INTRODUCTION

Fifty years after the first ‘Earth Day’ and nearly half a century after the scientific debate on forest dieback and decline began in Europe and the US (Siccama et al., 1982; Innes, 1987), which generated much social and some political action on global environmental issues (Lewis, 1990), the various direct and indirect effects of industrial pollution on the functioning and productivity of forest ecosystems are still poorly understood (Büntgen et al., 2014). This is particularly true for the boreal forests of the Northern Hemisphere (Gauthier et al., 2015; Girardin et al., 2016), a circumpolar biome that plays a major role in shaping the Earth’s carbon cycle and climate system (Bradshaw and Warkentin, 2015). Since the mid-20^th century, boreal forests in Eurasia and northern North America have become the dumping ground for large concentrations of anthropogenic air pollutants (McConnell et al., 2007).

Located c. 300 km north of the Arctic Circle in central Siberia (69°21’N, 88°11’E), Norilsk is the world’s northernmost city with over 100 000 inhabitants. Since the onset of Eurasia’s heavy industrialisation in the 1930s, pollution from the Norilsk mining complex (Kharuk et al. 1996) has destroyed vast areas of pristine taiga and tundra habitats on the Taimyr Peninsula (Bauduin et al., 2014; Korets et al., 2014). Experimental metallurgy in Norilsk began as early as 1938 (Table S1), refined nickel production started in 1942, and mining activities have continued to rapidly expand (Dolgikh, 2006). In 2018, the Norilsk complex released c. 1 805 200 tons of pollutants (Ministry of Ecology and Environmental Management of the Krasnoyarsk region, 2019), of which c. 98% was sulphur dioxide (SO₂), making this taiga region of approximately 24 000 km² the most polluted in the world (Blacksmith Institute and Green Cross Switzerland, 2013). Norilsk’s unrestricted emissions of highly noxious substances are not only extremely harmful to humans (Arutyunyan et al., 2014), but also to flora and fauna (Zhulidov et al., 2011; Telyatnikov and Prystyazhnyuk, 2014). The most recent release of more than 20,000 tons diesel oil in the summer of 2020 underlines the growing environmental threat of Norilsk’s industrial sector under Arctic warming and permafrost thawing (https://www.bbc.co.uk/news/world-europe-52977740).

Despite the unprecedented level of regional environmental destruction, and the obvious downwind effects of heavy metal pollution, a quantitative assessment of large-scale ecosystem devastation due to airborne pollutants from local emissions has not previously been undertaken. Moreover, there are no reliable measures on the rate and volume of pollutants from lower latitudes that are transported into the Arctic (Law and Stohl, 2007; Arnold et al., 2016). Nevertheless, we know that the highest atmospheric concentration of pollutants occurs in the once pristine and particularly vulnerable ecosystems of northern Siberia (Stohl, 2006; Arnold et al., 2016).

One prominent example of possible ecosystem dysfunction is the ‘Divergence Problem’ (DP) in dendrochronology: The idiopathic decoupling of tree growth to rising instrumental summer temperatures since around the 1970s (Smith et al., 1999; Briffa et al., 1998; D’Arrigo et al., 2008; Esper and Frank, 2009). Evidence for reduced sensitivity of tree growth to temperature has mainly been reported from forest sites at the high-northern latitudes. This alleged circumpolar phenomenon describes the apparent inability of formerly temperature sensitive tree-ring width (TRW) and maximum latewood density (MXD) chronologies to parallel the recent increase of instrumental temperature measurements. In addition to such low-frequency trend offset, some boreal sites indicate a failure of previously temperature-driven tree growth to reflect high-frequency, summer temperature signals in a warming world. Suggested explanations include drought stress, changes in stratospheric ozone concentration, non-linear growth-climate response, insufficient instrumental climate data for proxy calibration, as well as methodologically induced pitfalls associated with chronology development. Thus far, the causes, scale and consequences of the DP remain unknown (see Vaganov et al., 1999; Kirdyanov et al., 2003; Wilmking et al., 2005; Driscoll et al., 2005; Lloyd and Bunn, 2007; Pisaric et al., 2007; Büntgen et al., 2008, 2009; D’Arrigo et al., 2008; Esper and Frank, 2009 for further details). The DP therefore continues to cast a serious concern on the reliability of high-latitude tree ring-based temperature reconstructions. Despite some pioneering work by Stine and Huybers (2014), little to no attention has been paid to the role of anthropogenic air pollutants and/or atmospheric dimming to explain the DP. Moreover, forward modelling of tree growth under different pollution scenarios has never been applied systematically in pursuit of understanding the DP.

Here, we utilise hundreds of TRW samples from living and dead conifers surrounding the Norilsk industrial complex to reconstruct the sequence of events leading up to a historically unprecedented forest dieback and decline, in one of the world’s most polluted regions, since the 1930s. We expand our dendrochronological findings with high-resolution measurements of wood and soil chemistry to quantify the spatiotemporal extent of Norilsk’s devastating ecosystem damage. We develop a process-based forward model of boreal tree growth to detect and attribute the effects of changes in surface solar radiation (see Büntgen et al., 2020 for the methodological and conceptual extension of ‘Detection & Attribution’ to global change ecology). Considering reduced surface irradiance as a proxy for anthropogenic air pollutants, we find strong evidence to suggest ‘Arctic Dimming’ (AD) has substantially reduced tree growth since c. 1970. Similar to ‘Global Dimming’ (Wild, 2009), AD refers to a decrease in surface solar radiation in the second half of the 20th century, which is in line with independent long-term observations of sunshine duration, diurnal temperature range, pan evaporation, and, more recently, also satellite-derived estimates (Stanhill, 2005; Wild, 2016). In addition to major practical implications, reduced solar radiation at the Earth’s surface affects a large number of processes, including evaporation and associated hydrological components, snow and glacier melt, plant photosynthesis and related terrestrial carbon cycle dynamics, as well as the diurnal and seasonal course of surface temperatures (see Wild, 2009, 2016 for a review of changes in surface solar radiation). Causes for AD can either be external, such as changes in the amount of solar radiation incident on the planet at the top of the atmosphere, or internal, such as changes in the transparency of the atmosphere that modify the solar beam on its way to the Earth’s surface. Aerosols and clouds, which are not independent from each other (Ramanathan et al., 2001), have been considered the most likely factors for AD, and it is important to note that aerosols affect clouds in different ways depending on the levels of air pollution. Associated with anthropogenic pollution, aerosols can directly modify surface solar radiation by scattering and/or absorbing solar radiation in the atmosphere depending on their composition. Furthermore, aerosols can modify surface solar radiation indirectly through their ability to act as cloud condensation nuclei, thereby altering cloud optical properties and lifetime. As a semidirect effect, absorbing aerosols in heavily polluted regions are also known to heat and stabilise the atmosphere, which may inhibit cloud formation or dissolve existing clouds. In summary, all these factors and processes trigger dimming with increasing aerosol levels in the atmosphere (Wild, 2009), which are considered particularly high in the Arctic (Law and Stohl, 2007; Flanner, 2013; Stohl et al., 2013; Arnold et al., 2016).

MATERIALS AND METHODS

Dendroecology

To characterise the direct, downwind impact of airborne emissions from the Norilsk industrial complex on forest ecosystems, eight dendrochronological sites were established along a northwest-to-southeast transect through zones of different pollution levels (Fig. 1; Tables S2–S3). Across most of this transect, Siberian larch (Larix sibirica Ledeb.) is the main forest-forming species. At the southern end of the transect, Siberian spruce (Picea obovata Ledeb.) is intermixed with larch. The trees at site 1, located 50 km west of Norilsk, do not exhibit any visual evidence of damage or mortality due to the direct effect of pollution. At site 2, located 28 km west of Norilsk, the forest may be characterised as having c. 50% mortality, but at sites 3–7 the degree of tree mortality increases to 100%. At site 8, the south-eastern most location 85 km from Norilsk, c. 25% of larch and c. 90% of spruce trees were still alive when sampled.

Sites 1–4 were sampled in 2014 and 2015, and at sites 5–8, samples were collected in 2004. All eight sites may be described as open stands within the context of a forest-tundra ecosystem. At sites 2–8, sub-plots 1–3 of c. 0.15–0.2 ha were established where, in addition to core and disk samples, measurements of slope and exposure were also taken. At least 30 trees were sampled at each sub-plot. Located outside of the direct influence of Norilsk’s pollutants (Fig. 1), site 1 is considered the control/reference site, where only living dominant and co-dominant trees were cored at 1.3 m from the ground. Within the affected sites (2–8), both discs and increment cores were collected from every dead and living tree in each sub-plot at the same height. For each core or disc sample, TRW was measured at a resolution of 0.01 mm using a LINTAB measuring system (Rinntech-Metriwerk GmbH, Heidelberg, Germany). Depending on the integrity of discs, TRW was measured along multiple radii, including the longest radius (most annual rings), as well as the radius with the widest outermost rings. All TRW series were visually cross-dated and then averaged. To determine when a tree died, the TRW series from standing dead trees were cross-dated against distant site chronologies of similar species considerably south and west of Norilsk and presumed to be unaffected by Norilsk’s pollution plume while within the same dendro-climatic zone (Vaganov et al., 1996; Zubareva et al., 2003; Knorre et al., 2006; Kirdyanov et al., 2014). The year of tree death was defined as the next calendar year following the date of the outermost tree ring.

The number of larch and spruce trees analysed per site varied from 12 to 106 (Table S3). To estimate the pollution effect on radial tree growth, we developed TRW chronologies for each site. Depending on the course of the raw TRW measurements, we either fitted negative exponential functions, horizontal lines, or cubic smoothing splines with 50% frequency-response cut-off equal to 2/3 of the individual series length. We then divided the observed values by their associated curve values, thereby transforming each raw TRW series into a new timeseries of dimensionless indices (Cook and Peters, 1997). The final TRW chronologies were produced by computing the robust mean of the index TRW series from all trees per site (Fritts, 1976). To estimate possible effects of pollution on tree growth for regions not directly impacted by Norilsk’s emissions, we used a recently developed collection of Larix sibirica, Picea obovata and Larix gmelinii TRW chronologies from seven sites along a >1400 km long east-west transect that extends considerably south and west of Norilsk (Fig. 1; Table S4). These contributed chronologies share a significantly high degree of growth coherency (Table S5), permitting the use of their average as an expression of large-scale mean, boreal forest growth.

RESULTS

Here, we report on 88 724 annually resolved and absolutely dated TRW and wood chemistry measurements from 46 living and 503 dead Siberian larch and spruce trees that grew at eight sites along a c. 150 km northwest-to-southeast transect through the Norilsk complex (Fig. 1a). Each site represents a different pollution exposure level due to dispersal effects of the prevailing westerly winds (Table S2). The mean age of the analysed larch and spruce trees ranges from 71 to 216 years, with minimum and maximum ages 31–113 and 214–388 years, respectively. Varying from 0.31 to 0.84 mm, the site-specific, average annual radial stem growth rates are slightly lower for spruce than for larch (Table S3). A 45-day-long polar night, annual mean temperatures of circa −10°C, and daily cold extremes of up to −50°C that can occur from September to May (based on meteorological measurements at Dudinka, 1906–2012), constrain tree growth to just a few weeks between June and August.

The first year of larch decline leading to tree mortality coincides with the first year of smelter operation (Fig. 2a). Although tree mortality south-east of Norilsk started as early as 1938, the annual dieback rates of up to 5% remained relatively low at sites 3 and 4 until the 1960s. Mortality increased to 30% per year at site 4 in the second half of the 1960s, when the ‘Mayak’ mine opened and SO₂ emissions escalated (Table S1). By the early 1980s, all larch trees within 69 km east-southeast of Norilsk had died at sites 3–7, and at site 8 by the time of this reporting, only about 25% of the larch trees have survived. Protected from the dominant westerly airflow, site 1 and 2 west of Norilsk are less affected (Figs 1 and 2a). Forest dieback at site 2 gradually tracks SO₂ deposition until the ‘Nadejda’ complex started operation, after which dieback rapidly increased. Built in 1979, this latest smelter reached full capacity in the early 1980s, at which time Norilsk’s heavy metal production grew five-fold. By 1983, Norilsk’s annual industrial emissions peaked at 2 483 000 tons (Kharuk et al. 1996). Spruce dieback at site 7 started in the 1970s, about a decade after the larch decline, whereas at the furthermost site 8, spruce mortality rates of only 10% were reached as late as the 2000s (Fig. 2a). Almost all of the surviving conifers in the study area exhibit heavy crown damage, and the comparison of TRW near Norilsk with observations of regional summer temperature variability suggests an unprecedented decoupling in the second half of the 20th century (Fig. 2b). While TRW variations parallel June–August temperature means fairly well until around 1970 (r = 0.42, P < 0.01 for 1924–1969), the overall low growth rates of the surviving trees clearly depart from the recent warming afterwards (r = 0.02, P > 0.05 for 1970–2015).

The wood chemical analyses reveal a significant rise in S, Cu and Ni in all larch samples from site 4 (Fig. 3), with peak values during the late 1960s. Although the chemical profiles vary among individual trees, the elemental-specific patterns of site 4 exhibit a coherent picture of exposure to airborne pollution from Norilsk. Sulphur levels slowly start to increase in the 1930s, but do not significantly rise until the mid-1950s. Copper rises from background levels in the 1940s to peak concentrations in the mid-1960s. In contrast, none of the living trees from the ‘unpolluted’ reference site 1 show similar chemistry values (Fig. 3). However, substantial increases in S, Cu and Ni are found in the outermost rings of most of the dead larches at site 2. Spatiotemporal variation in the chemical concentration of the tree rings is corroborated by independent S, Cu and Ni measurements in the upper organic soil layer at each of the three sites (1, 2 and 4). The highest topsoil pollution is found at site 4 in 2014 (Table S2), when S, Cu and Ni concentrations peak at 2,660 (±188), 828 (±85) and 716 (±29) mg/kg (SD) respectively. The lowest soil contamination is found at site 1 (S = 359 (±27), Cu = 33 (±7) and Ni = 36 (±7) mg/kg). The medium level of soil pollution at site 2 agrees with the chemical concentrations found in the wood at the same site, consistent with the predominant west-northwest airflow that disperses polluting aerosols from Norilsk.

While trees in the vicinity of Norilsk have died of continuous exposure to uncontrolled industrial emissions since the 1940s, trees growing outside the area directly impacted by Norilsk’s industrial pollutants are suffering. Reduced growth rates across the boreal forest since around the 1970s are indicative of the DP and beg the question ‘what is hampering tree growth?’. To assess the putative large-scale, long-term effects of industrial pollution on the vigour of Siberia’s taiga, we compared the rates of tree growth from the VS-Lite model to those empirically derived from seven sites sufficiently far away from Norilsk that can be considered free from the obvious direct effects of industrial pollution (Fig. 1 insert; Table S4). The VS-Lite model reproduces the measured TRW until c. 1970 (Fig. 4a), after which the simulated and measured growth rates increase and decrease, respectively. While simulated and measured TRW during the 1901–1942 pre-industrial period is highly synchronous, correlation between the simulated TRW chronology and the measured TRW chronology is insignificant afterwards (P > 0.05). This obvious divergence between increasing simulated and decreasing measured TRW suggests the existence of a spatially extensive, negative forcing capable of dramatically countering the beneficial effects of the recent warming trend. After reducing incoming surface solar radiation (i.e. implementing dimming), the simulated TRW parallels the observed growth rates over the entire 20^th century and until 2018 (Fig. 4b; Tables S8). Correlations between simulated and measured TRW are now highly significant from 1901 to 2018 (P < 0.0001). After 10-year low-pass filtering, the measured TRW chronology correlates with the simulated TRW data (without dimming) at 0.48 and −0.38 over the 1901–1969 and 1970–2018 periods respectively (Fig. 4b). Both values, and particularly the later one, significantly increase to 0.62 and 0.79 when dimming is incorporated into the VS-Lite model.

DISCUSSION

Although Norilsk is an extreme example of industrial pollution, it is not atypical for the high-northern latitudes that are more polluted than anticipated (Arnold et al., 2016). Since the mid-20th century, continuous SO₂ emissions from high-latitude mining, oil and shipping activities (Fig. S1; Table S9) have affected biogeochemical cycles not only over central and eastern Siberia (Shevchenko et al., 2003), but also across much of the circumpolar Arctic (Hirdman et al., 2010; Bauduin et al., 2014; Panyushkina et al., 2016). Moreover, the high-northern latitudes have long been impacted by transported pollutants, including detailed reports of ‘Arctic Haze’ events since 1883 (Garrett and Verzella, 2008). ‘Arctic Haze’ results from the strong inversion often referred to as the ‘Polar Dome’ (Stohl, 2006). This circulation pattern is typical of the wintertime Arctic boundary layer, which facilitates transportation and accumulation of air pollutants from lower latitudes into the Arctic (Stohl, 2006; Arnold et al., 2016). Unprecedented concentrations of pollutants (McConnell et al., 2007; Smith et al., 2011) have driven both historic cooling and amplified recent warming in the Arctic (Acosta Navarro et al., 2016). Transport from lower latitude Asia and Eurasia increases their content in the upper atmosphere, whereas local emissions of aerosols substantially affect the transparency of the atmosphere and the formation and structure of clouds (Stohl et al., 2013; Zhao and Garrett, 2015). These have been shown to curtail surface irradiance and to cause light diffusion (Stine and Huybers, 2014), with the strongest impact over central and eastern Siberia, as well as parts of Alaska and Canada (Briffa et al., 1998). In addition to the antagonistic role of anthropogenic pollutants, the net primary productivity of the boreal biome is also vulnerable to other abiotic stressors (Beck et al., 2011; Gauthier et al., 2015; Girardin et al., 2016; Charney et al., 2016), such as the redistribution of nitrogen as a consequence of widespread wildfires that are expected to increase in frequency and intensity under future climate change (Ponomarev et al. 2018; Shvetsov et al., 2019; Knorre et al., 2019; Kirdyanov et al., 2020).

The most important Eurasian boreal forest species, larch and spruce, exhibit different leaf morphologies and survival strategies. The deciduous larch habit minimises winter desiccation, whereas evergreen spruce develops thick cuticle to reduce foliar water loss (Miranda and Chaphekar, 1980). The thickened spruce cuticle may also afford greater protection from foliar leaching due to the deposition of acidic pollutants. Since larch foliage has a greater specific leaf area, higher stomatal conductance and lower water use efficiency than spruce (Kloeppel et al., 1998), it might be particularly sensitive to ozone damage (Wieser et al., 2013). Finally, there is reason to believe that boreal tree growth will suffer from reduced nitrogen uptake as regional humidity increases and the vapour pressure deficit decreases (Lihavainen et al., 2016). Consequently, the expected large-scale ecosystem disturbance due to anthropogenic warming will have serious implications for the Earth’s climate system and carbon cycle.

Based on regional- to large-scale modelling, we now suggest that the recent failure of formerly temperature-sensitive boreal tree growth to track increasing instrumental summer temperatures can be attributed to reduced incoming surface solar radiation (Stine et al. 2014; Arnold et al., 2016; Wild, 2016). This finding is consistent with our understanding of the wide-spread ‘Arctic haze’ phenomenon (Law and Stohl, 2007), which has been observed since the beginning of the 20th century (Garrett and Verzella, 2008), and has been described as a key driver of AD. We interpret these results as a consequence of reduced solar radiation, attributable to increasing aerosol concentrations (Quinn et al., 2007) due to elevated industrial emissions since the 1930s both emitted in and transported to the Arctic (Law and Stohl, 2007). This assumption is supported by the temporal agreement between the evolution of the DTR and industrial emissions: Both were low before the mid-20th century (McConnell et al., 2007), then in the early-1970s DTR and industrial emissions started to simultaneously display clear parallel trends. The link between industrial emissions and surface solar radiation may be direct (Rinke et al., 2004), or may be mediated by the effects of aerosols on cloud cover and cloud residence time through their impact on condensation nuclei and droplet size (Twomey, 1977; Ramanathan et al., 2001). Long-term changes in cloud cover may be also associated with an increase in atmospheric concentrations of greenhouse trace gases or changes in sea-ice cover (Taylor et al., 2015). At high latitudes, future scenario simulations (Trenberth and Fasullo, 2009) indicate that a stronger greenhouse forcing leads to greater cloud cover. Greenhouse trace gases, therefore, would reinforce the effect of aerosol emissions on cloud cover at high latitudes. However, disentangling the effects of increased greenhouse trace gases from AD remains subject to large uncertainties as aerosol-cloud interactions and greenhouse-cloud feedbacks are difficult to simulate in climate models, and Arctic aerosol-cloud interactions are not fully understood (Browse et al., 2014). The lack of spatially explicit measurements of air pollutants, cloud cover and solar irradiance during the first half of the 20th century from northern North America, Russia and China, does hinder these modelling analyses.

In conclusion, our study represents, for the first time ever, the combined evidence from dendroecology, biogeochemistry and process-based forward modelling to address the direct and indirect effects of industrial pollution on the functioning and productivity of boreal forest ecosystems at varying spatiotemporal scales. In addition to the quantification of the exceptional rate of environmental devastation around Norilsk, the world’s most polluted Arctic region, we demonstrate that anthropogenic-induced AD can explain the yet unresolved DP in tree-ring research. By doing so, we provide an important seal of the quality for any tree ring-based temperature reconstruction, because the Principle of Uniformity as it applies, sine qua non, to dendroclimatology and thus to a substantial part of high-resolution palaeoclimatology, remains intact. Our findings are expected to generate widespread environmental and political interest, because we add a unique perspective to the extraordinary level of persistent Arctic pollution, for which scientific and governmental awareness is still lacking. Deeper insights into the effects of long-term industrial emissions on the functioning and productivity of forest ecosystems across the high-northern latitudes further improve understanding of large-scale climate dynamics and changes in the global carbon cycle. This study appears particularly timely in the light of Norilsk’s unprecedented release of more than 20,000 tons diesel oil in 2020; an environmental disaster that emphasises the threat of Norilsk’s industrial sector under rapid Arctic warming and permafrost thawing, and also stresses the ecological vulnerability of the high-northern latitudes.

AUTHORSHIP

UB, AVK and EAV designed the study. AVK, AIF, AVP, ASS and VVK collected field data. AVK, VSM, AIF, AAK, VVB, AVT, DAM and ASP performed soil chemistry and dendrochronology. MAK, VAR and AAO produced forest disturbance maps. VVS and VAI developed the forward model. AVK, PJK and UB analysed data. JE, JB, AP, KTS, MW and EZ contributed to interpretation and discussion. UB and AVK wrote the paper with input from PJK, JE, JB, VVS and KTS. Each author provided critical discussion and approved submission.