Forest Fire Smoke Exposure Assessment: Use of Remote Sensing Data in a Pollution Dispersion Model

Abstract

TM3-O-09 Introduction: Climate change analysts predict that wildfires will become more frequent and intense worldwide over the next 50 years. The health effects of smoke from these events are not well understood, largely due to lack of exposure data. Because most fires burn in areas with sparse air quality monitoring, novel exposure assessment methods are needed to facilitate epidemiologic studies on affected populations. Here we use physically based numerical models in combination with remote sensing data to estimate smoke-related elevations in 24-hour average PM10 concentrations across a 325,000 km2 region of Canada over a 92-day period in 2003. Methods: We initialized the CALMET meteorologic model with data from 99 surface weather stations and output from a numerical weather prediction model. Resulting fields were used as input for the CALPUFF dispersion model. Fire locations were remotely detected by 2 satellite-borne Moderate Resolution Imaging Spectroradiometers (MODIS). Three methods for estimating particle emissions were compared: 1) the U.S. Forest Service's Emissions Production Model (EPM), 2) Forestry Canada's Fire Behavior Prediction (FBP) System, and 3) calculations based on the radiative power of MODIS-detected fires (MRP). The spatial validity of model output was assessed by the horizontal overlap between CALPUFF-predicted and MODIS-imaged plumes. The 24-hour average concentration estimates were evaluated by comparison to surface PM10 (TEOM) measurements. Results: Test results for a 12,500 km2 area over a 14-day period indicate that CALMET/CALPUFF captured 65% of plume footprints on average; 2 examples are shown below. Surface concentrations at the only monitoring site in the test area were underestimated by the EPM emission rates and overestimated by the FBP and MRP methods. Regressions between modeled and measured 24-hour averages had slopes (R2) of 0.20 (0.70), 1.80 (0.70), and 6.93 (0.68), respectively. The high R2 values suggest that CALPUFF captures the temporal trend in all cases. Although MRP particle emission rates are unrealistic for highly radiative fires, these arithmetic estimates are more easily scaled than those from the EPM and FBP models. Techniques for further optimization and final results for the entire domain will be discussed. Discussion: Most simulations of wildfire plume dispersion require cumbersome models for estimating fire spread and particle emission rates. Here we demonstrate that readily available remote sensing data have the potential to produce results of equivalent quality. This will simplify and expedite smoke exposure estimates for future epidemiologic analyses.FIGURE.: Dark areas depict plume footprints as imaged by MODIS and gray lines depict CALPUFF plume contours at the surface. Location of the air quality monitoring site is shown as a white dot.