# Daily Temperature and Precipitation for 210 Long-Record US Cities, 1851-2018: GHCN-Daily ThreadEx Training Set (CMU Protostar Challenge)

## Facts

- Provider: Yuchuan Lai, David Dzombak (original data); CMU Open Source Programs Office (challenge cut)
- License: CC0-1.0 (https://api.figshare.com/v2/articles/7890488)
- Size: 6 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/CMU-OSPO/protostar-climate-train-to-2018
- Access: public, readable without an account
- Category: Climate & Weather
- Source: https://huggingface.co/datasets/CMU-OSPO/protostar-climate-train-to-2018
- Page: https://scigantic.com/archive/ad7053aa-9e4d-5f98-bdcd-d6e002f4483d
- Keywords: Climate & Weather, Parquet, Hugging Face

## About

Daily maximum and minimum temperature and daily precipitation for 210 US cities with climate records starting between 1851 and 1899, plus yearly averages, totals and extreme climate indices, cut at 2018-12-31 as the training set for Carnegie Mellon University's Protostar Challenge on climate risk forecasting (teams forecast 2019 to 2025). Each city record is the threaded combination of the weather stations that served it over time (ThreadEx), named by the city's current GHCN station ID, with observations from NOAA's Global Historical Climatology Network Daily (GHCN-D) via the Applied Climate Information System. Four Parquet subsets: daily (10.2 million city-days, tmax and tmin in degrees F, prcp in inches), annual_indices (28,052 city-years with average and extreme temperature, warm and cold days and nights, 1-day and 5-day maximum precipitation, wet-day precipitation and a cooling degree days column), city_info (stations combined per city) and coverage_and_stations. Loads with the datasets library. Built from KiltHub records by Yuchuan Lai and David Dzombak, DOI 10.1184/R1/7890488.v6 and 10.1184/R1/7961012.v6, both CC0; cite those records.

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("CMU-OSPO/protostar-climate-train-to-2018")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `annual_indices/train-00000-of-00001.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/ad7053aa-9e4d-5f98-bdcd-d6e002f4483d
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/ad7053aa-9e4d-5f98-bdcd-d6e002f4483d
- Over MCP: `get_schema_card` with id `ad7053aa-9e4d-5f98-bdcd-d6e002f4483d` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Climate & Weather datasets](https://scigantic.com/topics/climate-and-weather)
