Wildfires have grown increasingly destructive in recent decades, with impacts felt nationwide. Forecasters rely on fire weather metrics — such as wind speed and vapor pressure deficit (VPD), which indicates atmospheric dryness — to assess when and where conditions may intensify wildfire behavior and strain suppression capacity.
WPO selected the “Subseasonal Predictability of Fire Weather Metrics for Decision Support” project in its FY23 Innovations for Community Modeling funding competition. Researchers evaluated forecasting options within the Unified Forecast System (UFS) to strengthen wildfire preparedness and response.
UFS subseasonal forecasts show encouraging skill for key fire weather indicators, including temperature and VPD. However, wind forecasts across all forecast times tend to have speeds that are too low, and don’t vary enough, relative to observations across the Continental United States. Prior analyses show that using statistical and deep learning approaches to create finer resolution data — “statistical downscaling” — provided limited added value. Both forecast and reanalysis datasets underrepresent how much wind speeds actually vary and how high they get.
Incorporating a higher-resolution simulation, called dynamic downscaling, in the form of EPIC’s Short Range Weather (SRW) App shows improved representation of wind variability and higher wind speeds. Importantly, the enhanced hour-to-hour depiction of wind extremes better reflects influences from topography and aligns with periods of peak fire progression.
While work is ongoing, these results suggest that dynamic downscaling may meaningfully improve wind predictions when it matters most: during wildfires, to support decisions that protect lives and property.
Code developed through this project is publicly available on this GitHub page. Project datasets will soon be hosted by the Wildfire Interdisciplinary Research Center to reduce technical barriers and support broader research access. Project results will be shared through publications, webinars, and reports. The NWS Environmental Modeling Center may adopt these techniques in the future if further evaluation and demonstration show they can be successfully applied to operational forecast models.
This article was originally written by the project’s principal investigator, Samantha Kramer (Atmospheric Interdisciplinary Research Institute), and edited for clarity and context. Project co-PIs are Ben Kirtman (University of Miami) and Brian Potter (USDA Forest Service). Science contributors include Kayla Besong, Alejandro Valencia, Lindsay Nash, Tyler Fenske, and Nate Benson (Sonoma Technology); Natalie Perlin (University of Miami); Jay Charney (USDA Forest Service); ShihMing Huang (National Interagency Fire Center); and Hosmay Lopez (NOAA Atlantic Oceanographic and Meteorological Laboratory).
This story was published on March 3, 2026.
