Integrated fire-atmosphere prediction system with data assimilation

Most currently used operational fire spread/behavior forecasting tools are based on fire spread models driven by local weather observations or weather forecasted by numerical weather prediction (NWP) models run at relatively coarse resolution. The main benefits of these systems are their simplicity and low computational costs that allow for relatively easy operational implementation. They are, however, limited in their ability to represent local weather, topography, and fire conditions at the fire line. For instance, in complex terrain or under complex weather patterns, data from nearby weather stations may not represent conditions at the actual fire location, as wind speed and direction may change significantly across small distances and times, making a fire spread forecast driven by past observations prone to errors. Driving fire propagation by a numerical weather forecast may alleviate these problems to some degree, but only provided the weather forecast captures local meteorological processes adequately. As NWP models are run at relatively coarse resolutions (couple of kilometers), they often fail to provide a realistic local weather forecast when local meteorological conditions are strongly affected by the fire itself or when small-scale processes, complex topography, and land use characteristics significantly impact local meteorological conditions. Small-scale local circulations induced by thunderstorms or lake breezes, as well as topographical flows unresolved by coarse numerical weather prediction models, may significantly affect local weather conditions, impacting the fire behavior. We present an integrated fire spread modeling framework based upon a high resolution, multi-scale weather forecasting model, a semi-empirical fire spread model and a prognostic dead fuel moisture model (WRF-Sfire). The system renders fire progression, taking into account the fire-weather and fuel-weather feedbacks. The fire-released heat and moisture impact local meteorology, as the atmospheric model explicitly resolves fire-induced convective plumes. The high (hundreds of meters) model resolution enables detailed representation of complex terrain, and complex mosaics of surface properties. In computation of the fire propagation the system utilizes a fuel moisture model, driving the atmospheric component of the system, so no assumptions are necessary regarding the diurnal or spatial fuel moisture variability as the model simulates it.

The fuel moisture module assimilates surface observations of the 10h fuel moisture from RAWS stations, and generates spatial fuel moisture maps that are utilized by the fire spread model. Dead fuel moisture is tracked in 3 different fuel classes (1h, 10h and 100h fuel) that are integrated to provide total dead fuel moisture content at the fire model resolution (tens of meters). The combination of fuel moisture observation and the output from the fuel moisture model are used to generate the best estimate of the fuel moisture state which is used as an initialization for the fuel moisture forecasting.

The fire component of the system assimilates satellite active fires detection from VIIRS and MODIS. Even though the satellite active fires detection algorithms greatly advanced over the last decades, satellite fire detection has its limitations. There is no detection under cloud cover, the sensitivity is limited and false negatives are common, there are instrument malfunctions, and there may be false positives, for example from sun glint. The data assimilation system combines available satellite and model data to objectively generate the best estimate of the fire arrival time, which is then used to continue the simulation in the next assimilation cycle. The data assimilation mechanism enables also starting simulations of large, mature ongoing fires from observed perimeters rather than point ignitions.

We show preliminary results from the system, and comment on computational requirements, as well as model limitations in the context of operational application.