SCE wildfire models underestimated deadly Los Angeles fires by a factor of ten

Southern California Edison’s upgraded wildfire models misjudged the Eaton Canyon fire’s scale by tenfold, raising concerns over its disaster preparedness as the blaze became one of the most destructive in U.S. history.

Southern California Edison’s (SCE) internal wildfire forecasts vastly underestimated the scale of the January Eaton Canyon fire in Los Angeles, predicting only 1,000 acres would be affected—while the blaze ultimately consumed around 14,000 acres. The discrepancy raises questions over the utility’s fire modeling capabilities, which had been upgraded with AI, advanced computing, and extensive datasets following state-level initiatives launched in 2019.

According to documents reviewed by Reuters, SCE’s miscalculations suggest limitations in how the models simulated fire spread, particularly in urban environments. Despite using 13 billion simulations across millions of ignition points and weather scenarios, SCE’s predictions covered just eight hours of fire activity, whereas the most destructive damage from the Eaton Canyon fire occurred later. The blaze destroyed roughly 9,400 structures and killed 17 people.

Though the exact cause of the Eaton fire remains under investigation, lawsuits have claimed that SCE's decision to keep some power lines active in the Altadena area may have contributed. The utility has defended its fire modeling, with Raymond Fugere, SCE’s asset intelligence director, stating the forecasts remain “actionable” and useful for decision-making. However, he acknowledged that local fuel types and wind variations may have affected simulation accuracy.

Stanford wildfire policy expert Michael Wara criticised the models for being better suited to wildland environments rather than urban settings, noting that fires like Eaton Canyon rapidly transition from vegetation to structure-based conflagrations.

Technosylva, an SCE partner and state-funded firm, reported having forecast the scale of the Los Angeles wildfires five days ahead. Yet SCE’s models also significantly underestimated the size of the concurrent Palisades fire, predicting 1,000 acres when more than 23,000 were ultimately burned.

Together, the two wildfires destroyed over 16,000 structures and resulted in an estimated $250 billion in economic losses. SCE has stated it will review its modeling strategy and consider extending simulations to 24-hour timeframes, though it warned longer projections introduce greater uncertainty.

In filings submitted in May, the company admitted the January 2025 fires have underscored the challenges of modelling fire spread into dense, built environments. SCE plans to invest $8 million this year into improving fire science and modeling—a fourfold increase from 2018.

Copied