shourya01/ca_power_outages

California Outage & Weather Dataset

Overview

• File: california.nc
• Coverage: January 1, 2019 – December 31, 2020 (15-minute cadence; 70,176 timestamps)
• Hourly aggregate: california_hourly.nc aggregates the same period onto a 1-hour grid (outages summed, weather averaged).
• Spatial scope: 58 California counties (FIPS codes 06001–06115)
• Contents: County-level outage counts, a data-availability mask, and 24 ERA5-derived weather features collocated on the same grid as the outage observations.

This NetCDF file is organized to streamline joint modeling of outages and meteorological drivers at the county level. The layout mirrors common outage-modeling datasets: each data variable is indexed by location (county FIPS) and timestamp, while the weather tensor adds a feature dimension for atmospheric variables.

Coordinates and Dimensions

• location (58): Five-digit FIPS codes for California counties.
• timestamp (70,176): UTC datetimes spaced every 15 minutes.
• feature (24): Names of the weather variables listed below.
• state: Coordinate aligned with location, holding the California state FIPS code ("06").

Data Variables

Variable	Dims	Type	Description
tracked	(location, timestamp)	float64	Derived from the outage file’s availability flag; 1 indicates a real observation at that time/location, 0 indicates zero-padding.
out	(location, timestamp)	float64	Estimated customer outage counts per county.
weather	(location, timestamp, feature)	float64	Weather predictors sampled from ERA5 at each county centroid and broadcast to the 15-minute grid.

Outages

The outage counts were sourced from the EAGLE-I dataset developed by Oak Ridge National Laboratory. Any missing values in the outage counts stems from this upstream dataset.

Weather Feature List

blh, cape, d2m, gust, hcc, ishf, lcc, lsm, max_10si, mcc, pres, pwat, sdlwrf, sdswrf, slhtf, sp, sulwrf, suswrf, t2m, tcc, tcolw, tp, u10, v10.

Each feature is pulled from the ARCO-ERA5 public Zarr archive, subset to a California bounding box, associated with each county’s nearest grid cell, then interpolated from hourly to 15-minute resolution to match the outage cadence. Variables that contained any missing values across the two-year span (cin, orog, siconc, tcoli, vis) were dropped from the final NetCDF to keep every feature fully populated.

Using the Dataset

import xarray as xr

ds = xr.open_dataset("california.nc")
out_san_diego = ds["out"].sel(location="06073")
weather_subset = ds["weather"].sel(location="06073", feature=["t2m", "tp"])

With the shared coordinates, most modeling frameworks can join outages and weather simply by selecting along location and timestamp.

Notes

• File size is approximately 1 GB due to 64-bit precision and two full years of 15-minute data.
• The dataset is self-contained: no external references are needed beyond this repository to interpret the coordinates, variables, and weather feature names.