Wildland Fire Modeling and Prediction


RAL is developing a useful suite of methods to predict wildfires and their coincident weather. This is a long-term project engaging the expertise of several scientists and engineers across the organization.  Decision makers who deploy resources and strategize effective targets for firefighting need reliable, accurate, frequently updated, readily accessible, geo-referenced, current and predicted weather and fire behavior information. Timely information allows decision makers to better assess current conditions and future trends. Reliable information about the potential for rate of fire spread and extreme fire behaviors is essential to saving lives and property. 

Current operational wildland fire-spread prediction systems are not coupled to numerical weather prediction (NWP) models. These systems often rely on wind fields coarsely resolved in space and time. However, when flows begin rapidly evolving due to storm outflows, density currents, frontal passages, and other weather features, or are spatially variable due to complex-terrain effects, highly resolved wind fields in time and space are essential to accurately predict fire spread rate and direction. Furthermore, large wildfires result in significant surface heat fluxes generating strong updrafts and consequently intensifying local winds, which in turn cause more rapid fire-spread rates. Large wildfires also produce significant smoke plumes that can affect radiative transfer, while lofted particulate matter and moisture can form pyrocumulus clouds. All these phenomena can be predicted using coupled models. Therefore, developing an operational coupled wildland fire-spread capability is essential for accurate wildland fire spread prediction. To achieve this goal, RAL researchers are extending capabilities of the Weather Research and Forecasting (WRF) NWP model, based on the Coupled Atmosphere Wildland Fire Environment (CAWFE) model. In addition to gauging fire perimeter, the modeling system (which runs at a resolution of 100 m on its inner domain) produces predictions of a variety of fire and weather variables, including rate-of-spread, flame length, and smoke. These developments are being incorporated into the community WRF-Fire model.


During the last year, the wildland fire-prediction system was ported to a cloud-computing platform, which allows multiple simultaneous wildland fire simulations. Cloud-computing provides also greater flexibility and faster model execution due to availability of a range of hardware configuration. During 2019 fire season the cloud-computing instance of the system was extensively tested supporting operational wildland fire simulations. The system performed reliably for both automatic, nowcasting simulation as well as longer range high-resolution simulations. In addition, the wildland fire prediction system is was augmented by a fire spotting likelihood capability. The initial implementation of spotting capability was based on an online version of the Lagrangian particle dispersion model, HYSPLIT, integrated with WRF model. Due to inherent limitations of the online implementation of HYSPLIT in WRF, we have started development of a more flexible, massively parallel Lagrangian particle tracking algorithm.

The performance of the prediction system was assessed using observations from wildland fires in the State of Colorado from 2016 through 2018 fire seasons. The model tends to overpredict the size of most fires. One of the reasons for overprediction is that at present we are not able to account for the fire suppression efforts due to the lack of appropriate data. In previous years we carried sensitivity studies focused on the representation of terrain and fuel moisture content. This year we have also studied the effect of fuel types and related loads on the rate of spread prediction. The analysis indicated that even relatively minor errors in fuel type distribution can result in sizable errors in the rate of spread prediction. This result points to the need for more frequently updated fuel maps.

Because fuel moisture content is one of the significant parameters controlling the spread rate of wildland fires, having accurate estimates of the fuel moisture content are critical to firefighting managers. However, estimates of the fuel moisture content are currently based on spatially sparse observations. Furthermore, live fuel moisture is sampled infrequently because it has to be done manually. To address the need for a higher resolution fuel moisture data and also more frequent live fuel moisture observations, we have developed a real-time system for high-resolution, gridded fuel moisture-content data over conterminous United States (CONUS). Satellite reflectance data from MODIS Terra and Aqua platforms are used together with surface observations in machine-learning models to provide estimates of the dead and live fuel moisture content. We have selected random fores algorithm based machine learning model to produce daily, one-kilometer resolution maps of the dead and live fuel moisture over CONUS. The maps and associated data sets can be accessed through RAL’s web page: https://ral.ucar.edu/projects/improved-wildland-fire-spread-prediction, under the tap “Operations.”


Throughout next year, extensive evaluation of the system performance will continue based on wildland fires observed in Colorado during fire season 2019. The spot-fire prediction capability will be fully integrated in the wildland fire modeling system and its performance will be assessed using available spotting data. Spotting often causes the rapid rate of wildland fire spread (Jimenez et al. 2018). Projections about the likelihood of spotting can therefore be critical for more effective wildland fire management. The spotting likelihood is computed based on the likelihood of firebrand generation, transport, burnout, and deposition upon fuels that can be ignited. The spot fire capability will be assessed using observations of firebrand transport and deposition obtained using dual pol radar in recent wildland fires in Australia.


Jimenez, P. A., D. Muñoz-Esparza, and B. Kosović, 2018: A high resolution coupled fire-atmosphere forecasting system to minimize the impacts of wildland fires: Applications to the Chimney Tops II wildland event. Accepted for publication in Atmosphere.