The simulations of the isotopic composition of precipitation and water vapour in this study are conducted with isoCAM3. This is an atmospheric general circulation model (AGCM) enabled with the ability to trace the various species of water isotopologues. The model is based on the Community Atmosphere Model version 3 (CAM3) , and the isotope module was developed by David Noone, University of Colorado. More details of isoCAM3 can be found in The isoCAM3 model has been applied in several studies that investigated the isotopic response to past climate changes (; ; ; ; ; ; ).

The horizontal resolution of the model is T85 (∼ 1.4∘ × 1.4∘) with 26 hybrid-sigma levels in the vertical. In this study the SST and sea ice concentrations are specified; thus, the only surface temperatures that are calculated interactively are land and sea ice surface temperatures. This configuration allows for no feedback between atmospheric circulation and open-ocean SST. Greenhouse gases, vegetation, ice sheets are all set to modern conditions. More specifically greenhouse gasses are set to the following CAM3 default levels (year 1990): CO2 – 355 (ppmv), CH4 – 1714 (ppbv), N2O – 311 (ppbv). The solar constant is set to 1365 (W m-2) and orbital configurations are set to the year 1850.

We perform a set of four sensitivity experiments and one control simulation to investigate how observed variations in Arctic sea surface conditions influences δ18O. Every model integration is run for 15 years (following 1 year for spin-up). Each of the four sensitivity experiments simulates the δ18O response to sea ice concentration and SST for selected years in the time period 1979–2013 within the satellite era. The 12-month time periods are selected based on the four most extreme cases of high and low September sea ice extent recorded during the time period (1979–2012) by the National Snow and Ice Data Center (NSIDC) Sea Ice Index updated daily. The control simulation (CTRL) simulates the δ18O response using the 12-month climatology of sea ice concentration and SST for the full time period April 1979–March 2013. Only the Arctic oceanic surface boundary conditions differ between the runs. An overview of the model experiments are given in Table .

We force the model isoCAM3 with an annual cycle of monthly mean SST and sea ice conditions obtained from ERA-Interim . This annual cycle goes from April to March, thus spanning the full sea ice cycle related to the selected cases of September sea ice extent. Subsequently, the model runs for 15 years (following 1 year of spin-up) with repeated annual cycles. All re-analysis data are interpolated bilinearly from the ERA-Interim (1∘ × 1∘) to the CAM3 T85 resolution, and subsequently checked for consistency.

Changes in Arctic SST are in nature an inseparable part of sea ice changes. Keeping the SST constant and only simulating the atmospheric response to sea ice changes, would therefore lead to unrealistic temperature gradients (see , for further discussion on this topic). Therefore, we chose that these experiments are based on both changes in sea ice and SST. A masking of the SST data is applied to eliminate remote influences from extra-polar climate patterns (e.g. from the El Niño–Southern Oscillation or Pacific Decadal Oscillation). This masking is constructed so that only the conditions near the Arctic differ from experiment to experiment. Hence, this global ocean data are divided in an Arctic and a non-Arctic region. The Arctic region refers to the region of ocean/sea ice conditions expected to influence the Arctic climate and is therefore rather semi-Arctic. Due the geographical configuration of the continents it is chosen to confine this region with southern boundaries of 66∘ and 37∘ N for the Pacific and Atlantic sector respectively. The relatively southern definition of the semi-Arctic region in the North Atlantic is chosen to also include the southern-most position of sea ice export in the Newfoundland area.

Each experiment is forced by different SST and sea ice conditions in the (semi-)Arctic region corresponding to the values for the selected year. The non-Arctic part of the data set is identical for all the different experiments and has values from the mean climatology of ERA-Interim 1979–2012. The area between the Arctic and non-Arctic part in the North Atlantic has strong naturally occurring SST gradients. To avoid smoothing of natural SST gradients, no smoothing is applied to the constructed oceanic data set. The sea ice concentrations and SST used to force the model are shown in Figs.  and here displayed as annual mean anomalies between the respective experiment and the CTRL run.

Changes in sea ice concentration and SST force a strong local response in surface air temperature (T2m) (see Fig. ) with cooling where sea ice extent is increased, and warming where sea ice extent is decreased. The simulated temperature changes are in agreement with other modelling studies that have investigated the atmospheric response to prescribed re-analysis-based changes (, see also reviews ). Changes in annual mean precipitation amount are found negligible (see Appendix A).

The CTRL run is compared to values of δ18O observations from ice cores and GNIP stations for Greenland and a positive bias is found (see figure in Appendix A). As a consequence this study only investigates anomalies and not absolute values. All sensitivity experiments clearly show that changes in sea surface conditions influence the modelled δ18O of Arctic precipitation (Fig. ). Decreased (increased) sea ice concentration and connected SST results in enriched (depleted) δ18O values of precipitation (hereafter referred to as δ18Op). Annual means of δ18Op are computed as precipitation-weighted annual means. The spatial distribution of changes in δ18Op is similar to the spatial distribution of changes in simulated surface air temperature.

This shows that the spatial response of the simulated δ18Op to changes in sea surface conditions is controlled by the distribution of these changes. The distribution of the δ18Op response to the ocean conditions depends on the sea ice and SST configuration in the different experiments. As shown in Fig.  the δ18Op of the precipitation over central part of Greenland appears unaffected by the simulated changes in sea ice cover in all experiments whereas δ18Op changes over the Pacific–Arctic oceans and the Barents–Kara seas regions depend on the distribution of sea ice in the given experiment.

The experiments “1980” and “1996” both have increased sea ice extent and colder SSTs compared to the CTRL experiments, yet the spatial distributions of the sea surface conditions in the Arctic Ocean are very different. This is observed in the Barents/Kara Sea region, in Baffin Bay and near the northern coast of Greenland. The corresponding isotopic response match the differences in the spatial pattern observed in the sea ice cover.

The two experiments with low sea ice extent compared to the CTRL experiments (the “2007” and “2012” experiments) similarly show that the sea ice distribution is important for δ18Op. The Labrador–Baffin seas region does not experience any significant change in the isotopic composition of precipitation in the “2007” experiment. Conversely, significant changes are simulated in the “2012” experiment where the sea ice changes in this region are much more pronounced. For the Barents Sea region both experiments yield positive δ18Op anomalies, but the amplitude of the anomalies are different. Interestingly, this difference in amplitudes is also found in the sea ice concentration anomalies used to simulate the isotopic response. Thus, this suggests that both distribution and magnitude of the changes in sea surface conditions are important for the change in δ18Op.

From a climate reconstruction perspective it is interesting to examine whether the isotope-temperature relationship (δ18Op–T) is sensitive to changes in sea ice cover and SST. Scatter plots of annual mean anomalies of Δδ18Op–ΔT are shown in Fig. . Only grid points in the Arctic (60–90∘ N) are included in the analysis.

Linear regression shows that the spatial Δδ18Op–ΔT slopes for each of the experiment all are within the range of 0.38 to 0.53 ‰ ∘C-1. Linear regression for all experiments together (Fig.  “All experiments”) show a larger range of values of anomalies and yields a slope of 0.57 ‰ ∘C-1 with R2=0.761. For experiments with high sea ice extent, the slope is 0.38 ‰ ∘C-1 with R2=0.59 for “1980” and 0.53 ‰ ∘C-1 with R2=0.575 for “1996”. For experiments with low sea ice extent, the value of the slope is 0.42 ‰ ∘C-1 with R2=0.732 for “2007” and 0.48 ‰ ∘C-1 with R2=0.635 for “2012”.

In this study, the slope of δ18Op–T relationship is found to be insensitive to changes in the perturbation of sea ice. Differences in the intercept values of the regression are noted, most pronounced for experiment “2012” where the offset of Δδ18Op is -0.39 ‰.

The δ18Op response to sea ice changes (Fig.  ) shows that the response is predominantly local, yet with the “2012” experiment showing a more regional response. Here we broadly define a local response as a situation where the grid points in close proximity to regions of sea ice change experience large changes in δ18Op, and where grid points without sea ice change show no pronounced changes in δ18Op. Similarly, a regional response here is used to describe a response where changes in δ18Op both occur at grid points in close proximity to regions of sea ice changes and also at neighbouring grid points without sea ice changes.

Examination of the anomalies of isotopic composition of the water vapour yields insight into the isotopic composition of the Arctic moisture. Figure  shows the anomaly of isotopic water vapour composition at the 850 hPa level (hereafter referred to as δ18Ov). The anomaly is plotted with the 850 hPa level wind field anomaly overlayed. Similar to the isotopic composition of precipitation, the isotopic composition of water vapour at the 850 hPa level reveals, for all experiments, local anomalies at the same locations as anomalies of sea surface conditions occur. Locations with decreased (increased) sea ice extent and concentration are co-located with locations of enriched (depleted) water vapour. The wind vectors in Fig.  show that the changes in advection at the 850 hPa level cannot explain the change in δ18Ov. Interestingly the highest wind anomalies are found in the “2012” experiment, which is also the experiment that displayed a more widespread and regional isotopic response to sea ice changes. The slight increase in local wind anomalies could indicate that advection is responsible for the larger spatial extent of the isotopic response.

Changes in local evaporation here are investigated based on the surface latent heat flux over ocean and ice. To compare how changes in sea surface conditions change the amount of total local evaporation, only locations with grid points of strongly reduced sea ice (change bigger than 20 %) were selected and the amount of total latent heat flux per year for all grid points between 60 and 90∘ N was calculated for all experiments. To account for different numbers of grid points with sea ice change for each experiment, the comparison to the CTRL run is done using identical locations of the grid points, such that non-local effects in evaporation changes were excluded. As observed in Fig.  the amount of local evaporation is remarkably stronger for grid points where sea ice is reduced and weaker where sea ice is increased. The number of grid points of reduced sea ice in each of the experiments are 1980: 217, 1996: 444, 2007: 1148, 2012: 2116 and the number of grid points of increased sea ice are 1980: 1508, 1996: 1024, 2007:554, 2012: 437.

The two experiments with low sea ice extent (“2007” and “2012”) have warmer temperatures, more intense evaporation and higher values of δ18Ov than the CTRL experiment. This contrasts the two remaining experiments (“1980” and “1996”), which have sea ice extent larger than the CTRL run (based on 1979–2012 mean). In these experiments lower temperatures are observed as well as less intense evaporation and lower values of δ18Ov compared to the CTRL experiment. Our results confirm that sea ice concentration and SST control the ability of the ocean to evaporate water.

Based on the pronounced local structure of the isotopic response, and the evidence of an increase in ocean evaporation when sea ice is reduced and SSTs are warmed (see Fig. ), it is speculated that the perturbation in the isotopic composition are caused by changes in the contribution from local moisture sources. An increase in local Arctic Ocean evaporation would contribute with heavily enriched water to the ambient vapour resulting in vapour that has a higher value of δ18Ov. This could explain the simulated δ18Ov anomalies. In the case of an increased sea ice cover, the decrease in the contribution of local enriched water would result in ambient vapour with a lower value of δ18Ov. This hypothesis is supported by observational studies of the impact of Arctic sea ice changes on the isotopic composition of moisture . Changes in evaporation of local ocean water have also been suggested by modelling studies as important for sea ice induced changes in δ18Op . Furthermore, an analysis of future warming in the Arctic using state-of-the-art climate models showed changes in the hydrological cycle due to Arctic warming and sea ice changes . In that study it was found that moisture inflow from lower latitudes played a minor role, and the changes in the hydrological cycle were mainly caused by strongly intensified local surface evaporation.

An alternative explanation for the simulated changes in δ18Ov and δ18Op is that the changes occur as result of changes in air mass characteristics. Reductions in the poleward temperature gradient would reduce the cooling and condensation that air masses experience during the northward transport. This would cause isotopic composition of the air masses to be less depleted. In this study, the sea surface conditions effect on Arctic warming is clearly seen on the simulated T2m (see Fig. ). Additionally, also the vertical cross sections (Figs.  and ) cannot exclude that the changes in the δ18Ov are caused by changes in atmospheric temperature. Nevertheless, it is difficult to explain the spatially very local effects of δ18Op as a cause of reduction in the poleward temperature gradient. Yet, sea ice changes are connected to regions of cyclogenesis . Thus, regions of open and warmer ocean surfaces might potentially steer cyclones to follow these paths and precipitate over the these regions, thereby creating a local signal of δ18Op changes. Our experimental design cannot reveal the synoptical variability and the effects of changed wind patterns are not clear from analysis of annual mean advection in the 850 hPa layer in Fig. . The wind speed (u2+v2) at the 300 hPa level is weakened at midlatitudes in this study, which indicates that the changes in sea surface conditions are influencing atmospheric circulation, yet no clear connection to the changes in sea ice extent is found. Based on the considerations above it is difficult to separate the effects of changes in temperature and changes in evaporation, and consequently model simulations with moisture tracking features are suggested for further investigation of this study. However, independent of the cause of the changes, it is found that changes in sea surface conditions are important for the isotopic composition of non-Greenland δ18Op in the Arctic.

Changes in the isotopic composition of Greenland precipitation are of special interest due to the ice core research sites in this region. Interestingly, none of the sea ice perturbation experiments in this study display δ18Op changes over Greenland. Thus, the vertical distributions of T and δ18Ov near the location of the ice core drilling site NEEM, Greenland (∼ 77∘ N, 51∘ E), are used to investigate the differences in the response in Greenland and the rest of the Arctic. Figures  and show the circumpolar zonal vertical distribution of T and δ18Ov at the nearest grid point levels to NEEM. At non-Greenland locations the anomalies of T and δ18Ov are surface-based signals, sensitive to the local conditions. However, near NEEM, the Baffin Bay sea ice extent and associated simulated response in δ18Ov are important for the δ18Ov at NEEM. In experiment “2007” the Baffin Bay sea ice extent is increased compared to the mean values, while the near-NEEM δ18Ov displays negative anomalies of δ18Ov in the range 0.2–1 ‰, this in spite of an overall Arctic enrichment. This suggests that the local conditions at Baffin Bay, and not the general Arctic conditions, are relevant for studying the δ18Ov response to sea ice changes at NEEM. Modern observations of the isotopic composition of snow and water vapour from NEEM also show that variations in modern values of δ18O correlate with conditions in Baffin Bay sea ice extent .

The lack of sensitivity of the Greenland δ18Op to changes in Arctic Ocean surface conditions is argued to be related to the topography of Greenland. Specifically, the steep slopes of the ice sheet margin are associated with substantial orographic enhancement of precipitation and depletion of air mass water vapour content. Processes controlling the Greenland δ18Op might be decoupled from the processes influencing the δ18Op over the Arctic ocean. The Greenland katabatic wind blocking effect might also play a role in blocking of low level moisture to Greenland.

We note that our experiment does not exhibit the strong warming observed over Greenland in 2012. The observed 2012 Greenland melting was attributed to the following key factors: the North American heat wave, transitions in the Arctic Oscillation and transport of warm air and vapour via an atmospheric river . Forcing the model with only oceanic conditions can thus not create a similar atmospheric-induced warming.

In contrast to the results of this study, simulated 2–3 ‰ changes in central Greenland δ18Op for extremely warm climates with SST and sea ice conditions created from a coupled model experiment forced by large increases in CO2. The main differences between the simulations in this study and in the study by are related to the distribution and magnitude of sea ice and SST changes especially near northern Greenland.

In the study by sea ice and SST changes also occur in the region north of Greenland. Also the magnitude of Arctic SST anomalies are 8–10 ∘C whereas the simulations in this study have anomalies of 3–5 ∘C. These differences are compelling as our experiment “2012” with the largest prescribed SST anomalies and sea ice changes is also the only experiment that simulates a regional isotopic response. This indicates that the magnitude of SST changes might control not only the amount of local evaporation, but also the regional extent of the isotopic response. Hence, it is possible that the simulated changes of δ18Op by have a regional extent due to the same reasons as experiment “2012”.

Warming of the lower troposphere and associated weakening of the inversion layer might be important in controlling the extent of the isotopic response. As sea ice removal is connected to intense warming of the lower troposphere , it could be speculated that this warming is controlling the extent of the isotopic response. This would be possible as a weaker inversion layer allows atmospheric convection, and have shown that this can occur at high-latitudes in sea-ice-free regions in winter. Further investigation of the mechanism causing this change requires further idealised experiments following a similar design to , so that a systematic investigation of the atmospheric processes influencing the isotopic composition of moisture is possible.