Recent North Greenland temperature warming and accumulation

In a warming climate concise knowledge of the mass balance of the Greenland ice sheet is of utter importance. Speculations that current warming will increase the snow accumulation and mitigate mass balance losses are unconstrained as accumulation data across large regions of the northern ice sheet are scarce. We reconstructed the accumulation from six north 20 Greenland shallow firn cores (~10m) and eight snow cores (~2m) to constrain recent accumulation patterns in northern Greenland and calculated recent warming in the same area using borehole temperature measurements. We find an increase in temperatures in the north Greenland interior of 0.9 to 2.5 °C (method and site dependent) per decade over the past two decades in line with an Arctic amplified anthropogenic warming. We compare annual reconstructed accumulation from the firn cores (1966-2015) to radar estimates and to annual re-analysis data (1980-2016) of precipitation 25 subtracted evaporation from the regional climate model HIRHAM5, operated by the Danish Meteorological Institute. The spatial variability resembles that observed in earlier estimates with a clear increase west of the topographic divide and a low accumulation area across the north-eastern ice sheet. Our accumulation results are comparable to earlier firn core estimates, despite being larger in the east. We only find a positive significant trend in the accumulation for the period 2000-2010 to the northwest. 30 In the vicinity of the EGRIP deep ice core drilling site, we find variable accumulation patterns for two 15 km apart firn cores likely owing to local topographic effects as a result of the North East Greenland Ice Stream dynamics.


Introduction
The Arctic today is warming and automated weather stations from the Greenland Climate Network (GC-net) (Steffen and Box, 2018) and at Summit Station in central Greenland McGrath et al. (2013) have identified Greenland warming trends even in the 35 central part of the ice sheet of 0.9 °C/decade. From the Clausius-Clapeyron relationship one would expect an accumulation change following increasing temperatures, as was demonstrated by Box (2013) who found an increase of 6.8% K −1 .
Greenland's contribution to sea-level rise is currently estimated at ~10 cm by 2100 (Lewis et al., 2017;Muntjewerf et al., 2020). The uncertainty of this number is in part due to the surface mass balance (SMB, i.e. precipitation minus evaporation, sublimation, meltwater runoff and/or erosion). Precise knowledge of the accumulation is crucial to constrain the ice sheet 40 SMB. For the ice sheet interior the most precise data of accumulation are from ice cores, however, they are sparse and most do not cover the recent period 2010-present (Box et al., 2009;Montgomery et al., 2018), or the northern region of the ice sheet (Koenig et al., 2016;Montgomery et al., 2018). Satellite radar altimetry data sets over the Greenland ice sheet (GrIS) begin as https://doi.org /10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License. late as 1991. These data provide invaluable information on accumulation and temperature used for monitoring the current changes of the ice sheets due to the dense spatial and temporal coverage they offer (Koenig et al., 2016;Sandberg Sørensen et al., 2018;Simonsen and Sørensen, 2017). However uncertainties are large as only limited ground truth data is available in the form of snow and firn cores covering the satellite era Box, 2013). Given the large area it can cover, ground penetrating radar (GPR) offers a potential solution. However, it also relies on a reference point, normally in the form of an 5 accumulation layer of known depth and age, meaning that GPR data is seldom available in annual resolution. As one of the few methods available for constraining both satellite and GPR data, the collection of firn cores is therefore crucial .
With regard to the Greenland ice sheet's accumulation, studies including central Greenland are in disagreement, with results ranging from strongly decreasing to slightly increasing accumulation in recent times dependent on the exact region (Lewis et al., 2017). Mosley-Thompson et al. (2001) studied the period up to 1998 and mainly south of 72 o N and showed that 10 accumulation has indeed changed. They also found evidence for decadal timescale variability, speculated to be caused by the North Atlantic Oscillation (NAO) as opposite trends in northwest and central Greenland were observed. Burgess et al. (2010) expanded upon the study with additional ground truth data and found an interannual variability of 18% and an increase in accumulation of 7 mm w.eq. per decade. Miège et al. (2013) found no significant temporal trend in south-east Greenland using three 50 m firn cores and radar. Hawley et al. (2014) used snow and firn cores and ground penetrating radar to investigate the 15 trend in accumulation in northwest and central Greenland and, when compared to older snow pits in the area, found positive trends of 2.2-3.1 cm w.eq., equivalent a 10% change in the dry snow region over 52 yrs. Using a combination of modelling (RACMO2) and firn cores, Box (2013) found an accelerated increase in accumulation in most of Greenland between 1840 and 1996 compared to the period 1600 onwards, with the exception of localised decreases (5%) in the north-east. Buchardt et al., 2012 compared ice core accumulation with water isotopes (a proxy for temperature) and found no conclusive trends for 20 accumulation, but did demonstrate an increase in δ 18 O for the most recent decade, speculated to correspond with increasing temperature. Using a combination of radar, shallow cores and climate modelling, Lewis et al., 2017 observed a decrease in the accumulation rates in the western Greenland accumulation zone of 9 mm.w.eq. per year, while further analysis by Lewis et al. (2019) revealed decreasing accumulation of 2.5% annually for the period 1996-2016. Recently, the Surface Mass Balance and snow on Sea Ice working group presented a compilation of accumulation data in Greenland, known as the SUMup dataset. 25 (Montgomery et al., 2018). The SUMup compiled accumulation dataset includes many of the above mentioned datasets and covers 36 sites in Greenland. However, only 10% of the snow and firn cores in the dataset were retrieved after 1999 (excluding Summit station data) and just 5% of them are from north of 76N. Thus, there is a need for enhanced ground truth measurements to constrain the effect of recent temperature changes on accumulation, especially in northern Greenland. 30 Ice, firn and snow cores from the Greenland ice sheet offer a unique opportunity to constrain the recent trend in Greenland accumulation. Ice cores analysed in sufficient resolution e.g. by means of continuous flow analysis (CFA, (Bigler et al., 2011)) provide information about ancient atmospheric aerosol concentrations, most of which have annual peaks. These cyclic peaks can be used to constrain annual layers and, if the density is known, to reconstruct historic snow accumulation rates (Gfeller et al., 2014). 35 The principle of CFA includes the continuous melting of ice cores and the immediate analysis of separate chemical proxies by individual detection units. This happens within a sealed system, which ensures the lowest possible risk of contamination and sample loss (Röthlisberger et al., 2000). Such high resolution data allow for identification of annual layers and even sub annual events (Svensson et al., 2008;Winstrup et al., 2012). Water isotopes, a proxy for temperature, is commonly used for such annual layer identification, however in low accumulation areas diffusion can smooth the water isotope signal to an extent that 40 annual layers are no longer reliable. In such areas one can use other proxies with annual cycles. For example, in shallow firn cores hydrogen peroxide, which is naturally produced in the atmosphere by photochemical derived self-reaction of hydroperoxyl radicals (HO2), can be used to identify seasonality as the production of H2O2 mainly takes place during months https://doi.org/10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License. of intense insolation. Therefore, firn core records show maximum concentrations in the summer and minimum concentrations during the winter months when photochemical processes are absent at polar latitudes. This strong seasonal pattern creates an excellent proxy that is commonly used for dating polar ice cores (Sigg and Neftel, 1988) and thus can be used to reconstruct accumulation (Frey et al., 2006).

5
In this study accumulation have been reconstructed at 14 sites in northern Greenland. To do so, six shallow firn cores down to 14 m were collected during a traverse (N2E) in 2015 from the deep ice core site NEEM (77º25.219',51º09.588' W) to the deep ice core site EastGRIP (75º37.501' N,23º58.809' W), alongside eight snow cores in 2017 down to 2 m traversing from Kangerlussuaq in the west to the EastGRIP site. Snow and firn cores were analysed at the Niels Bohr Institute in Copenhagen for chemical impurities. The impurities, especially the H2O2, were used to constrain annual layers. Densities in 55 cm resolution 10 were also collected in the field for the 6 shallow cores. With the 6 new cores and snow cores we thus constrain the northern Greenland accumulation over the recent decades (up until 2016) and provide much needed ground truth information.
Additionally, to evaluate the sites temperature trend in the recent decades, we obtained the 10 m depth borehole temperature in each location, taken to represent the annual mean temperature, and compared it to previous measurements from NGT traverse in 1993traverse in -1995traverse in (Fischer et al., 1998Weissbach et al., 2016). 15 Both reconstructed accumulation and temperatures are compared to results from the Danish Meteorological Institute regional climate model HIRHAM5, forced at the lateral boundaries by re-analysis data (DMI, (Christensen et al., 2006)). The reconstructed accumulation is further compared to earlier available estimates from firn and ice cores (Montgomery et al., 2018;Schaller et al., 2016), recent radar estimates (Karlsson et al., 2020;Lewis et al., 2017) and precipitation-evaporation (P-E) reanalysis data from the HIRHAM 5 model. 20

Firn core drilling during the N2E traverse in 2015
The six firn cores were extracted using the American IDDO (U.S. Ice Drilling and Design Operations) hand auger (76 mm diameter) from the surface to a depth between 9.1 m and 14.0 m (Table 1).
In the field the cores were split into 0.55 m long segments,placed in a so-called "Danish bag" and weighed. The cores were 25 then stored in cooler boxes and transported on sledges to the EastGRIP site, flown to Kangerlussuaq and then shipped further to Copenhagen, Denmark for analysis.
The weight of each bag (2 g precision) and the length of each bag (0.5 cm precision) was noted. Thus the density of each 55 m section can be determined with a precision <4% assuming the cores are close to circular with a diameter of 75-76 mm. See supplementary Figure S1. 30 With the cores extracted, the borehole temperature at 10 m depth was determined by a single thermistor connected to a multimeter using the 4-wire method (+-0.1 ºC) (Clow, 1992;Gundestrup et al., 1994;Orsi et al., 2012). The temperature at 10 m depth is a good approximation of the in situ mean annual surface temperature in the absence of melting (Dahl-Jensen et al., 1998). Table 2 provides the temperature at 10 m depth for each site representing the annual mean temperature and one can calculate the annual cycle of temperature at such depths is dampened to 5% that observed at the surface (Johnsen et al., 35 1992).

Snow tube collection during the2017 Inuit Windsled Project traverse windsled ice river
During the 2017 Inuit Windsled Project traverse Greenland expedition from Kangerlussuaq to the EastGRIP deep ice core drilling site the upper two m of the snow pack was sampled at eight sites along the route (Table 5, Figure 1). The snow pack was sampled using the "liner technique" as described in Schaller et al. (2016), where 1 m-long hollow carbon tubes are pushed 40 https://doi.org/10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License. downward into the snow before excavation and retrieval. The snow from 1-2 m depth was sampled in a similar matter however only for sites north of 73 o N. The cores remained in the carbon tubes after sampling and were stored in cooler boxes carried by the wind driven sledge during transport to EGRIP. From EGRIP the samples followed same path as those collected in 2015.

Continuous flow analysis
Annual layers in the snow and firn cores were identified using the annual peak concentration of impurities, mainly H2O2. The 5 Copenhagen CFA system was used to analyse the impurity content in the snow and firn cores. For the 6 firn cores and 8 snow tubes the melt water conductivity, insoluble dust, H2O2, NH4 + , Ca 2+ and acid were determined (Bigler et al., 2011;Kjaer et al., 2016;Simonsen et al., 2018). Firn cores were cut into 3.6 x 3.6 cm squares prior to melting, whereas the snow cores were melted on a full size melt head. In the CFA system the inner section of the core is separated from the outer during continuous melting after which the inner decontaminated line is split into several lines. For each line an ion is determined after the addition 10 of specific reagent and in some cases a buffer by means of fluorescence (NH4 + , H2O2, Ca 2+ ). The soluble dust is measured by a laser instrument (ABAKUS LDS, Klotz, see also Simonsen et al., 2018) and the conductivity of the water is determined using an Amber Science 3082 conductivity metre as presented by Bigler et al. (2011). In addition to the system presented in Bigler et al. (2011) acid was also analysed (Kjaer et al., 2016). For every 4-5 metre of firn core, baselines of ultrapure milliQ water were determined and a set of standards were run to calibrate the measurements. Due to the density in the top 2.2 m of 15 each core calibration was carried out at 55 cm melting intervals. The melt speed was kept at 4 cm min -1 and with a maximum response time of 30s within the system, the final depth resolution of the ions measured are on the order of 2 cm (acidity, NH4 + , H2O2, Ca 2+ ), while for conductivity and dust a shorter response time for a depth resolution of 8 mm is achieved. The firn pore capillaries caused melt water to be pulled from the melt head into the firn creating some dispersion especially in the low density part of the dataset. The capillary effect was limited by placing a round circular coin between the melt head and the firn, 20 minimizing suction of air into the CFA system and percolation of melt water up into the firn. Even with percolation, the annual peaks in all proxies were well resolved.

Discrete water isotopes
For the firn cores 2015T-A1 (NEEM site) and 2015T-A5 (EastGRIP site) each bag was also subsampled in 2.5 cm discrete depth sections. Each sample was melted in separate air-tight metal containers and transferred to small vials for the purposes of 25 stable water isotope analysis (δ 18 O, δD). The measurements were performed using a commercial Cavity Ring-Down Spectroscopy (Picarro L2130) coupled to a high throughput-A0212 vaporiser (Gkinis et al., 2011). The analysis was conducted with a water concentration level in the range 18,000-23,000 ppm in order to minimize the measurement-induced humidity influence on the water isotopic composition. All samples are calibrated with respect to Vienna Standard Mean Ocean Water (VSMOW) and normalized to the VSMOW-SLAP (Standard Light Antarctic Precipitation) scale. The presented measurements 30 all have a precision better than 0.10 ‰ (δ 18 O) and 0.50‰ (δD), respectively.
The deuterium excess data (d_xs) were corrected for water-isotope specific diffusion which alters the d_xs signal unevenly in the upper firn column. This was achieved by separately back-diffusing the δ 18 O and δD records similar to the method outlined in Steen-Larsen et al. (2011). This study uses the diffusivity parameterization of Johnsen et al. (2000) where the magnitude of diffusion depends on isotopic species, surface temperature, density, accumulation rate and pressure at the drill site. 35

Core chronology and annual layer counting Techniques
A timescale for each firn and snow core was reconstructed by means of annual layer counting (Svensson et al., 2008;Winstrup et al., 2012). Annual cycles can be identified in the peroxide signal, which peaks in summer due to the photolytic reaction caused by sunlight; the ammonium signal, which peaks in late summer due to biogenic activity; the sodium signal, which peaks 40 https://doi.org/10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License. with winter storms; the dust signal, which peaks in spring; and the acid signal, which peaks a little later in spring due to anthropogenic activity and subsequent arctic haze (Gfeller et al., 2014;Kaufmann et al., 2008). In this study we predominantly relied on the seasonality of peroxide (Sigg and Neftel, 1988) and only used the seasonality in other compounds (mainly dust) in the cases where peroxide signals were poorly defined, such as the deeper part of the cores from low accumulation sites; 2015T-A6 , 2015T-A4 and 2015T-A5. The total age of each core and the uncertainty was defined as +/-0.5 year for each 5 uncertain year and can be found in Table 1

Reconstruction of accumulation
Density measurements were performed in the field for the N2E shallow firn cores on every 55 cm section, as described earlier.
The densification profile (supplementary Figure S1) was similar for all 6 sites and we use a linear interpolation to calculate 20 densification rates for each of the six cores.
The annual water equivalent accumulation (Ayr) was reconstructed using A=z*ρsample /ρwater, where A is the accumulation in m water equivalent per year (m. w.eq. a -1 ), z is the thickness of the annual layers in m and ρsample is their corresponding firn density in kg m -3 . The uncertainty associated with the firn density is on the order of 4% and the uncertainty on the peak summer position of the annual layer is on the order of ±1 cm for the N2E firn cores. The combined uncertainty on the annual 25 accumulation amount to less than ±1.1 cm w.eq.
As the windsled snow cores collected in 2017 were not weighed a model density profile was used to reconstruct accumulation.
In combination with the depths of the 2016 and 2015 summer peak of peroxide, a fixed density in the top 1 metre of 350 kg m -3 was used to reconstruct the accumulation form the snow cores. This value was chosen as Schaller et al. (2016) found densities between 300 and 400 kg m -3 in the top 1 m of snow by means of x-ray tomography performed on snow liners from 30 the 2015 N2E traverse.

Atmospheric modelling
The snowfall accumulation data obtained from the firn cores are compared with that simulated by the Danish Meteorological Institute HIRHAM5 regional climate model for a domain covering Greenland. HIRHAM5 is described in Christensen et al. (2006) and combines dynamics from the numerical weather prediction model HIRLAM7 (Eerola, K, 2006) and physics from 35 the global climate model ECHAM5 (Roeckner et al., 2003). The subsurface scheme in HIRHAM5 has been extended with an off line component that handles liquid water flow and retention on the Greenland ice sheet (Langen et al., 2017). On the lateral boundaries and over open ocean HIRHAM5 is forced with the European re-analysis dataset ERA-Interim from ECMWF (Dee et al., 2011). The HIRHAM5 experiment used in this study has a horizontal resolution at 0.05 × 0.05 degree (approximately 5.5 km) and covers 35 years (1980-2014). 40 https://doi.org/10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License.

Ice Bridge radar data
NASA Operation IceBridge radar data were collected during the 2013-2014 summer campaigns and is validated with ten deep ice cores from across the GrIS (Lewis et al., 2017). Accumulation rates were calculated for 18 epochs over 1712 -2014 to analyse SMB changes and correlations with both the Atlantic Multidecadal Oscillation and North Atlantic Oscillation. The accumulation radar operates in the 600 -900 MHz range, yielding an uncertainty of 12.7 cm w.eq. a -1 for any single epoch, 5 when taking into account the uncertainty on density and annual layer dated isochrones used for a calibration (see Lewis et al., 2017). Here, we choose to compare in situ firn core accumulation rates with the closest radar returns from IceBridge flight lines to validate our measurements in tandem with the HIRHAM5 output.

North Greenland temperatures 10
The temperature at ~10 m depth for each of the shallow core sites (Table 2)  For the cores A1 (NEEM) and A5 (EastGRIP) water isotopes were determined and are shown in supplementary material Figure   S3. Annual layers in the water isotopes can be clearly identified in A1, while for A5 the annual signal is lost below 5 m depth corresponding to year 1992. The A1 (A5) core's mean δ 18 O and δD values are respectively -32.17 ‰ (-36.16 ‰) and -247.2 15 ‰ (-281.6 ‰). Using the relationship between mean annual values of δ 18 O and temperature on the Greenland ice sheet (Temperature=(δ 18 O+13.7)/0.67 (Johnsen et al., 1992)) the observed mean δ 18 O concentrations in the two cores corresponds to -27.6 ºC at the northern NEEM site and -33.5 ºC east of the divide at EastGRIP and thus suggest much colder temperatures at the EastGRIP sites than those measured in the borehole, while at NEEM the two estimates are in correspondence. However, we recall that water isotopes are not only a function of temperature, but also of accumulation, especially so in North-East 20 Greenland (Buchardt et al., 2012). Orsi et al. (2017) used borehole temperature from the NEEM main core borehole and observed a mean temperature in the period 1900-1970 of -28.6 ºC and thus our estimates of -27.4 ºC (borehole) and -27.6 ºC (water isotopes) are within uncertainties.
The HIRHAM5 mean temperature in the period 1980-2014 is higher by 0.9 ºC to 2.9 ºC (Table 2) at the sites compared with 25 the observed 10 m borehole temperatures. However, recall that the borehole temperature is the seasonal cycle damped with time, thus influenced to some extent also by temperature of previous years and that the modelled temperature represents 2 m heights not skin temperatures.
We observe that mainly the winter temperatures from HIRHAM5 are biased compared to observations from weather stations and according to several model specifications can contribute to the winter warm temperature bias. First of all, the vertical 30 resolution of the model is insufficient to resolve the processes that leads to strong temperature inversion. Secondly, the temperature bias can be related to errors in the model parametrization of clouds and in that way related to a biased incoming longwave radiation. And finally, the turbulent exchange near the surface is poorly resolved in HIRHAM5.
In conclusion we find higher temperatures in north-west Greenland than in north-east in both observations and model and note that the HIRHAM5 model is generally warm biased. 35

Accumulation from firn and snow cores
The accumulation reconstructed from the 6 firn cores with time is shown in Figure 2 (coloured dotted lines) together with a five year running average trend (coloured full line). In Table 3  A clear east-west gradient is evident in the reconstructed accumulation ( Figure 3) with the highest accumulation found at the NEEM ice core site, 22.2 cm w.eq. yr -1 (core 2015 T-A1) and lowest accumulation, 11.8 cm w.eq yr --1 , at the ice divide site (core 2015 T-A6). East of the ice divide the accumulation again increases but only slight (13.7 cm w.eq. yr -1 -A5 ). The year to year variability (1 sigma) is 25% of the accumulation at the NEEM site (A1) and decreases towards the ice divide 16% (T-A3), 18% (A6) and increase again at the EastGRIP site (21%-A4 , 30 %-A5). 5 During the traverse also several snow core liners representing just the top 2 m of snow were also collected. The results from these are presented in Schaller et al. (2016) and the snow liners represent accumulation from 2 to 3 years. The comparable sites to the traverse cores from A1 to A6 are N2E02 (A1), N2E06 (A2), N2E09 (A3), N2E19 (A4), N2E22 (A5) and N2E12 (A6) with accumulation of 22. 45, 19.35, 15.59, 13.22, 13.96 and 12.47 cm w.eq. yr -1 respectively shown also in Figure 2 (grey line). Thus the results from both snow liners and firn cores agree within uncertainties. 10 The results from the snow cores collected during the windsled traverse in 2017 are shown in Table 5. Note that the accumulation presented from the 2017 snow cores represents just one year (summer 2015 to summer 2016) and a significant variability to the annual mean at the particular sites could be present. The accumulation vary from 24.5 cm w.eq. yr -1 at the central ice divide in the south (73.65N, 40.56 W) and decreases to 14.0 cm w. eq. yr -1 in the drier region near EastGRIP (75.40N, 27.03W) and we observe a clear decreasing trend from central-south to the north-east in line with other estimates from the region. The site 15 closest to EastGRIP (14.0 cm w.eq. yr -1 ) is well in line also with the shallow firn core results of 13.7 cm and 14.6 cm w.eq. yr -1 for the sites A5 and A4 respectively, though the longer firn cores are further north.

Accumulation from HIRHAM5 and comparison to firn cores
In general, the HIRHAM5 annual precipitation minus the surface water flux (mainly sublimation) slightly underestimates the 20 net accumulation observed in the firn cores ( Figure 3). Furthermore, a linear relationship between the annual accumulation in the firn core and in the model show that the model accumulation is just 66% (R 2 0.74) of the annual mean firn core accumulation ( Figure 5), though with better representation at the higher accumulation sites in the west (A1~ 106%). East of the divide HIRHAM5 again estimates a lower accumulation (56-67%) than that observed in the firn cores, suggesting that model performance decreases in the east. For the site A6, HIRHAM5 captures just 55%. However, we also note that the site A6 is 25 very close to where the central ice divide splits into a north-eastern and north-western branch (supplementary Figure S2). Here we speculate that the difference also relates to changes between these topographic features in the model domain and in the real world, as well as the ice divides influence on the dominant precipitation patterns. Further we note that a climate variable from a grid cell in HIRHAM5 represents the mean conditions over an area of 5.5 x 5.5 km and that comparing a variable from a grid point with a point measurement from a shallow ice core should be done cautiously. Pointwise measurements are sensitive 30 to immediate changes in the surrounding area and are thus not necessarily directly comparable to solid accumulation or other weather conditions simulated by HIRHAM5 on a grid of 5.5 km (Lucas-Picher et al., 2012). The annual variability in HIRHAM5 follows the accumulation trend linearly and is between 14 and 18% of the mean accumulation, whilst the annual variability in the firn cores is between 15 and 30% of the accumulation signal and not linearly related to the mean accumulation.
This suggests that the firn reconstructions are subject to depositional noise. 35 The site A4 is at the margin of the ice stream NEGIS. At the ice shear margin zone of the ice stream ( Figure 6) there are topographical changes over short distances of up to 40 m (Bamber et al., 2018;Joughin et al., 2010;Vallelonga et al., 2014).
The model does not incorporate such detailed topographical features, producing a flat surface at NEGIS in the model domain At site A4 HIRHAM5 underestimates the observed accumulation (63% captured). Topographic undulations cause more of the precipitation to accumulate in the depressions (Miège et al., 2013). Therefore, this difference may be explained by A4's 40 location in a valley on the ice shear margin. Traverse core A5 (just outside the ice stream) is subject also to some of this effect (67% captured) though not to the same extent as A4.
To estimate the area of accumulation that each firn core theoretically represents we compared the accumulated precipitation minus evaporation observed in the model for each site with that in all other model points (Figure 4). In Table 3 annual accumulation correlations between sites in both model and firn cores are shown for the period 1998-2015. It is evident that the three cores west of the divide (A1, A2 and A3) represent large areas and should be well correlated (Figure 4). This is also true 5 in both the model, where correlations (R 2 - Table 4) between A1 and A2 are 0.90 (p<10 -6 ) and A2 and A3 are 0.95 (p<10 -9 ), and in the firn cores (A1, A2 R 2 =0.56, p<10 -4 , and A2, A3 R 2 =0.54, p<10 -5 ), despite the influence of depositional noise not represented in the model. The site A6 close to the central divide is also well correlated with A1, A2 and A3 in the reconstructed accumulation from the firn cores, whilst in the model is correlates more strongly with eastern sites.
Correlations east of the divide are lower both in HIRHAM5 and in firn cores (Figure 4, Table 3). This may be due to the lower 10 accumulation rates in the east, which decreases temporal resolution in the core and increases uncertainty within and between reconstructed accumulation rates. In addition, precipitation across the north-eastern ice sheet is mostly cyclone related, whereas the western part is predominantly connected to the large scale circulations (Hutterli et al., 2005).
Core T2015-A4 and T2015-A5 , which are only 15 km apart, show in the model a very high correlation (0.97, p~10 -11 ), yet in the reconstructions from the firn cores they only correlate by 0.28. We speculate that the local topography caused by the north 15 East Greenland Ice Stream, which is not fully resolved in the model domain, is causing more variability in the accumulation reconstructed from the firn cores as a result of snow drift and depositional noise induced by such topographic hills (Vallelonga et al., 2014).

Increasing temperatures in northern Greenland since the early 1990's 20
In 1993-1995 several firn cores were drilled in northern Greenland as part of the North Greenland traverse (NGT (Fischer et al., 1998;Weissbach et al., 2016)). These can be used to evaluate if there has been a trend in the temperature with time by comparing the closest NGT site with our N2E firn core borehole temperatures representing 2014 (Table 2) (Figure 7), where a similar trend in the nearby NEGIS firn core is observed (Vallelonga et al., 2014). Variations in ice core deuterium excess are typically interpreted as reflecting changes at 40 the moisture source region through kinetic effects related to the relative humidity, wind speed and sea surface temperature https://doi.org/10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License. (Johnsen et al., 1989;Masson-Delmotte et al., 2005). Here we speculate that the decreasing trend is climatic and related to the moisture source region or unknown (and strong) post-depositional processes that cause the deposited snow's water isotopic composition to kinetically fractionate. The latter seems less likely as this effect would have had to linearly increase its influence on the d_xs signal over the past 15 years. While it is an interesting research question to quantify the cause behind this signal, it remains outside the scope of this paper as it would require studies with isotope -enabled atmosphere general circulation 5 modeling and/or vapor-ice water isotope fractionation experiments at the drill site. We also note that recent papers (Kopec et al., 2019;Madsen et al., 2019) have questioned the use of deuterium excess as a moisture source proxy.
In conclusion our results from both in situ snow temperatures at 10 m depth and from the HIRHAM model point to a northern Greenland warming trend over the past 2 decades of between 1.6 and 4.9 ºC (Table 2). However, the signature of such an 10 increase is not observed in the δ 18 O and δD (Supplementary material Figure S3). Furthermore, the deuterium excess results annual variability in trends. Most notably they observed the strongest warming in west and northwest coastal Greenland (total 6-6.5°C in winter for the period 1991-2019), however they also note a flattening of the temperature increase after 2010. The warming we observe is in line with these previous observations. It is worth also noticing that according to our results of the past two decades the warming trend of northern Greenland has already surpassed the temperature trend anticipated for central 20 Greenland from an RCP 8.5 scenario (NorESM model (Jansen et al., 2020)).

Annual mean accumulation across the northern Greenland ice sheet
We now compare the mean accumulation observed from both of the firn cores 2015 and the snow cores 2017 to those from other reconstructions (Figures 1 and 2). Figure 1 shows the annual mean accumulation reconstruction from Burgess et al. The average accumulation from both firn and snow core study sites are within uncertainties and in line with previous estimates.
A particularly strong comparison is observed with the radar lines by Karlsson et al. (2020) representing the period 1694-2015 and the snow cores estimates by Schaller et al. (2016), representing 2014-2015. However, the mean annual accumulation reconstructed as part of this study is more often than not increased compared to other previous estimates as evident in Figure  35 1 (especially east of the divide).
We observe no significant common trend in accumulation over time for the six firn cores and we find no evidence for the speculated accumulation change as a function of climate change for any of the six sites in the cores themselves. Whilst the trends are susceptible to the period investigated (10, 20, 30, and 40 yr periods), we do observe a decreasing trend for the period 2000-2010. This is equal to -1.40 cm w.eq. yr -2 in A1 in the west and -0.12 cm w.eq. yr -2 at site A4 in the east, although only 40 significant within 95% at sites A1, A2 and A3. https://doi.org/10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License.
The accumulation averages west of the divide (A1, A2) are in line with previous firn and radar estimates. Here we observe a period of increasing accumulation in the period 1988-1998 in the western cores (A1, A2 and A3), followed by a decrease towards 2015 for cores A1 and A2 , a trend also observed in the HIRHAM5 model. In the firn cores A1 and A2 the years around 2000 show up as extreme accumulation years and enhances the trend.
Closer to the central ice divide the trends with time in cores A3 and A6 are insignificant, however we notice that the period 5 1995-2004 is high compared to other periods ( Table 2). The firn accumulation reconstructions at these more central sites fall within previous estimates, whilst the model underestimates the accumulation.
East of the divide the extreme events seem more scattered and may be a result of local topography effects in combination with depositional noise amplified at low accumulation sites. We observe a clear decrease in accumulation when moving east off the 10 ice divide from the snow cores and an additional decrease when simultaneously moving northward. An increase compared to previous estimates is observed in the area around EastGRIP, but the site (A4 and A5) is highly influenced by the local topography of the north-east Greenland ice stream and additional spatial data are needed to constrain accumulation properly in this area.
Below the annual mean accumulation and their trends for each region is further discussed in detail. 15

North Greenland accumulation -West of the divide
The annual accumulation for the cores west of the divide (A1, A2 and A3) are highly correlated (>0.54, Table 4), a trend which is also observed in the HIRHAM5 model (>0.83). The accumulation means are close to previous firn and radar estimates, with a slight recent increase in accumulation at the NEEM site. We find in all three cores; A1, A2 and A3 that 1988-1998 is a period of increasing accumulation followed by a decrease towards 2015. High precipitation is observed in 2000 Close to the NEEM deep ice core drilling site we retrieved the firn core T2015 -A1 (Figure 2, blue line) with a mean accumulation of 22.2 cm w.eq. yr -1 , matching with the HIRHAM5 (6% difference on the mean). One could speculate that the NEEM ice core drilling camp structures and especially the dome in the period 2008 to 2015 have had an influence on the accumulation and satellite observations. This theory is supported by stake measurements taken around NEEM which suggest 25 an increase of accumulation around camp. We deliberately avoided such camp infrastructure interferences and drilled the core 5 km upwind from the main NEEM ice core camp location in the clean snow zone. Furthermore, we do not observe an increase nor decrease in accumulation compared to the remainder of the T2015-A1 record while NEEM was an active camp (2008)(2009)(2010)(2011)(2012), but note that the years 2009, 2010 and 2011 are higher than surrounding years.
The NEEM deep ice core project lasted for several years and in the process several shallow firn and snow cores were retrieved. 30 An average annual mean accumulation of 20.3±3.1 cm w.eq yr -1 over the past 3000 years is observed in the main NEEM ice core (Rasmussen et al., 2013) and a reconstruction using four firn cores retrieved close by covering the period 1727-2007 finds similar results (Masson-Delmotte et al., 2015). Our results are thus in the higher end of the previous ice core estimates. The radar lines from the traverse 2015 (Karlsson et al., 2020) show a similar mean to those from this study of 22.2 cm w.eq. yr -1 for the period 1694-2015 and when compared with other nearby radar lines (Karlsson et al., 2016(Karlsson et al., , 2020Lewis et al., 2017), 35 we find accumulation rates in close proximity to the NEEM drilling site that are also in line with our findings. This is particularly true when compared with the nearest Icebridge radar line, which shows for the period common with the T2015-A1 core an accumulation of 23.6±1.97 cm w.eq yr -1 and the snow core retrieved at NEEM during the N2E traverse, which has 22.7 cm w.eq yr -1 accumulation annually . The more recent high estimates point to the period 1998-2015 as one high in accumulation compared to the past 300 years as observed by Masson-Delmotte et al. (2015), yet not as compared 40 to the results from radar lines (Karlsson et al., 2020). However, Masson-Delmotte et al. (2015) did observe an increase in accumulation of 1.6 cm yr -1 per decade from 1980 to 2007. Such increasing trend is also supported by IceBridge radar and https://doi.org/10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License.
HIRHAM5 modelling for the same period. But, our firn core T2015-A1 does not cover all of that particular period. Conversely, in the overlapping period from 1998-2015, we observe a decreasing trend in the accumulation of the core T2015-A1 with a total of -3.7 mm w.eq. yr -2 , a value highly influenced by the extreme years in 2000 and 2001. That trend is also found in both operation IceBridge retrieved radar data (-2,8 mm w.eq. yr -2 ) and in the HIRHAM5 model (-1.3 mm w.eq. yr -2 ). In addition, we observe that 2009/2010 is extreme in accumulation in both model and reconstructed accumulation from the firn core at site 5 A1 .
In conclusion, at the NEEM site we find that the period 1998-2015 experienced increased accumulation compared to the previous ice core estimates by 9%, but not compared to radar and that throughout the periods 1998-2015 a decreasing trend was observed, mainly driven by extreme accumulation in 2000 and 2001 and that extreme accumulation was also found in 2010 in both firn and model. 10 At site T2015-A2 (Figure 2, red line) 100 km upstream from NEEM on the north western ice divide we reconstruct a mean accumulation of 17.6 cm w.eq. yr -1 , less than that found at the NEEM site. It however shares the same features of decreasing accumulation as observed near NEEM (core A2 -8.2 mm w.eq. yr -2 , IceBridge radar -9.3 mm w.eq. yr -2 ). In addition the core shows an increasing trend for the period 1988-1998 of 11 mm w.eq. yr -2 , a value also reflected to smaller extent in the HIRHAM 15 model (+5.3 mm w.eq. yr -2 ) and in the radar (+4.1 mm w.eq. yr -2 ). The three closest snow cores obtained by Schaller et al., 2016 show an accumulation of 20.5, 17.2 and 19.3 cm w.eq. yr -1 and thus within the range observed in the firn core A2 . Whilst the estimates from the IceBridge radar (19.09 cm w.eq yr -1 ) are slightly larger, the model estimates a comparable value of 17.9 cm w.eq. yr -1 over the full length of the core . The B26 firn core from the NGT traverse retrieved in 1995 in between our A1 and A2 sites, show for the period 1512-1994 an accumulation in line with the model and our results with a 20 value of 17.9 cm w.eq. yr -1 . Thus our recent accumulation reconstruction point towards a steady accumulation, with the caveat that the position of the NGT core and our N2E core are 37 km apart. We do not observe any sign of the extreme accumulation in 2010 in A2 despite the HIRHAM5 model results.
Site T2015-A3 (Figure 2, green line) covers the same temporal period as A2 and also shows an increase in the period 1988-25 1998 (+8.8 mm w.eq yr -2 ). However, unlike A1 and A2, no significant trend for the period following 1998 is observed. The site is an additional 94 km upstream from site T2015-A2 and is in general slightly lower in accumulation (17.5 cm w.eq. yr -1 ).
However, it is worth noting that at site A3 in general the HIRHAM5 model produces estimates of ~3 cm w.eq lower than those observed in the firn core, a constant offset that likely is a result of the ice divide position as discussed earlier. The IceBridge radar trend over time does not match those trends observed in the firn core, however mean values are comparable with one 30 another and also with the 2015 radar mean (Karlsson et al., 2020). The closest N2E snow core  point towards slightly lower accumulation, and is in line with the model despite only covering only few years. During the 1994 NGT traverse, between A2 and A3 a core was collected (B27/28) and they observe 18.0 cm w.eq. annually for the period 1783-1993 (Weissbach et al., 2016). Given their similarity to the firn core results, this suggests no recent accumulation increase has been observed at the T2015-A3 site. 35

North Greenland accumulation-at the central divide
In the dry central-Northern region we retrieved core T2015-A6 (Figure 2, yellow line) with 11.8 cm w.eq. yr -1 . At this low accumulation site the model underestimates the accumulation observed in the firn core by on average 2.5 cm w.eq. yr -1 . The closest IceBridge radar lines fit well in mean accumulation yet display slightly higher accumulation compared to the firn core estimate for the period 1996-2014 and slightly lower accumulation than observed in the A6 core reconstruction from the 70's 40 to the late 90's. The snow cores collected nearby also during the N2E traverse, which show accumulations of 11.7 and 12.7 cm w.eq. yr -1 , are in line with our results as is the radar 2015 results (Karlsson et al., 2020). There are https://doi.org/10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License. no NGT traverse cores nearby, however A6 is positioned almost half way between NGT core B18 and B29 (Figure 1). These earlier cores have annual accumulations of 10.3 and 14.9 cm w.eq. yr -1 respectively and align well with our results. At A6 we observe a small decreasing trend since the mid 1990's (-2.4 mm w.eq. yr -2 ), which is stronger when focusing on the past 10 years (-6.3 mm w.eq. yr -2 ). Such a trend is not captured in the model nor in the radar.

Northern accumulation east of the main ice divide (snow cores 2017). 5
Further south we observe a decreasing trend in accumulation when moving north and eastward from the central divide as a result of the divide shielding from the western precipitation as evidenced from the snow cores retrieved from the Inuit Windsled Project traverse in 2017. The snow cores cover a line of 280 km, ~265 km in a north-south direction and ~100 km in the east -west direction with an altitude decrease of 150 m when moving north-east (Figure 1, Table 5). The accumulation in 2015-2016 summer is spanning 24.5 cm to 15.4 cm w.eq. yr -1 . 10 The closest older points of observation are the NGT snow cores  and the radar line (Karlsson et al., 2020) all of which are east of our snow cores (Figure 1). The NGT results and radar lines support an eastern decreasing precipitation trend for this area of the Greenland ice sheet. Furthest south and 100 km east of the core-WP 726 (24.5 cm w.eq), a NGT core is found with 15.5 cm w.eq annually . Half way on the transect and about 80 km east of the snow core WS 2017-5 (18.6 cm w.eq), the longer NGT firn core B16 displays a mean accumulation of 14.1 cm w.eq annually for the 15 period 1640-1993 (Weissbach et al., 2016). Lastly, in the north 40 km east of the sites WS2017-6 (16.1 cm w.eq. yr -1 ) and WP 606 (15.4 cm w.eq. yr -1 ), NGT snow core has a value of 13.3 cm w.eq. yr -1 , supporting the decrease trend in accumulation when moving both east and north. We restrain from speculating on any accumulation trend over time for this area as the NGT snow cores, radar line and our snow cores are far apart and because our cores represent just one year of accumulation. 20

EastGRIP accumulation
Close to the EastGRIP ice core drilling site and only 15 km apart from the location where both the core T2015-A4 and T2015-A5 were drilled, an annual accumulation 14.1 and 14.7 cm w.eq. yr -1 was observed for the period 1997-2015. Similar observations were also produced by the three closest snow cores taken during the 2015 traverse with values of 13.2 14.6 and 14.0 cm w.eq. yr -1 . Similar results were also present within traverse radar lines (13.0 cm w.eq. yr -1 25 (Karlsson et al., 2020)) and from two 4 metre snowpits, where accumulations of 13.8 and 14.5 cm w.eq. yr -1 where observed (Nakazawa et al., 2020). These values are all larger than the older estimate of 11.6 cm w.eq. yr -1 from the North East Greenland Ice Stream within the similar period (NEGIS, (Vallelonga et al., 2014)) and also larger than the NGT traverse snow cores taken in the area in the early 90's (11.6 and 12.9 cm w.eq. yr -1 ). The discrepancy to recent firn and snow results could be interpreted as a recent increase in accumulation compared to the 1990's (NGT cores). However we note that when comparing the 30 overlapping time period with of our firn core  that NEGIS results also show a lower 11.2+-1.6 cm w.eq. yr -1 than the A5 core having 13.8 cm w.eq. yr -1 (Table 3). Furthermore, the trend observed in our A4 and A5 core shows increased accumulation in the 1990's compared to the mean and thus a discrepancy between our results and those from the NEGIS firn core exist in this area.
In addition we observe that the accumulation in A4 and A5 firn cores are poorly correlated (R 2 -0.28) and that the trend in both 35 cores is variable. The reconstructed accumulation from A4 and A5 are poorly correlated also with the HIRHAM5 model accumulation. However, in the model domain the correlation between the sites just 10's of km apart is excellent (0.97). We explain the discrepancies with the position of the cores close to the North East Greenland Ice Stream. The North East Greenland Ice Stream (NEGIS) is the largest ice stream in Greenland and the central velocity close to EGRIP is larger than 60 m yr -1 in the area. The strain from the ice stream compared to the surrounding stagnant ice forms local topographic hills both with 40 https://doi.org /10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License. minima on the shear margin and as undulating waves along the ice stream itself (Fahnestock et al., 2001;Joughin et al., 2010;Vallelonga et al., 2014). Figure 6 shows the area close to EGRIP, the location of A4 and A5 and the ice sheet surface altitude. The NEGIS core is at the centre of the ice stream (close to deep ice core EastGRIP being drilled these years), while core A4 is outside the main ice stream on the shear margin and A5 is outside the shear margin to the north in "stagnant" ice. All three NGT firn cores 5 (Weissbach et al., 2016) are similar to A4 on the shear margin to the northern side of the ice stream. Further upstream compared to A4 and the snow core WP506 retrieved in 2017 is even further upstream, but again on the shear margin. Thus we hypothesize that the variability between A4 and the older snow cores from the NGT traverse is caused by local topography influencing the shear margin and thus the local accumulation.
Site A5 is in the stagnant ice outside the ice stream and shows a more regular deposition from the mid 80's, except for the 10 extreme 2010 year. Prior to the mid 80's the core has increased variability compared to the latter period. This is contrary to the model and ice bridge radar, which show increasing accumulation (1 and 3 mm w.eq. yr -1 respectively) from the mid 80's.
Both in the core and in the radar a higher accumulation is observed in the late 60's. However, we note that the uncertainty associated with the timescales of cores A4 and A5 allow for changes that could influence the trends discussed, especially in the deeper part of the cores. 15 In conclusion the accumulation reconstructed from firn and snow cores around EastGRIP se em to be influenced by local topographic variations caused by the North East Greenland Ice Stream. As such any accumulation record from this site should be carefully analysed in relation to the local constraints. To truly resolve this areas accumulation structure and any temporal trend and to determine how the EastGRIP deep ice core can be analysed as a proxy of past climate, a fine grid of cores or radar 20 analysis for reconstructing accumulation will be necessary.

Conclusion
In this study we have increased the number of recent accumulation constraints in central northern Greenland by adding six shallow firn cores and eight snow cores. We find no joined trend in the northern accumulation, thought to be a result of the inter-annual noise in the reconstructions. In general our findings point to slightly larger accumulation in the northern Greenland 25 area than that mapped by Burgess et al. (2010), especially around EastGRIP in the Northeast. The results are generally in line with previous ice cores and radar estimates, especially for the most recent two decades. The comparison to the P-E from the HIRHAM model show that the model tends to underestimate the accumulation compared to our findings from the firn core in the central north and east of the divide, a relevant finding for people reconstructing the surface mass balance.
The three traverse cores taken west of the divide tell a similar story of increasing accumulation in the 90's with a subsequent 30 decrease from the millennia onwards. This trend is also supported by radar and reanalysis data though the trends in these are less. All three sites west of the divide are well correlated (>0.54) and a decrease in accumulation from the NEEM site (T2015-A1) is observed with a value of 22.2 cm w.eq yr -1 moving up the ice divide to site T2015-A3 (17.5 cm w.eq yr -1 ). In general, the mean accumulation from the 3 western cores show similar accumulation to previous estimates. We find increased temperatures of ~3ºC since 1993 based on borehole temperature evidence. However, when looking into the details of the 35 temperature versus the accumulation a Clausius-Clapeyron relationship trend is not observed on neither an annual basis nor in the mean trend in our reconstructed accumulation, despite a clear relation between temperature and precipitation in the reanalysis (R 2 >0.6). In addition, we find some evidence of 2000's extreme accumulation in the firn cores A1 and A2.
When moving from the centre of Greenland north-east off the divide in the northern central Greenland we observe a decrease in accumulation as also reconstructed from snow cores retrieved in 2017. Comparison with the older NGT snow cores 40 https://doi.org /10.5194/tc-2020-337 Preprint. Discussion started: 11 January 2021 c Author(s) 2021. CC BY 4.0 License. (Weissbach et al., 2016) and the radar line (Karlsson et al., 2020) further supports such a spatial variability. The HIRHAM5 models capture less of the observed accumulation when moving eastwards off the divide.
Close to EastGRIP the firn cores A4 and A5 are influenced by local topographic variations caused by the north east Greenland Ice Stream. Whilst A4 and A5 match well with previous accumulation estimates from snow pits (Nakazawa et al., 2020;Schaller et al., 2016) the previous results from the NEGIS ice core do not (Vallelonga et al., 2014). We advise that any 5 accumulation record from this site should be carefully analysed in relation to the local constrains. To truly resolve this ar ea's accumulation structure additional firn reconstructions are necessary and a fine grid of cores upstream will be crucial to better understand the topographic influence on the EastGRIP deep ice core, if climate is to be reconstructed from it. At the EastGRIP site in the model the relationship between accumulation and temperature is less evident than to the west (R<0.3).
In conclusion, despite our findings of a significant north Greenland ice sheet warming of 0.9 to 2.5C/decade no common 10 northern increasing trend in accumulation was observed in the firn cores as a result thereof. The overall spatial pattern of accumulation resembles previous estimates with the exception of sites close to the NEGIS ice stream, where we find larger accumulations than previous reported. The spatial variability observed in P-E from re-analysis data matches well, but our estimates of accumulation from the firn and snow cores were in general larger than the HIRHAM5 model estimates, except close to the NEEM ice core drilling site. This general higher accumulation should be accounted for when modelling the current 15 state of the Greenland ice sheet.

Data availability
The firn core T2015-A1 to T 2015-A6 and snow core from 2017 accumulation reconstructions will be made available at relevant depositories upon publication. CFA. MS and DDJ measured the water isotopes in firn cores A1 and A5. BV and CTH analysed the water isotopes and deuterium excess. HK and PZ made the annual layer counting and reconstructed the accumulation and further analysed the accumulation and chemistry data. MO and RM provided the re-analysis data and the HIRHAM modelling. GL provided nearest neighbour IceBridge radar data and NK radar from the traverse 2015. All authors contributed to the writing of the paper.        (Weissbach et al., 2016)), data collected as part of the sum up dataset (Montgomery et al., 2018) including snow and ice cores as well as stake measurements (diamonds) as well as radar lines from Karlsson et al., 2016;Lewis et al., 2017Lewis et al., , 2019Karlsson et al., 2020. 5