Arctic wildfire smoke has significant health impacts across borders

Model setup

We simulate global PM to estimate the impact of Arctic Council wildfires on human mortality._2.5 Under two scenarios:

• “FIRE ON”, control simulation using Quick Fire Emissions Dataset (QFED, version 2.5)³⁵ Fire emissions dataset. Includes daily fire emissions from all fires (wildfires and agriculture) and all regions.

• “ARCTIC WILDFIRE OFF” is a counterfactual scenario that does not include wildfire emissions within the Arctic Council, but includes other types of fires in the Arctic and all types of fires outside the Arctic. This scenario eliminates wildfires caused by both human-induced and natural combustion.

There are eight member states of the Arctic Council: Canada, Denmark, Iceland, Norway, Sweden, Finland, Russia, and the United States. Because we focus on wildfire impacts in high latitudes, we do not include the neighboring United States or Denmark as Arctic Council regions (although we do include Greenland).

The simulations are performed using Community Earth System Model (CESM) version 2.2.⁵¹ A community atmosphere model with chemistry (CAM-Chem) to simulate the chemistry of the troposphere and stratosphere, in a configuration that includes simplified representations of other components of the Earth (oceans, sea ice, etc.). His version of CAM-chem in CESM v2.2 is CAM6-chem, which uses his MOZART-TS1, the latest version of the Model of Ozone and Related Chemical Tracers (MOZART) chemical mechanism.⁵², Volatility Basis Set schemes for representing gas species and for representing aging of organic aerosols. The aerosol size distribution is represented using a four-mode version of the modal aerosol model (MAM4).⁵³, additional modes have been added to improve the representation of carbonaceous aerosols, black carbon (BC), and primary organic matter (POM). We use MERRA2 reanalysis data to fine-tune the meteorology and run CESM at a standard grid resolution of 0.9° x 1.25° (latitude x longitude).

Because chemical and meteorological processes are coupled in CESM, changes in fire emissions between the two scenarios result in small variations in meteorological parameters (e.g., temperature, wind speed), which affect secondary PM. It will be._2.5 Concentrations are high and can spread to areas far from the North Pole. These fluctuations could be mistakenly attributed directly to wildfire-derived PM._2.5 When comparing two scenarios. To focus on the direct impact of wildfires, we calculate PM due to wildfires._2.5 It is the difference in the proportion of POM and BC between the two scenarios, which is not influenced by secondary aerosol formation or meteorological feedbacks. These species account for on average >99.9% of the primary aerosol mass emitted by Arctic Council wildfires, so this method covers most of the expected sources of PM._2.5– Attributable health effects.

We also exclude regions where POM + BC PM increases._2.5 The fire outbreak scenario is not statistically significant when compared to the Arctic wildfire suppression scenario. To do this, run a one-sided paired sample. t Tests at each CESM grid cell between the monthly mean concentrations in the two scenarios for the period 2001-2020. The significance thresholds are: p = 0.01. If no significant increase is found, set the “FIRE FRACTION” term in the bias correction equation to zero (section “Correction for simulated PM”)_2.5 ), there were no health effects from the wildfires at that location. This ensures that statistically insignificant differences between his two scenarios in densely populated areas far from the North Pole do not contribute unduly to the calculated health effects.result of t The test is shown in Supplementary Figure 4.

Developing a “Wildfires Off” emissions scenario

Both CESM model simulations include fire emissions from the QFED dataset³⁵. In QFED, emissions of gaseous and particulate species emitted by biomass combustion are measured on a satellite, along with biome-specific emission factors that are calibrated by comparing modeled and observed aerosol optical depths (AODs). Based on detected fire radiant power (FRP) observations. We chose QFED over other biomass combustion emissions inventories (BBEIs) because using AOD observations can reduce emissions underestimation.^35,45has been shown to have the lowest negative bias for AOD among the six commonly used BBEIs.⁴⁴Therefore, it is less likely to underestimate PM._2.5 The main focus of this analysis is concentration.

However, QFED emissions estimates do not distinguish between agricultural and wildfires. Therefore, to construct a biomass combustion emissions scenario without wildfires, assume that emissions from agricultural land are not wildfires, and emissions from other land uses are wildfires.

The QFED emissions resolution used is regridded to CESM resolution (0.9° x 1.25°), while MODIS land use data are available at 0.01° x 0.01°. We use MODIS data at its native spatial resolution (Supplementary Figure 5) to calculate the percentage of agricultural land within each QFED grid cell. This is used to partially split QFED emissions between wildfire and agricultural emissions. Emissions attributable to wildfires and agricultural land are shown in Supplementary Figure 6. Errors caused by resolution mismatch between QFED and MODIS data have a relatively small impact on the analysis because only a small portion of the land area in the Arctic is covered by agricultural land. Council. Most of them are grouped into separate areas.

Modification of simulated PM_2.5 To observation

Comparison with few available PMs_2.5 Observations in the Arctic region show that simulated air pollutant concentrations are consistently underestimated by observations in both regions.⁵⁴ and global model^55,56,57. This is partly due to the difficulty of representing the plume at low resolution, as the plume becomes diluted within the model grid cells.Furthermore, emissions from biomass combustion are often underestimated.^43,44,47. A special cause of this in arctic regions may be peat soils. Emission-rich smoldering fires occur frequently in peat soils, but are typically poorly reflected in emissions inventories because they are difficult to detect by satellite.^58,59. Additionally, satellite-derived burned area products often allow fires to be omitted due to their high detection limits and limited transit time.⁶⁰.

We compared the CESM model to a dataset of reference grade PMs._2.5 Monitors from the AIRNOW network (north of 55 degrees north latitude). This includes monitoring the U.S. embassies in Alaska, Canada, and Almaty, Kazakhstan. The normalized mean bias (NMB) of CESM is negative 0.61, while the geographically weighted regression (GWR) PM is negative._2.5³³ NMB is −0.05 (Supplementary Figure 7). Similar results were found when comparing the CESM model with a low-cost dataset.Purple Air” CESM is negatively biased relative to monitors located north of 55°N. purple air data (Supplementary Figure 8), NMB is −0.72. GWR PM_2.5 The product is more consistent with purple airMonitor with NMB of 0.1. From these results, we conclude that GWR PM:_2.5 Can better express the size of PM in the Arctic_2.5 Higher concentration than PM simulated in CESM_2.5 concentration.

Use GWR PM_2.5 To get a more accurate estimate of PM_2.5 concentrated in areas affected by wildfires. This dataset uses satellite AOD observations and model-derived vertical profiles and surface monitoring measurements to estimate monthly mean PM._2.5 With a resolution of 10km. However, FIRE ON simulation is underrated compared to his GWR PM._2.5 They occur in most regions and seasons, not just those affected by wildfire plumes. Therefore, we use the ARCTIC WILDFIRE OFF CESM simulation as the counterfactual scenario compared to the GWR PM._2.5 Overestimating the proportion of PM_2.5 Arctic wildfires may be the culprit.

Estimate the PM fraction using CESM simulation._2.5 This is due to the Arctic Council wildfires and we refer to this as the “fire fraction” (Equation 1). Apply FIRE FRACTION to GWR PM_2.5 Data for estimating counterfactual PM_2.5 A scenario where no Arctic wildfires occur. We call this the “BIAS CORRECTED WILDFIRE OFF'' scenario (Equation 2).

GWR PM_2.5 The reanalysis is only done up to 68°N, so FIRE ON CESM simulations are used for more northern latitudes. Linearly interpolate the GWR PM to avoid hard boundaries._2.5 Use FIRE ON simulation north of 68°N.

Health impact assessment

Health effects of long-term (chronic) exposure to ambient PM_2.5 Concentrations are estimated using the Global Exposure Mortality Model (GEMM).^{twenty three} Following the method used in the previous work^61,62,63and other assessments of the health effects of wildfire smoke.^46,64. Briefly, the GEMM uses data from 41 epidemiological cohort studies to estimate the increased relative risk of health effects from chronic ambient PM._2.5 Exposure above the counterfactual level of 2.4 µg m^-3 Adults aged 25 and over divided into 5 year groups. We used a relative risk function for non-accidental mortality (including non-communicable diseases and lower respiratory tract infections) and used parameters including the Chinese cohort.^{twenty three}. We use the GEMM because there are currently no models in the literature that specifically consider the chronic health effects of wildfire smoke. We consider chronic health impacts because in many high latitudes wildfire smoke regularly degrades local air quality and accounts for a high proportion of annual average PM._2.5^65,66.

The Gridded Population of the World (GPW) dataset version 4 was used for population counts and distribution.⁶⁷. The GPW dataset contains population estimates for five years and is linearly interpolated to provide previous years.Global Burden of Disease 2019 Survey Data⁶⁸ Annual population age structures and baseline mortality rates from 2001 to 2019 were used, and 2020 health impacts were calculated using 2019 data.

Calculate excess mortality due to chronic PM using GEMM_2.5 Exposure in both control scenarios (GWR PM)_2.5) and a counterfactual scenario in which the Arctic Council wildfire contamination is removed (“bias-corrected wildfire off”, Equation 2). The difference in excess mortality in these two scenarios is the mortality burden due to Arctic wildfire smoke.Estimating trends

Trends are calculated using the Theil-Sen trend estimation tool.⁶⁹, which is a nonparametric trend estimator that is robust to outliers. The significance of trends was tested using the Mann-Kendall test, which detects monotonically increasing or decreasing trends.⁷⁰.

Report overview

For more information on the study design, please see the Nature Portfolio Reporting Summary linked in this article.