Modeling fire emissions: A sensitivity study using an advanced statistical emulator

We have demonstrated the use of an advanced Gaussian-Process (GP) emulator to estimate wildland fire emissions over a wide range of fuel and atmospheric conditions. The Fire Emission Production Simulator, or FEPS, is used to first produce a small subset of emissions data that correspond to some selected values in the domain of the input fuel and atmospheric parameters for the purpose of training the emulator. The emulated emissions are found to be within ±5% of the FEPS simulated emissions, providing confidence in the potential use of the GP-emulator for this and other similar applications. Cluster analysis for 1000 emulator-produced posterior samples spanning a wide-range of fuel and environmental conditions suggest that the emulator not only produces valid results but also preserves the physical relationships between the fire emission and the fuel and environmental conditions. Results show that the GP-emulator could be used as an alternative to the simulations from the FEPS modeling system when four or more input parameters related to fuel type, fuel moisture, and weather condition are allowed to vary. This work demonstrates the feasibility of constructing automated look-up tables relating fire emissions to fuel and environmental properties, which could then be coupled to operational smoke dispersion models as well as air pollution models driven by weather forecasts or observations.