Arimancy/caiso-demand-weather

CAISO Hourly Electricity Demand + Population-Weighted Weather (2015 to 2026)

A model ready hourly time series for California grid load forecasting: CAISO balancing authority demand (MWh) pre joined to population weighted weather, with calendar, holiday, degree hour, irradiance, and cloud cover features. 96,712 hours from 2015-07-01 to 2026-07-12, no missing hours, every imputed or preliminary value explicitly flagged, in Parquet and CSV with an easy chronological split.

The seasonal chart above is the dataset's signature. In the spring panel, metered midday demand sinks deeper each year from 2016 through 2022 as behind the meter rooftop solar grows: the midday belly bottoms out at 0.844 of the morning and evening shoulders in spring 2022. From 2023 the belly refills, reaching 1.014 of the shoulders by 2026, while overnight demand barely moves. Likely drivers of the refill include grid battery charging, daytime electric vehicle charging, and time of use rates shifting load into the solar hours, alongside a slowdown in new rooftop installations; the dataset records the shape, not the attribution. To be precise about what this is: this is metered demand, not net load. The famous duck curve is net load (demand minus utility scale wind and solar), which this dataset does not contain. What you see here is the behind the meter solar effect visible in demand, and its partial reversal.

Why this dataset

The hard part of load forecasting is not finding demand data, it is the join. This dataset does the error prone work correctly and documents it:

• UTC canonical join. Demand and weather are joined strictly on UTC, which has no DST pathologies. Local time (America/Los_Angeles) ships as a derived column; spring forward days have 23 local hours and fall back days 25, flagged via is_dst_transition_day rather than "fixed."
• Population weighted weather, seven points. ERA5 series at Los Angeles, Riverside, San Diego, San Francisco, San Jose, Sacramento, and Fresno, weighted by Census Vintage 2025 CBSA populations (0.4118, 0.1529, 0.1052, 0.1484, 0.0636, 0.0794, 0.0386). The Bay is deliberately split into coastal San Francisco and inland San Jose, and Riverside carries the Inland Empire cooling load, because California microclimates drive the demand response.
• Solar columns. Population weighted shortwave_radiation_pop_wt_wm2 (W/m2) and cloud_cover_pop_wt_pct (percent), weighted with the identical code as temperature. These serve the behind the meter solar story. In the baseline notebook's staged ablation they add about 0.03 MAPE points on top of calendar, demand history, and temperature features (2.74 to 2.71 percent on the 2025 plus test), a real but small increment; the larger value is the visible solar signal itself.
• Provenance you can filter on. The long backbone is Catalyst Cooperative's PUDL cleaned EIA-930 (anomalies imputed, with reason codes); the freshest roughly 2 months come straight from the EIA API and are flagged demand_source="eia_api" (preliminary, unimputed). Weather gap handling only ever interpolates interior gaps of 3 hours or less, flagged; and the most recent 3 months of ERA5 weather are ERA5T preliminary, flagged weather_is_preliminary.

Footprint

CAISO is roughly 80 percent of California load. Los Angeles Department of Water and Power (LADWP), SMUD (in BANC), IID, TID and other municipal balancing authorities are separate BAs whose load is not in this series, while their populations partially overlap the weighted metros (Los Angeles city proper is LADWP; Sacramento is SMUD). The metro weights are retained anyway because regional weather, not exact service territory population, is the demand driver being modeled: SCE load lives in the same LA basin weather as LADWP customers. Territory refined weights are a candidate v2 refinement.

Sources

Data	Source	License
Demand backbone	EIA-930 via Catalyst Cooperative PUDL (out_eia930__hourly_operations)	US public domain; PUDL processing credited below
Demand fresh tail	EIA Open Data API v2, electricity/rto/region-data, respondent CISO	US public domain
Weather	Open-Meteo Historical Weather API, Open-Meteo ERA5 reanalysis, 0.25 degree, models=era5	CC BY 4.0
Population weights	US Census Bureau CBSA estimates, Vintage 2025	US public domain

Methodology summary

Hourly CISO demand (hour ending UTC) from PUDL's cleaned parquet, topped up past PUDL's lag with the EIA API, deduplicated preferring PUDL. Open-Meteo ERA5 reanalysis (0.25 degree, models=era5) 2 m temperature, shortwave radiation, and cloud cover pulled at each of the seven grid points in GMT, each variable combined as X(t) equals the weight sum of the per point series with fixed Vintage 2025 population weights (a weighted value is NaN unless all seven points report). Left join on the UTC hour, reindexed to the complete hourly range, gaps flagged, short interior gaps (3 hours or less) linearly interpolated and flagged. Calendar features computed from local wall clock time; holidays ship as statutory and observed flags for both the US federal and the California calendar (is_holiday_statutory, is_holiday_observed, is_holiday_ca_statutory, is_holiday_ca_observed; observed marks the proclaimed weekday when a holiday falls on a weekend and carries the weekday demand effect); degree features are hourly degree hours and daily degree days, base 65 F, from the weighted temperature, with the daily degree days null on any local day without complete temperature coverage. Full column by column reference: see the data dictionary.

Files

• caiso_demand_weather_hourly_v1.parquet: recommended; typed, compressed, timezones survive round trip.
• caiso_demand_weather_hourly_v1.csv: universal access; timestamps are ISO 8601 with explicit offset.

Known limitations

• The most recent roughly 2 months of demand are preliminary EIA data (demand_source="eia_api", unimputed). Weather is ERA5 reanalysis (models=era5, 0.25 degree); values within the most recent 3 months are ERA5T preliminary and may be revised by ECMWF per Copernicus's 2 to 3 month finalization schedule, so they carry the weather_is_preliminary flag. ERA5T and final ERA5 have diverged only twice on record (both regional snow corrections), so the effect is small, but the rows are marked so you can filter them. The export is also trimmed a conservative few days below the latest ERA5T, whose freshest hours can carry missing timesteps.
• ERA5 is a reanalysis (model) estimate, not station observations.
• CAISO covers roughly 80 percent of California load; municipal BAs are excluded (see Footprint above).
• Population weights are static (Vintage 2025).
• A backtest on this data uses actual weather; do not compare its error head to head with CAISO's live day ahead forecast error, which uses live weather forecasts and analyst correction.

License and attribution

This dataset is published by Arimancy LLC under CC BY 4.0. The full license text and the attribution target also ship in the package as LICENSE.txt.

• EIA-930 demand data: U.S. Energy Information Administration, Form EIA-930 / Hourly Electric Grid Monitor (US public domain). Source: U.S. Energy Information Administration (Jul 2026).
• PUDL: cleaned and imputed EIA-930 demand via Catalyst Cooperative's Public Utility Data Liberation (PUDL) project. PUDL's data and documentation are published under CC-BY-4.0.
• Weather data by Open-Meteo.com (CC BY 4.0). ERA5: Hersbach et al. (2023), ECMWF, doi:10.24381/cds.adbb2d47.
• Population estimates: U.S. Census Bureau (US public domain).

Built and maintained by Arimancy (Arimancy LLC). Dataset page: arimancy.com/datasets/caiso-hourly-demand-weather. Also available on Kaggle and Hugging Face. Column-by-column reference: the included data_dictionary.md.