopandas as gpd
  gdf = gpd.read_file("boundaries/yellowstone.geojson")

Programmatic helper (Python)

The Python package ships a small Zenodo download helper. It pulls the archive on first use, caches under ~/.cache/nps_climate_data/, and returns ready- to-use objects:

# Pure-stdlib helper that pulls + caches the Zenodo archives, then
# returns a pandas DataFrame for any park's daily series:
import nps_climate_data as nps

df  = nps.fetch_daily("yellowstone")     # raw daily CSV → DataFrame
sj  = nps.fetch_summary("yellowstone")   # summary JSON → dict
arc = nps.fetch_archive("boundaries")    # downloads + extracts the zip

# Caches under ~/.cache/nps_climate_data so subsequent calls are local.

For agents

Point an AI assistant at /llms.txt for a structured pointer to the dataset, download URLs, Python helpers, schema, and limitations. The Python helper (nps.fetch_daily(slug), etc.) is the simplest path for an agent to actually fetch a park's series — it handles the Zenodo download, gunzip, and parsing in one call. The same information is also exposed as schema.org/Dataset JSON-LD on this page and the home page so structured-data crawlers pick it up too.

Single-park lookups (live site)

The site also serves individual files at predictable URLs — handy for fetching one park without downloading the whole Zenodo archive.

1. Re-derive the value. Re-run the pipeline from raw, or independently compute the variable of interest from DAYMET / ERA5-Land directly, and confirm it matches.
2. Cross-check stations. Compare against NOAA NCEI, NWS, or NPS monitoring records for the specific park and variable.
3. Correct your statistics. Apply autocorrelation handling (Hamed–Rao MK) and multiple- comparisons correction (Benjamini–Hochberg FDR or Bonferroni). The deployed significance flags do neither.
4. Read the variable-specific limitations. See methodology → limitations, especially pet_mm (ERA5-Land overestimate vs FAO Penman–Monteith), high-elevation cold bias, and area- naive multipart averaging.

Cite the methodology page, run your own QC for your use case, and treat the published numbers as a starting point.