qompass/protostar-climate-submission

Qompass AI: Protostar Climate Challenge Submission

Negative result: no

Forecasts of yearly weather for 210 U.S. cities, 2019–2025, made as of the end of 2018 (no data that would not have been available at that time).

This is the staged submission dataset for team Qompass AI in the Protostar Challenge: Climate Risk Intelligence. It was copied from the challenge submission template, CMU-OSPO/protostar-climate-submission-template. This dataset is staged only; it has not been pasted into the challenge Space's Submission tab.

Team

Name	Hugging Face username	Program
Matt Porter	@PhaedrusFlow	Qompass AI

Method

• Model: per-city 30-year climatology baseline. For each city and metric, the forecast for every year 2019–2025 is that city's 30-year climatology: the mean over 1989–2018, quality-screened as described in code/climatology_baseline.py, with fallbacks to all quality-screened years up to 2018, then the last valid value at or before 2018, then the cross-city median.
• Features: the last 30 years of each metric (1989–2018).
• Training data through 2018 only: only training data through 2018 was used. The baseline script includes a leakage fence and refuses to run if any input row is dated after 2018-12-31.
• How we checked it: submission.csv contains 5,880 forecast rows (210 cities x 7 years x 4 metrics) in the template format, and the format can be checked with code/check_submission.py.

Policy Idea

To be completed before any submission.

Feedback to the Ecosystem

To be completed before any submission.

Files

File	Content
submission.csv	Our forecasts: city_id,year,metric,predicted_value
code/	check_submission.py (template format checker) and climatology_baseline.py (the script that generates this baseline)

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 = USW00014762).

How to Reproduce

1. Install pandas and pyarrow.
2. Run code/climatology_baseline.py following the inputs described in its header (training data through 2018 and the template's 5,880-row skeleton) to regenerate submission.csv.

Data

• External data: none beyond the training dataset below.
• Training data: CMU-OSPO/protostar-climate-train-to-2018, years through 2018 only, used for the forecast.
• Lai, Y., & Dzombak, D. A. (2019). Compiled historical daily temperature and precipitation data for 210 U.S. cities.
• Lai, Y., & Dzombak, D. A. (2019). Use of historical data to assess regional climate change. Journal of Climate, 32(14), 4299–4320.