Introduction—a Review of Arctic Change and the Carbon Cycle

Arctic Climate Change

In recent decades, the Arctic mean annual surface temperature has increased at over twice the rate of the global average [1, 2]. This polar amplification of surface air temperature is due to a combination of surface albedo feedbacks due to losses in snow and sea ice cover, cloud-sea ice interactions, lapse-rate change feedback, increased northward transport of heat and moisture, and increasing cloudiness and atmospheric water vapor, but the importance of these individual processes is unclear [3,4,5,6,7,8,9]. Arctic surface air temperatures over 2014–2019 have exceeded all previous years in the observational record going back to 1900 [2]. Winter surface air temperatures are warming most rapidly; e.g., during the winters of 2016 and 2018, temperatures were 6 °C above the 1981–2010 average [1]. In 2019, winter surface air temperatures in the Alaskan sector of the Arctic were 4 °C above the baseline period of 1981–2010 [2]. Box et al. [10] used the NCEP/NCAR reanalysis to show that temperature has been increasing at 0.7 °C/decade during the Arctic cold season and more slowly during the warm season, 0.4 °C/decade. Arctic climate change has been linked to anthropogenic radiative forcing [11], and Najafi et al. [12] showed using CMIP5 climate models that anthropogenic aerosols could have offset a significant amount of warming that would otherwise have occurred.

Arctic sea ice has markedly declined with the annual minimum extent in September decreasing by 13% per decade between 1979 and 2018 [1, 13]. Multi-year ice coverage (> 5 years old) decreased to less than 2% of winter sea ice area by 2018 [14, 15] and sea ice is thinning over time [1]. These reductions are unprecedented since the fourteenth century [16]; Notz and Stroeve [17] directly linked these changes to anthropogenic carbon emissions. The Arctic Ocean may become ice-free during summer months by the middle of the twenty-first century unless anthropogenic emissions are significantly decreased [18]. Lack of sea ice results in increased absorption of solar radiation by surface ocean waters and, combined with transport of heat from lower latitudes, Arctic Ocean heat content is increasing and summer mixed layer temperatures are increasing by 0.5 °C/decade [1, 19, 20].

Terrestrial snow cover extent is also decreasing as the Arctic warms. Mudryk et al. [21] found a strong link between warming air temperature and reduction of snow cover extent. Using multiple data sets, they showed that losses in snow cover extent are greatest in autumn and spring. The trend in snow cover extent (1981–2019) during May is − 3.4%/decade and − 15.2%/decade for June [22]. The duration of snow cover has also decreased over the past several decades by 2–4 days/decade [23]. Estimates of maximum snow depth-averaged over the pan-Arctic region are also showing declines over the past ~ 4 decades as well as shifts from frozen to liquid precipitation in relatively warmer coastal and low altitude environments [23]. Precipitation in the Arctic is also increasing [24]. Using the NCEP/NCAR Reanalysis covering 1971 to 2017, Box et al. [10] found that cold season (October-May) precipitation north of 50° N increased by almost 7%, and by nearly 5% during the warm season. An analysis of weather station observations by Wendler et al. [25] found a 17% increase in precipitation for Alaska over the past 67 years, with decadal variability affected by the phase of the Pacific Decadal Oscillation. As precipitation and temperature increased, so have evapotranspiration and river runoff into the Arctic Ocean [10, 26, 27]. Increased river discharge results in increased nutrient and organic carbon input to the Arctic Ocean and freshening of ocean water. Increased precipitation will not necessarily lead to increased soil moisture, however, since the evolution of soil moisture will depend on the balance between precipitation and evapotranspiration [28] as well as drainage from Arctic soils.

Arctic hydrology is strongly influenced by the presence of permafrost and subsurface ice. Permafrost underlies about 25% of the Northern Hemisphere land surface [29]. Biskaborn et al. [29] used borehole observations distributed throughout the Arctic from the Global Terrestrial Network for Permafrost to quantify temperature changes over 2007–2016 for the depth at which the annual soil temperature variation is zero. They found that soil temperatures at sites with continuous permafrost have increased over this period by about 0.4 °C, with smaller increases in discontinuous permafrost zones (0.2 °C). Liljedahl et al. [30] examined the role of subsurface ice-wedge evolution in reducing inundation and increasing runoff in tundra permafrost regions. Subsidence by thawing permafrost can lead to formation of new shallow lakes and ponds [31, 32], and about 20% of the Arctic land surface is covered by this thermokarst landscape. Ice-wedge thawing can lead to drainage of tundra soils, and thawing of permafrost along slopes and coastlines can lead to slum**, increased runoff, and transfer of organic material into rivers and the Arctic Ocean [33, 34]. The evolution of soil hydrology has important implications for the carbon balance of the Arctic since wetter environments rich in organic material favor anaerobic respiration that results in slower soil carbon remineralization and increased methane emissions relative to aerobic respiration. Using Landsat imagery for 1999 to 2014, Nitze et al. [33] found that lake area decreased by 1.4% for the region covered by their continental scale transects. The largest losses were found in regions of discontinuous permafrost, while in the continuous permafrost zone, some regions had expanding lake area and some had decreasing lake area. Other studies found increasing lake areas due to initiation of permafrost thaw [35, 36]. As permafrost thaws, the column of overlying soil that freezes during the cold season and thaws during the warm season (the Active Layer Zone; [37]) deepens and more soil carbon is available for efficient remobilization to the atmosphere where it can contribute to further warming, an effect known as the “permafrost carbon feedback”. Chang et al. [38] showed that a complication for modeling regional ALD is bias and non-representativeness of driving climate data and that this source of uncertainty can be as large as that due to unresolved spatial heterogeneity of soil hydrology and permafrost distribution.

An enormous amount of carbon is stored in Arctic soils, with estimates ranging from 1137 to 1850 PgC [39,40,41] including deep soils, and an unknown amount of carbon in subsea sediments [42]. Surface permafrost carbon (0–3-m depth) is estimated to contain 1035 ± 150 PgC [41], a significant amount of soil carbon considering that the permafrost region covers roughly 20% of global exposed land area and that total global soil carbon is around 3100 PgC [42]. Tarnocai et al. [40] point out that the northern permafrost region may account for about 50% of global soil organic carbon.

Submerged relic permafrost exists along shallow continental shelves, especially in the Laptev Sea. This subsea permafrost was formed when the sea level was much lower during the last glacial period [43]. Models of subsea permafrost indicate the possible continued presence of subsea permafrost and stability of intra-permafrost gas hydrates [44]. Over many centuries, it is possible that these sediments will thaw as sea surface temperatures rise [45], and Ferré et al. [46] showed that even small temperature increases could lead to destabilization of hydrates. However, this in itself is not guaranteed to lead to significant carbon emissions since other factors, such as sediment permeability must also be factored in [47].

Vegetation has also been changing in response to changing Arctic climate [48,44,50]. Changes in vegetation over the last several decades have been inferred from comparison of satellite observations of surface reflectance of near-infrared light, which is reflected by vegetation, and red light, which is absorbed by vegetation. This difference is known as the normalized difference vegetation index (NDVI). Maximum summer values of NDVI have been increasing across the Arctic (“greening”, [51]), especially for 1982–1998, likely due to longer growing seasons, warmer temperatures, and a more intense hydrological cycle. From 1999 to 2015, some regions exhibit negative (“browning”) trends, which also shows up in NDVI integrated over the growing season. Although peak summertime productivity has been increasing, during other periods of the growing season, increases are not as prominent and in recent years do not occur for the entire Arctic. NDVI changes could be interpreted as an expansion of woody vegetation, such as shrubs, and disturbance in the case of browning. However, Myers-Smith [52] highlighted some of the complexities in interpretation of vegetation index data. Standing water, snow cover, and soil moisture can influence surface reflectance of vegetated land. Vegetation indices can also be nonlinear with vegetation biomass so that they become more or less sensitive to changes over time with vegetation changes. Myers-Smith et al. [52] also highlighted the fact that spatial heterogeneity of Arctic vegetation and lack of coverage of in situ data make it difficult to compare changes in satellite vegetation indices directly with ground observations.

Biomass burning is an important disturbance in boreal ecosystems, and more frequently in tundra regions. Large amounts of carbon can be quickly released into the atmosphere from both above-ground biomass and soil organic matter. Over 1997–2016, van der Werf et al. [53] found emissions of CO2 from fires to average 185 TgC/year with minimum and maximum yearly emissions of 57 and 408 TgC/year. Emissions of CH4 are a small fraction of this amount since the CO2 emission factor is 250 times larger than that of CH4 for boreal forests and 81 times larger for peat fires which tend to smolder and produce reduced carbon such as CH4. Combustion of moss, peat, and litter were found to make up 85% of total combusted fuels for Canadian forest fires [54], and removal of this surface insulating layer can affect soil respiration and permafrost stability with further implications for soil hydrology [55, 56]. Turetsky et al. [57] demonstrated that late season, upland fires burn deeper and longer, sometimes even for multiple years, into the ground-layer than wetter peatland and permafrost environments. 2019 was a severe fire year across Siberia and Alaska, with record-breaking heat, and fires may have continued to burn deep in peat soils over the cold season and break out again during the 2020 warm season. These holdover fires, also known as “zombie fires,” have been observed in the Arctic in recent years and could accelerate permafrost loss and carbon emissions ([58]; https://phys.org/news/2020-05-scientists-zombie-arctic.html). Indeed, 2019 and 2020 both appear to be record-setting years for CO2 emissions from fires north of the Arctic Circle (https://www.economist.com/graphic-detail/2020/09/07/this-years-arctic-wildfires-are-the-worst-on-record-again, https://phys.org/news/2020-09-co2-emissions-arctic-wildfires-eu.html). After a burn, it can take many decades for recovery with possible changes to entirely different types of vegetation than were there initially [59, 60], for example, a transition from needleleaf to deciduous trees.

An analysis of Canadian forests by Coops et al. [61] found no long-term trend in burned area from 1985 to 2015; however, a trend towards an increasing area burned since 2006 was noted. Ponomarev et al. [62] found an increasing number of fires and burned area over recent decades for a transect in Siberia. However, there is large interannual variability in burned area, so longer records are needed to obtain clarity on trends. Satellite burned area products have been available since the late 1990s, and small but not statistically significant increases in area burned were found by Andela et al. [63]. Lightning is the most common ignition source for high-latitude fires, and Veraverbeke et al. [64] showed evidence for increasing lightning ignition between 1975 and 2015 for the Northwest Territories and Interior Alaska.

The Future of Arctic Climate

Past emissions and long lag times in ocean response mean that changes in Arctic climate will continue at least into the mid-twenty-first century [23], longer if no climate mitigation occurs. Surface temperatures will continue to increase at about twice the rate as lower latitudes, and depending on future emissions, they could rise at another 4–5 °C above late 1990s’ temperatures by the mid-twenty-first century. For high-emission scenarios, climate models (CMIP6) predict ice-free summers in the Arctic Ocean after the mid-twenty-first century [65]. By the middle of the century, snow cover duration is expected to decrease by 10–20% from present-day [23]. Earlier snowmelt will have implications for summer soil moisture and fire ignition. Glaciers will continue to melt leading to increases in sea level rise, and more coastal erosion exacerbated by permafrost thaw and the loss of protective landfast sea ice [66]. The hydrologic cycle is expected to continue to intensify with 30–50% increases in winter precipitation over the ocean and increasing runoff from land. The CMIP5 climate model ensemble also suggests that extreme precipitation events will become more common as variability of precipitation increases due to changes in moisture transport from mid-latitudes, possibly linked to changes in modes of variability such as the Arctic Oscillation, and the Pacific Decadal Oscillation [67]. At the same time, more precipitation may fall as rain rather than snow with further implications for hydrology [68]. Evapotranspiration is expected to increase. For the worst-case high-emission scenario (e.g., Representative Concentration Pathway leading to a radiative forcing of 8.5 W m−2 (RCP8.5) [69]), CMIP5 models show that Arctic temperatures could soar above late-twentieth-century levels by 13 °C in winter and 5 °C in summer [70].

In response to projected increases in Arctic air temperature, the CMIP5 models show significant changes to the Arctic permafrost distribution [23]. For RCP4.5, areas where discontinuous permafrost currently exists will disappear. For the highest emission scenario (RCP8.5), permafrost may only remain in the top 3 m of soils in the northernmost regions and at highest elevations. McGuire et al. [71] analyzed an ensemble of models with relatively detailed representations of permafrost and found that the model average suggests that 90% of near-surface (< 3 m) permafrost will be lost by 2300 for RCP8.5. For RCP4.5, the models predict an average loss of 29%. Models suggest that much of the permafrost area loss will occur by 2100, but they do not include the effects of abrupt thaw and fires on permafrost, both processes that could lead to more widespread and rapid permafrost loss. Recent model simulations that include abrupt thaw suggest a three to twelvefold increase in the amount of carbon that may be affected, depending on emission scenario [72].

Fire and other disturbances are expected to increase in boreal regions and Arctic tundra [73, 74]. For RCP6.0, Young et al. [74] found that the area of tundra burned in Alaska could double. Walter Anthony et al. [75] project an increase in thaw lake area, an indicator of abrupt thaw, of over 50% for the high-emission RCP8.5 scenario. Decreases in snow cover and changes in drainage due to permafrost thaw can further dry Arctic ecosystems, leading to increased burning and changes in vegetation.

Vegetation in the Arctic is expected to change significantly in the future. A statistical approach associating climate and vegetation types and considering two distinct emission trajectories found that trees and shrubs are projected to cover 24–52% of present-day tundra area by 2050, replacing current vegetation communities [76]. Expansion of woody vegetation into tundra regions may have additional effects that will feed back on climate through more evapotranspiration, albedo changes, and impacts on soil temperature [77]. Pearson et al. [76] argue that these effects could serve as positive feedbacks on regional Arctic climate.

Effects of Changing Climate on the Carbon Cycle

Arctic climate change will have important effects on the carbon cycle. Figure 1 summarizes schematically climate change and its effects on the carbon cycle. Currently, the Arctic Ocean is thought to be a small net carbon sink of 0.1–0.2 PgC/year [23]. Increased open water should result in increased CO2 uptake by the Arctic Ocean both due to physical gas exchange and increased marine productivity [78]. Freshwater input from rivers and melting ice could, however, change this picture. Increased freshwater could increase thermohaline stability inhibiting transfer of carbon from surface to deep waters, and nutrient loss from stratified surface waters could decrease productivity, thereby limiting the biological sink. Increased inputs of freshwater can reduce the buffering capacity of the Arctic Ocean and limit increased uptake of CO2 by delivering carbon to surface ocean waters. Export of carbon and nutrients from land to the Arctic Ocean can also increase productivity, especially in coastal regions [79].

Fig. 1
figure 1

Schematic illustrating Arctic climate change and its effects on the carbon cycle. Purple labels indicate carbon cycle responses. (Figure produced by S. Masais, NOAA Global Monitoring Laboratory)

Full size image

Warming of ocean waters could lead to enhanced CH4 emissions from subsea permafrost and possibly destabilized hydrates. Shakhova et al. [80] estimated that ~ 8 TgCH4/year is being emitted to the atmosphere from submerged hydrates in the subsea permafrost of the Laptev sea. However, emissions this large are not supported by atmospheric observations of CH4 abundance. Berchet et al. [81] found that emissions were likely to be less than half of that proposed by Shakhova et al. [80] and that atmospheric observations were more consistent with terrestrial or marine biosphere emissions rather than methane hydrates. Shipborne eddy covariance observations suggest ~ 3 TgCH4/year for the Eastern Siberian Arctic Shelf [82], in agreement with Berchet et al. [81].

On land, longer growing seasons and higher temperatures are expected to drive increased productivity, and greening trends seen in NDVI data are consistent with this. However, many factors add to the complexity of how the terrestrial Arctic carbon cycle will change. Warming soils could drive an increase in respiration that partially offsets increases in productivity [83]. Long-term eddy covariance studies show that opposing fluxes of photosynthesis and respiration mostly cancel each other out, and changes in the net carbon balance are small [84, 85]. Increases in precipitation and earlier snowmelt could affect soil moisture over the growing season, benefitting some plant communities while disadvantaging others. Expansion of woody plants into tundra regions will increase evapotranspiration [76], while thawing of permafrost could initially increase inundated areas due to subsidence leading to expanded methane-producing environments. In areas of discontinuous permafrost, drainage, encroachment of vegetation, and filling of shallower lakes will all have implications for carbon exchange [35, 86]. With drier conditions and possibly increased ignition by lightning, fires could increase in the Arctic releasing built up carbon from organic-rich soils [57].

The amounts of CO2 and CH4 emitted to atmosphere as permafrost thaws depend on the decomposability of organic matter stored in the soil and whether the decomposition occurs aerobically or anaerobically. Aerobic and anaerobic conditions are dependent on soil hydrology, which is susceptible to change as permafrost thaws [87]. As permafrost thaws, carbon release will initially increase rapidly as easily decomposed material is broken down leaving less labile carbon with slower release rates. Rates of carbon release are dependent on hydrological status, and field studies have shown considerable spatial variability in this over relatively small scales as shown by Chang et al. [38] who also pointed out that uncertainty in temperature and precipitation driving data could cause biases in CO2 and CH4 fluxes that are comparable to variability due to landscape heterogeneity. Methane is consumed in dry soil and oxic water columns [88, 89], and this must also be accounted for. In organic-rich, inundated environments, anaerobic respiration dominates, a slow process in comparison to aerobic respiration that leads to higher CH4 emissions. This is important because of the larger radiative impact of CH4 with a 100-year global warming potential that is about 25 times greater than CO2 on a per mass basis [98] inferred increased emissions of CO2 early in the cold season. Natali et al. [107] synthesized both chamber and eddy covariance flux measurements for the cold season (October-April) from over 100 sites throughout the Arctic and Boreal high latitudes using machine learning techniques with a variety of environmental drivers (vegetation type, soil moisture, soil temperature). Their approach is similar to that used by Jung et al. [108] to produce the FLUXCOM product based on global eddy covariance flux tower observations. Natali et al. [107] found that 1.7 PgC/year is emitted as CO2 from permafrost region soils during the cold season, an amount exceeding their modeled estimate of uptake during the warm season giving a net CO2 source. They also noted that they did not find any trends in cold season emissions for 2003–2017, which they attribute to a lack of circumpolar trends in the reanalysis data used in their algorithm. They did, however, find increases in winter respiration for site-level data from Alaska.

Figure 7 shows annual net CO2 exchange and cold/warm season fluxes for latitudes north of 60° N estimated using top-down and bottom-up approaches. Results from two atmospheric inversion systems (CarbonTracker and CarbonTracker-Europe) suggest that the Arctic is taking up more CO2 than is respired over a year. The prior flux estimates used in these inversions are the CASA-GFED model for CarbonTracker [174, 175] and SiB-CASA for CarbonTracker-Europe [176], both dependent on remotely sensed NDVI for estimating GPP. In the annual mean, both priors are nearly annually balanced with uptake about equal to respiration. The SiB4 terrestrial ecosystem model [177, 178], which has prognostic phenology and does not use NDVI, produces annual net CO2 fluxes that are similar to the prior estimates. FLUXCOM’s annual net flux lies between the inversions and bottom-up models. During the warm season, FLUXCOM shows significantly less uptake than the other estimates and also less respiration during the cold season. Likewise, the inversions estimate less cold season respiration than their prior estimates or SiB4. During the cold season, emissions range from ~ 0.8 PgC/year for FLUXCOM to over 2.5 PgC/year for SiB4. The estimate of Natali et al. [107], 1.6 PgC/year, falls between the inversions and the priors; however, their calculated warm season uptake (~ − 1.0 PgC/year) is significantly smaller than the inverse estimates by at least a factor of 3. These results show the difficulty in interpretation of annual net CO2 fluxes since they are a balance of respiration and photosynthesis for which estimates from different methods can significantly diverge.

Fig. 7
figure 7

a Annual net CO2 fluxes north of 60° N for two inverse models and their prior emissions, the SiB4 terrestrial ecosystem model, and FLUXCOM. b Warm season (May-September) CO2 fluxes. c Cold season (October-April) CO2 fluxes. Inversions shown are CarbonTracker ([169], CarbonTracker CT2019, http://carbontracker.noaa.gov) and CarbonTracker-EU (CTE2018, [170], http://www.carbontracker.eu). The prior flux estimates for both inverse models are based on terrestrial ecosystem models that use remote sensing observations to constrain GPP (e.g., NDVI). FLUXCOM is based on flux tower observations scaled to regional and global scales used machine learning techniques and ancillary data sets (for example, soil temperature). Note that negative fluxes indicate removal of carbon from the atmosphere and positive fluxes indicate emission to the atmosphere

Full size image

Conclusions and Recommendations

Multiple lines of observational evidence show that presently the remote Arctic is undergoing rapid environmental change. These changes have been directly linked to anthropogenic emissions [12].

Our understanding of Arctic climate tells us that there are important feedbacks in operation, such as between the cryosphere and the atmosphere, and between vegetation and atmospheric energy/moisture budgets. It is certain that changes in Arctic climate will drive changes in the carbon budget of the Arctic as vegetation changes, soils warm, fires increase, and wetlands evolve with permafrost thaw. Massive amounts of carbon are stored in Arctic soils, and some fraction of this carbon is likely to be mobilized to the atmosphere and oceans with consequences that will feedback to affect global climate. This permafrost carbon feedback needs to be understood, quantified, and taken into account when considering climate mitigation (e.g., [179]), and formulating policy designed to ensure temperatures remain below particular thresholds. This will require a commitment to long-term pan-Arctic observations, as well as improvements in models used to help better understand the Arctic climate system.

Observations currently support a more active CO2 cycle in high northern latitude ecosystems with both enhanced productivity and increased respiration. Evidence points to increased uptake by boreal forests and Arctic ecosystems. Evidence also points towards increasing respiration, especially late in the warm season. On the other hand, there is currently no strong evidence of increased CH4 emissions, across the region, although we demonstrated here using atmospheric observations that small increases cannot be ruled out. Sweeney et al. [100] pointed out that given the widely observed temperature dependence of microbial production of CH4 in Arctic wetlands, the response of atmospheric CH4 to temperature increases appeared to be much smaller than expected. They attributed this lack of response to a lack of understanding of processes leading to emissions. However, it is also possible that microbial consumption of atmospheric CH4 in drier upland soils is also increasing as temperatures rise and that this process has been significantly underestimated as proposed by Oh et al. [88]. Alternatively, a reduction in wetland extent and increased lake drainage may have played a role.

Finally, we point out that long-term observations help us to understand what has or is changing, but they are also critical for improving projections of future emissions. Long-term observations provide the valuable opportunity to test predictive models and their sensitivity to change. Increasing observational coverage of flux measurements, in situ atmospheric sampling, and remote sensing data from satellites and committing to maintaining data records over decades will improve our understanding of the Arctic carbon cycle and how it is changing over time.