Electric power outages from 900k simulated hurricanes in a changing climate, for the United States and Puerto Rico

This dataset is described and explored in Rice et al. 2025, "Projected Increases in Tropical Cyclone-induced U.S. Electric Power Outage Risk", published in Environmental Research Letters (doi.org/10.1088/1748-9326/adad85)

This dataset collects peak outage levels modeled for 900,000 synthetic tropical cyclones (TCs; also commonly known as hurricanes) representative of a modeled historical (1980-2015) and future (2066-2100) period under SSP5-8.5 warming. Synthetic TCs are generated with the Risk Analysis Framework for Tropical Cyclones (RAFT; see Xu et al. 2024 and Balaguru et al. 2023), forced by climate simulation data from the Coupled Model Intercomparison Project phase 6 (CMIP6; see Eyring et al. 2016). Outages are modeled with the newly introduced Electric Power Outages from Cyclone Hazards (EPOCH) model, which was trained on county-level outage data from 23 historical TC events in the EAGLE-I dataset (Brelsford et al. 2024). 

The EPOCH model predicts outages based on county population and the maximum wind speed and rainfall rate experienced during the TC. Predicted outage levels are provided in the form of peak outage fraction: the maximum fraction of electricity customers expected to experience an outage at any one time during the storm's lifetime. Although we do not model outage duration, other research suggests peak outage level is strongly correlated with duration (Jamal and Hasan, 2023).

Data Format

The data is provided in NetCDF4 files, one for each CMIP6 model and time period. Each NetCDF4 files has the following:

Dimensions:

ncounties = 2715. The counties in the study domain

ntracks = 50000. The number of storms

Variables:

int pseudofips(ncounties). The FIPS code for each county. Puerto Rico data is not available at county level, but instead for six utility-defined regions. We assign "pseudo-FIPS" codes to these region starting at 100000

double centroid_lons(ncounties). Longitude of approximate center of county, in the range [-180, 0].

double centroid_lats(ncounties). Latitude of approximate center of county, in the range [0, 90].

float outage_prediction(ntracks, ncounties). The predicted peak outage fraction for each county, for each storm. Due to the particularities of ensemble models, some predictions may be slightly below zero or above one; we clip these values to the range [0,1] before any analysis in our study.

ubyte prediction_complete_flag(ntracks). A verification flag used during dataset generation. This flag should equal 1 everywhere for complete data.

Each file also contains the raw predictors at a county level for every storm, inside the 'predictors' group, for feature analysis.

Also provided for convenience is 'counties_pseudofips.csv', which maps the pseudo-FIPS codes to the the name and spatial extent (WKT format) of each county. It can be read easily by Python GeoPandas, or other software.