Vegetation plays an important role in mitigating climate change by absorbing atmospheric carbon dioxide, reducing planetary albedo, and influencing the aerosol composition of the atmosphere. Feedbacks between vegetation and climate are particularly conspicuous in the Arctic region because secondary organic aerosol (SOA) from vegetation constitutes a considerable proportion of the atmospheric composition. In addition, increasing vegetation cover reduces surface albedo and leads to more absorption of solar radiation. This process, called “Arctic greening”, has intensified over the past few decades, resulting in increased Arctic plant productivity that is coincident with increases in Arctic surface air temperatures. Tools to study historical vegetation changes include dynamic vegetation modelling, the use of plant fossils, and other proxy data and proxy modelling comparisons.

The Lund–Potsdam–Jena General Ecosystem Simulator (LPJ-GUESS) is one important dynamic vegetation model for studying past vegetation–climate feedbacks and is an updated version of the LPJ Dynamic Global Vegetation Model (LPJ-DGVM) (Smith et al., 2001). LPJ-GUESS is one of the most widely used models to simulate past vegetation dynamics from landscape to global scales using a forest gap model scheme (Sitch et al., 2003; Smith et al., 2001). The LPJ-GUESS consists of a number of equations describing the biogeography, biogeochemistry, and biophysical processes of ecosystems. The biogeographic features of vegetation are mechanistically represented by plant functional types (PFTs), which are distinguished by different bioclimatic limitations. In the Arctic, boreal forests are generally formed by one or two dominant species, and the PFTs can be individual species, meaning that LPJ-GUESS can simulate vegetation dynamics at species or community levels.

The LPJ-GUESS model not only simulates vegetation growth but also interactions with other components of the climate system, which helps to quantify feedback loops. The model simulates the biophysical and biogeochemical processes of energy and matter exchange between the atmosphere, soil, and biosphere (Hickler et al., 2012; Sitch et al., 2003), which in turn modulate net primary productivity, vegetation structure and composition, and the carbon and nitrogen soil and litter budgets, including soil water. The model has been applied in various investigations, such as carbon cycle studies and investigations of fire occurrence and aerosol changes (e.g. Fang et al., 2015; Schurgers et al., 2009).

Pollen data are the most widely used vegetation proxy for climate reconstructions. Species apportionment of pollen is used as an indicator of historical vegetation composition, and the chronologies of pollen data are determined by dating surrounding sediments. However, caution must be exercised when analysing pollen data because pollen undergoes significant aeolian transport and its presence does not always prove the local presence of the coinciding plant species. Therefore, it is also useful to use pollen accumulation rate records to investigate species presence and abundance. For example, threshold data above 500 and above 300 grains cm-2 yr-1 are employed as indicators of the local presence of Pinus and Picea forests, respectively (Hicks, 2006; Seppä and Hicks, 2006). Macrofossil and megafossil records are large remains of plants and can be found in small lakes and ponds in the Arctic areas. The presence of such fossil records is a robust indicator of the presence of local sources. Furthermore, due to the carbon composition of macrofossils and megafossils, their ages can often be determined using radiocarbon dating methods.

Quantitative representations of vegetation structure and function, as well as interactions with other process-based model components, allow for the quantification of feedback loops that include historical vegetation changes. Although proxy-based records are not able to fully quantify vegetation structure and function, they can provide benchmarks, which can be used for validating process-based simulations. Furthermore, comparisons between proxy data and model simulations are useful tools for exploiting the advantages of both proxies and vegetation models. For example, LPJ-GUESS has been utilized to simulate the European Arctic treeline during the Holocene and compared with proxy-based treeline reconstructions. Utilizing an Arctic treeline threshold biomass value of 2 kg C m-2 leads to an agreement of the simulated and proxy-based treeline, with mismatches seen in mountainous areas (Fang et al., 2013).

Environmental archives, such as ice cores, peats, and lake and marine sediments, chronologically encapsulate material, including material deposited from the atmosphere (e.g. Rose and Ruppel, 2015). These archives can preserve long-term records for up to millennia (e.g. Petit et al., 1999; Zachos et al., 2001) and are essential when assessing past, present, and future Arctic climate change, both for setting modern variations into a broader context and for model validation. Despite the importance of such data in climate change evaluations, very few records of some important Arctic variables are available (e.g. BC; McConnell et al., 2007). Before CRAICC commenced Arctic BC deposition records were available only from high-altitude Greenland (McConnell, 2010), which do not represent BC deposition in the rest of the Arctic located closer to sea level (e.g. McConnell et al., 2007). BC records are more readily available from Antarctica (e.g. Bisiaux et al., 2012a, b), the Himalayas (e.g. Xu et al., 2009; Kaspari et al., 2011), and the European Alps (e.g. Lavanchy et al., 1999; Painter et al., 2013).

Black carbon originates from the incomplete natural and anthropogenic combustion of biomass and fossil fuels, and due to myriad emission sources and formation conditions, the term “black carbon” covers a wide spectrum of carbonaceous particles ranging from charred biomass to soot formed by gas condensation in high-temperature flames. Currently, no single accepted definition for BC exists, particularly between disciplines (Rose and Ruppel, 2015). Moreover, BC is an operational term, which means that its precise definition depends on the method used for its quantification. Coincidingly, no standard BC quantification method exists, and analyses of identical samples have shown that measurement methods differ by up to a factor of 7 in concentration (Watson et al., 2005). Schmidt et al. (2001) even reported concentration differences of a factor of 500 between measured BC in a soil sample inter-comparison study. Thus, comparing results between different methods remains challenging, particularly between disciplines, as, for instance, atmospheric measurements may quantify BC particles based on light absorption properties, while snow or soil measurements may extract and quantify BC based on chemical and/or thermal properties (e.g. Hammes et al., 2007).

Within CRAICC, two well-established and widely used methods were employed for BC analysis from a Svalbard ice core and four northern Finnish lake sediments. The Svalbard ice core is a 125 m ice core collected from the Holtedahlfonna glacier, which dates from ca. 1700 to 2004. It was analysed using a conventional thermal–optical method for elemental carbon (EC), which is a proxy for BC (Birch and Cary, 1996). After subsampling, ice samples were melted and filtered through quartz fibre filters and EC was quantified with a Sunset Instrument (Sunset Laboratory Inc., Forest Grove, USA) using the EUSAAR_2 temperature protocol (Cavalli et al., 2010) for determining the carbonaceous aerosol fraction on the filters (Ruppel et al., 2014). Finnish lake sediments were radiometrically (137Cs, 210Pb) dated covering ca. 150 yr BP, and were analysed for soot BC (SBC) with a chemothermal oxidation method (CTO-375) developed specifically for BC quantification from sediments (Gustafsson et al., 1997, 2001). After thermal removal of organic material and chemical removal of carbonates from the samples, SBC concentrations were determined with an elemental analyser (Ruppel et al., 2015). This method detects condensed SBC formed at high temperatures in gas-phase combustion, irrespective of the combusted material (Elmquist et al., 2006). Soot BC particles represent the smallest size fraction of BC, whereas the filter-based thermal–optical method used for the ice core samples may most effectively determine bigger char-type BC and agglomerated soot particles.

During the last decade, there have been considerable developments with respect to new online and offline techniques, typically based on mass spectrometry, for investigating the chemical composition of atmospheric gases and particles (Nozière et al., 2015; Glasius and Goldstein, 2016; Laj et al., 2009). In this section some of the methods utilized within CRAICC to study atmospheric particles in cryospheric environments and relevant laboratory studies are presented. Further detail of the Soot on Snow project, the sea spray aerosol simulation tanks, and the SMEAR (Stations Measuring the Ecosystem–Atmosphere Relations) stations is also given.

Molecular tracers – levoglucosan. The chemical speciation of particles provides information on the composition and processes involved in the formation and growth of those particles. Atmospheric particles are composed of a multitude of organic compounds (Goldstein and Galbally, 2007), and thus it is not feasible to completely elucidate their chemical composition. Instead, molecular tracers for specific sources or processes can be identified and investigated. An example of this is the use of levoglucosan as a tracer for biomass burning emissions in aerosol particles collected on Svalbard in the European high Arctic (Yttri et al., 2014).

Molecular tracers – secondary organic aerosols. Few studies have explored the formation and distribution of SOA in the Arctic. Hansen et al. (2014) investigated molecular tracers of biogenic and anthropogenic SOA in both North Greenland and Svalbard using the filter collection of particles followed by extraction and analysis by high-performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry (HPLC–qTOF-MS) using an electrospray ionization inlet. This methodology is well suited for analysis of the polar organic compounds often found in oxidized SOA, while less polar constituents, such as the alkanes characteristic of emissions from fossil fuels and their combustion products, are not observed.

Within the last decade organosulfates and nitrooxy organosulfates have been identified as an important, novel class of SOA constituents (Surratt et al., 2007; Iinuma et al., 2007). Organosulfates and nitrooxy organosulfates are analysed using HPLC–qTOF-MS and are identified from the presence of HSO4- (m/z=97), the neutral loss of SO3 (80 Da), and in the case of nitrooxy organosulfates an additional neutral loss of HNO3 (63 Da; Surratt et al., 2007). The influence of temperature on the gas-particle distribution of semi-volatile compounds is a major challenge to building a holistic understanding of aerosols in cold climates. Temperatures may change tens of degrees from the ambient air to collection or detection, leading volatile species to evaporate within sampling inlets. This issue requires careful consideration, and potential artefacts should be investigated and avoided using separate gas and particle sampling for offline analysis (Kristensen et al., 2016).

Inorganic ions and BC in aerosols. Particle size distributions of inorganic anions during Arctic haze were determined using size-selective collection (by MOUDI) followed by ion chromatography (Fenger et al., 2013). Furthermore, long-term monitoring data were used for the source apportionment of particles over 2 years (Nguyen et al., 2013), as well as BC and sulfate (Massling et al., 2015) at the Villum research station, Station Nord (81∘36′ N, 16∘40′ W), Greenland.

INPs. Cloud processes can also be influenced by ice-nucleating particles, which assist in the heterogeneous nucleation and growth of atmospheric ice (Vali et al., 2015). Within CRAICC a combination of particle and surface measurements has been utilized to compare and contrast how quantifiable material and thermo-kinetic properties affect ice nucleation efficiency in relation to the thermodynamic driving force. For studies of the ice nucleation proclivity of particles the CRAICC partners participated in the development of the Frankfurt isothermal static diffusion chamber for ice nucleation (FRIDGE) and the complementary electrostatic deposition unit (PEAC7) used to collect particle samples for FRIDGE analysis (Schrod et al., 2016; Thomson et al., 2018). The PEAC7 is a sampling unit for the electrostatic deposition of aerosol particles onto silicon wafer substrates and subsequent characterization in the FRIDGE temperature- and humidity-controlled chamber (Schrod et al., 2016). Using high-resolution photography of the substrate surface, INPs are directly counted as a function of temperature and water vapour saturation. Multiple benefits of the PEAC7 sampling unit include (i) enabling sampling in clean environments with very low ambient concentrations and (ii) enabling the identification of single INPs with further characterization using scanning electron microscopy. In CRAICC this powerful tool was deployed to two locations to characterize ambient INP concentrations. Those locations included the Villum research station and Nyålesund (Svalbard). Each of these locations is characterized by low ambient aerosol particle concentrations and thus suited to the PEAC7 sampling technique.

Within CRAICC a wide range of instrumentation was used to determine aerosol physical properties in laboratory studies and during field measurements. Some are well-established techniques only briefly described below, while others are more recently developed and thus described in somewhat more detail.

Mass spectrometry. Online quantitative measurements of particle chemical composition for non-refractory submicron aerosol particles were performed using an Aerodyne aerosol chemical speciation monitor (ACSM; Aerodyne Research Inc.; Ng et al., 2011). Gas-phase precursors participating in new aerosol particle formation were measured with several mass spectrometers including an atmospheric pressure interface time-of-flight mass spectrometer (APi-TOF; Aerodyne Research Inc. and Tofwerk AG) for the molecular composition of naturally charged ions and clusters (Junninen et al., 2010), a nitrate chemical ionization atmospheric pressure interface time-of-flight mass spectrometer (CI-APi-TOF; Aerodyne Research Inc. and Tofwerk AG) for neutral clusters like sulfuric acid and organic vapour (Jokinen et al., 2012; Kürten et al., 2014), and a proton-transfer-reaction time-of-flight mass spectrometer (PTR-TOF; Ionicon Analytik GmbH) for organic vapours (Graus et al., 2010). Each technique allows chemical species to be identified by their mass signatures and isotopic fractions (Junninen et al., 2010; Ehn et al., 2010; Schobesberger et al., 2013).

DMPS and SMPS. Aerosol particle size distributions were measured in the field and the laboratory using standard electrical-mobility- and optical-based techniques. Electrical-mobility-based instruments rely on the size separation of charged particles in a differential mobility analyser (DMA) column. This is followed by condensation of a low-volatility liquid on the size-selected particles and optical counting in a condensation particle counter (CPC; see, for example, Wiedensohler et al. (2012) for a detailed description). The combination of a DMA and CPC is referred to as a scanning mobility particle sizer (SMPS) or differential mobility particle sizer (DMPS) system. These systems are used for the size classification of submicron-sized particles. For size measurements of larger particles, optical particle sizers (OPSs) are used.

V-TDMA and H-TDMA. Particle volatility was probed using thermodenuders in a tandem DMA set-up (V-TDMA), whereby particles are passed through an oven heated to a known temperature. The heating is followed by a cooling section in which gases volatilized from the particles are trapped. Particle size is measured before and after the thermodenuder to assess the contribution of volatile compounds to the aerosol condensed phase. Measurements are performed at a series of oven temperatures to extract information about the volatility distribution of the aerosol constituents. Similarly, particle hygroscopicity can be ascertained at subsaturated conditions using a hygroscopicity tandem differential mobility analyser (H-TDMA), wherein particle size is measured before and after exposure to well-defined relative humidity (Liu et al., 1978). Within CRAICC particles in the boreal forest environment were also characterized using a combination of the two instruments (i.e. the VH-TDMA; Hong et al., 2014).

Cloud condensation nucleus counter (CCNC). The ability of particles to form cloud droplets was measured using established CCNC techniques. For example, the continuous-flow thermal-gradient diffusion-type CCNC manufactured by Droplet Measurement Technologies. Such a CCNC operates by exposing particles to well-known supersaturations of water, generated by applying a temperature gradient over a wetted column, and using optical detection to monitor droplet formation (Roberts and Nenes, 2005).

PINCii. Instrumentation for the online analysis of INP was developed as a result of a formalized technology sharing agreement and collaboration by six institutional partners (Ulrike Lohmann group, ETH Zurich; Frank Stratmann group, TROPOS Leipzig; Markku Kulmala group, University of Helsinki; Merete Bilde group, Aarhus University; Erik Swietlicki group, Lund University; and the University of Gothenburg) that was initiated by the CRAICC partners. The six-partner group has worked to develop and build a next-generation portable ice nucleation chamber (PINCii; Fig. 2), which is a continuous-flow diffusion chamber (CFDC) designed to update earlier parallel-plate CFDCs (e.g. ∼ ZINC, PINC, SPIN; Stetzer et al., 2008; Garimella et al., 2016). The PINCii instrument is an ice-coated flow tube reactor system designed to stimulate and measure ice nucleation within a test aerosol flow. Dry particles sampled from ambient or laboratory-generated flows are injected into a chamber, which contains a controlled water vapour supersaturated environment with respect to ice. Thus, by monitoring both the input and output flow the fraction of INP can be directly determined. Papers describing the PINCii instrument and the initial results of its first field deployment and ambient measurements are currently in preparation (see also Sect. 4.1.4; Castarède et al., 2019; Brasseur et al., 2018; Wu et al., 2018).

Soot on Snow (SoS) project. As part of CRAICC, the SoS project was conducted to study the effect of light-absorbing particles on snow surfaces. It consisted of a series of field experiments for which BC and other light-absorbing impurities, including Icelandic dust, were dry-deposited onto the surface of natural snowpacks and the consequent effects on albedo, snow density, other physical properties, including melting, were measured during the spring season. The broadband albedo was measured with pyranometers, in addition to the directional reflectance of snow. Concentrations of EC in the snowpack were analysed using a thermal–optical method (as described for the Svalbard ice core; Sect. 3.1.2) and compared with the measured albedo and that modelled with the SNow, ICe, and Aerosol Radiative (SNICAR) model (Flanner et al., 2007, 2009). Details of the experiments were presented by Meinander et al. (2014), Peltoniemi et al. (2015), and Svensson et al. (2016).

Sea spray aerosol simulation tanks. During CRAICC significant effort was devoted to improving the understanding of sea spray aerosol. One important development was the design, construction, and use of new temperature-controlled sea spray aerosol simulation tanks (King et al., 2012; Salter et al., 2014). In these tanks air is entrained in real or artificial seawater via frits, diffusers, or plunging jets. The entrained air breaks up into bubbles, which rise to the surface where aerosols are generated by bubble-bursting processes. These tanks can be coupled with other aerosol characterization instrumentation, for example particle size and number concentrations (see above), and can thus be used to probe a variety of physical and chemical properties of sea spray aerosol, including cloud-forming ability, hygroscopicity, and volatility.

The CRAICC core permanent research infrastructure included 18 well-established field research stations, covering ecosystems from Arctic to boreal locations (Fig. 3). These stations provided the time-resolved datasets used by the CRAICC community and are reflected in many publications from the Nordic Centre of Excellence.

The CRAICC consortium contributed to the continuing development of a model for two-dimensional Lagrangian aerosol dynamics, gas- and particle-phase chemistry, and radiative transfer (ADCHEM; Roldin et al., 2011) with improved representations of biogenic secondary organic aerosol formation (Hermansson et al., 2014, Öström et al., 2017). New process-based schemes for aerosol dynamics, particle-phase molecular diffusion mass transfer limitations, organic and inorganic particle-phase chemistry, and gas-phase chemistry schemes were implemented and constrained based on laboratory smog chamber experiments (Roldin et al., 2014, 2015). The latest version of ADCHEM includes a detailed gas-phase chemistry scheme that is based on the Master Chemical Mechanism (MCMv3.3.1; Jenkin et al., 1997; Saunders et al., 2003). This scheme also includes a novel scheme for the formation of highly oxygenated organic molecules (HOMs) formed from the ozonolysis and OH oxidation of monoterpenes. The HOM formation scheme is based on experimental work by Ehn et al. (2014) that was recently used to evaluate the contribution of HOMs to the activation and growth of new particles during observed new particle formation events in subarctic forests (Öström et al., 2017). The non-equilibrium SOA formation scheme simulates size-resolved particle growth using concentrations of around 700 different organic molecules provided by the gas-phase chemistry scheme. The SOA scheme can also account for heterogeneous oligomerization and nonideal organic and inorganic particle-phase interactions as well as the impact of particle-phase mass transfer limitations on the formation and evaporation of SOA particles (Roldin et al., 2014, 2015; Öström et al., 2017).

To assess long-term BC concentrations and deposition in the Arctic an offline Eulerian chemical transport model was run for the period between 1980 and 2015. The System for Integrated modeLling of Atmospheric composition (SILAM) is documented in detail by Sofiev et al. (2006, 2014). The SILAM model has several chemical transformation modules, including gas-phase chemistry and secondary inorganic aerosol formation, linearized sulfur oxide chemistry, radioactive nuclide decay, and aerosol dynamics (condensation and coagulation) computed either from thermodynamic equilibrium or dynamically. The aerosol size spectrum is described with a sectional approach and a user-defined bin distribution. Mechanisms of dry deposition vary from primarily turbulent diffusion-driven removal of fine aerosols to primarily gravitational settling of coarse particles, depending on the particle size (Kouznetsov and Sofiev, 2012). Wet deposition distinguishes between below- and in-cloud scavenging by both rain and snow (Horn et al., 1987; Smith and Clark, 1989; Jylhä, 1991; Sofiev et al., 2006). BC and other fine anthropogenic particulate matter (PM) components are modelled as inert aerosol with 0.5 µm dry diameters. Emissions from natural sources (fires, sea salt, and desert dust) are parameterized in terms of continuous distributions and split into species-specific size bins. Deposition and settling of each bin are related to the mass-mean wet diameter of the bin. The SILAM model has been extensively evaluated against European and global air quality observations (Solazzo et al., 2012; Huijnen et al., 2010; Ruppel et al., 2017) and is driven by ERA-Interim (Dee et al., 2011) meteorological data with 3 h temporal and 0.72∘ horizontal resolutions. SILAM uses the MACCity emission dataset (Granier et al., 2011) for anthropogenic emissions, except for flaring emissions, which were taken from the ECLIPSE dataset (Stohl et al., 2013). Emissions are available every 5 years, beginning in 1980 for MACCity and in 1990 for ECLIPSE, with the remaining years estimated by linear interpolation. Global simulations utilized a horizontal resolution of 0.72∘ × 0.72∘ and a vertical grid consisting of nine unevenly spaced atmospheric layers. The lowest, thinnest layer was 25 m thick, with the top layer reaching into the stratosphere. The source contribution to BC deposited at Holtedahlfonna was investigated by tagging the different emission sectors while computing the atmospheric dispersion of BC.

To support the interpretation of dust storm observations in Iceland, simulations with the numerical weather prediction model HIRLAM (Unden et al., 2003) were used. The simulations were run on a 5 km horizontal grid at 65 vertical levels, with forcing at the boundaries from the operational suite of the ECMWF. The turbulence calculations were based on the CBR scheme (Cuxart et al., 2000) and the Interaction Soil Biosphere Atmosphere (ISBA) employed at the surface (Noilhan and Mahfouf, 1996).

NorESM1-M is the Norwegian Earth System Model version 1 (Bentsen et al., 2013; Iversen et al., 2013). This particular set-up of NorESM1, used in both CRAICC and in the Coupled Model Intercomparison Project Phase 5 (CMIP5), has an intermediate horizontal atmospheric resolution of 1.9∘ × 2.5∘ and 26 vertical levels. The ocean module is an updated version of the isopycnic ocean model MICOM, while the sea-ice (CICE4) and land (CLM4) models and the coupler (CPL7) are similar to CCSM4 (Gent et al., 2011), differing only in the tuning of snow grain size for fresh snow on sea ice within CICE4. The atmosphere module, CAM4-Oslo (Kirkevåg et al., 2013), is a version of CAM4 extended with advanced representation of aerosols, aerosol–radiation, and aerosol–cloud interactions.

The land model has its own carbon cycle model and includes the SNICAR model (Flanner et al., 2007, 2009). The latter facilitates calculations of the effects on radiative transfer from snow darkening by deposited light-absorbing aerosols, i.e. BC and mineral dust. The albedo effect of light-absorbing aerosols deposited on snow-covered and bare sea ice is also accounted for in the sea-ice model.

CAM4-Oslo calculates mass concentrations of aerosol species that are tagged according to production mechanisms in clear and cloudy air for four size classes (nucleation, Aitken, accumulation, and coarse mode). Furthermore, it includes primary emission, gaseous and aqueous chemistry (cloud processing), gas-to-particle production (nucleation), condensation, and coagulation of small particles onto larger pre-existing particles. Loss terms are dry deposition and in-cloud and below-cloud scavenging. Aerosol components included are sulfate, BC, organic matter, sea salt, and mineral dust distributed over 11 modes for externally and internally mixed particles that are emitted or produced in air, 9 of which are tagged to production mechanisms in air or in cloud droplets. In addition to these 20 transported tracers for aerosols, there are two additional tracers for the gaseous aerosol precursors SO2 and DMS. Optical properties are estimated from Mie theory, while supersaturation and hygroscopic growth for CCN activation are calculated from Köhler theory.

Internally mixed water from water vapour condensation is treated separately using look-up tables for the aerosol optical parameters. Other look-up tables are used to obtain dry size parameters (dry radii and standard deviations) of the aerosol population, which are used as input in the calculation of CCN activation, following Abdul-Razzak and Ghan (2000).

A separate post-CMIP5 NorESM1 version with explicit parameterizations of nucleation and secondary organic aerosols (Makkonen et al., 2014) has also been used in CRAICC. In this model version, hereafter referred to as NorESM1-CRAICC, the land model uses the MEGAN parameterization for interactive biogenic volatile organic compound (BVOC) emissions instead of the prescribed sources applied in the CMIP5 version of the model.

The atmosphere model CAM4-Oslo has also been extensively validated and compared with other models through the AeroCom initiative (Aerosol Comparisons between Observations and Models) by Jiao et al. (2014), Tsigaridis et al. (2014), Kipling et al. (2016), and Koffi et al. (2016). A separate evaluation of NorESM1-M and other CMIP5 models through the remote sensing of aerosols in the Arctic has been made by Glantz et al. (2013). NorESM1-M is also taking part in the ongoing Precipitation Driver Response Model Intercomparison Project (PDRMIP; see e.g. Samset et al., 2016; Myhre et al., 2017).

The atmospheric boundary layer in the Arctic is of key importance because it connects the atmosphere, cryosphere, and marine components of the climate system. Nevertheless, observational data from the Arctic boundary layer are sparse. This is partly due to the hostile climatic conditions with low and even extreme temperatures that are beyond the operational range of many meteorological instruments. Long-term climatological measurements are traditionally performed with rugged instruments such as sturdy cup anemometers and wind vanes located near the ground – typically at heights between 2 and 10 m. Except for the few sites with regular radio soundings, limited information on the vertical structure of the Arctic atmospheric boundary layer is available.

A preponderance of information on the structure of the Arctic atmospheric boundary layer has been obtained during intensive observations carried out over limited time periods, primarily during summer months (Gryning et al., 1985; Lampert et al., 2011; Burgemeister et al., 2013; Di Liberto et al., 2012; Batchvarova et al., 2014; Achert et al., 2015). However, recent technological improvements to ground-based lidar remote sensing may significantly benefit future Arctic meteorological research. Utilizing components developed for use in fibre optics communication, lidars have become more compact, reliable, and easier to use. The development of eye-safe fibre-based lidars started in the mid-1990s and the first commercial products became available around 2005. One such instrument, an aerosol lidar (CL51 ceilometer from Vaisala), was installed at Villum research station in spring 2011. The CL51 measures the backscatter profile of the aerosols in the atmosphere and has survived to measure in the harsh Arctic conditions for several years. The aerosol backscatter is measured as a function of height with a vertical resolution of 10 m up to 7 km of height. The backscattered signal depends on the number concentration, size, and optical properties of the particles in the air.

Satellites provide a plethora of environmental data, including information on atmospheric composition and surface properties over wide spatial areas. Their instruments provide coverage that is spatially and temporally delineated based on orbiting patterns and swath width. Satellite observations are highly complementary to ground-based in situ and remote sensing measurements, and they make observations possible over remote and difficult-to-access areas where in situ measurements are not available. However, satellites are typically confined to optical (UV–Vis), infrared (NIR–TIR), and microwave (radar) observations that do not necessarily contain the same detail as ground-based measurements. For atmospheric composition, different instrument platforms are used to detect trace gases, greenhouse gases, and aerosols and clouds. These include sensors such as TROPOMI, the Moderate Resolution Imaging Spectroradiometer (MODIS), VIIRS, and GOME-2 with daily global coverage at spatial resolutions varying from 250 m for aerosols and clouds (MODIS and VIIRS) to 3.5 km × 7 km for methane, NO2, and SO2 from TROPOMI. Other sensors such as SLSTR provide smaller but still useful coverage, especially at high latitudes. Together with information on forest fires and aerosol absorption such data provide information on the emission and transport of aerosols and trace gases and their effects on air quality, climate, and pathways to the Arctic. CO2 and SIF (solar-induced chlorophyll fluorescence) are retrieved from OCO-2 data with a revisit time of 16 days because of the rather narrow swath (10 km); the OCO-2 pixel size is 1.3 km × 2.2 km. Thus, the spatial resolution of OCO-2 allows an anthropogenic CO2 signature to be distinguished (Hakkarainen et al., 2016; Eldering et al., 2017). As for other GHG satellites, TanSat is comparable to OCO-2 in resolution, and GOSAT has a revisit time of only 3 days but a pixel diameter of 10 km. For many optical techniques satellite data are only collected when solar radiation is available and solar zenith angles are large enough, which is a serious limitation over polar regions.

The detection of gases requires high spectral resolution (spectrometers, interferometers) with specific wavelengths for the gases of interest. The design of such instruments implies a relatively low spatial resolution of tens of kilometres. Examples are SCIAMACHY, GOME-2, OMI, and TROPOMI, with in incremental improvement of the spatial resolution, which for the recently launched (2017) TROPOMI instrument on the Sentinel-5P satellite is 7 km × 7 km sub-nadir compared to the OMI footprint of 13 km × 24 km. TROPOMI extends the capabilities of OMI and measures column concentrations of ozone, formaldehyde, carbon monoxide, NO2, SO2, and methane, as well as aerosol layer height using the O–A band and the absorbing aerosol index, a qualitative parameter indicating the presence of absorbing aerosol particles, and UV radiation. Information on trace gases, such as NO2, SO2, and formaldehyde (a proxy for less volatile organic compounds), and near-surface UV radiation is also important for the formation of aerosol particles through gas-to-particle conversion.

Dedicated instruments for quantitative measurements of aerosol and cloud properties are radiometers with moderate spatial resolution, which do not require high spectral resolution but need a wide range of spectral bands (e.g. MODIS, VIIRS) and preferably also several viewing angles (MISR) and polarization (POLDER). Other instruments used for the retrieval of aerosol properties are e.g. SeaWiFS, MERIS, the AVHRR instruments, which together provide a long time series that started around 1981, and the dual-view ATSR series (ATSR-2, AATSR; 1995–2012) followed up by SLSTR launched in 2016 on Sentinel-3.

Some of the instruments mentioned above are also used, or even designed, to retrieve information on land and ocean surface properties. The AVHRR, ATSR, and SLSTR instruments were designed to measure land and ocean surface temperature. MODIS also provides land surface temperature, land surface albedo, and bidirectional reflectance distribution function (BRDF); several of the sensors mentioned provide information on forest fires by virtue of their TIR channels. SeaWiFS, MERIS, and OLCI (on Sentinel-3) were designed to measure ocean parameters and in particular ocean colour.

Satellite information has been used in the CRAICC feedback analysis, as described in more detail in Sect. 4. The use of satellites to study spatial changes in glaciers, sea-ice extent, dust storm occurrence, and deposition is discussed in Sect. 4.1. High-spatial-resolution sensors such as Landsat and Sentinel-2 provide detailed information on land surface properties and were used to detect high-latitude dust sources and the spatial extent of glaciers. MODIS was used to detect dust storms over Iceland.

Two MODIS instruments onboard the Terra and Aqua satellites have been routinely collecting information on multiple environmental parameters since the year 2000. CRAICC studies on large-scale snow-covered surface albedo changes were based on MODIS products of snow cover fraction and surface albedo (MCD34C3). The MODIS snow-mapping algorithm is based on the normalized difference snow index (NDSI), which utilizes the fact that snow has a high reflectance in the visible part of the solar spectrum and low reflectance in the infrared (Hall et al., 2002). The MODIS BRDF and albedo algorithm utilize multiple spectral bands to retrieve broadband albedo information (for detail see Lucht et al., 2000; Schaaf et al., 2002). MODIS data products are consistently validated, readily available, and widely used by the research community, including CRAICC.

MODIS data have been used by Atlaskina et al. (2015) to study the temperature dependence of the albedo of snow-covered land surfaces in the Northern Hemisphere as described in Sect. 4.2.3. MODIS and AATSR have been used to explore the retrieval of aerosol parameters in Arctic regions (Istomina et al., 2010; Mei et al., 2013). For an overview of aerosol remote sensing in polar regions using ground-based and satellite instruments, see Tomasi et al. (2015). The use of satellite data to study Arctic amplification has also led to a controversy in the literature as described in Sect. 4.2.2.

Iceland straddles the Arctic circle in the north Atlantic Ocean and is thus significant as a source and monitoring point for Arctic climate change. That said in situ aerosol observations from Iceland are minimal. The city of Reykjavik and the Environmental Agency of Iceland have maintained PM10 (particulate matter with diameter less than 10 µm) observations at a couple of fixed locations for 15 years in Reykjavik, far from the local dust sources (Thorsteinsson et al., 2011), while shorter-term monitoring installations have been established during volcanic eruption events (Thorsteinsson et al., 2012; Dagsson-Waldhauserova et al., 2014a). There is also a unique dataset of dust and volcanic ash observations that have been collected by the Icelandic Meteorological Office for nearly a century. Frequent volcanic eruptions increase the annual dust frequency, and the consequent dust–volcanic ash resuspension events prolong the impacts of eruptions (Thorsteinsson et al., 2012; Thordarson and Höskuldsson, 2008). For example, after the Eyjafjallajökull eruption in 2010, the erosion flux exceeded 11 000 kg m-1 during one dust event (Arnalds et al., 2013), which is one of the most severe wind erosion events ever recorded on Earth.

Given its location and ecosystem, much of the aerosol particulate in Iceland originates from wind–surface interactions. Icelandic surfaces are classified by the “AUI Nytjaland” land cover database into vegetation classes (Gisladottir et al., 2014), which shows that Iceland's total desert area is about 43 400 km2 (Arnalds et al., 2016). About 15 000 km2 of the total desert area is active aeolian surface, which includes super-active dust hot spots. Over the long term Iceland has been observed to have 34–135 dust days per year without strong seasonality (Dagsson-Waldhauserova et al., 2013, 2014a). Dust is produced at a rate of about 31–40 Tg yr-1 (Fig. 4) and that dust is redeposited onto the land, into the ocean, and onto glaciers (total area of >500000 km2), with some dust plumes capable of travelling over 1000 km (Fig. 4, MODIS data; Arnalds et al., 2016). Iceland is the largest Arctic and European desert with a dust event frequency comparable to major lower-latitude desert areas like the Gobi or Iranian deserts (Arnalds et al., 2016).

The lack of established observations led CRAICC to support several field dust–aerosol measurement campaigns between 2013 and 2016 (Dagsson-Waldhauserova et al., 2014b, 2015, 2016). During severe dust storms PM10 concentrations exceeding 6500 µg m-3 were measured (1 min averaging time) with median PM10 values > 1000 µg m-3 during 24 h intervals. During a moderate storm a PM1 maximum of 261 µg m-3 was measured. High values of submicron particles are typical of Icelandic volcanic dust, with resulting PM1 / PM2.5 ratios of >0.9, while PM1 / PM10 ratios range from 0.34 to 0.63. These values are comparable to urban air pollution rather than natural dust storm events observed in other regions. Particle number concentrations (PM ≈ 0.3–10 µm, OPS) are also high, with maximum concentrations of ≈150000 particles cm-3 recorded with a primary peak in the size distribution between 300 and 337 nm and a smaller peak for particles with diameters between 1.5 and 5 µm. Such extreme concentrations are reflective of ongoing volcanic eruptions. The dust in Iceland is primarily volcanic in origin and dark in colour, with many particles that include sharp-tipped shards and large bubbles. About 80 % of PM is volcanic glass that is rich in heavy metals, such as iron and titanium.

Satellite data collected since the 1970s yield detailed information on spatial changes in glaciers, sea-ice extent, dust storm occurrence, and dust deposition in Iceland. High-resolution imagery from the Landsat-8 (NASA) and Sentinel-2 (ESA) satellites allows for improved analysis of dust source areas, although at latitudes above 65∘ temporal resolution is limited to approximately twice weekly. Despite having a relatively mild and humid climate dust storms are visible in Iceland about 40 days a year on average (MODIS 250 m resolution images), which is a conservative estimate given that low light and clouds likely obscure events. About twice as many events were observed to occur annually between 2002 and 2012 (Georgsdóttir, 2012), and preliminary results for 2010 to 2016 show a similar frequency, especially with increases in 2010 after the Eyjafjallajökull eruption due to subsequent volcanic ash resuspensions. The prolonged melting season on Icelandic glaciers is captured by Sentinel-2 images, which show suspended dust as well as older ash layers in the glaciers, both of which affect the surface albedo (Fig. 5). Although MODIS observations correspond to the lowest dust frequencies relative to meteorological observations, daily images from MODIS captured high-frequency dust events during warm autumns in north-eastern Iceland in 2015 and 2016. Annual dust deposition of 17 g m-2 clearly affects albedo, which, for example, was observed to be reduced by 0.36 (albedo dropped from 0.86 to 0.5) after a Vatnajökull glacier dust event, leading to increasing snowmelt of ≈0.6 m annually (loop E → C in Fig. 1; Wittmann et al., 2017).

In addition to increasing frequency, Icelandic dust storms exhibit spatial and temporal variability. Storms are most frequent in the north in summer, while in the south the dust storms are more frequent in late winter. Additionally, there is significant inter-annual and decadal dust storm variability (Dagsson-Waldhauserova et al., 2013, 2014a). With a combination of in situ observations, numerical modelling (HIRLAM model with 5 km horizontal resolution), and meteorological time series analysis, Arnalds et al. (2014) have assessed individual dust storms to estimate the quantity of annual Icelandic dust emissions. They estimate that an average 31–40 Tg of dust is suspended annually, with 12 %–35 % of that reaching the ocean before being redeposited (Arnalds et al., 2014; see Fig. 4). Further numerical simulations of wind during dust storms has revealed the local nature of the windstorms, underlining the need for high atmospheric and topographic spatial resolutions for accurate simulations (Dagsson-Waldhauserova et al., 2016; Baddock et al., 2017). In another study using FLEXDUST and FLEXPART simulations, and meteorological reanalysis data from Groot Zwaaftink et al. (2017), annual Icelandic dust emissions were estimated to be an order of magnitude lower at 4.3±0.8 Tg with about 7 % of that reaching the high Arctic (>80∘ N). However, Groot Zwaaftink et al. (2017) likely underestimate dust activity from the hyperactive dust hot spots which produce high dust outputs, for example the redistribution of material after volcanic events and glacial outbursts. Dust uptake processes are also poorly understood, and Arnalds et al. (2016) calculated deposition rates which also included large particles with short travel distances < 100 km. That said, both estimates illustrate the importance of high-latitude dust production in the Arctic and the global context.

The Arctic Ocean is surrounded by vast regions of tundra and boreal forest, which during the summer period emit large quantities of BVOCs. In order to fully understand how warming-enhanced BVOC emissions will influence the BVOC–aerosol–cloud–climate feedback loop in the Arctic climate system in the future, a detailed understanding of the processes governing the formation and growth of new particles is crucial. Within CRAICC the Lagrangian model ADCHEM was used to simulate the formation and growth of new particles along air mass trajectories originating from the Arctic Ocean and travelling over tundra and boreal forests before arriving at the Pallas Atmosphere–Ecosystem Supersite (67.97∘ N, 24.12∘ E; 565 m a.s.l.) in Northern Finland. A main focus was to constrain the processes governing the growth of new particles into the climatically important cloud condensation nuclei size range. Model simulations together with measured particle number size distributions reveal that during observed new particle formation (NPF) event days in Pallas, new particle formation begins when the air masses move from the Arctic Ocean inland over the boreal forest. The analysed event days are characterized by lower concentrations of particles larger than 50 nm in diameter (N50nm), with median N50nm of 140 cm-3.

According to the model simulations, particle growth can largely be attributed to condensation of low-volatility and highly oxygenated organic compounds (Fig. 6; Öström et al., 2017). With the HOM condensation mechanism included, ADCHEM can capture the initial growth between 1.5 and 20 nm in diameter during the observed new particle formation events. However, the model still underestimates particle growth between 20 and 80 nm in diameter and therefore underestimates the concentrations of particles in the CCN size range (diameter > 50 nm) the day following a nucleation event. Both in the model and observations the N50nm peaks at around 06:00 UTC the day after new particle formation events with median N50nm of 1109 and 1674 cm-3, respectively (Öström et al., 2017). The more than 1 order of magnitude increase in the N50nm concentration during the morning after NPF event days, in the model and observations, indicates that growth from biogenic secondary material is an important process for maintaining CCN concentrations over subarctic forests, as long as primary particle emissions are low.

Kyrö et al. (2014) studied the effect of reduced sulfur emissions from the Kola Peninsula on aerosol growth, concentrations, and long-term NPF trends at SMEAR I in eastern Lapland, Finland. The frequency of NPF days (clear, class I events) was found to decrease by about 10 % per year concurrently with decreasing SO2 (also ≈10 % per year). High concentrations of SO2 and H2SO4 were found to promote the onset of nucleation by several hours and even catalyse some events without sunlight. In general, air masses coming over the Kola Peninsula were found to favour NPF, with sulfuric acid explaining up to 50 % of the observed growth of new particles. Decreasing sulfur emissions decreased the condensation sink by about 8 % per year and thus increased the event time J3 (particle formation rate at 3 nm particle diameter) by almost 30 % per year. The biggest decrease in NPF frequency was during spring and autumn, when the importance of H2SO4 for growth was maximum and H2SO4 concentrations were too low to grow the particles to potential CCN sizes. Simultaneously, during these seasons surface temperatures are on average too low to compensate for and enhance SOA formation. Collectively these effects have resulted in a 3 %–4 % per year decrease in potential CCN concentrations.

Secondary organic aerosol precursor sources and processes can be tracked using molecular tracers such as carboxylic acids and organosulfates. Carboxylic acids are typically of either biogenic or anthropogenic origin, while organosulfates are formed from the reactive uptake of organics on acidic sulfate aerosols (Surratt et al., 2007; Iinuma et al., 2007). The formation of organosulfates thus represents a mechanism for anthropogenic enhancement of biogenic SOA (Hoyle et al., 2011). Within CRAICC organosulfates were discovered in the Nordic and Arctic environment (Kristensen and Glasius, 2011; Hansen et al., 2014; Nguyen et al., 2014a; Kristensen et al., 2016). Hansen et al. (2014) measured the first year-round time series of Arctic organosulfates and identified elevated levels during the late winter haze period at both the Villum research station and in Svalbard, probably due to the influence of anthropogenic sulfate aerosols. Another recent study in Northern Europe during winter points to coal combustion (i.e. high-sulfur fuel) as an important precursor of organosulfates in aerosols (Glasius et al., 2018).

The first annual time series measurements of humic-like substances (HULIS) in Arctic aerosols were also obtained within CRAICC (Nguyen et al., 2014b). HULIS consists of complex, relatively high-molecular-weight polyacidic organic molecules (e.g. Decesari et al., 2000) and influences the light-absorbing properties of aerosols. The average concentration of HULIS was measured to be higher during months with less sunlight (November–April, 11 ng C m-3) than during the remainder of the year (4 ng C m-3) (Nguyen et al., 2014b).

A prominent feature of organic aerosols is their aqueous surface activity. Already 2 decades ago, Facchini et al. (1999) recognized the potential impact of surface tension on aerosol–cloud interactions, but the subject has remained a source of recurring debate in the aerosol chemistry and cloud microphysics communities (e.g. Ruehl et al., 2016; Ovadnevaite et al., 2017). Building on the work of Sorjamaa et al. (2004) and Prisle et al. (2008, 2010), researchers in CRAICC studied hygroscopic growth and the cloud-forming potential of laboratory-synthesized organosulfates (Hansen et al., 2015) and commercial modelled HULIS samples (Kristensen et al., 2014) using H-TDMA and CCNC chambers. These studies demonstrated the presence of complex surface tension effects, surface partitioning, and salting out, which significantly broadened the understanding of surface activity for water uptake and the cloud activation of atmospheric organics with very different origins and chemical characters.

Field measurement campaigns have been initiated in an effort to quantify the ice-nucleating capacity of Arctic and Nordic air masses. Simultaneously, idealized laboratory experiments have been conducted to examine the underlying physicochemistry and molecular-scale processes from water adsorption to heterogeneous nucleation. Results from such studies complement cloud-resolved modelling studies whereby CRAICC efforts have tried to connect the availability of ice nuclei to the production of ice crystals in clouds, advances which contribute to understanding how, for example, mixed-phase clouds form and evolve (Savre and Ekman, 2015a, b).

Idealized molecular beam and light-scattering experiments have focused on building a fundamental understanding of how critical-scale ice clusters initially form and grow vis à vis vapour deposition on well-characterized surfaces (Kong et al., 2012; Thomson et al., 2015). Observations demonstrate that surfactants play an important role due to their ability to enhance and/or suppress the adsorption and desorption kinetics of atmospheric particles (Kong et al., 2014a; Papagiannakopoulos et al., 2014; Lejonthun et al., 2014; Thomson et al., 2013; Johansson et al., 2017). In particular, organic hydrocarbon surfactant layers – that for experimental purposes are used to model atmospherically relevant organic layers – have size- and temperature-dependent effects on ice nucleation, growth morphology, and molecular water uptake (Kong et al., 2014b; Papagiannakopoulos et al., 2013). An important implication of such findings is that common atmospheric surfactants, such as VOCs, can promote and/or hinder water uptake, changing the hydrophilicity of atmospheric nanoparticles. Additional, laboratory investigations focused on how soluble salts may nucleate ice and whether or not at below-eutectic temperatures sea salts might act as ice-nucleating particles rather than deliquesce as CCN (Kong et al., 2018; Wagner et al., 2018; Castarède and Thomson, 2018). All of these processes may affect cloud evolution and lifetime and thereby impact fundamental environmental processes like the water cycle and radiative balance. Arctic clouds are known to be unique and often those that are most important to the regional radiative balance are mixed-phase clouds (Morrison et al., 2012), which include both liquid and solid cloud particles and therefore become inherently more unstable as temperatures diverge from the triple point (Savre and Ekman, 2015a, b). Thus, small effects, like those enumerated above, that influence the sensitive balance between liquid and solid cloud particle coexistence may have amplified environmental feedbacks in Arctic air.

The INP counting and characterization instrumentation (Sect. 3.2) introduced to the Nordic region through the CRAICC collaboration was used in various field studies. The PEAC7 electrostatic deposition collectors and subsequent FRIDGE analysis (Schrod et al., 2016) were used to initiate the acquisition of building time series data for INP concentration in the Arctic environment, where very few observational data currently exist. A 2-year time series of measurements was established in Svalbard in conjunction with a global data series initiated through the EU FP7 BACCHUS project (data currently being analysed). In an additional study the PEAC7 was used to sample seafaring ship emission plumes transiting the Port of Gothenburg, Sweden. In those measurements, an amplification of ice-nucleating capacity was observed from ship emissions (Thomson et al., 2018). Absolute INP activities observed from the ship plumes agreed with previous measurements of soot particles that suggest they are weak ice nucleators (Dymarska et al., 2006; Friedman et al., 2011; Schill et al., 2016). However, because these emissions are strongly concentrated, with INP emission factors of 105–107 per kilogram fuel, they still result in strongly increasing INP compared to ambient values. As such, the observed amplification may be important to consider as shipping routes open in Arctic waters and sea-ice loss increases (Peters et al., 2011; Fuglestvedt et al., 2014). The magnitude of cloud- and climate-scale effects will be a product of many additional factors, including how such INP observations may change with particle ageing and how microphysical feedbacks and cloud response manifest in non-obvious ways (for example, see Possner et al., 2015).

In a very recent field campaign (HyICE2018) aimed at investigations of ice nucleation within the boreal environment, the PINCii instrument developed within CRAICC was first field deployed. In tests leading up to the campaign, and during preliminary campaign data analysis, the observed PINCii operational parameters and observed INP measurements are in good agreement with other instruments (Fig. 7). In Fig. 7a the stability of the PINCii chamber background is illustrated, and in Fig. 7b initial tests of the conditions for water droplet breakthrough are shown to be in line with other existing CFDCs (Stetzer et al., 2008; Garimella et al., 2016; Kong et al., 2018). During HyICE2018 data for instrument inter-comparison were also collected using multiple existing INP measurement platforms (PINC, FRIDGE, SPIN, etc.). Campaign data analysis by all involved research groups is currently ongoing (Castarède et al., 2019; Brasseur et al., 2018; Wu et al., 2018).

The boundary layer in the Arctic differs in a number of ways from its counterpart in the mid-latitudes. Away from the Arctic (and Antarctic) the diurnal variation in surface heating controls the dynamics of the boundary layer. During daytime, atmospheric mixing is driven by the sun heating the surface and wind shear throughout the boundary layer (Batchvarova and Gryning, 1991). After sunset, turbulence decays and a new neutral and stable layer with little turbulence forms over the ground. In the Arctic, the diurnal variability is absent throughout most of the year. During the long dark winter period, a long-lived shallow stable boundary layer is expected to form. The layer is shallow, partly due to surface cooling during the Arctic night and partly due to the large Coriolis force at high latitudes. If resilient enough, the stable layer may even extend through the Arctic spring–summer after the sun is continuously above the horizon. Such conditions are favourable for the creation of atmospheric waves and low-level jets, both of which influence the atmospheric boundary layer and atmospheric exchange processes. However, very little is known about these processes and their importance for the climate system.

Convective boundary layers driven by upward heat flux due to ground warming are also expected in the Arctic. Even in inner Antarctica at the Dome C research station, which is characterized by very low temperatures, Argentini et al. (2005) observed convective boundary layers. The formation of the Dome C convective boundary layers is likely connected to the very low ambient temperatures – typically -30 ∘C. King et al. (2006) argue that at such low temperatures boundary layer formation is due to the partitioning of available energy into sensible heat flux.

Figure 8 shows an example of the boundary layer height evolution with time over 2 weeks during a period of continuous Arctic sun at the Villum research station in the high Arctic. The land surface is fully covered by snow during the observations and the sun is visible throughout the entire day with a 13∘ elevation angle that only varies diurnally by about 1∘. The diurnal cycle that is characteristic for the behaviour of the boundary layer at mid-latitudes is absent. Instead, the boundary layer depth remains ≈250 m over the entire 5-day period with occasional outbreaks of wave-like motions. Thus, it is clearly not representative of an idealized boundary layer but is rather influenced by effects that are not yet fully understood. Figure 9 illustrates the conditions at the same location in mid-July. For the selected days the boundary layer is about 100 m and again considerable wave-like motions are observed.

In Sodankylä, Finland (67∘22′ N, 26∘39′ E), BC and OC in snow were measured in connection with broadband and spectral snow albedo, as well as other ancillary parameters of seasonal European Arctic snow. Measured low UV albedo values were partly explained by the properties of melting snow and also UV absorption caused by impurities in the snow (Meinander et al., 2013). The results revealed that the OC in the snow could have a significant contribution to light absorption. Light absorption by OC may also partly explain the high mass absorption cross section (MAC) value needed for BC to match the measured low albedo values with the SNICAR (Flanner et al., 2007) simulated albedo. To derive optical properties for OC that is considered representative of the OC in the snow, spectrally resolved MAC estimates for the target are needed.

The SoS field experiment showed that soot has a negative effect on snow albedo (Svensson et al., 2016); however, the measured effect was not as pronounced in outdoor conditions as in a previous laboratory study (Hadley and Kirchstetter, 2012). In Fig. 10, the albedo of the snow from the first 3 days of the SoS experiments and the corresponding EC concentration are presented; the figure also contains a fit to the experimental data. In the experiments, it was observed that the absorbing contaminants on snow enhanced the metamorphism of snow under strong sunlight conditions (Peltoniemi et al., 2015). Immediately after deposition, the surface of the contaminated snowpack appeared darker than the natural snow in all viewing directions, but the absorbing particles sank deep into the snow within minutes. The nadir measurement remained the darkest, but at larger zenith angles the surface of the contaminated snow was almost as white as clean snow. Thus, for a ground observer the darkening caused by impurities can be completely invisible, causing overestimations of the albedo, while a nadir-observing satellite sees a more absorbing surface, thus underestimating the albedo (Peltoniemi et al., 2015).

Atmospheric deposition not only influences albedo but also other physical properties of snow and ice (e.g. Meinander et al., 2014). Experiments focused on physical characteristics of snowpack utilized artificially added light-absorbing impurities, which decreased the density of seasonally melting natural snow (Meinander et al., 2014). The data suggests that BC can decrease the liquid water retention capacity of melting snow. No relationship was found in the case of natural non-melting snow. The significance of these results comes via the fact that snow density multiplied by snow depth equals the important climate model parameter of snow water equivalent (SWE).

Dust aerosols interact with rain, snow, and ice during dust storms. Suspended dust was observed during moist and low wind conditions (Dagsson-Waldhauserova et al., 2014b) and the phenomenon of snow–dust storms has been documented (Dagsson-Waldhauserova et al., 2015). Icelandic dust is therefore likely to affect the cryosphere in Iceland and elsewhere (Meinander et al., 2016). We hypothesize that Icelandic dust on snow can be one of the causes for Arctic amplification. This effect is explained as a positive feedback loop in which Icelandic dust deposited on snow decreases surface albedo and increases solar radiation absorption, snowmelt, and ice melt (Fig. 11). Such effects are normally linked to soot but volcanic ash and dust from Iceland may actually have a larger influence in the Arctic region. This is in contrast to what is generally concluded on crustal dust climate effects in the latest IPCC report (IPCC, 2013).

About 50 % of the annual dust events in the southern part of Iceland take place at subzero temperatures or in winter (November–April), when dust can be mixed with snow. This implies that the impacts and significance of Icelandic eruptions and dust events may be seriously underestimated. In studies related to aerosol effects on Arctic climate the effect of mineral dust is often not explicitly considered (e.g. Yang et al., 2014; Najafi et al., 2015), although paleoclimate data archives indicate a substantial contribution of dust to radiative cooling in the Arctic (Lambert et al., 2013). Climate change can also increase extreme wind events and dust storms. Human-initiated deforestation has also enhanced the desert area in Iceland, which coupled with changing climate can increase the probability for increasing dust event occurrences (Sect. 12 in Arnalds, 2015).

Such positive feedbacks between snow and impurities were investigated during several outdoor and indoor experiments with applications of Icelandic dust, volcanic ash, and BC on snow surfaces in Lapland and Finland (Meinander et al., 2014; Dragosics et al., 2016; Peltoniemi et al., 2015; Svensson et al., 2016). The outdoor SoS 2013 experiment showed that volcanic dust on snow causes reductions in albedo and increases snow melting similarly to that of BC. The spectral reflectance of Icelandic dust was observed to be as black as soot and, measured at 400–2500 nm, remained <0.04, which is typical for highly absorbing particles (Peltoniemi et al., 2015). In fact, the snow albedo effect of light-absorbing aerosols and the use of satellites to measure the darkening of the Greenland ice sheet (GrIS) has been the subject of a recent controversy in the scientific literature (Polashenski et al., 2015; Tedesco et al., 2016). Polashenski et al. (2015) studied satellite-observed trends in the Greenland ice sheet albedo and found no evidence supporting a hypothesis that these are caused by the deposition of BC or dust. Instead, they attribute the albedo trend observed by MODIS Terra (but not by MODIS Aqua) to uncorrected sensor degradation. Tedesco et al. (2016) in turn explain that Polashenski et al. (2015) use a daily product, which has been shown to deteriorate with latitude. Instead, Tedesco et al. (2016) use the GLASS surface albedo product, a 16-day integrated product which is a combination of data from MODIS and AVHRR and accounts for directional (BRDF) effects at high solar zenith angles, and observed a statistically significant decrease in the surface albedo over the GrIS. They compare their observations with model results and conclude that discrepancies are due to the absence of light-absorbing impurities in the model. Furthermore, the observed albedo trend is confined to regions of the GrIS that undergo melting in the summer with the dry-snow zone showing no trend. Both papers point at the absence of trends in the concentrations and deposition of either BC or dust, which leads to the conclusion that the observed albedo trends are not caused by changes in the deposition of these species. However, rather than ascribing the trend to sensor degradation, as in Polashenki et al. (2015), Tedesco et al. (2016) suggest that the albedo trend is caused by the exposure of a “dark band” of dirty ice and to increased consolidation of light-absorbing material at the surface from melting. Within CRAICC, Meinander et al. (2016), addressing these concerns, concluded that an assessment of the effect of Icelandic dust on snow–ice surface darkening and melt is currently unavailable and therefore scientific research is critically needed. These authors hypothesize that, in the Arctic, Icelandic dust may have a comparable or even larger effect on the cryosphere than soot (Bond et al., 2013). Observations and modelling results on Icelandic dust and cryosphere interactions for the past, present, and future are urgently needed.

Ash from the 2010 eruption of Eyjafjallajökull was used to find an insulating threshold for snow and/or ice. If the layer thickness of ash on snow or ice is very thin, it has the potential to increase snowmelt, but for thicker layers the snowmelt can be decreased compared to clean ice due to thermal isolation. In experiments by Dragosics et al. (2016) and Wittmann et al. (2017), a 9–15 mm thick ash layer on top of snow had an insulating effect, whereas a thin layer of only 1 mm increased melting to a maximum.

Atlaskina et al. (2015), using 13 years (2000–2012) of MODIS observations, have conducted a study on snow-covered land surface albedo during spring in different geographical areas of the Northern Hemisphere. The study showed that in the territories where the snow cover fraction did not change and remained 100 % throughout the study period, albedo has changed by ±0.2 over the 12-year period. The effects of air temperature, summertime enhanced vegetation index, and precipitation amount and frequency on the surface albedo were investigated in an attempt to explain the observed changes. A clear effect of the air temperature was found where and when the monthly average air temperature exceeded -15 to -10 ∘C, which is colder than what is observed in the laboratory experiments. In laboratory experiments snow albedo is stable at temperatures colder than -10 ∘C (e.g. Aoki et al., 2003). When temperatures reach -5 ∘C and higher, snow albedo decreases. Laboratory experiments are widely used to describe snow properties in models, but CRAICC research showed that they might represent atmosphere–cryosphere interactions unrealistically. This finding indicates that positive snow–albedo feedback takes place even at relatively low temperatures and when snow cover is complete. The results indicate that some Arctic areas are more sensitive to warming and therefore can exhibit climate change at a faster rate. In the study, the relation between surface albedo and the other parameters was less clear in comparison to the temperature effect or even absent in most regions.

When European Arctic treeline LPJ-GUESS model simulations are compared with the locations of modern treeline detected from satellite imagery and past treeline determined using proxy-based reconstructions, the simulated treeline agrees with the spatial distribution of the actual treeline when a threshold biomass value of 2 C kg m-2 is used (Fang et al., 2013). Mismatches are primarily observed over mountainous regions, such as in northernmost Fennoscandia and in regions near the Ural Mountains. The likely cause is that the spatially distributed climate data used to drive the climate model are limited in terms of representing the varying climate conditions between valleys and mountain peaks. Inaccuracies and simplifications in species-specific simulations are also caused by the parameterization of coniferous species. For example, one significant factor contributing to low simulated biomass of pine in Fennoscandia modern treeline vegetation is that in the LPG-GUESS simulation pine is classified as a shade-intolerant species and is therefore suppressed by shade-tolerant spruce when the ranges of these two species overlap. Such comparisons suggest that needed improvements include the spatial resolution of climate data within regions with complex terrain, such as steep slope gradients, and vegetation models need to simulate mixed forests composed of species with close bioclimatic thresholds.

Due to the inherent difficulties of species-specific simulations the CRAICC focus was on simulating the dynamics of coniferous treeline instead of species-level treeline shifts. The simulations and proxy data both indicate a northward expansion of treeline in the middle Holocene and a retreat during the late Holocene. In Fig. 12 northward expansion towards the current tundra is indicated by the rise of spruce pollen values 9000 yr BP and the late Holocene retreat by the decline of spruce pollen values after 5000 yr BP. When the simulated treeline dynamics are compared with these data, it can be seen that the model correctly captures the northward expansion of the boreal forest during the mid-Holocene and realistically simulates a treeline retreat in response to climate cooling during the last 3000 years. However, there are data–simulation disagreements, particularly during the early Holocene, which mainly result from the differences between the two paleoclimate model scenarios used to drive the simulations. Thus, the results show that the LPJ-GUESS dynamic vegetation model does not perform particularly well for species-specific biomass simulations. However, the general advances and retreats of the Arctic treeline can be realistically simulated with the dynamic vegetation model. In the model, the northward advances of the treeline are positively related to temperature variations, leading to a positive feedback loop between temperature and treeline.

Boreal forest landscapes, including boreal soils, waters, and vegetation, are well-known sources of biogenic emissions (Penuelas et al., 2014). A CRAICC modelling platform was applied to assess the potential Earth system feedback linking BVOC emissions from boreal forests, and the resulting SOA formation, to direct and indirect aerosol and climate effects. Two experiments were performed with NorESM1-CRAICC to quantify the effects of BVOC feedbacks using a slab-ocean model and the Community Land Model (CLM) with interactive BVOC emissions calculated according to the MEGAN module. While including the BVOC feedback mechanism, the model was run for a pre-industrial CO2 level of 357 ppm (1 × CO2-FB) and a CO2 doubling (2 × CO2-FB) scenario. To achieve equilibrium the model was integrated for 70 years. In the boreal forest region, monoterpene emissions increased from 18 to 28 Tg yr-1 due to climate change from doubled CO2. The increase in monoterpenes led to nearly doubling the simulated SOA formation and a 10 % increase in total particulate organic matter (POM). NorESM1-CRAICC simulates the effect of SOA formation on (1) nucleation, (2) nuclei growth, and (3) bulk SOA mass formation (Makkonen et al., 2014). The effect of the 10 % increase in POM can be attributed to the changes in aerosol size distributions resulting from several competing pathways. Increased POM contributes to increased coagulation and is a condensation sink, possibly decreasing nucleation and subsequent growth. Furthermore, the aerosol size distribution is modified by simulated changes in cloudiness and precipitation. Indeed, cloud cover over boreal forest increases from 54.1 % to 56.9 % (0.5 % standard deviation) and mean precipitation intensity increases slightly from 0.09 to 0.10 mm h-1 (standard deviation 0.0017 ).

NorESM1-CRAICC was also used to simulate an alternative climate warming scenario, in which biogenic VOC emissions were not allowed to change (“no feedback” experiment, 2 × CO2-NOFB). Thus, the two simulations, one with and one without BVOC–aerosol–climate feedback, allow us to quantify the BVOC effect on aerosols and climate. In a control (1 × CO2) simulation, total particle concentrations over boreal forest regions averaged 820 particles cm-3, while in the climate change simulation without BVOC–aerosol feedback particle concentrations decreased to 790 particles cm-3, but the increased SOA formation in the 2 × CO2-FB simulation resulted in increased concentrations of 880 particles cm-3. Similarly, the vertically integrated cloud droplet number concentrations increased from 1.9×106 to 2.6×106 cm-2 with BVOC–aerosol feedback and only to 2.3×106 cm-2 without feedback when simulating a doubled CO2 scenario.

The strength of the simulated BVOC–aerosol–climate feedback is clearly weaker over tropical regions compared to boreal forest. In the tropics, monoterpene emissions are shown to increase 20 %, resulting in a POM increase of only 2 % due to simulated climate change. However, even at the global scale, the BVOC–aerosol–climate feedback can increase SOA formation by 45 % in a doubled CO2 experiment, showing the potential for a strong feedback mechanism during the 21st century.

The CRAICC modelling platform was also applied to assess the importance of land-use perturbations for atmospheric chemistry and aerosols. While the albedo and GHG effects of land use are widely recognized, accounting for the changing spectra of primary aerosol and BVOC emissions is largely missing. Anthropogenic land-use change at high latitudes could play a role in modifying aerosol–climate feedbacks through direct and indirect aerosol effects. A total of four simulations (Table 1) were performed using NorESM1-CRAICC to study the effects of land-use change on BVOC emissions, atmospheric chemistry, and SOA formation.

Since 1850 there has been significant tropical deforestation in South America, South Asia, and Africa and considerable extratropical deforestation of mixed and boreal forests in Eurasia and North America. NorESM1-CRAICC shows large reductions of BVOC fluxes over these regions as a result of deforestation. Globally, when isoprene and monoterpene emissions are modelled and all other factors controlling BVOC emissions are kept fixed, the emissions are modelled to be about 10 % lower in 2000 than in 1850 due to land-use change (comparing simulations 1 and 2 in Table 1). The global reduction of BVOC emissions leads to an increase in OH concentrations in the boundary layer of up to 40 % over regions marked by strong deforestation because BVOCs are important compounds controlling OH reactivity and abundance through direct reactions. Simulations 1 and 2 also show that boundary layer ozone concentrations are reduced (up to 10 %) because BVOCs serve as ozone precursors when NOx concentrations are high.

Simulations 3 and 4 indicate that lower BVOC emissions lead to a reduction in SOA formation, lower loadings of organic aerosol (Fig. 13a), and lower cloud droplet concentrations (Fig. 13b) if oxidants are fixed at present-day levels. The effects of deforestation are seen as decreased organic aerosol burden and cloud droplet number concentration. The changes in organic aerosol are particularly marked in the Northern Hemisphere, including large portions of boreal forests in Eurasia and North America.

The thin veneer of sea ice which covers much of the Arctic Ocean is an important variable for the energy and ecosystem balances of northern latitudes. A historical minimum in sea-ice extent occurred in 2012, during which time CRAICC joined the CHINARE Arctic expedition to carry out various observations on sea ice and other atmospheric and oceanographic conditions (Lei et al., 2014, 2015). Additional satellite imagery analysis using data from 1979 to 2012 to analyse inter-annual, seasonal, and spatial changes in the sea ice shows that general thinning of the ice cover and delayed freeze-up result in high variability during the month of October (Lei et al., 2015). Changes to the freezing and melting cycles of that ice are also important for the survival of multi-year sea ice (e.g. Rothrock and Maykut, 1999; Wadhams, 2016; Johannessen et al., 2019). Seasonal ice forms annually due to sub-freezing winter temperatures, while multi-year sea ice is connected to summer warming because the equilibrium thickness is highly sensitive to summer melt. Land-fast ice around the Arctic Ocean's coasts and islands is seasonal and its mechanics are sensitive to the ice thickness and forcing by winds and tides (Yang et al., 2015). Sea-ice transport out of the central Arctic Ocean is a significant sink for ice that is sensitive to large-scale atmospheric circulation, for example the Arctic Oscillation. Even just a few decades ago most of the central Arctic was covered by multi-year ice 3–4 m thick. Now a larger and larger fraction of the central Arctic has experienced ice-free periods, and the thickness of multi-year ice has decreased by 1 m (Wadhams, 2016). In a future warmer Arctic, sea-ice extent and thickness may also experience qualitative changes, for example more snow ice and frazil ice may form as a result of different freezing pathways. Present Arctic sea-ice models do not represent the physics and uncertainties of such future sea-ice cover scenarios well.

Changes in sea-ice cover and sea temperatures also affect marine emissions to the atmosphere, including sea salt aerosols, organic sea spray, and DMS (Struthers et al., 2011; Nilsson et al., 2001; Browse et al., 2014). There are a number of secondary processes that are also likely to effect marine emissions under large-scale melting. For example, bubbles are generated on melting sea ice (Norris et al., 2011), leading to enhanced sea spray formation because they protrude from the ocean surface (de Leeuw et al., 2011). The melting of sea ice also, at least temporarily, causes a brackish surface layer of water that has been observed to produce fewer but bigger bubbles and therefore fewer sea spray particles (Mårtensson et al., 2003). The issue of how the surface melts is also important. Sea spray particle flux measured over open leads was found to be an order of magnitude smaller than measured over the open sea (Nilsson et al., 2001). Thus, while an overall trend of increasing marine emissions is expected with decreasing sea ice, the spatial geometry of the melting will be important in addition to the effects of changes in salinity, fetch, and biological activity. Taken together, these changes will alter the character of the Arctic atmospheric aerosol and thus likely play a role in Arctic cloud formation and cloud processes.

Struthers et al. (2013) combined global climate model output with an emission parameterization to estimate the change in regional and global sea salt aerosol number emissions from 1870 to 2100. Globally averaged, a general increase in sea salt aerosol number emissions was found due to increasing surface wind speed. However, emission changes were not uniform over the aerosol size spectrum due to an increase in sea surface temperature. From 1870 to 2100 the modelled emissions of coarse-mode particles increased by approximately 10 % (global average), whereas no significant change in the emissions of ultrafine-mode aerosols was found over the same period. Significant regional differences in the number emission trends were also found. Based on global climate model output from CAM-Oslo (Seland et al., 2008; Kirkevåg et al., 2008), the predecessor of CAM4-Oslo in NorESM1, no straightforward relationship was found between the change in the number emissions and changes in the sea salt aerosol burden or optical thickness. This was attributed to a change in the simulated residence time of sea salt aerosol. For the 21st century, a decrease in residence time leads to a weaker sea salt aerosol–climate feedback than what would be inferred based on changes in number emissions alone. It should be noted that the above-cited simulations of sea spray in CAM4-Oslo and NorESM1, including Struthers et al. (2013), did not apply an improved temperature-dependent sea spray source parameterization (Salter et al., 2015) built on data from a new Stockholm sea spray simulation tank (Salter et al., 2014). The Salter et al. (2015) parameterization is likely to result in more temperature-sensitive emissions compared to the previously used parameterization by Mårtensson et al. (2003).

Sea spray simulation chambers like the Stockholm simulation tank developed and/or deployed within CRAICC (Sect. 3.2.3) have enabled investigations of the physical and chemical properties of sea spray aerosol. The cloud-forming ability of sea spray aerosol has been targeted in a series of studies coupling CCN counters to sea spray tanks containing real and artificial seawater samples. King et al. (2012) established a baseline for artificial sea salt CCN activity and found that the particle generation method (diffuser vs. plunging jet) can affect particle size distributions and particle properties. During two spring campaigns Rasmussen et al. (2017) investigated the CCN activity of particles generated from seawater samples collected in the bay of Aarhus, DK. The measured CCN activity was similar to particles generated using artificial sea salt and did not vary significantly over the sampling period. Another laboratory study dedicated to the effect of saturated (solid) and unsaturated (liquid) fatty acids on sea spray aerosol properties showed that a significant organic volume fraction of saturated fatty acid (>60 %) was needed to decrease CCN activity compared to that of a purely inorganic sea salt particle of the same size (Nguyen et al., 2017). Salter et al. (2016) presented observations of the size-dependent enrichment of calcium in sea spray aerosol generated from both artificial seawater and natural seawater collected in the North Atlantic. A significant result from CRAICC has been to clarify the possible role of hydrates in dried sea salt particles on climate-relevant properties. Rasmussen et al. (2017) used a sea spray simulation tank coupled with a thermodenuder to show that sea spray volatility can to a large extent be ascribed to the presence of hydrates, while Zieger et al. (2017) used the Stockholm apparatus and found it likely that the presence of hydrates reduces the hygroscopic growth of inorganic sea spray aerosol by 8 %–15 % compared with pure sodium chloride.

Connecting the fundamentals of sea spray from bubble formation and bursting to atmospheric aerosol processes is key to understanding the complex feedbacks associated with changing Arctic surface conditions, as enumerated in Fig. 1, loop C → A.

Iceland represents an underutilized source for Arctic sea-ice recording. Meteorological observations have been carried out in Iceland since the early 1800s although the Icelandic Meteorological Office was not established until 1921. In place of scientific observations, regional historical sources indicate the extent of glaciers and the presence of sea ice centuries back in time (Björnsson, 2016; Ogilvie and Jónsdóttir, 1997). For example, information on monthly sea-ice extent goes back to 1850 based on farmer diaries and ship captain logbooks (Jónsdóttir, 1995). Systematic observations begin in the early 1900s (DMI, 2018) and are more recently complemented by higher-temporal- and spatial-resolution data obtained from the 1970s onward during the satellite era. Geologic indicators such as sediment records also provide insight into environmental change over the last millennia (Knudsen et al., 2009; Larsen et al., 2011).

Sea-ice extent off the Icelandic coast shows considerable variation with time. Conditions were severe throughout the second half of the 19th century and into the 1920s, with sea ice blocking the north coast for weeks or months in most years, often with serious implications for fishing, farming, and transport (Jónsdóttir, 1995). Much milder conditions followed and were observed up to 1965, when harsh conditions returned for more than a decade. The 21st century has seen dramatic changes in Northern Hemisphere sea-ice cover with decreased ice extent, thickness, and proportion of multi-year ice within the ice pack (NSIDC, 2018; Wadhams, 2016; Lei et al., 2014; Jónsdóttir and Sveinbjörnsson, 2007).

Most lakes on Earth are located at high latitudes in the Arctic and boreal regions where they are covered by ice during winter (Verpoorter et al., 2014). Changing cryospheric conditions are critical to how these lakes evolve and influence the global climate system. High-latitude lake–atmosphere interactions are largely governed by surface temperature, waves, and the presence and character of ice cover. Lake processes can act on timescales of days to months, depending on lake size, while longer memory effects can extend over the length of the ice season. Lakes are active players in weather phenomena and budgets of gases, and large lakes and lake districts show up in regional climatology (Yang et al., 2013; Leppäranta, 2015). Numerous publications have shown that recent climate warming has affected lake icing, resulting in later freezing dates and earlier ice break-up dates, both on the order of 5–7 days, over the last century (Kirillin et al., 2012). In addition to higher air temperatures, break-up is forced by solar radiation, which causes melting throughout the ice column, where internal melting also contributes to mechanical weakening and ice cover breakage. Overall ice thickness may decrease due to milder autumn–winter conditions or less cloudy spring conditions that directly affect the radiation balance at lake surfaces. Trends in winter precipitation that affect snow amount can also affect ice thickness and quality.

In contrast to land surfaces, turbulent heat transfer in lakes provides a strong smoothing mechanism for temperature variations. Surface roughness is less over lakes than over land and the water surfaces are continuously striving to achieve water vapour equilibrium vis à vis ongoing evaporation and sublimation, resulting in locally humid conditions. Ice cover has a major impact: turbulence decays, circulation becomes thermohaline, and lake water becomes decoupled from the atmosphere. Cold, freshwater ice is largely non-permeable to gases but becomes permeable as it deteriorates during melting. Compounds deposited from the atmosphere are stored within the ice cover during the ice season and released during the short melting period.

The annual cycle of boreal and tundra lakes can be divided into stratified summer conditions, the ice season, and mixing periods between them (Fig. 14). The formation of the summer stratification necessitates that air temperatures rise to more than 4 ∘C, while freezing takes place only with sustained below-freezing air temperatures and is delayed depending on lake depth. Lake ice response to climate warming is expected to manifest as reduced ice cover periods but also a change in the quality of ice seasons. Early and late winter are characterized by periods when ice is weak and may be broken by wind. During the stable phase, ice cover is safe and thick enough to support on-ice traffic. In a warming climate, the stable period is expected to shorten due to the decreasing ice thickness, and eventually the entire ice season may become unstable.

Although lakes only comprise about 3.7 % of the Earth's non-glaciated land area (Verpoorter et al., 2014) they are very efficient in emitting greenhouse gases. Recently, it has been shown that on a global scale as much as 2.1 Pg C is annually emitted from CO2 reservoirs in lakes, ponds, and running waters (Raymond et al., 2013). This amount is comparable to the annual CO2 uptake by oceans (≈2.0 Pg C yr-1; Wanninkhof et al., 2013). Thus, fresh water must also be considered as an important regulator within the global carbon cycle and thereby also influences climate.

A large fraction of the CO2 emitted from inland waters has its origin in surrounding terrestrial ecosystems. However, a substantial amount is also produced by microbial and photochemical mineralization within the water column (Cole et al., 1994; Humborg et al., 2010; Weyhenmeyer et al., 2015). The production of CO2 within the water column depends upon the availability of dissolved organic carbon (DOC), which enters inland waters from the surrounding watersheds. Kasurinen et al. (2016) developed models to quantify the DOC export from soils in northern watersheds. The modelling approach showed that DOC concentrations depend on watershed water storage as well as on soil temperature. Consequently, changes in land use will have a major impact on DOC export and therefore CO2 production in lakes.

Each year during the cold winter months substantial amounts of CO2 are produced and accumulated below lake ice cover (Denfeld et al., 2015, 2016a). Similarly, CH4 is produced below ice cover, in particular when nutrients do not limit its production (Ricão Canelhas et al., 2016; Denfeld et al., 2016b). At ice break-up (melting stage) these gases can be emitted into the atmosphere. However, gas production in water below ice is still less than production during the open water season (Denfeld et al., 2016a). Thus, given sustained nutrient conditions, similar groundwater hydrological connectivity, and sustained biological CO2 in the water column, it is also likely that gas production and emissions from inland waters will increase with longer and stronger cryosphere melt seasons (Denfeld et al., 2016a). It has been further suggested that these and other physical manifestations of global warming will impact ice phenology including time-delayed effects and feedbacks, which inspires continued research.

CRAICC has greatly increased the available BC deposition data from the Arctic. Previous observations of BC deposition were mainly from North American emission-influenced and high-elevation Greenland sites (McConnell et al., 2007; McConnell and Edwards, 2008; McConnell, 2010; Keegan et al., 2014), which are not necessarily representative of overall BC deposition trends at lower-elevation Arctic locations. Many of those receive the majority of their BC loading from Eurasia (McConnell et al., 2007; Hirdman et al., 2010). The Greenland records show increasing BC deposition from 1850, peaking around 1910, followed by a decline to almost pre-industrial levels after 1950 (McConnell et al., 2007; McConnell and Edwards, 2008; McConnell, 2010; Keegan et al., 2014). The CRAICC work includes a pioneering determination of depositional BC fluxes from the atmosphere to a Svalbard ice core (Ruppel et al., 2014) and four northern Finnish lakes sediments (Ruppel et al., 2015), questioning the prevailing conception of declining or stable Arctic BC values during the last decades.

The Svalbard and Greenland ice cores show similar temporal BC trends between 1750 and 1950, after which the Svalbard ice core indicates an increase in EC (a thermal–optical proxy for BC) deposition post-1970 (Fig. 15a), deviating strongly from the Greenland records. This increase is hypothesized to be partly caused by increased flaring emissions from gas and oil extraction in northern Russia (Ruppel et al., 2014). Flaring has been a previously underestimated Arctic BC emission source but is suggested to account for 42 % of mean annual atmospheric surface BC concentrations in the Arctic (Stohl et al., 2013). These within-Arctic BC emissions do not reach the high-elevation Greenland ice core sites due to restricted isentropic uplift in the atmosphere between the emission sources and Greenland summit (Stohl, 2006; Stohl et al., 2013), which helps to explain the differences between Greenland and Svalbard BC deposition trends.

Four northern Finland lake sediment records also show a similar standardized average SBC deposition trend in agreement with the Svalbard ice core with increasing deposition since 1970 (Fig. 15b; Ruppel et al., 2015). Lake sediment BC records are more sensitive to external factors (e.g. delayed BC import from catchment areas) than ice cores, and thus only general trends observed in multiple records can be considered significant (Rose and Ruppel, 2015). The observed SBC deposition increase is likely caused by regional BC emissions from the Kola Peninsula, as BC deposition from long-distance sources would have been expected to clearly affect all the studied sediment records (Ruppel et al., 2015). Thus, although the observed results cannot be extrapolated to wider regions, the fact that increasing BC deposition trends were observed in different environmental archives analysed with different methods suggests a robust result. Furthermore, the implication of the findings is that BC sources are considerable and that similar BC deposition trends may be observed in other parts of the Arctic.

Consequently, the CRAICC data suggest that BC deposition has increased since 1970 in some parts of the Arctic, coinciding with an increasing climatic impact of BC. This highlights substantial gaps in the previous (AMAP, 2011, 2015; Bond et al., 2013) understanding of Arctic BC trends. Conclusions that Arctic warming has occurred during the last 20 years despite decreasing BC concentrations (AMAP, 2011) may have been premature. More observational BC deposition data are urgently needed to confirm the BC trends across other parts of the Arctic.

The Svalbard ice core and Finnish lake sediment BC records both underline the potential existence of strong BC emission sources within the Arctic (Ruppel et al., 2014, 2015). The presence of high-latitude BC sources is important because within-Arctic BC emissions have a higher probability to remain in the region and are known to result in an amplified Arctic surface warming (5×) compared to the same emissions originating at mid-latitudes (Sand et al., 2013a). If the BC deposition trend has increased since the 1970s over wider Arctic areas, BC may have hastened the retreat of the Arctic sea ice, contrary to what has been previously thought (e.g. Doherty et al., 2010) based on temporal snapshots derived from BC in snow measurements, which show lower values between 2005 and 2009 than during the 1980s.

SILAM chemical transport modelling results developed within CRAICC to investigate the Svalbard ice core data have the potential of increasing the general understanding of meteorological processes driving BC deposition trends. Modelled BC deposition trends at the Svalbard glacier show variations that appear independent from BC emissions or trends in atmospheric BC concentration between 1980 and 2015. According to the model, about 99 % of BC mass is wet-deposited on the glacier, indicating that meteorological processes such as precipitation and scavenging efficiency have a stronger influence on BC deposition than the emissions or trends in atmospheric concentration (Ruppel et al., 2017). Additionally, the model results indicate that, contrary to expectations, dry deposition does not follow a similar trend to atmospheric concentrations across the Arctic, in particular at sites located closer to anthropogenic sources (Soares et al., 2019). Thus, the model results do not point to a specific within-Arctic BC emission source being responsible for the observed post-1970s increase in European Arctic BC deposition, but highlight the fact that temporal trends in atmospheric BC concentrations and deposition may diverge due to meteorological processes. Consequently, Arctic BC deposition trends should not directly be inferred based on atmospheric BC measurements, and more observational BC deposition data are required to comprehensively assess the climate impact of BC in Arctic snow (Ruppel et al., 2017).

With the reduction of atmospheric sulfate concentrations since the 1980s in Europe and North America, the aerosol cooling effect has likely decreased. Using the CMIP5 version of the climate model NorESM1-M, Acosta Navarro et al. (2016) found that the reduction of sulfate in Europe between 1980 and 2005 can explain as much as half of the warming observed in the Arctic during the same period. Air quality emission(s) regulations have succeeded in reducing soil and water acidification and have therefore also removed a large portion of the aerosol dampening effect, and thus the Arctic is experiencing warming more in line with increased global greenhouse gas levels. However, over continental Europe itself the modelled warming due to the same European emission regulations is only one-fifth of the observed warming (Kirkevåg et al., 2016). Acosta Navarro et al. (2016) found that a redistribution of the energy input to the Arctic over the year appears to be a critical factor in explaining the stronger temperature response in the Arctic. In short, the reduction of European anthropogenic sulfate emissions leads to increased summertime atmospheric heat transport (+0.25 W m-2) and increased yearly oceanic heat transport into the Arctic region (+0.48 W m-2). Together with the increased ocean heat transport, the summertime energy surplus is mainly used to melt sea ice, thus increasing the amount of available energy in the Arctic and delaying the onset of autumn freezing. Delayed freezing is reflected in a larger heat transfer from the ocean to the atmosphere during fall and winter, causing strong lower tropospheric warming.

Most aerosol particulate reaches the Arctic through long-range transport. Arctic sites have measured surface concentrations of EBC (“equivalent” BC determined using light absorption methods) and sulfate for over 3 decades. At these monitoring sites, northern Eurasia has been identified as the largest EBC source region (Sharma et al., 2004: Eleftheriadis et al., 2009; Hirdman et al., 2010), with contributions peaking in winter and spring. During summer, BC emissions from agricultural and boreal fires have a primary influence on Arctic concentrations (Stohl et al., 2006, 2007). Within CRAICC, Sand et al. (2015a) studied feedback effects of increasing BC in the atmosphere. The modelling experiment increased BC to unrealistically high levels to show that increased BC concentrations may change cloudiness and precipitation in ways that will feed back onto the spatial distribution and atmospheric residence times of BC. In extreme cases, increased BC in the upper parts of the troposphere may contribute to stabilizing the atmospheric column below and destabilizing the column above, thus causing a positive feedback which mixes BC to higher altitudes and further from emission sources, e.g. to polar latitudes. These feedback effects are not accounted for in many model experiments wherein BC concentrations are directly controlled.

Using NorESM1 with an emission inventory that includes flaring and seasonal variations from domestic sources, Sand et al. (2013a) investigated how mid-latitude BC emissions impact the Arctic in comparison with local BC emissions. They find that BC emitted within the Arctic has a 5× surface temperature response in the Arctic compared to the same amount of BC emitted at mid-latitudes. A large fraction of this warming is due to increased absorption from BC deposited on snow–sea ice. BC emitted at mid-latitudes, on the other hand, enters the Arctic at higher altitudes and is therefore less likely to be deposited on the surface. BC at higher altitudes may warm surrounding atmospheric layers, but the high static stability of the Arctic atmosphere suppresses any heat exchange between the upper and lower atmosphere (Sand et al., 2013b). Motivated by the BC mitigation potential, Sand et al. (2015b) split the emissions into different sectors and regions and provided detailed numbers for the surface temperature sensitivity in the Arctic per kg emission of BC, taking co-emitted species into account. They found that domestic emissions from Asia have the largest warming effect in the Arctic due to the large absolute amount of these emissions and the relative high BC-to-SO2 fraction. However, the Arctic is most sensitive, per unit mass emitted, to Russian flaring emissions, followed by forest fires and flaring emissions in the Nordic countries.