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
3
5.24k
USC00042863
2,019
annual_avg_temp
66.51691
USC00042863
2,019
total_precip_in
13.994677
USC00042863
2,019
hottest_day_temp
104.014471
USC00042863
2,019
cdd
1,698.586156
USC00042863
2,020
annual_avg_temp
66.560645
USC00042863
2,020
total_precip_in
13.873714
USC00042863
2,020
hottest_day_temp
104.01563
USC00042863
2,020
cdd
1,710.760372
USC00042863
2,021
annual_avg_temp
66.619448
USC00042863
2,021
total_precip_in
13.340707
USC00042863
2,021
hottest_day_temp
104.01808
USC00042863
2,021
cdd
1,722.657533
USC00042863
2,022
annual_avg_temp
66.662716
USC00042863
2,022
total_precip_in
13.262757
USC00042863
2,022
hottest_day_temp
104.018268
USC00042863
2,022
cdd
1,728.285942
USC00042863
2,023
annual_avg_temp
66.716132
USC00042863
2,023
total_precip_in
13.262221
USC00042863
2,023
hottest_day_temp
104.018705
USC00042863
2,023
cdd
1,737.027726
USC00042863
2,024
annual_avg_temp
66.758604
USC00042863
2,024
total_precip_in
13.261859
USC00042863
2,024
hottest_day_temp
104.018933
USC00042863
2,024
cdd
1,741.747924
USC00042863
2,025
annual_avg_temp
66.785373
USC00042863
2,025
total_precip_in
13.261689
USC00042863
2,025
hottest_day_temp
104.018968
USC00042863
2,025
cdd
1,742.726187
USC00166584
2,019
annual_avg_temp
67.36093
USC00166584
2,019
total_precip_in
56.47907
USC00166584
2,019
hottest_day_temp
100.537603
USC00166584
2,019
cdd
2,748.573339
USC00166584
2,020
annual_avg_temp
67.398765
USC00166584
2,020
total_precip_in
55.270561
USC00166584
2,020
hottest_day_temp
100.54325
USC00166584
2,020
cdd
2,758.509118
USC00166584
2,021
annual_avg_temp
67.444514
USC00166584
2,021
total_precip_in
55.622086
USC00166584
2,021
hottest_day_temp
100.544465
USC00166584
2,021
cdd
2,767.427698
USC00166584
2,022
annual_avg_temp
67.487328
USC00166584
2,022
total_precip_in
55.666078
USC00166584
2,022
hottest_day_temp
100.541677
USC00166584
2,022
cdd
2,775.416037
USC00166584
2,023
annual_avg_temp
67.530097
USC00166584
2,023
total_precip_in
55.667396
USC00166584
2,023
hottest_day_temp
100.543001
USC00166584
2,023
cdd
2,782.019151
USC00166584
2,024
annual_avg_temp
67.565346
USC00166584
2,024
total_precip_in
55.668958
USC00166584
2,024
hottest_day_temp
100.543781
USC00166584
2,024
cdd
2,785.885153
USC00166584
2,025
annual_avg_temp
67.5897
USC00166584
2,025
total_precip_in
55.670428
USC00166584
2,025
hottest_day_temp
100.544062
USC00166584
2,025
cdd
2,787.233422
USC00280734
2,019
annual_avg_temp
51.419606
USC00280734
2,019
total_precip_in
45.363023
USC00280734
2,019
hottest_day_temp
93.859503
USC00280734
2,019
cdd
783.740318
USC00280734
2,020
annual_avg_temp
51.455265
USC00280734
2,020
total_precip_in
45.160228
USC00280734
2,020
hottest_day_temp
93.859503
USC00280734
2,020
cdd
787.772952
USC00280734
2,021
annual_avg_temp
51.504654
USC00280734
2,021
total_precip_in
45.353394
USC00280734
2,021
hottest_day_temp
93.859503
USC00280734
2,021
cdd
790.99906
USC00280734
2,022
annual_avg_temp
51.552066
USC00280734
2,022
total_precip_in
45.317186
USC00280734
2,022
hottest_day_temp
93.859356
USC00280734
2,022
cdd
794.120026
USC00280734
2,023
annual_avg_temp
51.598251
USC00280734
2,023
total_precip_in
45.317186
USC00280734
2,023
hottest_day_temp
93.859356
USC00280734
2,023
cdd
796.184735
USC00280734
2,024
annual_avg_temp
51.635539
USC00280734
2,024
total_precip_in
45.317186
USC00280734
2,024
hottest_day_temp
93.859356
USC00280734
2,024
cdd
797.836502
USC00280734
2,025
annual_avg_temp
51.65941
USC00280734
2,025
total_precip_in
45.317186
USC00280734
2,025
hottest_day_temp
93.859356
USC00280734
2,025
cdd
799.157915
USC00286055
2,019
annual_avg_temp
54.575894
USC00286055
2,019
total_precip_in
49.059895
USC00286055
2,019
hottest_day_temp
97.770068
USC00286055
2,019
cdd
1,123.023232
USC00286055
2,020
annual_avg_temp
54.614266
USC00286055
2,020
total_precip_in
48.996426
USC00286055
2,020
hottest_day_temp
97.773465
USC00286055
2,020
cdd
1,128.374016
USC00286055
2,021
annual_avg_temp
54.665853
USC00286055
2,021
total_precip_in
49.121054
USC00286055
2,021
hottest_day_temp
97.77684
USC00286055
2,021
cdd
1,133.414105
USC00286055
2,022
annual_avg_temp
54.71529
USC00286055
2,022
total_precip_in
49.13056
USC00286055
2,022
hottest_day_temp
97.772364
USC00286055
2,022
cdd
1,138.614236
End of preview. Expand in Data Studio

HeatReady v10 | Team Cloudbreak

Team Cloudbreak · Premices Irakoze · Carnegie Mellon University

This retrospective edition forecasts four annual weather measures for 210 U.S. cities over 2019–2025. Cloudbreak developed v10 after the challenge closed. The organizers scored v7. V10 has no official score.

Reproduce the forecast

Extract the ZIP. In a Python 3.12 Jupyter environment, install the packages listed in code/requirements.txt. Open code/HeatReady_v10.ipynb and run all cells in order from the root or the code directory. The notebook reads only local files, writes submission.csv at the package root and writes historical evaluation tables into results/. Run python code/check_submission.py from the root to check the grid. The expected SHA-256 of submission.csv is 8f2ee7587f37a49c65eb404e8cfa8a9fe53264d4244a83eba768d5038c17e4b2.

Forecasting method

V10 starts with v8 and compares each city with nearby stations whose weather changed similarly over the past 50 years. It estimates the city’s usual departure from those neighbors and gives more weight to recent departures. Another estimate uses the other weather measures, and a third adjusts for the difference between the historical mean and median. Completed earlier forecasts determine the weights for years 1–3 and 4–7 separately. V8 always receives at least half the weight.

Historical evaluation

Nonoverlapping verification periods v5 v7 v8 v10 v10 wins / ties / losses against v5
23 three-year blocks, 1950–2018 3.71 4.02 4.10 4.19 17 / 6 / 0
10 seven-year blocks, 1949–2018 3.57 3.87 3.88 3.98 7 / 3 / 0

Each period uses the official score on identical eligible rows. In a separate check of all 64 overlapping seven-year starts covering 1949–2018, v10 beat v5 in 51 and tied in 13. Across 68 overlapping three-year starts, the largest decline versus v5 was 0.30 points in 2015–2017. The periods share climate regimes and training histories. Earlier model development used some of the verification years. Challenge rules and scoring

Policy idea

Screen Pittsburgh public buildings for cooling upgrades. At Pittsburgh International Airport, v10 projects an average of 912.8 cooling degree days per year over 2019–2025, compared with the eligible 1989–2018 average of 811.2. The city could begin with libraries and community centers, recording indoor temperatures on hot afternoons, hours of cooling access and equipment condition. At buildings that repeatedly overheat, facilities staff could compare shading, insulation and efficient cooling systems using lifetime cost and emissions.

Feedback to the ecosystem

Per-metric eligibility flags would make it easier to reproduce which city-years count. Station histories linked to municipal boundaries would help planners interpret local forecasts. For external climate indices, recording both the last observation and its publication date would make historical replays easier to audit.

AI tools and agents

Premices supplied the team's prior forecasts and evaluation criteria. ChatGPT/Codex helped implement and compare candidates, check the date cutoffs, reproduce historical benchmarks, and verify the final CSV. The notebook computes every forecast from the packaged inputs. Cloudbreak used the leaderboard to verify competition status and the scored commit. No 2019–2025 weather observations were retrieved or used to fit or select v10.

Attribution and prior exposure

The historical station data originate from the Lai and Dzombak CMU collections. The data card identifies the frozen inputs and their sources. The earlier Pittsburgh policy comparison viewed 2019–2025 NOAA observations after the original forecast was frozen and before later revisions. That descriptive comparison was outside the v10 forecasting and evaluation calculations. Original forecast snapshot · Data card

Downloads last month
235