Southern California Edison (SCE) significantly misjudged the potential size of the Eaton Canyon wildfire in Los Angeles earlier this year, underestimating its scale by a factor of ten, according to documents reviewed by Reuters. Despite investing in enhanced computing, datasets, and artificial intelligence, the utility’s internal forecasts failed to anticipate the severity of the January blaze.

As wildfires ravaged both ends of Los Angeles—from Santa Monica to Eaton Canyon—over 34,000 acres (approximately 53 square miles) were burned, reducing entire communities to ash. Although the exact cause of the Eaton Canyon fire remains officially undetermined, multiple lawsuits claim that SCE’s decision to keep some power lines energized in Altadena contributed to the disaster.

SCE has defended the accuracy of its forecasting tools, with Asset Intelligence Director Raymond Fugere stating that while localized wind conditions and fuel sources may not have been fully represented, the models still provide valuable insights. According to an internal SCE forecast, the company predicted that the Eaton Canyon fire, which began on January 7, could burn 1,000 acres in eight hours without intervention. In reality, the blaze consumed roughly 14,000 acres, destroyed around 9,400 buildings, and claimed 17 lives—becoming one of the costliest natural disasters in U.S. history.

Experts pointed out flaws in SCE’s modeling approach. Joseph Mitchell, a scientist and wildfire expert, noted that the utilitys simulations only ran for eight hours post-ignition, missing the most destructive phases of the fire. Meanwhile, Stanford Law School’s Michael Wara argued that the models were more suited to predicting wildfires in natural landscapes than in urban environments, where fire spreads from house to house. SCE acknowledged the limitations of its current system and is now exploring 24-hour simulation models to better capture extreme events. In a May 16 regulatory filing, the utility said the January wildfires exposed key challenges in forecasting fire behavior in urban areas.

The Eaton Canyon and Palisades fires together destroyed over 16,000 structures and accounted for the bulk of an estimated US$250 billion in damages, according to AccuWeather. Notably, the Palisades fire, which also started on January 7, far exceeded SCE’s 1,000-acre forecast, ultimately scorching over 23,000 acres, killing 12 people, and leveling nearly 7,000 buildings.

SCEs fire modeling capabilities have been under development since Governor Gavin Newsom’s 2019 “Wildfire Innovation Sprint,” which aimed to enhance disaster prediction using AI. The utility now operates four supercomputing clusters that generate 13 billion simulations across 400 weather patterns and 29 million ignition points. It also uses forecasting tools developed by Technosylva, a California-based company that received state funding to assist emergency response planning. Technosylva’s CEO claimed their models accurately predicted the fire’s magnitude five days ahead, enabling better preparation. In response to these shortcomings, SCE plans to invest US$8 million this year to further improve its fire science and modeling—up from US$2 million in 2018.

Disclaimer: This image is taken from Reuters.