Estimated Impacts of Prescribed Fires on Air Quality and Premature Deaths in Georgia and Surrounding Areas in the US, 2015–2020

Smoke from wildfires poses a substantial threat to health in communities near and far. To mitigate the extent and potential damage of wildfires, prescribed burning techniques are commonly employed as land management tools; however, they introduce their own smoke-related risks. This study investigates the impact of prescribed fires on daily average PM2.5 and maximum daily 8-h averaged O3 (MDA8-O3) concentrations and estimates premature deaths associated with short-term exposure to prescribed fire PM2.5 and MDA8-O3 in Georgia and surrounding areas of the Southeastern US from 2015 to 2020. Our findings indicate that over the study domain, prescribed fire contributes to average daily PM2.5 by 0.94 ± 1.45 μg/m3 (mean ± standard deviation), accounting for 14.0% of year-round ambient PM2.5. Higher average daily contributions were predicted during the extensive burning season (January–April): 1.43 ± 1.97 μg/m3 (20.0% of ambient PM2.5). Additionally, prescribed burning is also responsible for an annual average increase of 0.36 ± 0.61 ppb in MDA8-O3 (approximately 0.8% of ambient MDA8-O3) and 1.3% (0.62 ± 0.88 ppb) during the extensive burning season. We estimate that short-term exposure to prescribed fire PM2.5 and MDA8-O3 could have caused 2665 (95% confidence interval (CI): 2249–3080) and 233 (95% CI: 148–317) excess deaths, respectively. These results suggest that smoke from prescribed burns increases the mortality. However, refraining from such burns may escalate the risk of wildfires; therefore, the trade-offs between the health impacts of wildfires and prescribed fires, including morbidity, need to be taken into consideration in future studies.


2002)
. For example, a 10-ppb increase in the daily average ozone concentration corresponds to approximately 18.75-ppb increase in the MDA8-O3.S1.1.Data-Fusion: Here, we breakdown the data-fusion approach (Friberg et al., 2016): Ambient monitor observations provide very limited spatial information and information on temporal variation that decreases with increasing distance from monitors.CMAQ simulations, on the other hand, provide information that is independent of observations.The approach to fusing observations and CMAQ simulations involves three steps.

Interpolated Observation Method (FC1):
To obtain pollutant estimates over space (x) and time (t) with temporal variation driven by monitor data, we first normalize daily observations (OBSm) at each monitor (m) to annual mean levels ( !%%%%%%%% ).Next, normalized data are spatially interpolated by the krig method described below.
Lastly, the interpolated field is denormalized using the CMAQ annual field adjusted to the annual mean observations ( %%%% ), also described below.Eq 1 describes this procedure for estimating daily concentration fields with temporal variation driven by observations and spatial structure governed by the adjusted annual mean CMAQ field.
Daily spatial interpolation of the normalized observations was performed by ordinary kriging.
Normalization prior to kriging provides a smoother surface necessary with limited monitor coverage of the spatial domain.A more detailed spatial structure is obtained in denormalization using the mean CMAQ field.
Regression of annual mean measurements( !%%%%%%%% ) and CMAQ simulations at monitor locations ( !%%%% ) provides parameters for an annual mean pollutant field model that captures the spatial pattern of emissions and annual effects of meteorological variables, correcting for CMAQ annual biases (eq 2).
() %%%%%%%% =  ./0+× () %%%%%% 1 (2) Here, the overbar indicates temporal averaging (annual), β is a parameter derived for all years, and αyear is a regression parameter derived for each year.Spatial misalignment of measurements (point) and CMAQ simulations (4 × 4 km) and a changing network of monitors over time are factors that contribute to model instability and led to the use of more constrained models (e.g., nonvarying β) and zero intercepts.Linear models with slopes and intercepts varying each year provided similar results, although negative intercepts and greater variability across years were found.
Here, βseason is the seasonal correction function.Analysis of the differential pattern between the observations and CMAQ simulations across species showed the seasonal component could be described and minimized using a sinusoidal cycle.The seasonal bias, which follows a sinusoidal variation, was modeled as a smooth trigonometric function (eq 4) with two fitted parameters: amplitude (A) and day of peak correction (tmax).
Use of this correction factor removed seasonal trends in the residual errors of the FC2.

Optimized Fused Fields (FCopt):
Third, we combine these to produce optimized fused concentration fields ( C3 ) by computing a weighted average with the weight depending on the spatial autocorrelation of observations (which governs how well FC1 predicts temporal variance) and the correlation between observations and CMAQ simulations (which governs how well FC2 predicts temporal variance).
W is an average A weighting factor, spatial fields for the study period.The optimized field resembles the FC1 field near observations (where the weighting is large) and the FC2 field far from observations (where the weighting is small).
Finally, we combine eq(1) to eq(4) put it in the eq(5) and represent in the manuscript.
Normalized mean bias (NMB) = Fig S3.Spatial distribution of yearly prescribed burned area by adjusted FINN (unit: acres).
Fig S8.Spatial distribution of daily average prescribed burn smoke PM2.5 concentration (µg/m 3 ) during maximum burned area days.

Figure S10 .
Figure S10.The top figures depict the spatial distribution of daily average PM2.5 and MDA8-O3 resulting from prescribed burns over 2015-2020.The middle figure illustrates the population density across the study region.The bottom figures show the spatial distribution of total all-cause premature deaths attributable to prescribed burn-induced PM2.5 and MDA8-O3 during 2015-2020.

Table S2 .
100 Note.Subscript j represents the pairing of N observations O and Model predictions M by site and time.Overbars signify means over site and/or time.Yearly prescribed burn area and estimated emitted pollutants.

Table S5 :
Statistics of the prescribed burn smoke pollution concentrations over the study period

Table S6 :
Statistics of prescribed burn smoke pollution concentrations over the burning season (January-April)

Table S7 :
Statistics of the prescribed burn smoke pollution concentrations over the cold season (October-December)

Table S8 :
Statistics of the prescribed burn smoke pollution concentration over the no-fire season (May-September)

Table S10 :
Means and standard deviations of total MDA8-O3 concentrations (ppb) and burn impacts at monitoring sites during 2015-2018.