Data-enabled modeling of wildfire smoke transport
Agency : NSF
Type : NSF-DMS: #2111585
Amount to BSU : $549K
In the Western United States, Australia and many other parts of the world, wildfires are now a seasonal occurrence. Wildfires emit pollutants into the air creating poor air quality that is hazardous to people’s health and the environment. Communities use results from high resolution global scale simulations of wildfire smoke to prepare for poor air quality. This project will quantify the uncertainty in operational smoke forecasts due to incomplete knowledge of the smoke plume, wind and other weather conditions. Uncertainty estimates provide a more complete understanding of smoke forecasts, and can be communicated along with the predictions. These estimates have the potential to improve weather prediction models that are affected by smoke, and planning efforts by rural and downstream communities. This project will support two graduate students and one undergraduate student per year for each year of the three year project.
Weak constraint four dimensional data assimilation (4DVAR) will be implemented to combine wind field, emission and concentration data with a partial differential equation that describes transport of PM2.5 concentrations generated by wildfire smoke. Data from numerical weather prediction (NWP) models, including NCEP and EMCWF, smoke emission models from NOAA and US Forest service, and concentration data from EPA will be used. The representer method will be developed for 4DVAR to reduce the search space for the optimal estimates from the state space to the data space. The computational cost of 4DVAR will be further improved by developing algorithmic advances for adaptive mesh refinement (AMR) in parallel with storage and checkpointing of adjoints. Approximation of the Dirac delta distributions, appearing in the adjoint method, will be improved with a new formulation inspired by the Immersed Boundary Method. Estimates of PM2.5 concentration, wind field and emission estimates arising in the transport model will fit observations within specified error covariances. This data assimilation procedure will quantify the uncertainty in operational smoke forecasts from historical wildfire events which can be used to estimate uncertainty in smoke forecasts for future wildfire events.
PI: Jodi Mead, Boise State University; Co-PI : Donna Calhoun
Funding period : 5/2021-4/2024