Dataset Viewer
Auto-converted to Parquet Duplicate
city_id
stringclasses
210 values
year
int64
2.02k
2.03k
metric
stringclasses
4 values
predicted_value
float64
1.7
5.02k
USC00042863
2,019
annual_avg_temp
66.4
USC00042863
2,019
total_precip_in
11.06
USC00042863
2,019
hottest_day_temp
103.89
USC00042863
2,019
cdd
1,571.4
USC00042863
2,020
annual_avg_temp
66.4
USC00042863
2,020
total_precip_in
11.06
USC00042863
2,020
hottest_day_temp
103.89
USC00042863
2,020
cdd
1,571.4
USC00042863
2,021
annual_avg_temp
66.4
USC00042863
2,021
total_precip_in
11.06
USC00042863
2,021
hottest_day_temp
103.89
USC00042863
2,021
cdd
1,571.4
USC00042863
2,022
annual_avg_temp
66.4
USC00042863
2,022
total_precip_in
11.06
USC00042863
2,022
hottest_day_temp
103.89
USC00042863
2,022
cdd
1,571.4
USC00042863
2,023
annual_avg_temp
66.4
USC00042863
2,023
total_precip_in
11.06
USC00042863
2,023
hottest_day_temp
103.89
USC00042863
2,023
cdd
1,571.4
USC00042863
2,024
annual_avg_temp
66.4
USC00042863
2,024
total_precip_in
11.06
USC00042863
2,024
hottest_day_temp
103.89
USC00042863
2,024
cdd
1,571.4
USC00042863
2,025
annual_avg_temp
66.4
USC00042863
2,025
total_precip_in
11.06
USC00042863
2,025
hottest_day_temp
103.89
USC00042863
2,025
cdd
1,571.4
USC00166584
2,019
annual_avg_temp
67.17
USC00166584
2,019
total_precip_in
54.32
USC00166584
2,019
hottest_day_temp
100.91
USC00166584
2,019
cdd
2,726.96
USC00166584
2,020
annual_avg_temp
67.17
USC00166584
2,020
total_precip_in
54.32
USC00166584
2,020
hottest_day_temp
100.91
USC00166584
2,020
cdd
2,726.96
USC00166584
2,021
annual_avg_temp
67.17
USC00166584
2,021
total_precip_in
54.32
USC00166584
2,021
hottest_day_temp
100.91
USC00166584
2,021
cdd
2,726.96
USC00166584
2,022
annual_avg_temp
67.17
USC00166584
2,022
total_precip_in
54.32
USC00166584
2,022
hottest_day_temp
100.91
USC00166584
2,022
cdd
2,726.96
USC00166584
2,023
annual_avg_temp
67.17
USC00166584
2,023
total_precip_in
54.32
USC00166584
2,023
hottest_day_temp
100.91
USC00166584
2,023
cdd
2,726.96
USC00166584
2,024
annual_avg_temp
67.17
USC00166584
2,024
total_precip_in
54.32
USC00166584
2,024
hottest_day_temp
100.91
USC00166584
2,024
cdd
2,726.96
USC00166584
2,025
annual_avg_temp
67.17
USC00166584
2,025
total_precip_in
54.32
USC00166584
2,025
hottest_day_temp
100.91
USC00166584
2,025
cdd
2,726.96
USC00280734
2,019
annual_avg_temp
50.23
USC00280734
2,019
total_precip_in
46.79
USC00280734
2,019
hottest_day_temp
94
USC00280734
2,019
cdd
656.8
USC00280734
2,020
annual_avg_temp
50.23
USC00280734
2,020
total_precip_in
46.79
USC00280734
2,020
hottest_day_temp
94
USC00280734
2,020
cdd
656.8
USC00280734
2,021
annual_avg_temp
50.23
USC00280734
2,021
total_precip_in
46.79
USC00280734
2,021
hottest_day_temp
94
USC00280734
2,021
cdd
656.8
USC00280734
2,022
annual_avg_temp
50.23
USC00280734
2,022
total_precip_in
46.79
USC00280734
2,022
hottest_day_temp
94
USC00280734
2,022
cdd
656.8
USC00280734
2,023
annual_avg_temp
50.23
USC00280734
2,023
total_precip_in
46.79
USC00280734
2,023
hottest_day_temp
94
USC00280734
2,023
cdd
656.8
USC00280734
2,024
annual_avg_temp
50.23
USC00280734
2,024
total_precip_in
46.79
USC00280734
2,024
hottest_day_temp
94
USC00280734
2,024
cdd
656.8
USC00280734
2,025
annual_avg_temp
50.23
USC00280734
2,025
total_precip_in
46.79
USC00280734
2,025
hottest_day_temp
94
USC00280734
2,025
cdd
656.8
USC00286055
2,019
annual_avg_temp
54.43
USC00286055
2,019
total_precip_in
49.59
USC00286055
2,019
hottest_day_temp
98.7
USC00286055
2,019
cdd
1,132.32
USC00286055
2,020
annual_avg_temp
54.43
USC00286055
2,020
total_precip_in
49.59
USC00286055
2,020
hottest_day_temp
98.7
USC00286055
2,020
cdd
1,132.32
USC00286055
2,021
annual_avg_temp
54.43
USC00286055
2,021
total_precip_in
49.59
USC00286055
2,021
hottest_day_temp
98.7
USC00286055
2,021
cdd
1,132.32
USC00286055
2,022
annual_avg_temp
54.43
USC00286055
2,022
total_precip_in
49.59
USC00286055
2,022
hottest_day_temp
98.7
USC00286055
2,022
cdd
1,132.32
End of preview. Expand in Data Studio

Ocarina of Ten: Protostar Climate Challenge Submission

The 10-year mean: the example notebook's model

Negative result: no

A complete example entry for the Protostar Challenge: Climate Risk Intelligence, made by the organizers. It is on the leaderboard as Ocarina of Ten¹ (tagged reference): rerun it, then try to beat it. It follows the submission template, so you can also use it as a filled-in example of every section.

Forecasts of yearly weather for 210 U.S. cities, 2019–2025, made as of the end of 2018 (no data that reveals 2019–2025 weather).

Team

Name Hugging Face username University, Program
Ocarina of Ten CMU-OSPO Carnegie Mellon University, Open Source Programs Office

Method

  • Model: the 10-year average. For each city and metric, every year from 2019 to 2025 gets the average of the city's last 10 valid years (up to 2018). The climate is warming, so the last decade is closer to the next few years than a 30-year average is; NOAA's Climate Prediction Center has long used such short "optimal climate normals" (Huang, van den Dool & Barnston, 1996, Journal of Climate 9:809).
  • Features: none beyond each city's own history; no other cities, no external data.
  • How we checked it: practice run, trained up to 2010 and tested on 2011–2018 with the challenge's scoring: Score 5.5 (skill +0.09 on average temperature, +0.13 on CDD; close to zero or negative on precipitation and hottest day).
  • What it ignores (ways to beat it): the trend within the decade, year-to-year cycles such as El Niño, and what neighboring cities share.

AI Tools and Agents

  • Tools and models: Claude Code (Claude Opus 5.5)
  • What they did: prepared this README and the submission (the notebook code that writes submission.csv)
  • What a person reviewed: the model, the data used (challenge dataset only, up to 2018) and the final file
  • Agent instructions file: none used; the template's code/AGENTS.md is included for teams that start from this example

Policy Idea

  1. What risk did we notice? Cooling degree days rise faster than average temperature: a small warming of the mean pushes many more days above 65 °F, so air-conditioning demand grows faster than the thermometer suggests.
  2. What would we do about it, and who would do it? Utilities and city energy offices plan summer peak capacity on the last 10 years of CDD instead of a 30-year normal, and pair it with cool-roof and tree-canopy programs in the neighborhoods with the highest cooling demand.
  3. Why is it sustainable? Planning on recent conditions avoids under-built capacity and emergency purchases on hot days, and cooling demand that is avoided (shade, reflective roofs) costs less and emits less than demand that is met.

Feedback to the Ecosystem

  • What got in the way: the 355-day completeness rule has to be applied by hand from the "missing days" columns; a ready "valid year" flag in the dataset would prevent silent mistakes.
  • What would fix it: a valid column per metric in annual_indices, documented in the dataset card.

Files

File Content
submission.csv Our forecasts: city_id,year,metric,predicted_value
code/example_challenge_notebook.ipynb The whole entry: practice run, refit, submission.csv, file check
code/check_submission.py The template's check script
code/requirements.txt The package versions used to produce submission.csv
code/AGENTS.md The template's instructions for AI agents (the data rule)
code/LICENSE MIT license for the code
data/ No curated data: the notebook reads the challenge dataset from the Hub (data/README.md)

Metric codes: annual_avg_temp (°F), total_precip_in (inches), hottest_day_temp (°F), cdd (°F-days). City IDs: the station ID from the city_info table of the training dataset (e.g., Pittsburgh = USW00094823).

How to Reproduce

  1. Install the packages: pip install -r code/requirements.txt (Python 3.10 or later).
  2. Run code/example_challenge_notebook.ipynb from the code/ folder, top to bottom. It reads the challenge dataset from the Hub and writes submission.csv to the dataset root.
  3. Check it: python code/check_submission.py.

License

Data and text (this card, submission.csv): CC BY 4.0, as in the header. Code in code/: MIT (code/LICENSE).

Data

Training data: CMU-OSPO/protostar-climate-train-to-2018, from:


¹ Ocarina of Ten is not to be confused with Ocarina of Time, whose Switch 2 remake comes out on November 5. Save the princess after the deadline: in Hyrule, pulling the Master Sword jumps you 7 years ahead; here, you have to forecast them.

Downloads last month
124