An Ignition Point Sensitivity Study of the WRF-Fire Model: An Analysis of Wildfire Area and Location for the Indian Valley Fire

The state of Colorado has approximately 24.4 million acres of forestland which are at an increasing risk of being burned by wildfires each year. Although some wildfires are a natural occurrence in Colorado, the National Park Service declared that nearly 85 percent of wildfires are caused by human actions. As the population in Colorado grows in regions where forestland and wildland fuels exist, so does the annual average of acreage burned by wildfires per year. Increased temperatures and long drought seasons create favorably dry conditions for wildfires to propagate. As the annual amount of Colorado land being burned by wildfires increases, discussions around how to best extinguish fast-spreading fires circulate between firefighters, the state government, and other stakeholders. In 2015, the Governor of Colorado passed a house bill to begin the creation of the Colorado Fire Prediction System (CO-FPS), a system that integrates the Weather and Research Forecasting (WRF) model’s wildland fire behavior module (WRF-Fire) into a decision-support framework. Wildland fuels, atmospheric conditions, terrain elevation, and ignition point location are inputted into the WRF-Fire model to simulate fire propagation. There is uncertainty in the model’s output with regard to change in the location of ignition point due to observations of reported starting locations often being delayed and inaccurate in time and space. To increase understanding of the model’s output uncertainty, an analysis of how varying ignition point location influences the fire size and location of the 2018 Indian Valley wildfire was conducted. For the Indian Valley wildfire, the reported ignition point location, just northwest of Meeker, CO, was located outside of the fire perimeter, making it a suitable case study from which to conduct a sensitivity analysis. In this study, seventeen simulated ignition points were inputted into the model: 1 was at the reported ignition location, 8 were distanced 100 m away from the reported ignition location, and 8 were distanced 200 m away. Both distances were calculated in each of the cardinal (N, S, E, W) and inter-cardinal (NE, SE, NW, SW) directions. The resulting fire spread forecasts were compared against each other and the observed fire spread to determine the sensitivity of the WRF-Fire model to small changes in ignition location. Results concluded that ranges of uncertainty for the model’s output of fire area and location increased as distance from the reported ignition point increased. In addition, preliminary results found that ignition points south of the reported ignition point simulated below average with regard to area. The reason for why ignition points south of the reported ignition point under-forecasted will be discussed at the talk.