An analysis of WRF-Sfire's hindcast of Utah's 2013 Patch Springs Fire

WRF-Sfire is a wildfire-atmosphere integrated modeling system for wildfire simulation. It is based on a multi-scale weather forecasting model, coupled with a semi-empirical fire spread model and a prognostic dead fuel moisture model. It aims to directly resolve plume rise and smoke dispersion, and is easily integrated with operational WRF and WRF-Chem products. The system operates on a high resolution atmospheric grid (i.e., horizontal grids 10s to 100s m) to explicitly resolve fire-induced convection and to capture the effects of complex topography and land use characteristics on the local weather. Fire behavior and fuel moisture are therefore driven by realistic meteorological forcing, while heat and moisture fluxes released at the fire line are fed back into the model atmosphere, altering air temperature, humidity, and flow. Although fire-induced circulations are considered a local meteorological condition with respect to weather prediction, this two-way coupling may have a significant impact on the larger-scale weather prediction. The predictive fuel moisture model, with meteorological components, provides in-line temporal and spatial evolution of fuel moisture. This enables simulation of diurnal changes in fire activity associated with wind speed variations and nighttime fuel moisture recovery. No assumptions are necessary regarding the diurnal or spatial fuel moisture variability, as the model simulates it based on WRF forecasted weather conditions. Fire activity is therefore a consequence of the weather and fuel moisture conditions. The system has certain known prediction challenges, and requires testing and evaluation to determine its limitations.

Here the Patch Springs Fire is used as a benchmark simulation to demonstrate the capabilities of WRF-Sfire to hindcast wildfire spread and plume behavior for a wildfire that burned in complex terrain under weak synoptic forcing. For this hindcast, WRF was configured with 5 nested domains of, respectively, horizontal grid sizes 36km, 12km, 4km, 1.33km, and (fire domain) 444m. A refined surface grid of horizontal cell size 22 meters was used to model/track fire perimeter propagation through the 5th WRF domain. The simulation was initialized with NARR (North American Regional Reanalysis) data (at 32 km horizontal resolution). The fire model, Sfire, used 30m resolution elevation and fuel data, whereas the atmospheric model, WRF, used 1.5km resolution MODIS land-use representation. The fire was initialized by a single point ignition. Convection was resolved in the 1.33km and 444m WRF domains, while cumulus parameterization was employed for the 36, 12, and 4 km domains. Smoke tracers were active. The WRF was provided with boundary conditions only at the initial stage, was not nudged towards the observations, and no updates of the model state with current meteorological observations were done during the run. The spatially variable fuel moisture was initialized based on observations at the nearest weather station.

The Patch Springs Fire started (lightning caused) at 8 pm local time, 10 Aug, 2013, 2 miles north-west of Terra, Utah. It burned along the west facing Stansbury Mountain Range for the next 20 days, and was 80% contained by 21 August and finally 100% contained by 23 October, when the fire received rain. In this study we will concentrate on the time between 14 August to 15 August when the fire's perimeter increased unexpectedly. We will present an integrated overview of the relationships between the observations, fire fighting decisions, and WRF-Sfire model predictions, in an attempt to discover the reason for this sudden fire spread. We will investigate how well the fire's actual behavior and history coincided with the multi-scale WRF-Sfire forecast, and present reasons for differences.