Evaluating wildfire emissions projection methods in comparisons of simulated and observed air quality
Climate warming has been implicated as a major driver of recent catastrophic wildfires worldwide but analyses of regional differences in US wildfires show that socioeconomic factors also play a large role. We previously leveraged statistical projections of annual areas burned (AAB) over the fast-growing southeastern US that include both climate and socioeconomic changes from 2011 to 2060 and developed wildfire emissions estimates over the region at 12 km × 12 km resolution to enable air quality (AQ) impact assessments for 2010 and selected future years. These estimates employed two AAB datasets, one using statistical downscaling (“statistical d-s”) and another using dynamical downscaling (“dynamical d-s”) of climate inputs from the same climate realization. This paper evaluates these wildfire emissions estimates against the U.S. National Emissions Inventory (NEI) as a benchmark in contemporary (2010) simulations with the Community Multiscale Air Quality (CMAQ) model and against network observations for ozone and particulate matter below 2.5 µm in diameter (PM2.5). We hypothesize that our emissions estimates will yield model results that meet acceptable performance criteria and are comparable to those using the NEI. The three simulations, which differ only in wildfire emissions, compare closely, with differences in ozone and PM2.5 below 1 % and 8 %, respectively, but have much larger maximum mean fractional biases (MFBs) against observations (25 % and 51 %, respectively). The largest biases for ozone are in the fire-free winter, indicating that modeling uncertainties other than wildfire emissions are mainly responsible. Statistical d-s, with the largest AAB domain-wide, is 7 % more positively biased and 4 % less negatively biased in PM2.5 on average than the other two cases, while dynamical d-s and the NEI differ only by 2 %–3 % partly because of their equally large summertime PM2.5 underpredictions. Primary species (elemental carbon and ammonium from ammonia) have good-to-acceptable results, especially for the downscaling cases, providing confidence in our emissions estimation methodology. Compensating biases in sulfate (positive) and in organic carbon and dust (negative) lead to acceptable PM2.5 performance to varying degrees (MFB between −14 % and 51 %) in all simulations. As these species are driven by secondary chemistry or non-wildfire sources, their production pathways can be fruitful avenues for CMAQ improvements. Overall, the downscaling methods match and sometimes exceed the NEI in simulating current wildfire AQ impacts, while enabling such assessments much farther into the future.