Highly unusual open fires burned in western Greenland between 31 July and
21 August 2017, after a period of warm, dry and sunny weather. The fires
burned on peatlands that became vulnerable to fires by permafrost thawing.
We used several satellite data sets to estimate that the total area burned
was about 2345 ha. Based on assumptions of typical burn depths and
emission factors for peat fires, we estimate that the fires consumed a fuel
amount of about 117 kt C and emitted about 23.5 t of black carbon (BC) and
731 t of organic carbon (OC), including 141 t of brown carbon (BrC). We used
a Lagrangian particle dispersion model to simulate the atmospheric transport
and deposition of these species. We find that the smoke plumes were often
pushed towards the Greenland ice sheet by westerly winds, and thus a large
fraction of the emissions (30 %) was deposited on snow- or ice-covered
surfaces. The calculated deposition was small compared to the deposition from
global sources, but not entirely negligible. Analysis of aerosol optical
depth data from three sites in western Greenland in August 2017 showed strong
influence of forest fire plumes from Canada, but little impact of the
Greenland fires. Nevertheless, CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization) lidar data showed that our model
captured the presence and structure of the plume from the Greenland fires.
The albedo changes and instantaneous surface radiative forcing in Greenland
due to the fire emissions were estimated with the SNICAR model and the uvspec
model from the libRadtran radiative transfer software package. We estimate
that the maximum albedo change due to the BC and BrC deposition was about
0.007, too small to be measured. The average instantaneous surface radiative
forcing over Greenland at noon on 31 August was 0.03–0.04 W m
In August 2017 public media reported unprecedented fire events in western Greenland (BBC News, 2017; New Scientist Magazine, 2017). These events were documented with airborne photographs (SERMITSIAQ, 2017) and satellite images (NASA, 2017b) and raised public concerns about the effects of climate change and possible impacts of soot emissions on ice melting. Historically, wildfires have occurred infrequently on Greenland, because three-quarters of the island is covered by a permanent ice sheet and permafrost is found on most of the ice-free land (Abdalati and Steffen, 2001). Permafrost, or permanently frozen soil, lies under an “active” soil layer several meters thick that thaws seasonally. But in certain areas, where the permafrost layer starts melting, it can expose peat, a material consisting of only partially decomposed vegetation that forms in wetlands over the course of hundreds of years or longer. Peatlands, also known as bogs and moors, are the earliest stage in the formation of coal. Globally, the amount of carbon stored in peat exceeds that stored in vegetation and is similar in size to the current atmospheric carbon pool (Turetsky et al., 2014). When peatlands dry, they are often affected by fires burning into the peat layers. Peat fires are difficult to extinguish and they often burn until all the organic matter is consumed. Smoldering peat fires already are the largest fires on Earth in terms of their carbon footprint (Turetsky et al., 2014). For Greenland, it has been suggested that degradation of peat will accelerate towards 2080 (Daanen et al., 2011) and that the area affected by the fires in August 2017 is particularly vulnerable to permafrost thawing (Daanen et al., 2011).
Fires in the high northern latitudes release significant amounts of
Here we study transport and deposition of BC, OC and BrC over the Greenland ice sheet from the fires that occurred in western Greenland in August 2017, which likely represent the largest fires that have occurred on Greenland in modern times (Fig. S1 in the Supplement). Since the fires occurred in an area entirely lacking ground-based observations, we use satellite data and a Lagrangian atmospheric dispersion model for our study. Finally, we evaluate the changes in the albedo of the Greenland ice sheet from the respective deposition of BC and BrC and present instantaneous radiative forcing calculations for these two atmospheric constituents released from the 2007 fires in Greenland.
Remote sensing has been useful for delineating fire perimeters,
characterizing burn severity and planning post-fire restoration activities in
different regions. The use of satellite imaging is particularly important for
fire monitoring in remote areas due to difficult ground access. The method
that is presented in this section has been already used to calculate burned
area in the highly contaminated radioactive forests of Chernobyl
(Evangeliou et al., 2014, 2015, 2016). Coordinates of fire locations (hot spots) were
downloaded from FIRMS (Fire Information for Resource Management System; NASA, 2017a). For the mapping of the burned area, Sentinel-2A
images were used. To delineate fire perimeters and define burn severity
precisely, we used Landsat 8 Operational Land Imager (OLI) (resolution:
30 m
Map of Greenland
Injection heights into the atmosphere of the emitted smoke were simulated
with version 2 of the plume rise model (PRM; Paugam et al., 2015) which is
implemented in the Global Fire Assimilation System (GFAS) emission inventory
(Rémy et al., 2017). The model (hereafter referred to as PRMv2)
is a further development of PRM (Freitas et al., 2007, 2010) and has already
been used in previous studies of fire events (Evangeliou et al., 2015, 2016).
The model simulates a profile of smoke detrainment for every single fire,
from which two metrics are extracted: (i) a detrainment layer (i.e., in which the
detrainment rate is
Wildfires in boreal peatlands in the Canadian Arctic and in Alaska typically
have (shallow) burn depths of 1–10 cm and consume 20–30 t C ha
Estimation of the emissions of BC, OC and BrC,
The emissions of BC, OC and BrC obtained from Eq. (1) were fed to the
Lagrangian particle dispersion model FLEXPART version 10.3
(Pisso et al., 2019) to simulate transport and deposition.
This model was originally developed for calculating the dispersion of
radioactive material from nuclear emergencies, but since then it has been
used for many other applications (e.g., Fang et al., 2014; Stohl et al.,
2011, 2013). The model has a detailed description of particle dispersion in
the boundary layer and a convection scheme to simulate particle transport in
clouds (Forster et al., 2007). The model was driven by hourly
To compare BC and OC concentrations in Greenland due to the emissions of the
Greenland fires to those due to emissions occurring elsewhere, we used the
so-called “retroplume” mode of FLEXPART for determining the influence of
other sources. For only a few receptor points, this mode is computationally
more efficient than forward simulations. Computational particles were tracked
30 days back in time from four receptor regions: northwestern (
Start and end date of releases, source of data, type of sensor, burned area and daily increment of burned area, fuel consumption and calculated BC emissions from Eq. (1) during the Greenland fires in 2017.
RS – remote sensing. MSI – multispectral images. SAR – synthetic aperture radar.
The IRF of the emitted substances of interest was calculated using the
uvspec model from the libRadtran radiative transfer software package
(
The IRF was calculated for three scenarios: (a) BC-only, (b) BC and BrC and
(c) BC and BrC, for which all OC is considered to be BrC. The BC-only scenario
demonstrates the impact of BC alone, while the two other scenarios provide an
estimate of the additional impact of BrC in the plume, with the last scenario
considered to be a maximum estimate. We calculated both the bottom-of-the-atmosphere (BOA) and top-of-the-atmosphere (TOA) instantaneous radiative forcing
(IRF) due to the Greenland fires at
To confirm the presence of the emitted substances from the Greenland fires
and elsewhere in the atmosphere over Greenland, we used the AERONET (AErosol
RObotic NETwork) data (Holben et al., 1998). AERONET provides globally
distributed observations of spectral aerosol optical depth (AOD), inversion
products and precipitable water in diverse aerosol regimes. We chose data
from three stations that were close to the 2017 fires and for which
cloud-free data exist for most of the simulated period, namely Kangerlussuaq
(
To examine in particular the vertical depth of the smoke, we used data from
the CALIOP (Cloud-Aerosol Lidar with Orthogonal Polarization) lidar on the
CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations)
platform (Winker et al., 2009). CALIOP provides profiles of backscatter at
532 and 1064 nm, as well as the degree of the linear polarization of the
532 nm signal. For altitudes below 8.3 km lidar profiles at 532 nm are
available with a vertical resolution of 30 m. We have utilized the level 1
data products (version 3.40) of total attenuated backscatter at 532 nm. This
signal responds to aerosols (like BC, OC and BrC) as well as water and ice
clouds, which in most cases can be distinguished based on their differences
in optical properties. The data were downloaded from the ICARE Data and
Services Center (
Table 1 reports burned areas in August 2017 calculated for Greenland. In total, 2345 ha burned between 31 July and 21 August 2017 (Fig. 1). We estimate that about 117 kt of carbon was consumed by these fires. The area burned is not large compared to the global area burned each year (464 million ha), or the areas burned in boreal North America (2.6 million ha) or boreal Asia (9.8 million ha) (Randerson et al., 2012), but it is still highly unusual for Greenland.
It is not yet known how these fires started. Fires on carbon-rich soils can be initiated by an external source, e.g., lightning, flaming wildfire and firebrand, or self-heating. The fires burned relatively close to the town of Sisimiut, so it is quite possible that humans started the fires. Self-heating is another possibility as porous solid fuels can undergo spontaneous exothermic reactions in oxidative atmospheres at low temperatures (Drysdale, 2011; Restuccia et al., 2017b). This process starts by slow exothermic oxidation at ambient temperature, causing a temperature increase, which is determined by the imbalance between the rate of heat generation and the rate of heat losses (Drysdale, 2011). Fire initiated by self-heating ignition is a well-known hazard for many natural materials (Fernandez Anez et al., 2015; Restuccia et al., 2017a; Wu et al., 2015) and can also occur in natural soils (Restuccia et al., 2017b). Southwestern Greenland was under anticyclonic influence during the last week of July, and according to the MODIS EOSDIS Worldview tool, direct sunshine occurred for 8 consecutive days before the fires started at the end of July 2017. It might be possible that this long period of almost continuous insolation at these latitudes in July heated the soil enough to self-ignite. In any case, the continuous sunshine had dried the soil, making it susceptible to fire.
The fact that these fires were burning for about 3 weeks but spread
relatively slowly compared to above-ground vegetation fires indicates that
the main fuel was probably peat. The predominant vegetation in western
Greenland varies from carbon-rich
Literally no fires should be expected in Greenland, since there is little available fuel as it has been suggested by global models and validated by observations (Daanen et al., 2011; Stendel et al., 2008); the only way to provide substantial amounts of fuel in Greenland is permafrost degradation. However, it has been suggested that significant permafrost loss in Greenland may occur only by the end of the 21st century (Daanen et al., 2011; Stendel et al., 2008). The fires in 2017 might indicate that significant permafrost degradation has occurred sooner than expected.
We estimate that about 23 t of BC and 731 t of OC, including 141 t of BrC,
was released from the Greenland fires in August 2017 (Table 1). According to
the FLEXPART model simulations, these emissions were transported and
deposited as shown in Fig. 2. Due to the low injection altitude of the
releases within the boundary layer, transport was relatively slow, and thus
the emitted substances initially remained quite close to their source. Slow
transport was also favored by mostly anticyclonic influence during the first
half of August. It seems that even though katabatic winds from the Greenland
ice sheet occasionally transported the plume westwards, most of the time the
large-scale circulation pushed the plume back towards Greenland (see
animations in the Supplement). Consequently, a large fraction of the emitted
substances were deposited in southwestern Greenland. On 3 August a small
portion of the emitted BC, OC and BrC (0.5, 16.1 and 3.1 t, respectively)
were lifted higher into the atmosphere and were transported to the east and
deposited in the middle of the ice sheet over the course of the following 2 days (4 and 5 August). From 5 to 8 August, when the fires were particularly
intense, the emitted aerosols were transported to the south, where they were
deposited on the southern part of the ice sheet and close to the coastline.
At the same time, another branch of the plume was moving to the north
depositing BC, OC and BrC over Greenland's western coastline up to
80
Total (wet and dry) deposition of
The total deposition of BC, OC and BrC from the fires in Greenland was estimated to be 9, 280 and 54 t, respectively, or about 39 % of the total emissions. About 7 t of BC, 218 t of OC and 42 t of BrC were deposited on snow- or ice-covered surfaces, which is equivalent to 30 % of the total emissions. Most of the rest was deposited in the Baffin Bay between Greenland and Canada and in the Atlantic Ocean. With 30 % of the emissions deposited on snow or ice surfaces, Greenland fires may have a relatively large efficiency for causing albedo changes on the Greenland ice sheet.
By comparison, the respective BC deposition on snow and ice surfaces over
Greenland from global emissions of BC (from ECLIPSEv5) was only 0.4 %
(39 kt) of the total emissions. Even the total deposition of BC in the
Arctic (
We also calculated the concentration of the deposited carbon aerosols in
Greenland snow (Fig. 3) by taking the ratio of deposited quantities and the
amount of water deposited by rain or snowfall during the same time period
(31 July to 31 August 2017). As expected, snow concentrations show the same
general patterns as the simulated deposition with the highest concentrations
obtained close to the source (western side of Greenland). High snow
concentrations were also computed in some regions of the ice sheet due to
relatively intense precipitation events. By contrast, dry deposition (example
for BC) over the ice sheets was low (Fig. S4). Dry deposition was only responsible
for a major fraction of the deposition in regions where the plume was
transported during dry weather, and in most of these regions total deposition
was low. A notable exception is the region close to the fires, where dry
deposition was relatively important due to the generally dry weather when the
fires were burning. It can be also ascribed to the fact that dry deposition
occurs in the quasi-laminar sublayer close to the surface. A fraction of
the aerosols can be quickly deposited close to the sources before they are
transported to higher altitudes and away from the sources (Bellouin and
Haywood, 2014). The average calculated snow concentration of BC on the ice sheet was estimated to be
Calculated snow concentrations of
It has been reported that the size of rapidly coagulated aerosol particles
produced by different types of fires ranges between 0.1 and 10
In summertime 2017, intense wildfires were reported in British Columbia, Western Canada (NASA, 2017c), and fires also burned at midlatitudes in Eurasia, as is typical during spring and summer (Hao et al., 2016). Previous studies of wildfires have shown that the produced energy can be sufficient to loft smoke above the boundary layer by supercell convection (Fromm et al., 2005), even up to stratospheric altitudes (Leung et al., 2007). As a result, emitted aerosols can become subject to long-range transport over long distances (Forster et al., 2001; Stohl et al., 2007). To examine the impact of these fires in Greenland, average footprint emission sensitivities were calculated for four compartments of Greenland (northwestern, southwestern, northeastern and southeastern Greenland) for the period 31 July to 31 August 2017, and the results are shown in Fig. S6 together with the active fires in the Northern Hemisphere from 10 July to 31 August 2017 adopted from the MODIS satellite product (MCD14DL) (Giglio et al., 2003). As can be seen in Fig. S6, fires in Alaska and in Western Canada might have affected BC, OC and BrC concentrations in Greenland, as the corresponding emission sensitivities are the highest in North America. In contrast, emissions from fires in Eurasia seem to have affected Greenland less.
Using gridded emissions for BC and OC, the contribution of both biomass
burning and anthropogenic sources to surface concentrations in the four
different regions over Greenland (northwestern, northeastern, southwestern
and southeastern Greenland; Fig. S7) was calculated (see Sect. 2.3). Fires
affected the northern part of Greenland more than the southern part, with an
average BC concentration of about 30 ng m
As an example of the importance of northern hemispheric biomass burning
emissions for the air over Greenland, we present time series of surface BC
concentrations in northwestern, northeastern, southwestern and southeastern
Greenland from the fires in Greenland and from all the other wildfire
emission sources occurring outside Greenland (Northern Hemisphere) for the same
period of time (Fig. 4). The calculated dosages (concentrations summed over a
specific time period) for the same time period were also computed. The fires
in Greenland affected mainly its western part with concentrations that
reached up to 4.8 ng m
There are few observations available that can be used to evaluate our model results. We use the AERONET and CALIOP data for some qualitative comparisons. We present only BC here, but similar plots can be generated for OC, considering that we used the same scavenging coefficients as for BC to represent the similar lifetimes of BC and OC (Bond et al., 2013; Jo et al., 2016; Lim et al., 2003). Contours of simulated vertical distribution of BC and column-integrated simulated BC from fires inside and outside Greenland are plotted together with time series of measured AOD (fine- and coarse-mode AOD at 500 nm and total AOD at 400 nm) for the AERONET stations Kangerlussuaq, Narsarsuaq and Thule (Fig. 5). It can be seen that observed AOD variations were in very good agreement with the variation of simulated column-integrated BC from fires outside Greenland (mainly in Canada), confirming that the transport of these fire plumes was well captured by FLEXPART. Good examples are the peaks at Kangerlussuaq on 24 August, at Narsarsuaq on 19 August and at Thule on 21 August (Fig. 5) that are attributed to the Canadian fires. The simulated contribution of the Greenland fires to simulated BC burdens was negligible by comparison, except at Kangerlussuaq in the beginning of August when the Greenland fire emissions were the highest. This station is less than 100 km away from where the fires burned, but not in the main direction of the BC plume transport. It seems the period of simulated fire influence corresponds to a small increase of the observed AOD values of up to 20 % (Fig. 5).
Contour plot of the vertical distribution of simulated BC (altitude
a.g.l. shown on left
To evaluate the smoke plume's vertical extent, we used the CALIOP data. These
data were only available from 5 August 2017 onward and frequent dense cloud
cover inhibited lidar observations at the altitudes below the clouds. High
aerosol backscatter was only found in the close vicinity of the fires.
Figure 6a shows NASA's EOSDIS view of the plume on 14 August 2017 at
06:00 UTC (available at
BOA IRF due to (a) BC-only, (b) BC and BrC and (c) BC and BrC when all OC was
assumed to be BrC (extreme scenario) for noon on 31 August 2017 is depicted
in Fig. 7a–c. This day is shown because almost all the aerosols emitted by
the fires had been deposited, thus giving a high IRF via albedo reduction due
to snow contamination. The IRF is the largest over ice close to the fire site
and at locations where relatively large amounts of BC and BrC were deposited.
For BC-only, the maximum BOA (TOA) IRF is 0.63 W m
The instantaneous direct BOA RF due to
The IRF depends on the optical properties of the smoke from the fire, which are not known. Hence, a sensitivity analysis was performed in which the single scattering albedo (SSA) was perturbed in contrast to a “medium case” (Fig. S8a) that was adopted from the SNICAR model (Flanner et al., 2007, 2009) and was used for the discussion in the previous paragraph. To estimate the uncertainty due to the choice of BC optical properties, additional calculations were made by scaling the SSA (red solid lines in Fig. S8a). The choices of these scaled SSA values were based on the SSA reported for various modified combustion efficiencies (MCEs) by Pokhrel et al. (2016).
Pokhrel et al. (2016) reported an MCE of 0.9 for peatland. As such, our adopted SSA may be considered low (compare black solid line and red line with upward-pointing triangles). Figure S8b shows the IRF as BC is deposited for the three cases. It suggests that the IRF ranges between 40 % and 130 % of our above-assumed medium-case values for realistic variation of the aerosol optical properties.
Figure 7d depicts the temporal behavior of the cloudy TOA IRF averaged over
Greenland (daily averages) for BC-only (red line), for BC and BrC (blue line)
and for BC and BrC, when all OC is assumed to be BrC (black line; extreme
case scenario). The daily averaged IRF is seen to increase as the plume from
the fires spreads out and starts to decline after the fires were extinguished
at the end of the month. The fact that the reduction towards the end of August is
relatively slow is caused by the effect of the albedo reduction, which
persists until clean snow covers the polluted snow. Overall, albedo reduction
dominates the total IRF averaged over Greenland for the period of study, contributing between 85 % (in the beginning of the study period) and 99 %
(at the end of the study period) and increasing in relative importance with
time as atmospheric BC and BrC are removed. The largest IRF differences
between the BC-only case IRF and the two
According to Hansen et al. (2005) the TOA IRF of BC approximates the adjusted
radiative forcing (RF) as reported by
Myhre et al. (2013). In their Table 8.4, Myhre et al. (2013) estimated the global
averaged RF due to BC between the years 1750 and 2011 to be
The albedo reduction at 550 nm for the three scenarios (BC-only,
We studied atmospheric transport, deposition and impact of BC, BrC and OC
emitted as a result of unusual open fires burning in Greenland between
31 July and 21 August 2017. Our conclusions can be summarized below.
The fires burned on peatlands that became vulnerable by permafrost
thawing. The region where the fires burned was identified previously as being
susceptible to permafrost melting; however, large-scale melting was expected
to occur only towards the end of the 21st century. The 2017 fires show that
at least in some locations substantial permafrost thawing is already
occurring now. The total area burned was about 2345 ha. We estimate that the fires
consumed a fuel amount of about 117 kt C and emitted about 23.5 t of BC
and 731 t of OC, including 141 t of BrC. The Greenland fires were small compared to fires burning at the same time in
North America and Eurasia, but a large fraction of BC, OC and BrC emissions
(30 %) was deposited on the Greenland ice sheet. Measurements of aerosol optical depth at three sites in western Greenland in
August 2017 were strongly influenced by forest fires in Canada burning at the
same time, but the Greenland fires had an observable impact, doubling the
column-integrated BC concentrations at the closest station. A comparison of the simulated BC releases in FLEXPART with the vertical
cross section of total attenuated backscatter (at 532 nm) from CALIOP lidar
showed that the spatiotemporal evolution and particularly the top height of
the plume was captured by the model. We estimate that the maximum albedo change due to the BC deposition from the
Greenland fires to be about 0.006, whereas adding deposited BrC increases
albedo to 0.007 at maximum, which is too small to be measured. The average
instantaneous BOA radiative forcing over Greenland at noon on 31 August was
between 0.03 and 0.04 W m We conclude that the fires burning in Greenland in the summer of 2017 had a small
impact on the Greenland ice sheet, causing almost negligible extra radiative
forcing. This was due to the – in a global context – still rather small
size of the fires.
The very large fraction of the emissions deposited on the Greenland ice sheet (30 %) makes these fires very efficient climate forcers on a per unit emission basis. Thus, while the fires in 2017 were still relatively small on a global scale, if the expected future warming of the Arctic (IPCC, 2013) produces more and larger fires in Greenland (Keegan et al., 2014), this could indeed cause substantial albedo changes and thus contribute to accelerated melting of the Greenland ice sheet.
All data used for the present publication can be obtained from the corresponding author upon request.
The supplement related to this article is available online at:
NE performed the simulations and analyses, wrote the paper and coordinated the paper. AK performed the radiation calculations and wrote parts of the paper. VM and SZ performed GIS analysis for the burned area calculations. RP made all the runs for the injection height calculations using the PRMv2 model. KS analyzed satellite data for AOD and CALIOP. SE and AS coordinated and commented on the manuscript. All authors contributed to the final version of the paper.
The authors declare that they have no conflict of interest.
This study was partly supported by the Arctic Monitoring and Assessment
Programme (AMAP) and was conducted as part of the Nordic Centre of Excellence
eSTICC (Nordforsk 57001). We acknowledge the use of imagery from the NASA
Worldview application
(