On 7 January 2025, a series of devastating wildfires started in Los Angeles. Aided by extensive drought conditions and Santa Ana winds, the fires quickly spiralled out of control. The fires had significant impacts in the wildland–urban interface, where human development meets natural vegetation. The flames devastated thousands of buildings. Lives were tragically lost, and nearly 200,000 residents were forced to evacuate. The Palisades and Eaton fires burned for weeks, overwhelming firefighting efforts. With economic losses projected at $150 billion, this disaster ranks among the costliest in modern United States history.
ECMWF has been producing fire danger forecasts since 2018 as part of the Copernicus Emergency Management Service (CEMS). In recent years, ECMWF has been exploring new approaches to fire forecasting by leveraging advances in machine learning and ECMWF’s own weather prediction model, shifting from predicting fire danger to forecasting fire activity. In this analysis, we evaluate how the new kind of fire forecasting performed during these extraordinary events.
Dry vegetation
To fully understand why the Los Angeles fires were so extreme, we need to rewind to spring 2023. This was the start of an unusually wet period that lasted until late summer 2024, facilitating rapid vegetation growth. This wet period was followed by the exceptionally dry autumn and early winter of 2024 (Figure 1). An alternating pattern of wet and dry conditions – termed ‘hydroclimate whiplash’ – is not new in California but is being amplified by climate change and is likely to become more common. The sequence of wet and dry spells is visible in the Standardised Precipitation Evapotranspiration Index (SPEI), which is based on ECMWF’s ERA5 reanalysis. In southern California, this whiplash effect led to an abundance of dry and very flammable vegetation (Figure 1).
Figure 1: Timeseries of the Standardised Precipitation Evapotranspiration Index (a drought index) with a 6-month accumulation period (SPEI-6) over southern California since January 2022 (left-hand panel), indicating the hydroclimate whiplash effect that led to high fuel load and caused particularly low dead fuel moisture content in some areas around Los Angeles at the beginning of January 2025 (the right-hand panel shows the dead fuel moisture anomaly). The drought index is calculated using ERA5, and the total fuel load and dead fuel moisture anomaly are outputs from ECMWF’s fuel model, SPARKY.
The ECMWF fuel model SPARKY monitors changes in vegetation state in response to weather. The model uses a combination of observations, weather forecasts, and land surface modelling to describe the moisture content of both live and dead vegetation, as well as the abundance of multiple components of that vegetation, including leaf and wood mass. At the time of the fires, fuel was very dry. SPARKY’s prediction of dead fuel moisture content, which is important to identify ignitions, was anomalously low specifically in the areas where the wildfires began.
Real-time monitoring of fuel abundance and moisture content is challenging because of uncertainties in biochemical processes related to vegetation growth. As a result, monitoring of fuel is an active field of research that will greatly benefit from future observations from satellite missions focussing on vegetation (e.g. FLEX, BIOMASS). Such missions will help SPARKY to generate the most accurate fuel characteristics prediction. ECMWF is collaborating with the European Space Agency (ESA) in this field. As fuel status is highly uncertain at the global level, it is important to extract information from these planned missions to pinpoint fuel dryness, which usually sets the stage for extreme fire events like the ones in California.
Weather conditions
Long-term dry conditions were not the only drivers of the 2025 California fires. Between 5 and 8 January 2025, a large high-pressure system formed over the Great Basin, an arid desert region, whilst pressure fell over northwestern Mexico. Strong pressure gradients developed, resulting in a stream of exceptionally dry air, known as Santa Ana winds, rushing towards southern California, creating conditions that enabled the rapid spread of flames.
ECMWF model predictions of the broad-scale weather patterns that led to the Santa Ana winds were accurate up to eight days in advance (Figure 2). Complementing these large-scale predictions, ECMWF’s model forecasts of extreme local wind risk, represented through an Extreme Forecast Index (EFI), also intensified as the devastating events drew near.
Figure 2: Forecast from 12 UTC on 31 December 2024 of the upper-level weather pattern (here the 500 hPa geopotential height from the control run, in decametres) for 12 UTC on 8 January 2025 (left) against the verifying analysis (right).
The Santa Ana winds are driven by a specific weather pattern in the southwestern United States, where the air, already dry from its desert origins, becomes even drier as it descends the mountains, accelerating in certain areas due to topographic interactions. Acting like a massive hairdryer, these winds strip vegetation of its remaining moisture, leaving it highly flammable. Also, once a fire starts, they not only spread the flames, but they also impede even the most aggressive suppression actions.
Fire danger models
Meanwhile, fire-prone weather conditions were forecast days in advance by the local agency and were also well captured by the CEMS Global Wildfire Information System (GWIS) forecast, which uses ECMWF fire danger predictions. Warnings for extreme fire conditions were issued across much of the state, but the challenge remained – how do you pinpoint where fires are most likely to ignite?
Traditional fire danger indices often fail to clearly identify areas at risk of ignition, relying solely on weather conditions to assess fire risk. While they are excellent at highlighting broad regions of concern, they tend to over-predict fire occurrence and fail to provide the precision needed for targeted action. This was the case during the Los Angeles fires, where critical fire weather conditions were flagged across a wide area using traditional fire weather forecasts (EFFIS/GWIS), but these lacked the specificity to identify ignition hotspots. ECMWF's new data-driven fire danger model, the Probability of Fire (PoF), is designed to go beyond using just weather (Figure 3).
Figure 3: Southern California 1-day forecast of fire danger from the Probability of Fire model (left) and the traditional Fire Weather Index, which is the headline metric used in GWIS and many other early warning systems regionally (right) for 7 January 2025. Active fire observations from MODIS (triangles) and VIIRS (circles) for the same day are overlaid.
The cutting-edge PoF model incorporates essential factors like fuel availability and vegetation dryness as provided by the SPARKY model. It also incorporates information on human presence, road density, and detailed maps of the urban–wildland interface – factors that could help to understand fire probability in a complex landscape like California. In this event, the PoF model provided a far more localised and accurate assessment of high fire danger in areas where fires occurred than the Fire Weather Index. By accounting for variables such as the abundance of vegetation and its condition, and the proximity to human activity, the model captured the intricate dynamics that drive fire risk. Of course, fire ignition remains a fundamentally stochastic process, but the level of precision that PoF achieves marks a significant leap forward in fire forecasting, offering better tools to protect lives, homes, and ecosystems.
Impact on atmospheric pollution
As a result of the intense burning, surface concentrations of atmospheric pollutants, including PM2.5, volatile organic compounds, and other toxic air pollutants, surged well above recommended levels. While the strong Santa Ana winds dispersed much of the smoke over the Pacific, significant impacts on local air quality were observed in surface measurements around Los Angeles. The emissions were also monitored by the Copernicus Atmosphere Monitoring Service (CAMS) operated by ECMWF through its Global Fire Assimilation System (GFAS), which reported that pyrogenic emissions from wildfires in California were by far the highest for January in the CAMS 22-year record.
CAMS forecasts are based on fire emissions assuming burning vegetation and are further constrained by satellite observations of the main pollutants in the atmosphere. They do not account for the different emission profiles arising from fires burning in urban areas. Pollutants from these fires may persist in the environment long after the flames have been extinguished. This is an area of ongoing research within the wildfire science community.
Conclusion
The devastating Los Angeles fires highlight the importance of improving the capability of forecasting systems to predict extreme events. In this example, ECMWF's fire danger forecasts were particularly effective in capturing the ‘whiplash’ effect, and they capitalised on accurate predictions of strong Santa Ana winds. This event underscores the need for ongoing innovation in predictive tools to better mitigate the impacts of extreme fires as they are becoming more frequent in a warming climate.
Further information
ECMWF, in conjunction with the UK Met Office, the UK Centre for Ecology & Hydrology, and the UK University of East Anglia, coordinates the annual State of Wildfires report. The next release at the end of July/early August 2025 will provide up-to-date information on the causes of wildfires from the previous 12 months, including Los Angeles and how climate change and human activity were involved. For more information, see the article on ‘State of Wildfires 2023–2024’ in Earth System Science Data.