The intensity, frequency, and duration of fires is rapidly changing globally1, altering the global carbon cycle and climate1,2,3. High latitude regions of the Northern Hemisphere (>50°N) have dense boreal forests and peatlands subject to major wildfire activity, emissions from which have approximately doubled (north of 60°) over the last decade4. The Arctic Oscillation-induced temperature increase appears to be critical for driving earlier snowmelt and fire activity, particularly in southeastern Siberia5. Aerosols and gases emitted from wildfires are predominantly carbonaceous in composition, but smoke plumes also carry significant amounts of bio-essential nutrients such as phosphorus6,7, nitrogen (N)8, and iron9,10. Although Russian observation stations do not routinely record information about N species, N deposition in other northern high latitude regions (e.g., North American High Arctic) is enhanced by wildfire smoke11,12. Consequently, wildfires can impact the Earth’s biosphere by altering plant productivity, biodiversity, and ultimately ecosystem carbon storage13,14.

Over the past two decades, wildfires have released substantial amounts of carbon in North America (60 Tg C year−1) and Asia (124 Tg C year−1)15. Major Arctic wildfire source regions include Canada, Scandinavia, and Russia4, but also Greenland16. Russia accounts for approximately two-thirds of the total burnt area within these countries4, highlighting the importance of understanding how changing fire activity in Russia under a warming climate could impact marine biogeochemical processes17.

Depending on the type of vegetation burned, the return interval for boreal forest fires ranges from a few decades to many centuries18. Return intervals are projected to decrease in the future, leading to more frequent and severe fires. More severe fires have the potential to release more aerosol to the atmosphere per unit time burned, and thus nutrients deposited within their plumes can be expected to increase. The coupling of wildfires and marine biogeochemical cycles is a recent development in our understanding of the Earth System19,20, and the impact of increasing boreal wildfires is yet to be assessed for Arctic Ocean marine primary production. Here, we suggest that boreal wildfires directly affect Arctic Ocean primary production by providing a new source of N, the macronutrient primarily limiting biological productivity in these waters21,22, and thus stimulating phytoplankton growth23.


In summer 2014, ocean color satellites captured one prolonged, or several short, phytoplankton blooms (reaching 1–2 mg chlorophyll a m−3, Fig. 1) up to 850 km south of the North Pole in the eastern Eurasian Basin. The pervasive cloud cover did not allow for continuous characterization of bloom dynamics, but snapshots of the ocean surface clearly revealed anomalously high chlorophyll a concentrations in the eastern Eurasian Basin, north of the Laptev Sea interior shelf. Given the spatial extent and short periods between missing retrievals, it was likely one single prolonged summer bloom. Also captured by satellite remote sensing were the exceptional sea ice conditions; although the onset of melt was two weeks later than the climatological mean, the consolidated ice pack disappeared in July and August at the most rapid rate ever recorded for this region and much further north (Fig. S1).

Fig. 1: Large summer phytoplankton bloom near the North Pole (eastern Eurasian Basin) in summer 2014.
figure 1

Satellite-derived mean chlorophyll a concentration within the region of the bloom (28–155°E, 80–85°N) during the summer of 2014 (a). Dot color represents which satellite sensor (MODIS Aqua, Terra, or VIIRS) is used. Dot size is relative to the number of observations obtained (i.e., pixels). The blue line is the climatological daily average of surface chlorophyll a concentration over the period 2003–2019 (except 2014) with the shading envelope corresponding to the interval between the first and third quartiles. Sea-ice concentration and sea surface temperature, for the full period July 28–August 31 (b), and for the three time periods July 27–28, August 13 –15, and August 29–31 (ce, respectively). Sea-ice concentration and chlorophyll a concentration, for the same dates as be, shown in panels fi. For bi: location of the bloom is within the dotted box (28–155°E, 80–85°N) and the continental shelf (bottom depth <50 m) is shown by cross-hatching.

To ascertain the potential source of new nutrients fueling this high latitude bloom we assessed several plausible mechanisms (see the supplementary results for a comprehensive evaluation of all potential mechanisms) of new N supply to the N-depleted surface ocean in summer (Fig. S2). Summarizing, storm events can mix N-rich deeper waters to the surface; however, winds remained weaker than the 10 m s−1 threshold generally required to induce rigorous vertical mixing24,25 (Fig. S3). Likewise, upwelling in the Arctic can transport deep nutrient-rich waters to the surface that support intense marine production26,27,28. The relative importance of this mechanism depends on regional factors such as sea ice cover, shelf depth, and wind direction with respect to the shelf break29. While upwelling favorable southeast winds over the Laptev Sea slope were observed in July–August 2014 (Fig. S3) they were moderate (max ~9 m s−1). Furthermore, the strong topographically-controlled eastward boundary current (positive zonal component, Fig. S4) clearly inhibited shelf break upwelling in this region, as confirmed by temperature and salinity mooring data (Fig. S4). On the contrary, the temperature and salinity sections show a downwelling event along the shelf slope. Thus, we argue that neither storm-induced mixing nor shelf break upwelling provides the N that stimulated the observed bloom north of the Laptev Sea in summer 2014.

Ocean dynamics may trigger changes to phytoplankton growth rates. Polyakov et al. (2017)30 argued that in recent decades (including 2014) increased vertical mixing in winter months (peaking in April) was driven by weakened halocline stratification and enhanced sea-ice production in the eastern Eurasian Basin. However, surface winter nutrient inventories are typically exhausted within two weeks by the spring bloom at the time of sea-ice retreat (which in 2014 began in early July; Fig. S1). Thus, any vertically mixed winter nutrients were likely exhausted prior to the onset of this late summer bloom. Further analyzing available records, we found a massive sub-surface anticyclonic eddy capable of introducing nutrients from deeper depths into surface waters (see section 2.3 in the Supplementary Notes; Figs. S5 and S6). However, its period of activity, low intensity in the upper water column, and position downstream of the current that influenced the bloom, suggest it had no effect on the bloom. The lateral advection of nutrients, particularly from river inflows, can be an additional source of new N to the open ocean. However, in situ studies clearly show that the Laptev Sea slope serves as a strong barrier preventing continental shelf waters from escaping into the deeper basin (especially in summer 2014 in the Laptev Sea26,27,28).

Excluding these physical mechanisms of N supply that increase phytoplankton growth promotes the hypothesis that an increased nutrient supply from the atmosphere supported the observed phytoplankton bloom. During the summer intense wildfires extend over large areas of boreal forests and peatland (Fig. S7), producing extensive smoke plumes that include N compounds (i.e., nitrous oxide, nitrite, and ammonia). Analysis shows Arctic wildfire activity and pollution enhancements are most pronounced in July and August31,32, when the 2014 bloom was observed. The region 115–125°E and 60–70°N, within the Sakha (Yakutia) Republic in Russia, is directly upwind of the observed bloom and MODIS recorded a burnt area (1486 kha) in July and August (Fig. 4a) that was approximately three-fold higher than the decadal average (Fig. S8). Furthermore, this examined Sakha Republic region is 7.5× smaller than Canada above 60°N, but the area burnt there in 2014 was two-thirds of the total area burnt in Canada above 60°N (Fig. S8 and Supplemental Text). The 2014 fires in Sakha, Russia were exceptionally large, and thus may provide an indication of future fire behavior.

Atmospheric circulation can transport continental wildfire smoke from Russia northwards to the Arctic Ocean. The Copernicus Atmosphere Monitoring Service (CAMS) reanalysis dataset (Fig. 2) and satellite images combined with back-trajectory analysis (Fig. 3b) both show how wildfire smoke emitted from Russia (Fig. 3a) traveled over the Laptev Sea and the eastern Eurasian Basin to reach the anomalous Arctic bloom of August 2014. Tiksi (128.9°E, 71.6°N; 1 m above sea level) is a remote coastal Siberian measurement station situated along this transport path. During the summer of 2014, four clear aerosol peaks (aerosol optical density; AOD) were recorded at Tiksi. All AOD peaks are dominated by fine mode aerosols indicative of a wildfire source, as opposed to coarse mode aerosol which would indicate other local natural aerosol sources (e.g., sea spray or dust). High elemental carbon (EC) concentrations measured at Tiksi (Fig. S9) further support the domination of AOD by fire aerosols. CAMS AOD over the bloom, and over the Laptev Sea en route to the bloom, shows three August peaks (Fig. 2). A time-lag between the original Tiksi AOD peak and two subsequent AOD peaks en route suggests that the smoke was transported from the continent to the bloom over a couple of days (see also: Fig. S10). However, the mid-July AOD peak at Tiksi is not observed over either Arctic Ocean region of interest, suggesting that smoke from fires in early July may have taken a different trajectory than later in July and August.

Fig. 2: Aerosol content over ground reference site and region of interest.
figure 2

Locations of the different regions of interest (a). Comparison of the aerosol optical depth at 550 nm based on AERONET measurements (level 2.0), CAMS archive, and CESM simulations with FINN fire emissions over the Tiksi-AERONET site (b). The color scale indicates the fraction of fine mode aerosol contributing to AOD as retrieved from the AERONET-Solar processing. Spatially averaged AOD for the regions of interest (c).

Fig. 3: Wildfires in Siberia, from nitrogen emissions to deposition in the eastern Eurasian Basin.
figure 3

Modeled FINN fire emissions of N (July 15–August 14 2014 anomaly versus the 2002–2019 mean) (a) with a box showing the location of panels (c, d). MODIS visualization of the aerosol plume and corresponding HYSPLIT back-trajectories highlighting atmospheric transport pathways (in red) to the location of the bloom (b). Simulated N deposition flux in 2014 (c) and the related deposition anomaly versus the 2011–2016 mean N deposition flux (d).

The Community Earth System Model (CESM; see “Methods”)33 includes N deposition from wildfires and is used here to estimate N deposition fluxes to the Artic Ocean. Simulations suggest that in late July–August 2014, N deposition fluxes were anomalously high (Fig. 3c), reaching over double that of the preceding and following years (Fig. 3d). Fires and plume transport directions are episodic34 (Fig. S10), and results show five consecutive deposition pulses, including one strong pulse in late July, three smaller pulses in early August, and one final large pulse on 21 August (Fig. 4b). Since Arctic Ocean waters are relatively stratified compared to other ocean basins, the estimated residence time of these atmospherically deposited aerosols could be sufficiently long to allow any associated N to accumulate, over a period of days35, to a sufficiently high enough concentration for phytoplankton uptake and subsequent growth.

Fig. 4: Nitrogen stock in boreal peat reserves and deposition in the region of the bloom (longitude: 128–155°E; latitude: 75–85°N).
figure 4

Stocks and spatial distribution of nitrogen contained in boreal peat reserves (a) and 2014 wildfire locations (a; teal dots). Time series of the estimated magnitude of nitrogen wet and dry deposition integrated over the region (128–155°E, 75–85°N) from CESM simulations (b). Uncertainty in deposition fluxes is estimated from missing peat burning emissions and the fire inventory uncertainties (factor of 2–5). Values given in Table 1 are the integrated deposition fluxes as shown here. Nitrogen in peat from Hugelius et al. (2020)46. Wildfire locations from MODIS where burnt area was > 0 in July and August 2014.

To assess whether the fires provided sufficient exogenous N to enhance production, we established a nutrient budget for the region based on current models of atmospheric N deposition (Table 1). Several lines of evidence suggest that the Fire INventory from NCAR (FINN) emission dataset36, used here in CESM simulations, likely underestimates boreal fire emissions. First, McCarty et al. (2021)4 showed that Arctic FINN fire emissions are between a factor of 2–5 lower than all other major fire emission inventories. Next, while the CAMS reanalysis dataset reproduces AOD observations at Tiksi (Fig. 2), both in magnitude and timing, atmospheric modeling of aerosol transport using FINN emissions in CESM captures only the timing of the AOD peaks; the AOD magnitude is lower than both CAMS and observations by a factor of 2–5, with the best-captured peak being the first one which may not have reached the bloom. In addition, Eckhardt et al. (2015)37 showed that global modeling, on average, underpredicts measured aerosol observations of EC taken at Tiksi by a factor of 3. Finally, CESM CO concentrations (a tracer of combustion) are lower than pan-Arctic observations by the same factor of ~2–5 (Fig. S11). Applying adjustments of 2–5× suggests that wildfires provided between ~12 and 30% (Table 1: standard model ×2 and ×5) of the N required to fuel the enhanced biomass associated with the bloom. However, the atmospheric deposition model we used is missing a significant source of N in sub-polar environments from N-rich peat fires (Fig. 4a), which are not represented in the FINN emission dataset. Assuming a ×3.5 factor is required to account for additional N emissions from peat fires (see “Methods”), the successive N deposition events we observed in the summer of 2014 could then support between ~40 and 100% of the total N content of the bloom (Fig. 4b and Table 1). Despite uncertainties in our knowledge of N emissions, and subsequent atmospheric deposition to aquatic ecosystems, the budget reconciles N transport when including peat fires in this region. Thus, in very N-depleted and highly stratified conditions in the Arctic Ocean, especially in the Central Arctic, significant deposition of N from peat fires will likely impact Arctic Ocean phytoplankton growth and productivity.

Table 1 Nitrogen budget for wildfire aerosol fueled Eastern Eurasian Basin phytoplankton bloom.

Discussion and conclusions

Here, our objective is not to demonstrate that this unusual summer bloom is entirely explained, or triggered, by the wildfire aerosol N deposition, but rather that it is most likely a significant contributor to its development and/or duration. Other mechanisms have been investigated and excluded as potential significant N sources, but it is plausible that other unidentified N sources may contribute to this unusually long bloom in the highly stratified and oligotrophic eastern Eurasian Basin. Nevertheless, the potential importance of wildfire aerosol deposition as an emerging mechanism in modulating Arctic Ocean biogeochemical cycles needs to be highlighted. Under the effect of climate change, these deposition events can supply surface waters with significant levels of nutrients, as in other ocean basins38,39. It is very likely that future fire-related deposition events will become more frequent and intense, with increasingly severe wildfires in peatlands and boreal forests. However, episodes of high fire activity remain unpredictable and are thus difficult to measure; from the deposition of aerosols to their impact on biogeochemical cycles, which may have a time-lag of several days35. The impact of wildfires (and more specifically aerosol deposition) on sea ice thermodynamic and dynamic properties is not investigated here, although they are likely to be interconnected40,41. Soot and associated nutrients from wildfires deposited on ice can (1) lead to increased melting by reducing its albedo, (2) be transported over large distances with sea ice (even in the Central Arctic) and, (3) when the ice melts, increase nutrient and freshwater fluxes that stratify the water column and make the upper ocean more suitable for phytoplankton growth. Given the degree to which shifts in sea ice surface properties can drive large-scale changes in under-ice ecology42,43, understanding the effects of wildfire-driven changes to sea ice, and the related effect on Arctic ecosystems, ought to be explored in future work.

Knowledge about atmospheric aerosol nutrient sources and deposition patterns is scarce at the pan-Arctic scale, with nearly all marine aerosol nutrient observations being collected below the Arctic circle44,45. Significant reserves of all peatland N (~80%) are currently stored within northern high latitude peatlands46 (Fig. 3a). Global warming is projected to result in reductions to the permafrost of peatlands by 50–100% with a warming of 2–6 °C relative to preindustrial times46. If increases in precipitation do not offset soil moisture losses through warming, then vegetation water stress increases and peatlands dry further and faster, making fuel more combustible and leading to an overall elevated risk in sustained major fire outbreaks. In this study, global climate model simulations, including N species contained in wildfire smoke, are used to test the hypothesis that Arctic biogeochemical cycles are sensitive to changes in boreal wildfire activity. However, the 2014 wildfires only occurred within regions of moderate peatland N stocks (Fig. 4a). Wildfires have been detected within regions containing peatland with a higher N store (Fig. S7), and thus future increased boreal wildfire activity combined with a more readily mobilized N stock from thawed peatland may rapidly amplify impacts of human perturbations to Arctic ecosystems. Russian burnt area accounts for ~2/3 of all high latitude regions combined in the present day4, yet it is understudied when compared to North America. Different N species have different atmospheric lifetimes and thus transport potential. The N species contained in smoke plumes is thus likely to be an additional important consideration in determining the changing contribution of wildfires to aerosol burdens and marine fluxes in different Arctic regions47. Increased observational efforts will thus improve model simulations and provide a better understanding of the impact of boreal wildfires on Arctic Ocean productivity.

Considering the N-depleted nature of Arctic Ocean surface waters, N-bearing aerosol deposits originating from wildfires will undoubtedly have repercussions on the nutrient and carbon cycles. Especially during the summer, when phytoplankton growth is severely N-limited4, these new N inputs could stimulate phytoplankton productivity and may partly explain the ongoing increase in annual primary production in the Arctic Ocean48,49. Phytoplankton growth is controlled by many factors, both physical and biogeochemical. Aerosol deposition, including from wildfires, is a source of new nutrients in many remote ecosystems50. In this study, global aerosol transport modeling suggests that Siberian wildfires supplied between 12 and 100% of the required N to support a large Arctic bloom in the summer of 2014, with the mass of N emitted from peat fires identified as a main uncertainty. Yet, nutrient aerosol addition may not always result in increased primary productivity17. Addressing the question of what initial conditions support aerosol-mediated phytoplankton growth should be explored further and will aid in understanding the evolution of the biogeochemical couplings between the land, ocean, and atmosphere under human-mediated climate change. These amplifying climate-driven changes, in addition to late summer/fall storms, will certainly promote secondary/fall blooms and thus contribute to the potential borealization of Arctic marine ecosystems25,51. The cascading effects of wildfire aerosols on different components of the Arctic ecosystem (land, atmosphere, sea ice, and ocean) create multiple questions that need to be assessed, quantified, and integrated into Arctic studies, in order to understand their implications on marine biogeochemistry in a changing global climate.


Ocean color and surface parameters

Ocean color

Chlorophyll a concentration, was inferred from retrievals of the three main ocean color satellite missions in orbit over the time period of this study (2014): MODIS-Aqua, MODIS-Terra, and VIIRS-SNPP; note that similar processing was used to generate the daily chlorophyll a climatology using satellite data available between 2003 and 2019. The composite products of the three missions were generated based on the recommendations of the Ocean Color - Climate Change Initiative52, but for a finer spatial resolution corresponding to a pixel edge of ~1 km. The remote sensing reflectance Rrs and chlorophyll a were obtained after atmospheric correction performed through the SeaDAS software53 and the combination of the spectral ratio and color index algorithms for chlorophyll a54,55. Note that the latter parameter was also estimated through regionally-tuned algorithms showing similar results (see Fig. S12). Image reprojection, binning, and aggregation were performed using the Sentinel Application Platform (SNAP) software, developed by Brockmann Consult.

Thanks to the high number of data acquisitions in this period (up to 8 for a given satellite per day), bad quality pixels were filtered out, and basic statistics were performed to provide quality controlled daily values. First, for each individual image, pixels were filtered out if one of the following quality flags provided by the SeaDAS algorithm is true: ATMWARN, ATMFAIL, HIGLINT, HILT, HISATZEN, STRAYLIGHT, CLDICE, CHLFAIL, and MAXAERITER. Due to particularly complex environmental conditions in Arctic seas for optical remote sensing, another pixel filtering procedure was performed when at least one the spectral bands exhibited Rrs values smaller than 0.0003 sr−1 or if the ratio between Rrs at 412 and 443 nm was >2.5. Note that pixels with aerosol optical thickness >0.2 were also filtered out to avoid misinterpretation of pixel information in the presence of aerosol plumes from the wildfires. The second step consists of calculating the daily median and standard deviation values for each pixel location of a given satellite mission. Once the daily product is obtained, the composite image was constructed by averaging the pixel values of the three distinct satellite missions.

Sea surface temperature

The SST was obtained from the MODIS-Aqua level-2 NASA products derived from long-wave (11–12 µm) thermal radiation. Daily SST images were generated following the similar approach to that applied for the ocean color products but using the following flags to filter out bad quality pixels: BTDIF, SSTRANGE, BTNONUNIF, CLOUD.

Atmospheric transport modeling

CESM version 1.533 was used with the interactive chemistry version of the Community Atmospheric Model (CAM-chem) as the atmospheric component56 following the set-up described in Bernstein et al. (2021)57. Aerosols in CAM-chem are represented by four modes (Aitken, accumulation, coarse, and a primary carbon)58. All simulations were performed on a horizontal resolution of 0.9 × 1.25° and 56 vertical levels with offline meteorology nudged to GEOS559 meteorological analysis. Dust and sea salt are prognostically calculated following the MAM4 default configuration58. Anthropogenic emissions are taken from HTAP-260. Daily fire emissions (from wildfires, agricultural fires, and prescribed burning) are taken from the Fire INventory from NCAR (FINN) dataset36 version 1.6 and prescribed vertically following AeroCom recommendations61 up to a maximum plume height of 5 km.

The CAMS reanalysis dataset incorporates satellite-derived AOD reanalysis using the ECMWF Integrated Forecasting System62. In this way, CAMS reduces bias with observations, and in the context of this study, provides an AOD reference with which to compare CESM results. The Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT)63, developed by NOAA’s Air Resources Laboratory, was used to simulate back trajectories from the location of the bloom (82°N, 130°E, August 12, 2014).

Current fire emission inventories combine data from different sources, on fire fuel loading, biomass consumed, and species-specific emission factors to estimate emissions over the area burned. This results in substantial uncertainties between commonly used datasets which differ by a factor of 4 globally and by factors of 3–15 regionally64,65. A second uncertainty is that when fire emissions are prescribed in global model simulations, there is a well-known low bias compared to aerosol optical depth measurements2,66, leading to regional adjustments of fire emissions commonly between a factor of 2–3.

The fire emission database FINN used here does not include peatland fire emissions, and thus model simulations do not transport any N from peat fires, despite fires occurring within regions containing peatlands (Fig. 4a). For our study region, FINN currently contains the lowest estimates of fire carbon emissions between all inventories: the highest estimates being a factor of ~5 larger14. In addition to peat fires being a large missing carbon source they are also a significant missing source of nitrogen owing to some species (e.g., HCN and NH3) being emitted from smoldering peat fires at a factor of ~10 higher, per unit biomass consumed, than for flaming savanna fires8 or a factor of 3.5 higher than boreal fires8. Therefore, to estimate the missing source contribution of peat fires to the N deposition flux, an emission ratio of 3.5 (peat fires:boreal fires) is used as a bias correction, based on emission factor differences between boreal forests and peatlands for some N species8 and the missing PM2.5 contribution from tropical peatland fires when using the FINN dataset67. Assuming linearity in transport of peat fire emissions with those from forest fires, we applied this factor 3.5 bias correction factor to the ‘inclusion of peat biomass’ nitrogen deposition flux in Table 1. Burned area was estimated using the collection 6 MODIS Global Burned Area Product MCD64CMQ (climate model grid)68.