Volume 126, Issue 6 e2020JC017074
Research Article
Free Access

Persistent Multidecadal Variability Since the 15th Century in the Southern Barents Sea Derived From Annually Resolved Shell-Based Records

Madelyn J. Mette

Corresponding Author

Madelyn J. Mette

U.S. Geological Survey, St. Petersburg Coastal and Marine Science Center, St. Petersburg, FL, USA

Department of Geological and Atmospheric Sciences, Iowa State University, Ames, IA, USA

NORCE Norwegian Research Centre, Bjerknes Centre for Climate Research, Bergen, Norway

Correspondence to:

M. J. Mette,

[email protected]

Search for more papers by this author
Alan D. Wanamaker Jr

Alan D. Wanamaker Jr

Department of Geological and Atmospheric Sciences, Iowa State University, Ames, IA, USA

Search for more papers by this author
Michael J. Retelle

Michael J. Retelle

Department of Geology, Bates College, Lewiston, ME, USA

University Centre in Svalbard, Longyearbyen, Norway

Search for more papers by this author
Michael L. Carroll

Michael L. Carroll

Akvaplan-niva, FRAM–High North Research Centre for Climate and the Environment, Tromsø, Norway

Search for more papers by this author
Carin Andersson

Carin Andersson

NORCE Norwegian Research Centre, Bjerknes Centre for Climate Research, Bergen, Norway

Search for more papers by this author
William G. Ambrose Jr

William G. Ambrose Jr

School of the Coastal Environment, Coastal Carolina University, Conway, SC, USA

Search for more papers by this author
First published: 09 June 2021
Citations: 6


In the North Atlantic Ocean, multidecadal variability in sea surface temperatures (SSTs) over the past several centuries has largely been inferred through terrestrial proxies and decadally resolved marine proxies. Annually resolved proxy records from marine archives provide valuable insight into this variability, but are especially rare from high latitude environments, particularly for centennial timescales. We constructed continuous, absolutely dated records of shell growth (1449–2014 CE; 564 years) and oxygen isotope ratios (δ18Oshell; 1539–2014 CE; 476 years) from shells of the bivalve Arctica islandica from coastal northern Norway, a location sensitive to large-scale North Atlantic Ocean dynamics. An annual (January–December) SST reconstruction derived from δ18Oshell for the past five centuries suggests an increase of at least 2°C from the mid-18th century to 2014. The SST reconstruction correlates significantly with instrumental records and with other proxy reconstructions in the southern Barents Sea region. Spectral analysis of the shell growth and isotope records supports evidence for Atlantic multidecadal variability (65–80 year periodicity) extending into polar and subpolar latitudes for the past five centuries. These results provide additional evidence that multidecadal variability in SSTs are a persistent feature of the North Atlantic marine system.

Key Points

  • Continuous shell growth and geochemistry proxies from northern Norway (southern Barents Sea) reveal marine climate change for the past five centuries

  • Multidecadal periodicity in southern Barents Sea temperatures persists for at least the past five centuries

  • Temperature in the southern Barents Sea has increased by at least 2°C since the 18th century

Plain Language Summary

Ocean temperatures in the North Atlantic over the past 150 years have alternated between relatively warm and cool periods that last around 65–80 years. Much of the evidence for these fluctuations in past centuries comes from natural archives on land (e.g., tree rings and lake sediments). In our study, we investigate these fluctuations in the marine environment in northern Norway using shells from the ocean quahog clam, Arctica islandica. Like trees, these clams produce yearly growth layers in their shells and they live for centuries. The growth rate and geochemistry data collected from the growth layers suggest the temperature fluctuations seen across the North Atlantic Ocean (∼65–80 years) are also present farther north in the southern Barents Sea to at least ∼1750 CE, and have likely persisted for at least five centuries. These results provide evidence that alternating warm and cool periods are a persistent feature of the North Atlantic marine system. Our reconstruction using the shell's chemistry indicates that temperatures in the southern Barents Sea have increased by at least 2°C since the mid-18th century.

1 Introduction

The North Atlantic Ocean plays a key role in modulating global climate (e.g., Marshall et al., 2001). Surface currents of the North Atlantic, including the Gulf Stream and its northern extension, the North Atlantic Current, bring relatively warm, salty waters northward as the surface components of the Atlantic Meridional Overturning Circulation (AMOC). At higher latitudes, these waters exchange heat with the atmosphere and cool to form dense, sinking water masses (deep-water formation) that return southward at depth, contributing to global ocean circulation (e.g., Broecker, 1991; Lynch-Stieglitz et al., 2007). The strength and structure of these currents, the rate of deep-water formation, and ocean-atmosphere interactions across the North Atlantic have influenced major climate events during the deglacial and Holocene (e.g., the Younger Dryas; Barber et al., 1999) and smaller-scale transitions within the Late Holocene (Broecker, 2000; Lund et al., 2006). For example, recent work emphasizes the role of ocean circulation in the onset and magnitude of the Little Ice Age (LIA) in the North Atlantic region (approximately 1400–1850 CE), suggesting it may have been amplified by a weakened Gulf Stream and North Atlantic Current (Lund et al., 2006; Wanamaker et al., 2012) and by sea ice expansion in the Arctic Ocean (Miles et al., 2020; Miller et al., 2012). In the Modern Climate period (post-1850), increasing temperature and flow of North Atlantic surface currents into the Arctic Ocean have contributed to recent sea ice losses and air temperature increases in this climatically sensitive area (Årthun et al., 2012; Asbjørnsen et al., 2020; Drange et al., 2005; Kinnard et al., 2011). Variability in North Atlantic sea surface temperatures (SSTs; Atlantic multidecadal variability [AMV]; Enfield et al., 2001; Trenberth & Shea, 2006), decadal-scale variability in atmospheric patterns (e.g., North Atlantic Oscillation [NAO]; Hurrell & Deser, 2009), and variability in overturning circulation (AMOC) are factors that significantly impact North Atlantic climate on decadal to centennial timescales.

Despite the known significance of North Atlantic Ocean processes to larger-scale climate variability, there remains uncertainty in key aspects of the timing, extent, and mechanisms involved. Climate modeling results for the last millennium, for example, do not generally support a systematic change in the AMOC system during LIA times (e.g., Drijfhout et al., 2013; Moreno-Chamarro et al., 2017). The role of ocean transport and ocean-atmosphere processes in driving AMV during the instrumental period has also received considerable debate (Clement et al., 20152016; Mann et al., 2021; Zhang et al., 2016). Approaches to the debate have most often utilized climate models to understand the relative roles of such processes, employing both slab-ocean and fully coupled climate models (e.g., Delworth et al., 2017; Li et al., 2020; Oeldmann et al., 2020). Whereas climate models can provide unparalleled spatial, temporal, and mechanistic detail in exploring climate variability and forcing mechanisms, they can be hindered by complex dependencies prescribed by the model design (i.e., boundary conditions, parameter estimates, structural uncertainties, etc.; Knutti et al., 2010). Empirical approaches driven by real-world observations are hindered by limited spatial and temporal availability of both instrumental and proxy records (Moffa-Sánchez et al., 2019). The persistence, or lack thereof, of AMV explored through development and analysis of high-resolution, long-term proxy records of marine climate variability can provide additional evidence toward understanding the driving forces behind the observed AMV during the instrumental period (O'Reilly et al., 2019). The debate regarding the role of ocean transport in driving AMV thus provides motivation for the development of proxy records relevant to North Atlantic surface temperature variability.

Annually resolved marine proxy records, especially those from high-latitude environments, are rare (Jones et al., 2009). Terrestrial-based proxies are often used to infer SSTs (Gray et al., 2004; Mann et al., 2009; Wang et al., 2017), which can be a problematic approach given the complex interactions between ocean and atmosphere and potential instability in the proxy-target relationship through time (Alexander et al., 2014; D'Arrigo et al., 2008). Available marine proxy records often exhibit decadal and lower resolution, which are not ideal for detecting the full magnitude and timing of climate variability (Moffa-Sánchez et al., 2019). Fortunately, developments in the field of sclerochronology (the study of accretionary hard tissues of organisms and the temporal context in which they form; Oschmann, 2009) have provided additional avenues for understanding marine paleoclimatic variability through the lens of skeletal marine organisms (e.g., Black et al., 2019; Trofimova et al., 2020). Shells from the marine bivalve, Arctica islandica, are high-quality archives because of their long lifespan (up to five centuries; Butler et al., 2013) and wide distribution across the North Atlantic Ocean (Dahlgren et al., 2000). A. islandica produces annual growth increments within its shell structure (Jones, 1980) that can be crossdated within populations (Black et al., 20162019; Wanamaker et al., 2019) and precipitates its aragonitic shells in oxygen isotopic equilibrium with seawater (e.g., Mette et al., 2018; Wanamaker & Gillikin, 2019; Weidman et al., 1994). This established proxy archive is thus well suited to investigating long-term, high-resolution marine climate dynamics in the North Atlantic Ocean. To this end, we used A. islandica to develop annually resolved, multicentury records of shell growth and oxygen isotopic composition from northern Norway. We apply these records to investigate variability in the marine environment across major climate transitions of the past in a climatically sensitive region at the Atlantic-Arctic gateway. The time series and their spectral properties are compared to AMV dynamics and nearby long-term proxy records (Figure 1) to assess ocean connectivity at high latitudes and the persistence of multidecadal variability through time. We thus add to a more detailed understanding of ocean climate dynamics and interactions for the past several centuries.

Details are in the caption following the image

Northern North Atlantic Ocean currents and locations of proxy data used in this study. Yellow star indicates the bivalve collection site (Ingøya, Norway). Dashed yellow lines show hydrographic monitoring sections at the Barents Sea Opening (BSO) and Kola Section (Kola) maintained by the Institute for Marine Research (Norway) and the Polar Research Institute of Marine Fisheries and Oceanography (Russia), respectively. Yellow circle indicates the location of sediment record used in the Hald et al. (2011) Atlantic Water temperature reconstruction (Malangen Fjord). White circles indicate the locations of Fennoscandian tree-ring records used in the McCarroll et al. (2013) summer (JJA) temperature reconstruction. Base map created in GeoMapApp (version 3.6.10; North Polar Projection; elevation data from General Bathymetric Chart of the Oceans [GEBCO 2014] and Advanced Spaceborne Thermal Emission and Reflection Radiometer [ASTER]). Schematic ocean current pathways styled after Loeng and Drinkwater (2007).

2 Study Location

Ingøya is a small island (18 km2) located about 15 km off the mainland of northern Norway in the southern Barents Sea, strategically located for studying marine climate and ocean circulation dynamics (Figure 1). The Norwegian Atlantic Current, an extension of the North Atlantic Current and major surface component of AMOC, flows northward along Norway before bifurcating into two main pathways as it approaches the BSO (Schauer et al., 2002). The West Spitsbergen Current continues northward directly toward the Arctic Ocean, while the Barents Sea Branch follows deeper contours of the Norwegian coastline near Ingøya before continuing to the interior Barents Sea (Yashayaev & Seidov, 2015). Along coastal Norway, these southerly sourced ocean currents flow parallel to the Norwegian Coastal Current, a cooler and fresher water mass occupying the shallower coastal zone (Skagseth et al., 2011). The properties (i.e., temperature, salinity, and velocity) of these two currents are increasingly coupled dynamically as they progressively mix along their pathways to the Barents Sea (Helland-Hansen & Nansen, 1909; Skagseth et al., 2015). Ingøya is exposed to these ocean current pathways, being situated among the northernmost islands of Norway. Local hydrographic variability and ecosystem processes at Ingøya are thus subject to these larger-scale ocean currents, justifying using proxy records from this location to explore long-term marine climate change.

A. islandica is present around Ingøya, presenting the opportunity to construct long-term, high-resolution records of paleoclimatic change (Schöne, 2013; Weidman et al., 1994). Dense populations of living A. islandica (Figure 2) and abundant dead shell material can be found in a shallow bay on the eastern side of Ingøya (Sanden; 5–15 m) and on adjacent raised beaches dating to 7 kyr BP. Previous work established that shell growth and geochemical records from this location reflect large-scale North Atlantic SST for the modern interval and thus have the potential to inform understanding of the North Atlantic Current system in the past (Mette et al., 2016). The influence of local environmental factors controlling shell growth and activity in this population has also been studied in detail at Ingøya by Ballesta-Artero et al. (20172019).

Details are in the caption following the image

Arctica islandica as a proxy archive. (a) Left valve of a shell showing axis of maximum growth (dashed line) along which the specimen is bisected for further processing. (b) Photomicrograph of an acetate replica peel of the polished shell cross-section. (c) Magnified view of photomicrograph showing annual growth lines. The typical measurement axes for individual increments are superposed (black-capped lines).

3 Methods

3.1 Data

Instrumental data near Ingøya were sourced from in-situ measurements taken during the study period (2009–2016) in addition to publicly available datasets described in subsequent text. Local hydrographic data from a monitoring station 5 km north of Ingøya were obtained through the Norwegian Institute for Marine Research (IMR; see: http://www.imr.no/forskning/forskningsdata/stasjoner/). These data contain submonthly SST and sea surface salinity (SSS) measurements up to 300 m depth since 1938, but with several monthly to multiyear hiatuses. The data from 10 m depth were used for analysis because (a) they contain fewer outliers compared to the surface (1 m depth) time series and (b) 10 m approximates the depth at which we collected the live shells. Local moorings were installed in Sanden (the shell collection site), recording temperature and conductivity at 5 m depth (Onset HOBO Conductivity Logger) and temperature at 30 m depth (Onset HOBO Conductivity Logger) from June 2012 to August 2016. Because the local IMR station temperature record is very similar to the temperature measured from 5 and 30 m depths on the moorings (Figure S1), we focus our analysis on the comparison to the IMR data due to its longer time span. It should be noted, however, that the shell collection site experiences slightly cooler winter temperatures than temperatures on the moorings (<1°C cooler on average in January–April from 2012 to 2016).

BSO temperature anomalies represent coastal (<100 km from shore) upper layer (0–50 m depth) temperature anomalies as presented in Yashayaev and Seidov (2015) from data available through the IMR (http://www.imr.no/forskning/forskningsdata/stasjoner/). Kola Section temperature anomalies were obtained from the Russian Polar Research Institute of Marine Fisheries and Oceanography (PINRO; http://www.pinro.ru). Data from the 0 to 200 m layer across the entire Kola section were used (as opposed to the coastal upper layer as in the BSO data). Gridded SST products used are from the Met Office Hadley Centre Sea Ice and Sea Surface Temperature product (HadISST v1.1; 1° resolution; Rayner et al., 2003) and the National Climatic Data Center Extended Reconstructed SST (ERSST v5; 2° resolution; Huang et al., 2017) as obtained through The Royal Netherlands Meteorological Institute's (KNMI) Climate Explorer web application (https://climexp.knmi.nl/start.cgi). Data from the gridded SST products (HadISST and ERSST) are generally reflective of the IMR station record (Figure S1), and provide an even longer temporal length and consistent monthly resolution available for comparison with the shell-based data.

The AMV index is typically calculated as the January–December mean of North Atlantic SST anomalies with the global (or tropical/subtropical) trends removed. For this work, we consider the AMV index both without and with this detrending procedure, referred to herein as the “raw” and “detrended” AMV indices, respectively. In both cases, we compute an ensemble mean of four members taken from typically used methods for AMV index calculation. This approach was taken to reduce the impact of differing methodological biases on our analysis (AMVens; as in Rossby et al., 2020). The four members comprise two different spatial domains applied to two different SST datasets (e.g., see Rossby et al., 2020). The spatial domains are 0°W–80°W, 0°N–60°N and 7°W–70°W, 25°N–60°N, as used by Trenberth and Shea (2006) and van Oldenborgh et al. (2009), respectively. The data sets comprise the gridded SST products HadISST and ERSST (described above). For the detrended AMVens index, the four members are detrended following the literature source of the respective spatial domain. For the 0°W–80°W, 0°N–60°N domain, the index represents anomalies of mean SST minus global mean SST (Trenberth & Shea, 2006). For the 7°W–70°W, 25°N–60°N domain, the index represents anomalies of SST minus the regression on global mean temperature (van Oldenborgh et al., 2009). The member indices as described were obtained using the KNMI Climate Explorer and manually averaged as the AMV ensemble index (AMVens).

Several publicly available proxy reconstructions of particular geographic (i.e., located nearby) or climatological (i.e., related via ocean or atmosphere processes/interactions) relevance were obtained for comparison with the shell-based records. The AMV reconstruction of Wang et al. (2017; WangAMV) uses 46 annually resolved proxy records from the circum-North Atlantic and Arctic regions. The AMV reconstruction is interpreted to represent the May–September AMV index and was produced using a nested principal component regression method. The Fennoscandian summer temperature reconstruction of McCarroll et al. (2013) uses four annually resolved, absolutely dated tree-ring chronologies from northern Fennoscandia. The Fennoscandian reconstruction is interpreted to represent June–August temperatures and was produced from regression and variance scaling of nine z-scored series (including ring width, density, and tree height). The PAGES2k Arctic annual temperature anomaly reconstruction of McKay and Kaufman (2014) uses climate field reconstruction techniques to compile annually resolved paleoclimate records across the Arctic region. The subdecadally resolved, November bottom water (218 m) temperature reconstruction of Hald et al. (2011) uses sediment records (δ18Oforam) from Malangen Fjord (northern Norway). Their Atlantic Water reconstruction was wiggle-matched against an instrumental SST record (Kola section, Barents Sea) and atmospheric temperature record (Ona, Norway). The subdecadally resolved, NE North Atlantic proxy SST reconstruction of Cunningham et al. (2013) uses high-resolution (subdecadal) sediment records from across the northeast North Atlantic basin (Cunningham et al., 2013). The records were screened based on correlation with broad summer temperatures in their respective regions and were compiled using a composite plus regression approach.

3.2 Sample Collection, Preparation, and Development of the Master Shell Chronology

Living and dead specimens of A. islandica were collected from Sanden Bay, Ingøya, Norway (71°03.734′N, 24°05.895′E). All samples were collected within a 1 km2 area in June-2009, May-2013, August-2014, and June-2015 using a small bespoke dredge in approximately 5–10 m water depth. Additional dead specimens were hand-collected on the beach at the head of the bay. Soft tissues were removed and shells were washed in freshwater before open-air drying. Further processing of shell material for visual analysis, growth increment measurement and crossdating, and geochemical sampling employed standard sclerochronological procedures using the previously published methods from the same location (Figure 2; Mette et al., 2016). A master shell-growth chronology (MSC) was constructed from 39 shells using the software program ARSTAN v44 (Cook, 1985). The Expressed Population Signal (EPS; Wigley et al., 1984) statistic, produced by ARSTAN, represents a measure of the common variability in a chronology, calculated as the correlation between the sample chronology and the theoretical population chronology based on an infinite number of samples. The running series of average correlation (rbar), also produced by ARSTAN, represents a measure of percent common variance or common signal strength, calculated as the average correlation between all series (Cook et al., 2000; Speer, 2010). EPS and rbar were calculated in 30-year windows with 29-year overlap throughout the length of the chronology. Individual detrended growth series were calculated by removing the first 40 (juvenile) growth increments, applying an adaptive power transformation to each series, and calculating residuals after removing a negative exponential curve fitted to each series to produce indexed and normalized values of shell growth, commonly termed the “shell growth index” (SGI; see also Butler et al., 2010). In cases where negative exponential detrending produced a poor fit (n = 16), an alternative detrending curve was used (see Text S2).

3.3 Isotopic Analysis of Shell Carbonate

Stable carbon and oxygen isotope analyses were performed at Iowa State University's (ISU) Stable Isotope Laboratory in the Department of Geological and Atmospheric Sciences (Ames, Iowa). Carbonate samples were collected using a Merchantek MicroMill mounted with a Leica GZ6 microscope. Crossdated annual increments were milled from the outer margin of the shell to a depth of 200–800 μm using Brasseler USAV scriber point (item #H1621.11.008) and round (item #H52.11.003) carbide drill bits set to 100% drill speed, 3–4 passes at 55 μm/s scan speed, 100–150 μm depth per pass, and 55 μm/s plunge speed. All micromilled shell samples were analyzed for oxygen isotopic composition (δ18Oshell) and stable carbon isotopic composition (δ13Cshell) on a Thermo Finnigan Delta Plus XL mass spectrometer coupled with a GasBench II and CombiPal autosampler. The long-term precision of the mass spectrometer was ±0.09‰ for δ18O and ±0.06‰ for δ13C, respectively (1σ standard deviation) based on reference standards NBS18 and NBS19. Values are reported in delta notation (δ) relative to the Vienna PeeDee Belemnite (VPDB). Sample replication was performed where feasible (i.e., depending on availability of material) and at key intervals (i.e., ensuring overlap where one series ends and another begins) to inform an estimate of intrashell isotopic variability (Mette et al., 2018). To create a single δ18Oshell series from replicated years (i.e., where multiple shells were sampled over the same time period), values were averaged per year among replicates.

3.4 Salinity and Oxygen Isotope Analysis of Water

Water samples were collected from Ingøya intermittently between 2012 and 2016 and analyzed for salinity using a YSI Pro Plus Handheld Multiparameter Meter (precision = 0.3 PPT). Oxygen isotopic composition of water (δ18Ow) was measured on a Picarro L1102-i Isotopic Liquid Water Analyzer (ChemCorrect software; long-term precision = 0.07‰, 1σ standard deviation; reference standards OH-1, OH-2, and GISP). Values are reported in delta notation (δ) relative to Vienna Standard Mean Ocean Water (VSMOW).

3.5 Statistical Methods

Linear relationships and shared variance between shell records and environmental indices were evaluated using the Pearson correlation coefficient (r) and coefficient of determination (r2). Correlations of smoothed time series were explored to remove noise from local environmental influences (filtered out through smoothing) and assess coherence at lower frequencies (higher periodicities). Running means were applied to the records to highlight patterns in graphical representation and for correlation with other smoothed records where specified. Because the Hald et al. (2011) record from Malangen does not contain data for every year, this operation smoothed over the missing values, resulting in an in-filled series. Associated p-values were adjusted for smoothing by dividing the original degrees of freedom (dof) by the length of the smoother where specified. P-values were not adjusted for autocorrelation in the time series. Lagged correlations were explored to consider the potential time delay in atmosphere-ocean processes and general ocean surface transport from south/west to north/east, which can occur on the order of several months to several years (Chafik et al., 2015).

Two independent techniques of spectral analysis were used to assess temporal characteristics in the shell growth and isotope time series: (a) continuous wavelet transformation (CWT; Torrence & Compo, 1998) and (b) multitaper method spectral estimate (MTM; Mann & Lees, 1996; Thomson, 1982). By comparing the spectral characteristics of the shell growth record in frequency (MTM) and time-frequency (CWT) space, possible forcing mechanisms can be further considered. The CWT has the advantage of revealing periodicities that may not be stationary in time (i.e., vary in amplitude or in periodicity through time; Grinsted et al., 2004), in other words, dominate only a small portion of the record. The MTM power spectra identify frequencies within the data that express high enough power overall to characterize the spectral nature of the time series. Notably, however, MTM may also identify significant concentrations of variance associated with single events (i.e., not related to oscillations; Delong, 2008). Significance for the CWT and MTM analyses was assessed at the 95% confidence level relative to a red noise background (AR1). Cross wavelet analysis was also performed to compare the shell records and nearby long-term proxy records in time-frequency space (Grinsted et al., 2004). The MTM spectral estimates were computed using kSpectra version 3.4.3 (resolution = 2, number of tapers = 3). The secular trend was removed from the δ18Oshell series before computing the MTM and CWT estimates. CWT and cross wavelet transforms were performed using Matlab 9.4 version R2018a (code from Grinsted et al., 2004).

3.6 Temperature Reconstruction

The aragonite temperature equation of Grossman and Ku (1986), as modified by Dettman et al. (1999), was used to derive paleotemperature estimates from δ18Oshell:
The oxygen isotopic composition of water (δ18Ow = −0.166‰) was determined using direct measurements of water samples from Ingøya (Table S1) and derived estimates from the local instrumental salinity records using the salinity-mixing relationship of Mette et al. (2016), updated for added decimal precision to show rounding units:

To evaluate the fidelity of the Ingøya δ18Oshell-temperature reconstruction in relation to the ERSST (1854–2014) and HadISST (1870–2014) annual means from the region 71°N –73°N, 23°E–25°E, we divided the common timespan between the reconstruction and target temperature series in half into calibration and verification periods. Each series was standardized to zero mean and unit standard deviation before the calibration and verification procedure. The Reduction of Error (RE) and Coefficient of Efficiency (CE) were used to measure the skill of the reconstruction to the gridded SST products. These statistics compare the performance of the proxy model to that of the mean climatology between a calibration period (first or second half of the target temperature series) and a verification period (second or first half of the target temperature series). See Cook et al. (1994) for more detailed explanations of the RE/CE procedure.

We calculated standard error for 30-year periods throughout the length of the δ18Oshell-temperature reconstruction in the form:
(following Flannery et al., 2018) where the individual uncertainties include 1σ standard deviations from intershell variability (a2; 1.32°C) and long-term precision of the mass spectrometer (b2; 0.39°C) and root-mean-squared errors from the oxygen isotope-salinity mixing line (Equation 2; c2; 2.21°C) and the Grossman and Ku (1986) paleotemperature equation (Equation 1; d2; 1.37°C; determined from data available in Table 2 of Grossman and Ku, 1986).

4 Results

4.1 Master Shell Growth Chronology

The MSC (Figure 3a) contains measurement series from 39 individuals (5 live caught and 34 dead collected shells; Figure 3b). Ages of the individuals in the chronology averaged 239 years, ranging from 128 to 394 years. The oldest individual is the longest-lived specimen reported thus far from Norway. It was collected dead (empty shell) from the beach at the head of the bay. The mean length of individual series used for chronology construction, after removal of the first 40 juvenile growth increments from each series, was 207 years. Mean EPS and rbar statistics (see Section 3.2) averaged 0.90 and 0.43, respectively. The population growth pattern, taken from the “standard” MSC computed by ARSTAN, is presented as a shell growth index (SGI; Figure 3a).

Details are in the caption following the image

Shell-based records from Ingøya, Norway, and associated series statistics. (a) δ18Oshell for individual shells (light blue), averaged for replicated years (medium blue), and 21-year running mean (dark blue). The shell growth chronology is shown as annual indexed values (Shell Growth Index) for individual shell series (gray), the standard MSC (thin black), and a 21-year running mean (thick black). (b) Expressed population signal (EPS; black; right axis) of the MSC calculated over a 30-year running window and the number of shells (left axis) included in the MSC (dark gray line) and δ18Oshell series (blue line) through time. The dashed line shows the 0.85 threshold value for EPS.

4.2 Oxygen Isotopic Composition of Shell Increments (δ18Oshell)

The δ18Oshell series (Figure 3a) contains isotopic measurements from 23 individuals. One shell was resampled due to the loss of some samples through instrument issues and is included as a separate series (for a total of 24 series). Years were assigned to samples by crossdating the sampled growth increment series with the MSC. The longest series collected from a single individual spans 120 years (1539–1659), with all other single measurement series ranging from 3 to 97 years. Mean δ18Oshell for annually compiled measurements was 3.15‰. Mean standard deviation among replicate samples (where n ≥ 3) was 0.31‰ (1σ standard deviation).

4.3 Oxygen Isotopic Composition of Seawater (δ18Owater)

A total of 13 water samples were collected within 1 km of the study site in May, June, and August of 2012–2016 (Table S1; Figure 4). Measured δ18Ow values ranged from −0.50‰ to 0.00‰, with a mean of −0.17‰. After applying the salinity−δ18Ow mixing line equation of Mette et al. (2016) (Equation 2) to salinity records from the nearby IMR monitoring station, the resulting mean monthly δ18Ow ranged from −0.202‰ to −0.113‰, with an annual mean of −0.166‰ (Figure 4). This value (δ18Ow = −0.166‰) was used for temperature reconstruction.

Details are in the caption following the image

Annual variation in salinity (left axis) and δ18Ow (right axis) from long-term records near Ingøya (1968–2019; box-and-whisker plots) and from intermittent sampling at the bivalve collection site (2012–2016; blue diamonds). Long-term data (box-and-whisker plots) represent median (horizontal black line), interquartile range (IQR; gray box), extended range (within 1.5*IQR; gray vertical lines), and outliers (gray points) of mean-monthly salinity measurements at a monitoring station 5 km north of Ingøya (10 m depth; data source: Norwegian Institute for Marine Research; http://www.imr.no/forskning/forskningsdata/stasjoner/view?station=Ingoy). Values on the right axis (δ18Ow) align with salinity as derived from the local mixing line for Ingøya (Equation 2). δ18Ow values for bivalve collection site water samples (blue diamonds) were measured in the laboratory. Error bars show 1σ estimate for uncertainty due to instrumental precision of the Picarro Liquid Water Analyzer (0.07‰) and the equivalent uncertainty in salinity (0.51 PSU) derived from the mixing line.

4.4 Comparison of the Shell-Based Records With Environmental Records

4.4.1 MSC, δ18Oshell, and Local-Scale to Large-Scale SST

The MSC (Figure 3a) significantly correlates with the detrended AMVens, ranging between r = −0.37 (p < 0.001, dof = 162) and r = −0.43 (p < 0.001, dof = 162) at 0–6-year lags with no clear maximum. The MSC shows no significant correlations (p > 0.05) with the raw (undetrended) AMVens. In contrast, the δ18Oshell record does not significantly correlate with the detrended AMVens, but does significantly correlate with the raw AMVens. Annual correlations between the δ18Oshell record and the raw AMVens ranged from r = −0.44 (p < 0.001, dof = 162) to r = −0.46 (p < 0.001, dof = 162) at 0–6-year lags with a maximum (r = −0.46; p < 0.001, dof = 162) at a 3-year lag. A similar pattern is found for correlations between the MSC and δ18Oshell records and gridded instrumental SST products nearby the collection site (71°N–73°N, 23°E–25°E). The MSC shows few significant correlations with monthly, annual (January–December), and growing season (February–September, April–October, and June–August) SST, with correlations ranging between r = −0.13 (p = 0.10, dof = 158) and r = −0.29 (p < 0.001, dof = 158). The δ18Oshell record significantly correlates with all local monthly, annual, and growing season SST means tested using the ERSST and HadISST datasets, ranging between r = −0.22 (p < 0.01, dof = 160) and r = −0.54 (p < 0.001, dof = 160).

4.4.2 Temperature Reconstruction and Regional Marine Records

The 476-year temperature reconstruction (1539–2014) based on δ18Oshell (Figure 5) exhibits annual values ranging from 1.12°C to 8.90°C with a mean value of 5.02°C. Evident in this reconstruction are notable extended (multidecadal to centennial) warm and cool periods (Table 1; Figure 5). RE and CE for the ERSST target were RE = 0.53 (0.51) and CE = −0.29 (−1.54) using 1935–2014 (1854–1934) as the calibration period. The HadISST target produced RE = 0.24 (0.35) and CE = −0.83 (−0.43) using 1943–2014 (1870–1942) as the calibration period.

Details are in the caption following the image

Sea surface temperature reconstructed from δ18Oshell from Ingøya, Norway. Annual (thin black line) and climatological means (30-year periods; gray bars) are shown. Background shades for climatological means are scaled from coldest (blue) to warmest (red). The height of the shaded area represents the ±0.54°C standard error for each climatological mean calculated from combined uncertainty of intershell variability (1σ standard deviation, 1.32°C), long-term precision of the mass spectrometer (1σ standard deviation, 0.39°C), the oxygen isotope-salinity mixing line (Equation 2 in text; root-mean-squared error, 2.21°C), and the Grossman and Ku (1986) paleotemperature equation (Equation 1 in text; root-mean-squared error, 1.37°C). Extended cold and warm periods in the late 18th century and late 20th century, respectively, are highlighted by horizontal brackets above the plot area.

Table 1. Summary Statistics for δ18Oshell and Reconstructed Temperature Over Periods of Interest
Period Mean δ18Oshell (‰) Reconstructed temperature (°C)
All years 3.15 5.02
1705–1824 (coolest) 3.32 4.30
1915–2014 (warmest) 2.86 6.30
[difference] [−0.46] [+2.00]
  • Note. Reconstructed temperature is estimated using Equation 1 (Grossman and Ku, 1986 as modified by Dettman et al., 1999) and δ18Ow = −0.166‰. A change of −0.46‰ in δ18Ow, assuming a constant temperature, is equivalent to a salinity change of −1.69 PSU. For reference, average annual temperature at Ingøya over the period 1936–2018 measures 6.28°C.

The δ18Oshell-temperature reconstruction positively correlates with local temperatures represented by the HadISST and ERSST gridded products for the region 71°N–73°N, 23°E–25°E as well as the IMR temperature records from 10 m depth (filtered to years that contain at least one measurement in every month of the period considered). Year on year (annual) correlations with HadISST and ERSST (Table 2) range between r = 0.22 (p < 0.01, dof = 160) and r = 0.54 (p < 0.001, dof = 160) and with IMR 10 m temperature range between r = 0.00 (p > 0.1, dof = 50) and r = 0.30 (p = 0.03, dof = 52). While correlations with the IMR record are weaker and generally statistically insignificant, smoothing both records with a 5-year moving average results in r = 0.44 (adjusted p > 0.1, dof = 8), similar to correlations found between proxy temperature and the gridded SST products. Limiting the comparisons to months considered to be representative of the shell growing season (e.g., beginning in February–April, ending in September–November; Ballesta-Artero et al., 2017; Mette et al., 2016) does not produce higher correlations. We therefore elect to consider the reconstruction to represent annual (January–December) temperature means. The δ18Oshell-temperature reconstruction correlates weakly, yet significantly, with temperature anomalies along the BSO and Kola Sections (Figure 6). Excluding data post-2008, for which the proxy temperature exhibits a sharp negative excursion, annual correlations with BSO and Kola are r = 0.38 (p < 0.01, dof = 57) and r = 0.31 (p < 0.05, dof = 56), respectively.

Table 2. Correlations Between Ingøya Temperature Reconstruction and Local SST
Jan 0.54 0.32 0.13
Feb 0.54 0.24 0.17
Mar 0.49 0.22 0.15
Apr 0.44 0.27 0.21
May 0.41 0.29 0.08
Jun 0.37 0.27 0.00
Jul 0.41 0.31 0.14
Aug 0.41 0.38 0.30
Sep 0.37 0.42 0.11
Oct 0.32 0.36 0.16
Nov 0.42 0.29 0.08
Dec 0.51 0.32 0.06
Jan–Dec 0.49 0.42 0.07
Feb–Sep 0.47 0.39 0.24
Apr–Oct 0.44 0.41 0.14
Jun–Aug 0.42 0.35 0.15
  • Note. Values are given as the Pearson correlation coefficient (r). Gridded data ERSST and HadISST are extracted from the region 71°N–73°N, 23°E–25°E. ERSST, National Climatic Data Center Extended Reconstructed SST v5 (Huang et al., 2017). HadISST, Met Office Hadley Centre Sea Ice and Sea Surface Temperature v1.1 (Rayner et al., 2003). IMR 10 m = Norwegian Institute for Marine Research long-term monitoring station at Ingøya (temperature at 10 m depth; see Section 4 for details regarding treatment applied to these data).
  • SST, sea surface temperatures.
  • Significant correlations above 95% (p < 0.05) are bolded.
Details are in the caption following the image

Comparison of regional annual marine instrumental records and Ingøya δ18Oshell-temperature reconstruction. Thick lines represent 7-year running means. (a) Kola section and BSO temperature anomalies (Yashayaev & Seidov, 2015). (b) Sea surface temperature reconstruction from A. islandica18Oshell) from Ingøya, Norway (this study) and local instrumental temperature (black) from the Norwegian Institute for Marine Research monitoring station at Ingøya at 10 m depth (IMR 10 m). (c) Local temperature records extracted from (black) the Met Office Hadley Centre Sea Ice and Sea Surface Temperature data set v1.1 (HadISST; (23°E–25°E, 71°N–73°N; Rayner et al., 2003); and (gray) the National Climate Data Center Extended Reconstructed Sea Surface Temperature data set version 5 (ERSST; 23°E–25°E, 71°N–73°N; Huang et al., 2017). (d) Atlantic Multidecadal Variability (AMVens) instrumental ensemble anomaly.

4.4.3 Temperature Reconstruction and Long-Term Proxies

The 476-year δ18Oshell-temperature reconstruction shows significant correlations with other long-term proxy records in the region (Figure 7). It matches most closely with the Arctic surface temperature reconstruction from proxy data contained within the PAGES 2k Arctic Database (Figure 7b; McKay & Kaufman, 2014). Correlations are positive and significant (r = 0.39, p < 0.001, dof = 460), increasing to r = 0.75 when smoothed with a 21-year moving average (adjusted p < 0.01, dof = 21). These relationships are strengthened when a 6–10-year lag is applied to Arctic temperature. The correlation between the Ingøya δ18Oshell-temperature reconstruction and the Fennoscandian proxy summer temperature reconstruction of McCarroll et al. (2013) (Figure 7c) is nearly as strong (r = 0.33, p < 0.001, dof = 466), increasing to r = 0.65 (adjusted p < 0.01, dof = 21) when smoothed. Applying a 1-year lag to the Ingøya reconstruction increases the annual correlation with proxy summer temperature to r = 0.37 (p < 0.001, dof = 466), but does not improve the smoothed relationship. The AMV reconstruction of Wang et al. (2017; Figure 7d) correlates weakly but positively with the Ingøya δ18Oshell-temperature reconstruction, ranging between r = 0.23 (p < 0.001, dof = 471) to r = 0.28 (p < 0.001, dof = 469) with a 0-year to 4-year lag (AMVWang lagging proxy shell temperature), increasing to r = 0.47 (adjusted p < 0.05, dof = 21) when smoothed. Correlations between the Ingøya δ18Oshell-temperature reconstruction and Malangen Fjord Atlantic Water temperature reconstruction of Hald et al. (2011; Figure 7e), are significantly positive, ranging from r = 0.36 (p < 0.001, dof = 188) to r = 0.42 (p < 0.001, dof = 188) with a 0–5-year lag (shells lagging Malangen), and increasing to r = 0.66 (adjusted p = 0.05, dof = 21) to r = 0.69 (adjusted p < 0.05, dof = 21) when smoothed. Correlations with the NE North Atlantic proxy SST reconstruction of Cunningham et al. (2013; Figure 7f) are comparable to those obtained for the AMVWang record, being weak but significantly positive (r = 0.20, p < 0.001, dof = 435), increasing to r = 0.37 (adjusted p = 0.09, dof = 20) when smoothing is applied to the Ingøya δ18Oshell-temperature reconstruction (and no further smoothing applied to the Cunningham record).

Details are in the caption following the image

Comparison of Ingøya δ18Oshell-temperature reconstruction (a) with regional surface (b)–(c) and marine (d)–(f) proxy temperature reconstructions. Thin lines represent highest resolution data (annual in A through D, subdecadal in E, decadal in F) and thick lines represent 21-year running means. (a) Sea surface temperature reconstruction from A. islandica δ18Oshell (blue) from Ingøya, Norway (this study), also shown in panels B through (f) (b) PAGES 2k Arctic temperature reconstruction (McKay & Kaufman, 2014). (c) Surface temperature summer (JJA) reconstruction from tree ring records across Fennoscandia (McCarroll et al., 2013). (d) Atlantic Multidecadal Variability proxy reconstruction from North Atlantic terrestrial records (Wang et al., 2017). (e) Bottom water (218 m) temperature reconstruction from sediment records (δ18Oforam) in Malangen Fjord (Hald et al., 2011), Norway. (f) Sea surface temperature reconstruction from marine records across the northeast North Atlantic basin (Cunningham et al., 2013).

4.5 Spectral Analysis

Several significant periodicities were identified in the MSC and δ18Oshell records (Table 3, Figure 8). The MTM identified significant multidecadal periodicities between the 30- and 85-year range for both records, however, none of the significant ranges overlap. The MTM and CWT results show there may be some commonality at higher periodicities (>>100 years), however, this range of frequencies should be interpreted with caution due to the length of the record in relation to the period identified. The MTM and CWT identified similar spectral characteristics inherent to each of the MSC and δ18Oshell records, providing robust evidence for the identified periodicities, with the CWT providing more detail on the temporal variability (see date ranges in Table 3).

Table 3. Periodicities Identified in Shell-Based Records Used in This Study
Multitaper method (MTM) Continuous wavelet (CWT)
SGI δ18Oshell (detrended) SGI δ18Oshell (detrended)
Periodicity 100–215* >185 90–130* (1725–1950) 75–100* (>1750)
45–55* 70–85* 40–70* (<1550, >1925) 60–65 (1740–1780)
10 30* 5–10* (<1500, 1560–1610, 1980–2000) 20–30* (1660–1750)
  • Note. Significant periodicities (in years) ≥10 years within Ingøya shell-based records (SGI, shell growth index from master chronology; detrended δ18Oshell) were identified through spectral analysis using the Multitaper Method (kSpectra version 3 using a red noise background (autoregressive lag 1), resolution = 2, number of tapers = 3) and continuous wavelet analysis (using Matlab 9.4 version R2018a; code from Grinsted et al., 2004). Periodicities were tested at the 95% significance level and estimated to the nearest 5-year bins for MTM and CWT, with years present noted in parenthesis for CWT. Periodicities identified through Multitaper Method also detected through continuous wavelet analysis are starred and in bold.
Details are in the caption following the image

Spectral analysis of the northern Norway shell-based records. Time series plots of the shell growth index (a; SGI) and δ18Oshell (d). Continuous wavelet power spectra (b), (e) showing significant periodicities through time (Matlab 9.4 version R2018a, Morlet wavelet; code from Grinsted et al., 2004). Secular trend was removed from the δ18Oshell series for (e) Color bar denotes spectral power with units of unit2 and °C2 for (b) and (e), respectively. The 95% significance level is noted by black outlines. Faded area denotes cone of influence within which results are interpreted with caution (Torrence & Compo, 1998). Multitaper Method spectral estimate (c, f; resolution = 2, number of tapers = 3; kSpectra version 3.4.3). Shaded regions (gray) show significant ranges of periodicities exceeding a red noise background (autoregressive lag 1) at the 95% significance level (gray line) and are labeled with the corresponding period in years.

Cross wavelet analysis between the Ingøya δ18Oshell-temperature reconstruction and nearby proxy reconstructions for SATJJA (Fennoscandian summer (JJA) temperature based on tree rings; McCarroll et al., 2013) and AMVWang (multiproxy; Wang et al., 2017) show covariation between the time series at several frequencies that varies with time (Figure 9). The strongest and most persistent coherence is evident between the Ingøya δ18Oshell-temperature reconstruction and Fennoscandian summer temperature reconstruction (SATJJA) at a ∼90–100-year periodicity, with significance above 95% for nearly the entire period analyzed (1550–1925 CE). Additional covariation is observed with nearby proxy reconstructions around the ∼30 and ∼60-year bands that modulate in covariance with time.

Details are in the caption following the image

Cross wavelet spectra between the Ingøya proxy SST reconstruction and (a) reconstruction of Atlantic Multidecadal Variability (AMVWang; Wang et al., 2017) and (b) Fennoscandian tree ring summer (JJA) surface temperature reconstruction (SATJJA; McCarroll et al., 2013). Color bar denotes spectral power. Cross wavelet was computed using Matlab 9.4 version R2018a (code from Grinsted et al., 2004). The 95% significance level is noted by black outlines. Faded area denotes cone of influence within which results are interpreted with caution due to the influence of edge effects (Torrence & Compo, 1998). Relative phase relationships are depicted as arrows (in-phase pointing right; anti-phase pointing left; any other direction = out-of-phase). SST, sea surface temperatures.

5 Discussion

The shell growth chronology (564 years) and the oxygen isotope series (476 years) from northern Norway represent the second-longest records of each type available at the time of this publication (Butler et al., 2013; Reynolds et al., 2016). This exceptional chronology length at a climatically significant area at the Atlantic-Arctic boundary provides a strong basis for investigating past climate periods through identifying changes in mean conditions and forcing-scale periodicities within the data set and in comparison with other proxies. Marine climate of the Barents Sea is known to closely follow the dynamics of the North Atlantic, mainly through its direct ocean connection via the North Atlantic Current and atmospheric circulation (Asbjørnsen et al., 2020; Levitus et al., 2009; Yashayaev & Seidov, 2015). The magnitude and stability of this relationship can be tested through analysis of the shell-based records from Ingøya (southern Barents Sea, northern Norway) due to absolute age control via crossdating of the shell growth series (Black et al., 2019), providing an error-free age model for the δ18Oshell series and associated temperature reconstruction. While the shell-based records are sourced from near-surface waters, local instrumental records show that deep-water dynamics, representing core Atlantic Water, closely parallel those at the surface (Figure S2). Since the absolute temperature reconstructed at the surface is several degrees warmer than core Atlantic Water at depth, we make the assumption that the reconstruction reflects parallel changes in the mean state and captures decadal to multidecadal variability throughout the water column.

Relationships between the MSC, δ18Oshell, and a multiproxy index with local to regional instrumental records over a shorter time frame (1900–2012) were previously presented in Mette et al. (2016) and briefly summarized here. All shell records were found to negatively correlate with a broad swath of North Atlantic SSTs and are maximized when combining the shell records into a Multiproxy Index (1982–2012; r = −0.54 to −0.90; p < 0.05). The updated analysis presented herein supports these findings, as AMV is negatively correlated with the MSC and δ18Oshell records. We now consider the effects of detrending and statistical treatment on these relationships and therefore exclude further consideration of a combined (multiproxy) index. The consequences of detrending individual shell growth series during MSC construction (i.e., the “segment length curse,” Cook et al., 1995), potential for nonlinear relationships between growth proxies and temperature (e.g., Emile-Geay & Tingley, 2016), and influence of phenotypic plasticity in biological organisms (e.g., Rodríguez et al., 2019), can result in the loss of decadal and longer timescale signals in long-term records of growth rates within an A. islandica population. In our case, and consistent with this consideration, the MSC correlates better with the detrended AMVens as opposed to the raw AMVens, likely because the former is computed with removal of the larger-scale (global or subtropical) temperature trend (Trenberth & Shea, 2006; van Oldenborgh et al., 2009). The δ18Oshell record is the result of equilibrium isotopic precipitation, a time-stable geochemical process, and understood to be independent of any biological interference (i.e., vital effects). Thus, it is expected that δ18Oshell correlates more strongly with the raw AMVens, representing Atlantic surface water conditions without any trend removal, sharing between 19% and 22% of the variance between a 0 and 2-year lag (Figure 6d; proxy lagging AMVens).

5.1 Temperature Reconstruction Based on δ18Oshell

5.1.1 Comparisons With Instrumental Records

The Grossman and Ku (1986) paleotemperature equation (Equation 1) is a commonly used transfer function for converting δ18O measurements from bivalve shell aragonite to paleotemperature estimates (e.g., Bonitz et al., 2018; Reynolds et al., 2017; Schöne et al., 2004; Wanamaker et al., 2008, and many others). Visually, the reconstructed values show a range similar to that of the local IMR records over recent decades, yet show a considerably wider range of values before 1970, and a notable decreasing excursion after 2008 (Figure 6b). The local IMR station data show weak and mostly insignificant statistical relationships with the δ18Oshell-temperature reconstruction on an annual basis. The irregular sampling and, thus, poor representation of monthly means inherent to the IMR data set likely influence this finding (Anderson & Gough, 2018). The 5-year smoothed records share 20% of the variance, a level of coherence that is more evident upon visual inspection of the time series (Figure 6b). The temperature derived from gridded products HadISST and ERSST (Figure 6c), which represent statistically infilled estimates of local temperature (i.e., there are no gaps in the data), share between 17% and 29% of the variability with the δ18Oshell-temperature reconstruction. The notable divergence between the HadISST and ERSST datasets before ∼1930 (Figure 6c) highlights the questionable reliability of any instrumental data in this region before 1930. Indeed, gridded SST products are often considered unreliable before 1950 (DeLong et al., 2014; Deser et al., 2010) due to low availability of direct observations of ocean temperature (Figure S3).

We rely on positive correlation coefficients over the entire period (Table 2), the long-standing use of δ18O as a reliable temperature proxy, and additional explorations in pseudoproxy modeling (Figure S4) to suggest the reconstruction does offer considerable value in understanding long-term regional temperature change while also containing a significant local signal at higher frequencies. The target temperature data sets show remarkably differing patterns, especially before 1930, suggesting available data products are either too sparse and/or uncertain in this region for a robust characterization of local temperatures and thus, may contribute to unreliable RE and CE results with our reconstruction. The post-satellite era (>1980), where temperature data are considered of higher quality, is too short for proper calibration and verification procedures. Still, at least 18%–25% of the variance in the Ingøya δ18Oshell-temperature reconstruction is shared with the gridded temperature targets. The variance increases to 34%–45% when a 5-year moving average is applied to the data.

The nearby hydrographic sections at Kola and the BSO (Figure 6a) provide additional useful targets for comparison due to their length (>50 years) and common use for understanding marine dynamics and transport of Atlantic Water into the Barents Sea (e.g., Chafik et al., 2015; Herbaut et al., 2017; Yashayaev & Seidov, 2015). The δ18Oshell-temperature reconstruction from Ingøya shows some similarity with those records, explaining between 7% and 15% of the variability, excluding the most recent years (post-2008) for which the proxy exhibits a steep negative excursion. The standard deviation among replicated δ18Oshell samples in this post-2008 (1σ = 0.42‰) period is greater than average (1σ = 0.31‰), suggesting higher uncertainty in the data for this portion, and should be further explored in future work (e.g., through resampling and analysis).

5.1.2 Temperature Reconstruction Key Features and Long-Term Trends

A notable characteristic of the δ18Oshell-temperature reconstruction over the past century (Figure 6b) is a decade-long cool period centered around 1955 CE. The Kola, BSO, and gridded temperature records appear to exhibit a similar, although greatly muted, anomaly. This may suggest significant influence from near-shore hydrography and dynamics within the Norwegian Coastal Current on the δ18Oshell-temperature reconstruction. While the reconstruction shares significant year-to-year variability with nearby and regional instrumental records, local dynamics within the shallow bay display a clear overprint. The lower shared variance between the annual temperature reconstruction with regional records compared to their filtered (smoothed) versions further suggests a local overprint. Multidecadal patterns within the proxy reconstruction are more consistent with nearby and regional instrumental records.

The δ18Oshell-temperature reconstruction from Ingøya (Figures 5 and 7a) shows notable extended warm and cool periods between 1539 and 2014 (assuming constant δ18Ow). The coolest period spans from 1705 to 1824 and the warmest period spans from 1915 to 2014, representing an increase of 2.0°C from the mid-18th century to the mid-20th to late-20th century (Table 1). This long-term warming may be indicative of increased Atlantic Water influence in the Barents Sea, possibly due to a strengthening or warming North Atlantic Current, as suggested by many other studies (Figure 5; Asbjørnsen et al., 2020; Dijkstra et al., 2015; Hald et al., 2011; Reynolds et al., 2016; Spielhagen et al., 2011; Wanamaker et al., 2012; Yashayaev & Seidov, 2015). Combined with the prevalence of AMV-like periodicity detected through CWT analysis of the Ingøya δ18Oshell-temperature reconstruction (Figure 8), evidence for the so-called “Atlantification” of the Barents Sea in recent times (Asbjørnsen et al., 2020), and perhaps over longer timescales, is supported by these shell-based records from Ingøya, Norway.

Because δ18Oshell is a function of both water temperature and δ18Ow, the potential for changing salinity (and thus, δ18Ow) influencing the δ18Oshell-temperature reconstruction should be examined. If changing δ18Ow was the sole contributor to the trends observed in the 0.46‰ decrease in δ18Oshell between the coolest and warmest periods (Table 1), the local salinity−δ18Ow mixing relationship (Equation 2) would predict a −1.69 PSU change in salinity from the mid-18th century to the early 20th century. Given the low seasonal variation in mean monthly (2σ standard deviation = 0.26 PSU) and yearly (2σ standard deviation = 0.24 PSU) salinity with only a minor (statistically insignificant) decreasing trend since record-keeping began in the 1930s (0.02 PSU per decade), a change in salinity of this magnitude is unlikely. On shorter timescales (i.e., over the past 80 years), instrumental measurements across the northern North Atlantic and Nordic Seas reveal a notable 30-year period of freshening between the 1960s and 1990s (Holliday et al., 2008), and an overall freshening of 0.04 PSU between 1951 and 2010 (Mork et al., 2014). Little is known, however, about changes in salinity along the Norwegian coast within the Atlantic or coastal waters for the past five centuries. A hypothesized weakened North Atlantic Current during LIA times (e.g., Dylmer et al., 2013; Wanamaker et al., 2012) suggests that northward surface ocean transport has increased since the mid-18th century. An associated increase in Nordic Seas salinity (and δ18Ow) is, therefore, also plausible. While previous work illustrates that changes in δ18Oshell at this location are likely dominated by temperature as opposed to δ18Ow (and salinity) variations (Mette et al., 2016; see also Table S4), considering the competing influences of δ18Ow (and salinity) and temperature on the δ18Oshell signal, and the hypotheses suggesting increasing salinity on longer timescales, we further argue that the 2°C temperature increase since the mid-18th century likely represents a minimum estimate.

5.1.3 Comparison of the δ18Oshell-Temperature Reconstruction With Other Proxy Records

Previously published high-resolution, multicentury North Atlantic proxy records provide appropriate benchmarks to compare with the northern Norway δ18Oshell-temperature reconstruction (Figure 7). The AMV reconstruction of Wang et al. (2017; Figure 7d) represents the most comprehensive and updated terrestrial proxy record of AMV available at the time of this publication. Though this AMVWang reconstruction is statistically robust, it is sourced primarily from terrestrial records of marine variability and thus, more likely to suffer from non-stationarity in the relationships between atmospheric-oceanic variability, or proxy response to that variability. The positive correlation between the Ingøya δ18Oshell-temperature reconstruction and the AMVWang record (annual r = 0.23, p < 0.001; smoothed r = 0.47, adjusted p < 0.05), highlights the common variability especially on multidecadal timescales. The network of proxies used in the AMVWang reconstruction contains most of the tree-ring records used in the Fennoscandian summer temperature reconstruction (Figure 7c; McCarroll et al., 2013). Due to the proximity of the Fennoscandian records, it is therefore not surprising that the annual correlation between the Ingøya δ18Oshell-temperature reconstruction and this record is stronger (annual r = 0.33, p < 0.001; smoothed r = 0.65, p < 0.01). The Ingøya δ18Oshell-temperature reconstruction most closely matches with the PAGES 2k Arctic temperature reconstruction (Figure 7b; McKay & Kaufman, 2014; annual r = 0.39, p < 0.001; smoothed r = 0.75, p < 0.01), highlighting the importance of Barents Sea dynamics to Arctic variability. The strengthened relationships found with a 6–10-year lag applied to PAGES 2k Arctic temperature are congruent with the timescales for propagation of temperature anomalies (Polyakov et al., 2020) and also through ocean influences on Arctic air temperatures via impacts on sea ice extent (e.g., Årthun et al., 2012; Carmack et al., 2015).

For comparison with a nearby marine record, we consider the Malangen Fjord Atlantic Water temperature reconstruction (Figure 7e; Hald et al., 2011). Although this record does not exhibit annual resolution, it represents the most highly resolved marine temperature reconstruction in the northern Norway coastal region. Correlations are significantly positive (annual r = 0.34, p < 0.001; smoothed r = 0.66; adjusted p = 0.05), however, the decreasing temporal resolution of this record through time renders the smoothed version of this record and resulting correlations less meaningful. Further, the reconstruction has been interpreted to best reflect November bottom water temperatures, limiting the usefulness of the comparison. The NE North Atlantic proxy SST anomaly reconstruction of Cunningham et al. (2013; Figure 7f) represents another reconstruction from a compilation of marine proxies exhibiting subdecadal resolution over the larger region with the Malangen Fjord as its northernmost member. Correlations with this record are comparable to those obtained for the AMVWang record, being weak but significantly positive (annual r = 0.20, p < 0.001; r = 0.37 when smoothing is applied to the Ingøya reconstruction). Hald et al. (2011) attribute variations in the Malangen Fjord sediment core benthic foraminiferal δ18O to influx of Atlantic Water. They highlight the gradual decreasing temperature trend until the 14th century (also noted in Cunningham et al., 2013) and a distinct LIA characterized by several periods of cooler than average temperatures, followed by unprecedented warming after 1800. A similar pattern can be identified from the Ingøya δ18Oshell-temperature reconstruction.

5.2 Spectral Properties of the Shell-Based Records

The spectral properties identified both in the MSC and δ18Oshell records are similar to that of AMV (65–80 years) in modern times and in the past (Schlesinger & Ramankutty, 1994; Wang et al., 2017). Notably, however, AMV-like periodicities are more significant in the latter half of the δ18Oshell record, which may suggest the spectral character of the AMV is not uniform through time (Alexander et al., 2014). Before ∼1800, the δ18Oshell record exhibits additional, somewhat higher frequencies (20–30 year periodicity) compared to post-1800. Cross wavelet analysis between the Ingøya δ18Oshell-temperature reconstruction and the Fennoscandian summer temperature reconstruction shows high power in the ∼90–100 year band for almost the entire length of the series (Figure 9b), suggesting it is a prominent feature among proxy reconstructions in the nearby region. This supports the close coupling of atmosphere and marine temperature dynamics since the late 16th century. It also lends further evidence to suggest such multidecadal variability in the Barents Sea (and likely, North Atlantic Ocean) has persisted for at least five centuries. Another notable detail from the cross wavelet analysis is the out-of-phase relationship at the 20–30 year band, indicating the leading nature of the Fennoscandian record. The significance of this periodicity, and the phase relationship, are consistent during mid- to late-LIA times (∼1625–1750 CE; Figure 9b). The lead-lag relationship suggests an additional feature of atmosphere-ocean coupling during a key climate interval that should be investigated in future work.

5.2.1 Implications for Understanding AMV Forcing at High Latitudes

Southern Barents Sea variability has long been understood to be influenced by North Atlantic processes through the direct ocean connection via the North Atlantic Current (Loeng, 1991). Because of this, evidence for persistent multidecadal variability in southern Barents Sea temperatures may suggest a dominant role of ocean dynamics in driving AMV-like periodicities in the region (Miles et al., 2020; Zhang, 2017). From the perspective of long-term proxy records compared in our study, temperature variability at Ingøya in the southern Barents Sea shares more variance with regional atmospheric dynamics represented by the Fennoscandian summer temperature reconstruction (11%) and reconstructed Arctic temperature (15%) than with the larger-scale marine dynamics represented by the AMVWang reconstruction (5%–8%). This highlights a strong ocean-atmosphere connection in the region and, together with the potentially significant lead-lag relationships identified through correlation and cross wavelet analyses, suggests a role for atmospheric forcing in driving SST patterns (Clement et al., 2015; Mann et al., 2021). All proxy records compared herein share fairly low levels of variance, and as previously noted, the AMVWang reconstruction is sourced from terrestrial-based proxy records considered sensitive to SST. Further complexity arises in seasonal aspects of ocean forcing. Atlantic inflow via the North Atlantic Current is strongest in winter months, imprinting on regional temperature for the remainder of the year (Årthun et al., 2012; Ingvaldsen et al., 2002; Ottersen et al., 2000). Additionally, the wintertime Norwegian Coastal Current is deeper and narrower than in summertime due to weakening cyclonic winds (Skagseth et al., 2011), influencing upwelling conditions and potentially shell growth. A more complex analysis of the δ18Oshell-temperature reconstruction from the southern Barents Sea presented is thus required to lend insight into the forcing mechanisms of AMV (i.e., separating internal vs. external forcing components). In summary, no clear resolution can be offered on the competing influences of oceanic-forcing versus atmospheric-forcing from our analysis of the Ingøya δ18Oshell-temperature reconstruction.

6 Summary and Conclusions

The shell growth and shell geochemical records from northern Norway provide marine-sourced, annually resolved proxy data that elucidate marine climate dynamics across the past five centuries, with most insight gained in understanding the timing and magnitude of temperature change and frequency characteristics of temperature variability. The δ18Oshell-temperature reconstruction from Ingøya suggests coldest water temperatures in the region were experienced throughout the 18th and early 19th centuries, transitioning by at least 2°C into the warmest period represented by 1915–2014 CE. The shell-based records also support evidence for dominant multidecadal, AMV-like periodicity (i.e., 65–80 year periodicity) in southern Barents Sea temperatures persisting for at least five centuries. Before ∼1800, slightly higher frequency variability (20–30 year periodicity) is also evident in the temperature reconstruction. The multidecadal variability detected in the southern Barents Sea supports the view of larger scale (Atlantic) multidecadal variability as a persistent, and potentially internally driven feature of the climate system, improving confidence in decadal-scale predictions of future climate (Zhang, 2017). This is especially relevant given the predominance and more frequent recognition of terrestrial-based AMV reconstructions in the literature (e.g., Mann et al., 2021).

Proxy records from the marine realm are commonly collected from sediment cores and other archives exhibiting subdecadal to multidecadal resolution. As a result, current understanding of marine variability is limited to longer-term, slower-operating climate dynamics. More abrupt climate changes that can happen within a decade, for example, thus tend to be “smoothed” over several years or decades in sediment records, masking the true nature of some significant marine climatic events. The annually resolved δ18Oshell-temperature reconstruction presented in our work provides an absolutely dated and robust estimation of the magnitude and pacing of climatic changes. The δ18Oshell-temperature reconstruction from the southern Barents Sea exhibits periods of especially rapid change (especially ∼1730 and 1930), coincident with the terrestrial-based proxy record of McCarroll et al. (2013), for example, warranting further investigation. Finally, our record contributes to an increasing network of shell-based records across the North Atlantic Ocean with excellent absolute annual dating on par with terrestrial-based tree-ring reconstructions, enhancing the opportunity to assess spatial variability in the marine system through the lens of benthic ecosystems (Black et al., 2019; Reynolds et al., 2018). Continued development of the oxygen isotope and shell growth records by extending further into the past will provide nearly unparalleled perspective on marine environmental conditions of the past millennium.


This study was supported by the U.S. National Science Foundation Grants #1417636 to Alan D. Wanamaker and #1417766 to Michael J. Retelle and William G. Ambrose and Geological Society of America Graduate Student Research Grants to Madelyn J. Mette. Michael L. Carroll was supported by the Research Council of Norway (Grant 227046) and by ARCEx partners and the Research Council of Norway (Grant 228107). Postdoctoral support for Madelyn J. Mette was provided through funding by the Icelandic Research Fund Grant 173906-051 (hosted by Carin Andersson at NORCE Norwegian Research Centre, Bergen, Norway). The authors thank Thorleif Hanssen, Ann Hansen, Erlend Hesten, and Dimitri Barjitski for extensive logistical support at Ingøya. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government. The authors thank two anonymous reviewers for providing useful comments for improving the manuscript. The authors thank Kristine DeLong for providing a thoughtful and thorough review that greatly improved this work.

    Data Availability Statement

    The shell-based data presented in this study have been deposited in the NOAA Paleoclimatology Database (https://www.ncdc.noaa.gov/paleo-search/study/33212). Data for other proxy records presented for comparison are publicly available in online databases: AMV reconstruction of Wang et al. (2017), https://www.ncdc.noaa.gov/paleo/study/22031 (accessed November 3, 2017); Fennoscandian summer temperature reconstruction of McCarroll et al. (2013); https://www.ncdc.noaa.gov/paleo/study/19943 (accessed August 21, 2020); PAGES2k Arctic temperature anomaly reconstruction of McKay and Kaufman (2014), http://ncdc.noaa.gov/paleo/study/16973 (accessed September 16, 2020); Malangen Fjord November bottom water temperature reconstruction of Hald et al. (2011), https://doi.pangaea.de/10.1594/PANGAEA.810470 (accessed October 30, 2017); and NE North Atlantic SST reconstruction of Cunningham et al. (2013), https://www.ncdc.noaa.gov/paleo/study/14193 (accessed October 30, 2017).