A study covering all the counties in the contiguous United States quantifies the losses incurred by planning three to five days in advance, and shows that early and comprehensive deployment remains more effective than late reinforcement.
After a storm, U.S. utilities rely on outside self-help teams to restore power. Requesting them early provides more, but relying on less reliable forecasts. Researchers quantified this dilemma by using operational forecasts from two AI-based weather models, GraphCast and FourCastNet, to anticipate outages 24 to 120 hours in advance in contiguous U.S. counties, and then measuring how much of the outages were actually covered by the counties where the crews would have been pre-positioned.
This share decreases with the deadline, well between three and five days. However, an early deployment with all the requested teams outweighs a late deployment with fewer teams, and the study calculates the availability rate that tilts the arbitration. The authors also warn against a methodological bias: area indicators calculated on the date of validity of the forecast include the weather for the forecast period, which artificially inflates their performance by 42 to 53%. Finally, the prediction intervals covered only 47.6% of tropical cyclone episodes, compared to the 90% targeted.
Read the full article on the original source
Go to the source →