The Greenland Ice Sheet has undergone persistent mass loss since the turn of the century (Hanna et al., 2014, 2024; Khan et al., 2015; Kjeldsen et al., 2015; Mouginot et al., 2019; Otosaka et al., 2023; The IMBIE team et al., 2020; Velicogna et al., 2020). While highly variable, mass loss from the Greenland Ice Sheet has consistently raised global mean sea level by over 0.5 mm yr⁻¹ in recent decades, contributing to sea-level rise at approximately twice the rate of the Antarctic Ice Sheet (Cazenave et al., 2018; Horwath et al., 2022; Smith et al., 2020; The GlaMBIE Team et al., 2025; The IMBIE team et al., 2018, 2020). Estimates place the total contribution of Greenland at over 10 mm of sea level rise since the 1990s (Mouginot et al., 2019; The IMBIE team et al., 2020). Moreover, most of the impact of recent climate change on total mass balance has yet to be realized, as the timescale at which ice sheet dynamics adjust to a climate perturbation is an order of magnitude or greater than that of the surface mass balance (SMB) response (Box et al., 2022). Recent work conservatively estimates another ∼274 mm of committed sea level rise before the ice sheet achieves balance with the current climate state – i.e., even without considering the additional impact of any future warming scenario (Box et al., 2022).

Over Greenland, there has been both an increase in solid-ice discharge and a decline in SMB over the past few decades (van den Broeke et al., 2009a; Mankoff et al., 2019; The IMBIE team et al., 2020); however, increased runoff has caused SMB to decline at a rate twice that of the observed increase in dynamic ice loss (Box et al., 2022; Fettweis et al., 2017, 2020; Mote, 2007; Noël et al., 2017). Consequently, SMB reductions have surpassed discharge as the largest source of mass loss (Mouginot et al., 2019) and, according to dynamically downscaled CMIP5 and CMIP6 global climate model (GCM) output, SMB losses are projected to exceed mass accumulation on their own by the year 2100 unless the most ambitious mitigation efforts are implemented (Noël et al., 2021). This timeline, however, contains considerable uncertainty as differences in prescribed bare-ice albedo and firn model parameterizations have been shown to yield a twofold spread in simulated runoff between the three leading polar Regional Climate Models (RCMs) under a common future radiative forcing, with RACMO – the RCM used to derive this estimate – underestimating future runoff relative to HIRHAM and MAR (Glaude et al., 2024).

This change in ice sheet surface conditions has been associated with a recent shift in summer atmospheric circulation over the North Atlantic. The decline in SMB has coincided with a more persistently negative North Atlantic Oscillation (NAO) and an increase in atmospheric blocking episodes over Greenland (Bevis et al., 2019; Fettweis et al., 2013; Hanna et al., 2015, 2016, 2018b, 2022; Hofer et al., 2017). Indeed, previous studies have shown a statistically significant increase in summer Greenland blocking between the 1990's and early 2010's using a variety of blocking detection methods (Davini and D'Andrea, 2020; Hanna et al., 2022; Lee and Polvani, 2026; Luu et al., 2024; Woollings et al., 2018). While multi-year mean summer Greenland blocking frequency has declined since its peak around 2010, interannual variability has remained anomalously high with extreme blocking episodes again facilitating record-setting melt of the ice sheet in 2019 and 2023 (Bonsoms et al., 2026; Cullather et al., 2020; Lee and Polvani, 2026; Luu et al., 2024; Tedesco and Fettweis, 2020). More frequent blocking has played a key role in encouraging melt via multiple contrasting mechanisms. The increase in surface melt has been ascribed to the suppression of cloud cover by large-scale subsidence within the blocking ridge (Hofer et al., 2017). This reduction in cloud cover has allowed for anomalously high downward shortwave radiation over the southern ice sheet which, owing to its lower surface albedo, is more sensitive to changes in shortwave radiation than other regions of the ice sheet (Hofer et al., 2017; Wang et al., 2019). Other studies, however, have demonstrated the importance of cloud longwave radiative effects, particularly in regions where albedo is high, such as the northern ice sheet and over the high-elevation accumulation zone (Gallagher et al., 2018; Lenaerts et al., 2019; Noël et al., 2019; Orsi et al., 2017; Wang et al., 2019). Southerly moisture transport upstream of a blocking anticyclone in July 2012 supported the formation of low-level cloud cover that produced melt over the highest elevations of the ice sheet for the first time in over a century (Bennartz et al., 2013; Clausen et al., 1988; Fausto et al., 2016b, a; Mattingly et al., 2018; Meese et al., 1994; Neff et al., 2014; Nghiem et al., 2012). Additionally, the high-amplitude Omega blocking patterns that have increased most in recent summers deliver moisture farther poleward, generating above-normal downward longwave radiation over northern Greenland (Preece et al., 2022) – conditions which have caused pronounced growth of the ablation zone and spurred a disproportionate increase in runoff from the northern drainages of the ice sheet (Noël et al., 2019).

A natural question is whether this shift in summer circulation may be a symptom of climate change. At the hemispheric scale, there are several theoretical frameworks that postulate a link between changes in the meridional temperature gradient under Arctic Amplification and more frequent persistent weather extremes (Cohen et al., 2014; Coumou et al., 2018; Francis and Vavrus, 2012), and there is mounting evidence of such a link during summer (Cattiaux et al., 2016; Coumou et al., 2015; Di Capua and Coumou, 2016; Kornhuber and Tamarin-Brodsky, 2021; Vavrus et al., 2017). However, not only have GCMs failed to capture the positive trend in Greenland blocking, they consistently predict a decline in blocking frequency in the region (Delhasse et al., 2021; Hanna et al., 2018a) – a critical source of uncertainty regarding a causal link between anthropogenic climate change and the observed shift in summer circulation over Greenland. Conversely, the change in the background thermodynamic environment, and its resulting impact on SMB represents a more robust signal of climate change than the potential dynamical response outlined above. Multiple well-documented radiative feedbacks have helped warm the Arctic at four times the global average rate (Pithan and Mauritsen, 2014; Rantanen et al., 2022; Serreze and Barry, 2011). This constitutes a likely contributor to the nonlinear decline in SMB, as surface melt would be expected to increase in frequency and magnitude in a warmer, more humid atmosphere.

The thermodynamic environment over Greenland is surely influenced by changes occurring more broadly throughout the Arctic. Indeed, Box et al. (2013) linked increased precipitation over Greenland to higher Northern Hemisphere surface air temperature and showed that this statistical relationship was more robust than when considering local near-surface temperatures over Greenland or North Atlantic SST – a result that emphasizes the importance of remotely-sourced heat and moisture to the SMB. However, local sea-surface conditions (SSCs) may also play an important role. Specifically, sea ice reductions over adjacent waters could further contribute to elevated temperatures over Greenland through the water vapor feedback, wherein a warmer atmosphere together with an ice-free ocean increases atmospheric water vapor, which then enhances longwave radiative forcing at the surface (Pithan and Mauritsen, 2014; Trenberth, 2011). One of the more intuitive ways that sea ice loss could impact the ice sheet is through advection of warm, moisture-enriched air from the neighboring seas. Studies have revealed a relationship between changes in sea-ice concentration near Greenland and ice sheet SMB (Pedersen and Christensen, 2019; Rennermalm et al., 2009); however, these studies fail to separate direct marine influence from any indirect effects via alteration of the large-scale circulation by oceanic thermal forcing (Ballinger et al., 2019; Liu et al., 2016) or relationships that might arise as a byproduct of mutual forcing of local sea ice concentration (SIC) and ice sheet melt by the large-scale synoptic setting (Ballinger et al., 2018; Stroeve et al., 2017).

Several modeling efforts have demonstrated minimal contribution from local SSCs during summers of pronounced melt due to the persistent katabatic outflow over the ice sheet that acts as a barrier to onshore advection (Hanna et al., 2009, 2014; Noël et al., 2014); however, observational evidence suggests that local SSCs may play an important role earlier in spring. While melt events during summer and fall are primarily a product of large-scale atmospheric conditions (Ballinger et al., 2019; Hermann et al., 2020; Noël et al., 2014), recent work demonstrated elevated atmospheric moisture and enhanced downward longwave radiation over the ice sheet approximately one week following sea ice retreat in the Baffin Bay and Davis Strait in years of early melt, suggesting that local sea ice anomalies precondition the ice sheet for early melt onset (Stroeve et al., 2017).

Disentangling the relative contributions of atmospheric dynamics versus thermodynamics is an intractable problem using observations alone. Previous studies have utilized RCMs to examine the sensitivity of the SMB to perturbations in SSCs (Hanna et al., 2009, 2014; Noël et al., 2014) or atmospheric thermodynamic fields (Delhasse et al., 2018); however, none of these efforts examined the combined influence of atmospheric and SSCs. Furthermore, these studies either applied arbitrary perturbations to the targeted boundary fields or examined a single melt season and, therefore, did not aim to measure the existent contribution of observed changes in these fields to mass loss.

Here, we provide an estimate of the relative contributions of dynamical versus thermodynamic change to recent Greenland ice sheet surface mass loss using the pseudo-global warming (PGW) method of dynamical downscaling. The PGW method uses adjusted reanalysis data for the initial and lateral boundary conditions of an RCM (Kawase et al., 2008; Kimura and Kitoh, 2007; Schär et al., 1996). To obtain the adjusted boundary conditions, this method applies a climate change perturbation signal that is estimated from GCM output by assuming a linear change in the boundary fields between the control period (i.e., the period of observed reanalysis data) and some alternative period of interest with a contrasting thermodynamic background state (Kawase et al., 2008; Rasmussen et al., 2011; Schär et al., 1996). Thus, the PGW technique effectively isolates the impact of the long-term thermodynamic component of climate change by assuming that the timing and structure of synoptic disturbances along the RCM's boundaries will be the same in the alternative period as during the control (Lackmann, 2015).

While the PGW method is typically utilized to simulate future conditions, it can also be used to investigate how recent periods of climate or individual weather events would have behaved under past conditions. For example, Lackmann (2015) estimated the thermodynamic contribution of recent climate change to the evolution of Hurricane Sandy by comparing a control run to a PGW simulation using boundary conditions that were adjusted to reflect the climate of the late 19th Century. Likewise, Kawase et al. (2008) used a similar approach in a climate change attribution study of the Mei-yu rain band in southern China. Here we use the PGW method to quantify what the magnitude of surface melt would have been if the recent dynamical forcing of the ice sheet had occurred in a preindustrial thermodynamic background state by answering the following questions: (1) How much of the recent SMB decline can be attributed to the combined influence of increasing background temperatures and contributions from local SSCs? (2) What portion of the thermodynamic influence is due to adjacent SSC change alone and do SSCs have a discernible impact on the timing or duration of the melt season?

A model schematic outlining our approach is presented as Fig. 1. Atmospheric conditions and the SMB and surface energy balance (SEB) response were modeled using the RCM, Modèle Atmosphérique Régional (MAR) (Fettweis et al., 2005, 2020; Lefebre et al., 2005). MAR includes a surface-atmosphere energy and mass transfer scheme with a one-dimensional snowpack model that represents snow grain metamorphism and its impact on albedo, and accounts for the percolation and refreeze of meltwater within the snowpack (Amory et al., 2021; Brun et al., 1989, 1992). For a more detailed description of MAR, see Amory et al. (2021). Here, we employed MAR version 3.12 initialized and forced at its lateral boundaries with 6-hourly ERA5 reanalysis data and integrated over a 120×180, 20 km grid with 24 vertical atmospheric levels.

As a control, we forced MAR with ERA5 data spanning 2000 to 2019 to represent historical conditions during the recent period of anomalous Greenland blocking (Fig. 1, gray components). For the control, all boundary fields including the SSCs of sea-surface temperature (SST) and SIC were sourced from ERA5. To simulate the preindustrial thermodynamic state (Fig. 1, blue components), we adjusted the boundary conditions of air temperature and specific humidity using perturbations obtained from the NCAR Community Earth System Model-Large Ensemble (CESM-LE) project (Kay et al., 2015), while zonal and meridional winds at the model boundaries were left unaltered to minimize differences in the large-scale atmospheric circulation between the experiment and the control. For the pre-industrial simulations, we adjusted ERA5 air temperature and specific humidity using a climate change perturbation derived from the 40 ensemble members of the NCAR CESM-LE project as follows: Δx=x‾p-x‾c Where Δx is the climate change perturbation for variable x, x‾p is the ensemble-averaged, long-term monthly mean of variable x for a preindustrial reference period of 1880–1899, and x‾c is the ensemble-averaged, long-term monthly mean of variable x for a control period of 2000–2019. We then linearly interpolated the monthly climate change perturbations derived from CESM-LE temporally to a 6-hourly timestep, vertically to ECMWF's L137 hybrid sigma-pressure levels, and horizontally to the ERA5 grid before adding them to the corresponding ERA5 boundary fields that were used to force MAR. A comparison of 500 hPa geopotential height over the study area shows strong temporal covariability between simulations, indicating that the large-scale circulation of the control was well preserved in our application of the PGW method (Fig. S1 in the Supplement).

While GCMs aim to capture the periodicity of internal climate variability, they may not accurately resolve the magnitude of said variability and the precise timing of a particular mode of variability differs between individual ensemble members. Considering a single GCM simulation alone would risk enhancing or suppressing the magnitude of the climate perturbation signal if the control and alternative periods are characterized by opposing phases of a relevant mode of internal variability. For example, an anomalously negative summer NAO since the turn of the century has favored the advection of warm, moist air over Greenland (Fettweis et al., 2013; Hanna et al., 2013; Henderson et al., 2021; Mattingly et al., 2018; Mote, 1998; Tedesco et al., 2016). Since our intent is to isolate the contribution of background thermodynamic change to mass loss, deriving a perturbation signal from observations would overestimate the baseline change in temperature and humidity around Greenland because it would also include the dynamical contribution of an anomalously negative NAO during our control period. Taking an ensemble mean of the CESM-LE effectively removes the noise of internal climate variability by averaging across the differing phases resolved by each ensemble member for a given date, thereby providing a more appropriate estimate of the change in the mean climate state.

Figure 2 shows a subset of the monthly surface air temperature and specific humidity perturbation fields that were applied at the lateral boundaries of MAR. Seasonally, CESM-LE simulates the greatest temperature difference in fall and winter (Fig. 2, top row), consistent with what should be expected under Arctic amplification, which is largely driven by sea ice loss (Screen and Simmonds, 2010). Differences in near-surface atmospheric moisture are largely reflective of the Clausius-Clapeyron relation, with drier conditions mirroring locations of cooler temperatures in the preindustrial climate (Fig. 2, bottom row).

The use of a GCM-derived perturbation signal presents issues when dealing with sea ice. The change in SIC is greatest along the sharply defined sea ice front and the GCM's representation may not geographically align with observations. Figure 3 presents one such comparison for 15 June 2018, which occurs during a month of exceptionally low SIC in the Greenland Sea. There is a considerable gap between the observed sea ice front and area of greatest SIC change according to CESM-LE, such that the application of this perturbation signal would result in a locally high SIC stretching from Iceland to Svalbard that is separated from the main body of sea ice. To avoid this unrealistic distribution, and to ensure consistency between SIC and SST, we prescribed both SIC and SST in our experimental simulations using 1880–1899 long-term monthly means calculated from the merged Hadley-OI observational dataset (Shea et al., 2020) and interpolated to a 6-hourly timestep.

Contrary to global SST trends, there are extensive areas around Greenland where SST during the preindustrial period was higher than during the current period (Fig. S2). This is most apparent during winter and spring when higher preindustrial SST is observed throughout the northern subpolar gyre to the southeast of Greenland and extending from the southern Greenland coast along the sea ice edge to Svalbard. The spatial and seasonal pattern of lower SST since the preindustrial period matches the fingerprint of the so-called North Atlantic warming hole – an observed decrease in subpolar North Atlantic SST that has been attributed to a weakening of the Atlantic meridional overturning circulation and associated poleward oceanic heat transport as a consequence of climate change (Caesar et al., 2018). In summer, SST throughout much of the region was lower during the preindustrial period (Fig. S2).

After interpolating the Hadley-OI fields to a 6-hourly timestep, we applied the following adjustments based on the work of Hurrell et al. (2008) to further ensure consistency between SST and SIC:

If an interpolated grid cell had a SIC >90 %, we set the SST of that cell to the sea ice freezing point of -1.8 °C.

The design of PGWSSC also allows us to test how much of the mechanism identified by Stroeve et al. (2017) – that low spring SIC in the seas surrounding Greenland preconditions the ice sheet for melt in early melt onset years – is due to direct thermodynamic forcing alone. Following Stroeve et al. (2017), we define melt onset as the first instance of five or more consecutive days of melt. The date of freeze onset is then defined as the first day following the last instance of five or more consecutive days of melt. We calculated all measures of the melt season at each MAR grid pixel, then tested for significant differences between the PGW simulations and the control using a paired Wilcoxon signed-rank test (Wilcoxon, 1945) with a predetermined significance level of α=0.05 (i.e., 95 % confidence level).

Much of the work examining the recent increase in Greenland Ice Sheet meltwater runoff has rightfully focused on the role of atmospheric dynamics (Bevis et al., 2019; Fettweis et al., 2013; Hanna et al., 2015, 2016, 2018b, 2022; Hofer et al., 2017). While some have presented evidence of a relationship between global climate change and the shift in summer atmospheric circulation that has promoted melt of the ice sheet (Liu et al., 2016; Preece et al., 2023b; Screen, 2013), a conclusive link remains a subject of investigation. In contrast, the accelerated rate of warming in the Arctic represents a robust climate change signal that has undoubtedly contributed to recent SMB trends (Boers and Rypdal, 2021; Hanna et al., 2008). This work represents, to our knowledge, the first systematic attempt to quantify the contribution of the local change in background thermodynamic conditions to recent surface mass loss.

Our results indicate that had the large-scale atmospheric circulation that was observed from 2000–2019 occurred under preindustrial thermodynamic background conditions, the cumulative SMB anomaly would have been reduced by over 62 % (Fig. 4). The mechanisms by which local thermodynamic background conditions contribute to SMB change appear to be dominated by longwave radiative effects stemming from the water vapor feedback. The amplified rate of warming in the Arctic has augmented surface runoff by promoting an increase in atmospheric moisture and associated downward longwave radiation (Figs. 9a, e and 7b, f), which disproportionately increases daily minimum temperatures (Fig. 9c, g). These results are consistent with Orsi et al. (2017), which identifies a positive trend in 1982–2011 surface air temperature reconstructed from borehole temperature measurements at the North Greenland Eemian Ice Drilling site in northwest Greenland. They point to an increase in downward longwave flux and associated feedbacks as the primary contributor to the warming trend. Likewise, Noël et al. (2019) show that an increase in downward longwave radiation has caused a disproportionate increase in surface runoff from the northern drainages by promoting melt and expanding the ablation zone in this region of high albedo, and by increasing daily minimum temperatures, which reduces meltwater refreeze within the firn layer. While the authors point to the advection of moisture-rich air to the northern ice sheet by anomalously anticyclonic summer circulation over Greenland, the results of this analysis suggest that the increase in background temperature constitutes an important contribution to this mechanism on its own.

The ∼62 % reduction, which is conditional on the ERA5 2000–2019 circulation occurring under preindustrial thermodynamic conditions, does not imply that atmospheric circulation is only responsible for 38 % of the observed impact on SMB, as the individual contributions of atmospheric dynamics and thermodynamics should sum to the total change in SMB relative to what it would have been if neither an increased frequency of Greenland blocking nor anthropogenic warming had occurred. Given that the ice sheet maintained a positive cumulative SMB anomaly through 2009 under the preindustrial thermodynamic background conditions imposed in PGWSSCT,Q, it is possible that the cumulative anomaly may have remained positive through the end of the study period if not for the increased frequency of anomalous anticyclones. Thus, relative to this hypothetical preindustrial climate with more typical atmospheric circulation, the total change in SMB would be greater than the magnitude of the negative anomalies presented in Fig. 4 and the contribution of atmospheric circulation to this total change would exceed 38 %. Indeed, using an earth system model to nudge the wind field toward observed conditions while maintaining constant external forcing, Topál et al. (2022) showed that changes in atmospheric circulation explained 56 % of the increase in surface air temperature over Greenland from 1990 to 2012. Using historical data and a circulation analogue technique, Fettweis et al. (2013) found that the shift in summer circulation explained ∼70 % of the 1993–2012 warming at 700 hPa over Greenland. Focusing on mass loss, Delhasse et al. (2018) compared output from MAR forced by perturbed reanalysis data from the recent period of increased Greenland blocking against simulations forced by output from GCMs which have collectively failed to capture this change in circulation (Delhasse et al., 2021; Hanna et al., 2018a). Their results suggest that if the recent anomalous circulation persists into the future, the ice sheet will undergo more than twice the surface mass loss that is currently projected by GCMs. Thus, understanding this circulation change and why it is not represented in climate models must be a top priority for accurate projections of Greenland Ice Sheet mass loss.

The contribution of local thermodynamic background conditions to total surface mass loss during the exceptional melt years of 2012 and 2019 was less than half that which was observed for the entire 2000–2019 study period (cf. Figs. 4 and 11a, e), suggesting that the relative thermodynamic contribution is reduced during strong large-scale atmospheric forcing. This is also evident over synoptic timescales, where the difference in surface air temperature between PGWSSCT,Q and the control is minimized on days of exceptional surface runoff; rather, the greatest differences in surface temperature emerge during periods encompassing temporal minima in air temperature (Fig. 11). This likely reflects the increased contribution of remotely sourced heat and moisture during periods of strong large-scale forcing, which would reduce the relative importance of the longwave radiative effects that typify the response to changes in the thermodynamic background state. In other words, recent local thermodynamic change around Greenland appears to have promoted surface runoff by raising the floor of the temperature distribution more so than by exacerbating warm extremes. This signal may be due in part to biases inherent to the PGW approach. The application of a monthly mean climate perturbation may underrepresent the true change in air temperature and water vapor concentration during extreme events such as the bocking episodes and attendant atmospheric rivers that have promoted melt of the ice sheet. While the thermodynamic fields are free to adjust in accordance with any relevant nonlinear processes within the MAR integration domain, it is likely that any biases at the model boundaries would be conveyed to the ice sheet to some extent. Despite these shortcomings, the PGW method of downscaling is recognized as an effective means of isolating the thermodynamic component of climate change (Gutmann et al., 2018; Lackmann, 2015; Mallard et al., 2013; Rasmussen et al., 2020) – an approach that has been advocated particularly in cases of extreme events for which the governing dynamics are not well represented in the models (Lloyd and Oreskes, 2018; Trenberth et al., 2015), which is true of both atmospheric rivers and atmospheric blocking (Delhasse et al., 2021; Hanna et al., 2018a; Wang et al., 2023; Woollings et al., 2018).

The same large-scale atmospheric conditions that typify our control period and have encouraged mass loss have also fostered below-normal sea ice in the region (Ballinger et al., 2018; Ogi and Wallace, 2007; Stroeve et al., 2017). Thus, the 1880–1899 sea ice climatology that we prescribe here may often exceed the SIC that would have occurred if the recently observed atmospheric circulation had occurred under preindustrial conditions. Recognizing this potential bias, our results likely represent an aggressive estimate of the contribution of SSCs to recent surface mass loss. Even so, this analysis reveals minimal direct thermodynamic contribution by local SSCs, supporting previous work showing low SMB sensitivity to adjacent SSCs due to the barrier to onshore advection from the marine layer presented by consistent katabatic outflow over the ice sheet (Hanna et al., 2009, 2014; Noël et al., 2014).

Regarding the hypothesis of Stroeve et al. (2017) that declining SIC may promote earlier melt onset, thereby preconditioning the ice sheet for greater meltwater production later in the season, our results suggest limited direct thermodynamic impact of local SSCs on recent surface melt and no discernable impact on melt timing (Figs. 4 and 10). There are several plausible reasons for this apparent disparity. First, our interpretation assumes that SSCs and ocean-atmosphere coupling are accurately represented by the model. Gridded climate datasets, with their relatively coarse spatial resolution, are less accurate in areas that rely more heavily on spatial interpolation, such as along coastlines and near the sea ice front where in-situ observations are less frequent (Hanna et al., 2006; Hurrell et al., 2008; Yang et al., 2021). This could degrade model representation of ocean-atmosphere coupling; however, the disparities among various gridded SST and SIC datasets are generally much smaller than the long-term trends (Yang et al., 2021) and, therefore, should exert minimal influence on the signal that we seek to quantify. Second, our experimental design did not examine the isolated contribution of changes in the atmospheric fields alone, and it is possible that nonlinear interactions between changes in SSCs and atmospheric thermodynamic fields caused the combined influence in PGWSSCT,Q to be greater than the sum of their individual contributions. Lastly, our analysis did not examine any indirect effects via alteration of the large-scale circulation by changes in SSCs. Both model- and observation-based studies have yielded evidence of a link between declining sea ice in Baffin Bay and the observed increase in summer Greenland blocking (Liu et al., 2016; Screen, 2013; Sellevold et al., 2022; Wu et al., 2013). Indeed, Stroeve et al. (2017) found that the statistical relationship between ice sheet melt and Baffin Bay SIC weakened considerably and, consistent with the PGWSSC response in Fig. 5c, became more confined to the periphery of the western ice sheet after correcting for the influence of Greenland blocking. Considered in conjunction with our results, this suggests that this indirect pathway of influence could help explain the statistical relationship between SSCs and early melt onset.

Because MAR assumes a fixed ice sheet geometry, the results presented herein strictly describe the thermodynamic influence on the SMB of the ice sheet; however, surface runoff and solid ice dynamics are not independent. Strong pulses of meltwater can cause rapid drainage through moulins that overwhelms the subglacial drainage network (Chu, 2014; Schoof, 2010), causing a surge in ice velocity that increases glacial discharge and accelerates ice sheet thinning (Andersen et al., 2011; Chu, 2014; Schoof, 2010). Thus, it is likely that the thermodynamic influence on surface melt documented here has indirectly contributed further to sea-level rise via its impact on ice sheet dynamics. Regardless, these results demonstrate that while the shift in summer atmospheric circulation over Greenland has been key to the acceleration of runoff from the ice sheet, the change in the background thermodynamic state under Arctic amplification has markedly enhanced the observed surface mass loss beyond that which would have occurred if not for anthropogenic climate change.