This study analyzes fire-induced winds from a wind-driven fire (Thomas Fire) and a plume-dominated fire (Creek Fire). Two numerical experiments, one without the fire present and the other with the fire, were used. The fire-induced perturbations were then estimated by subtracting a variable value in the “No Fire Run” from the “Fire Run” (Fire - No Fire). For this study, spatial and temporal variability of winds, geopotential height, and convergence were analyzed. Furthermore, cloud water mixing ratio, precipitation, and fuel moisture were analyzed during the Creek Fire to assess fire-induced rainfall and its impact on fuel moisture. It was found that the wind-driven Thomas Fire created more widespread and generally stronger fire-induced winds than the plume-dominated Creek Fire. In addition, fire-induced wind speeds during the Creek Fire followed a diurnal cycle, while the Thomas Fire showed much less temporal variability. When analyzing geopotential height, the results were very similar to other idealized simulations. A localized low-pressure region was observed in front of the fire front, with a preceding high-pressure area. When analyzing precipitation, it was found that the fire increased precipitation accumulation in the area surrounding the active fire. This created an increase in fuel moisture which could have helped locally decelerate the fire spread. Further research into the processes behind fire-atmosphere interactions will lead to a better understanding of fire behavior and the extent to which these interactions can impact the fire environment. These studies will help assess the limitations of uncoupled operational models and improve fire modeling overall.

We present a new statistical interpolation method to estimate fire perimeters from Active Fires detection data from satellite-based sensors, such as MODIS, VIIRS, and GOES-16. Active Fires data is available at varying temporal and spatial resolutions (375m and up several times a day, or 2km every 15 minutes), but pixels are often missing due to clouds or incomplete data. The question arises how to fill in the missing pixels, which is useful, e.g., to distinguish in an automated fashion between a single large fire visible as separate clusters of detection pixels because of cloud cover, and separate fires. We process the satellite data into information when was fire first detected at a location, and when was clear ground without fire detected at the location last. We are then looking for the most likely fire arrival time, which satisfies such constraints. Models at various levels of complexity are possible. Our base assumption in the absence of information to the contrary is that the fire keeps progressing without change, which is expressed as the assumption that the gradient of the fire arrival time is approximately constant. The method is then formulated as an optimization problem to minimize the total change in the gradient of the fire arrival time subject to the constraints given by the data. We consider probabilistic interpretations of the method as well as extensions, such as soft constraints to accommodate the uncertainty of the detection and the uncertainty where exactly the fire is within the pixel. This method is statistical in nature and it does not use fuel information or a fire propagation model. The results are demonstrated on satellite observations of large wildfires in the U.S. in summer 2018 and compared with ground and aerial data.