The 79°N Glacier cavity modulates subglacial iron export to the NE Greenland Shelf

Approximately half of the freshwater discharged from the Greenland and Antarctic Ice Sheets enters the ocean subsurface as a result of basal ice melt, or runoff draining via the grounding line of a deep ice shelf or marine-terminating glacier. Around Antarctica and parts of northern Greenland, this freshwater then experiences prolonged residence times in large cavities beneath floating ice tongues. Due to the inaccessibility of these cavities, it is unclear how they moderate the freshwater associated supply of nutrients such as iron (Fe) to the ocean. Here, we show that subglacial dissolved Fe export from Nioghalvfjerdsbrae (the ‘79°N Glacier’) is decoupled from particulate inputs including freshwater Fe supply, likely due to the prolonged ~162-day residence time of Atlantic water beneath Greenland’s largest floating ice-tongue. Our findings indicate that the overturning rate and particle-dissolved phase exchanges in ice cavities exert a dominant control on subglacial nutrient supply to shelf regions.


Quality assurance
Validation of method accuracy for seawater analyses was achieved through the reference material 7602a (National Metrology Institute of Japan) for macronutrients (given in the datasheet), and SAFe S for soluble, dissolved and total dissolvable trace metals (Supplementary Table 1) 1 . Method sensitivity for seawater trace metals was monitored through SeaFAST-ICP-MS procedural blanks (Supplementary Table 2) following ref. 2 . Labile particulate analyses were validated against BCR-414 reference and indicative values 3 , and monitored against process blanks (filter digest).

Nutrient fluxes, dilution and sinks across salinity gradients
From a freshwater perspective, a glacier outflow will always result in a positive flux of any chemical component into the marine environment 4 . Yet this can still result in a negative change in the availability of a nutrient in the ocean either by dilution 5,6 , as a result of prolific nonconservative removal in estuaries 5 , or because the stratification driven by freshwater discharge decreases vertical mixing and thereby suppresses marine primary production 7 . This results in inconsistent terminology between research fields because what always constitutes a nutrient flux at 0 salinity can also be a nutrient sink when considering a flux gate after the major nonconservative mixing processes have occurred or using a box model. Similarly, what always constitutes a nutrient source into the ocean as a whole (especially when considering geological timescales), can still reduce nutrient availability on annual timescales. This is exemplified by the behavior of phosphate (PO4). Freshwater PO4 concentrations are generally low across the cryosphere relative to saline waters 4,5 . Furthermore, PO4 seems to be subject to a degree of non-conservative removal downstream of glaciers related to high turbidity 8,9 . This results in negative PO4 anomalies (e.g. Supplementary Figure 4), i.e. the concentration of PO4 is lower than can be explained by mixing processes. Glacier outflows can therefore both dilute marine PO4 concentrations 5 and result in local net PO4 removal 8 despite constituting a measurable PO4 flux based on freshwater measurements 4 .

Estuarine removal
The change in salinity, and other properties such as temperature and turbidity across the salinity gradient, affect chemical components to varying degrees. Conservative components are those that scale linearly with salinity during mixing whereas non-conservative components show pronounced positive or negative deviations from the expected conservative mixing line. Dissolved Fe invariably shows a classic non-conservative estuarine mixing behavior with dFe concentrations always lower than expected from conservative mixing due to prolific removal of dFe onto particles 10 .
Whilst most literature concerning estuarine mixing of (micro)nutrients concerns temperate river estuaries, non-conservative losses of dFe are also well documented downstream of glacier outflows [11][12][13][14] with 76-99% removed at intermediate salinities. Dissolved silicic acid (Si(OH)4) behavior is more variable; some glacier catchments show close to conservative mixing over the observed salinity gradient 15 , and others indicate non-conservative addition at salinities lower than ~10 16,17 . Deriving the net fluxes of dFe and Si(OH)4 arising from glacier outflows is challenging, particularly for large marine-terminating systems, because of the general paucity of data close to glaciers where subglacial discharge first enters the marine environment and the few case studies where extensive data is available spanning the salinity gradient 18 . As elsewhere, in glacier fjords nutrient distributions are also affected by uptake by biota and benthic processes that act to add/remove nutrients from solution over the same spatial/temporal scale where inorganic processes arising from the change in salinity, turbidity, temperature and pH occur.
The non-conservative aspects of dFe and Si(OH)4 across the salinity gradient account for practically all of the order-of-magnitude variation in flux estimates from Greenland into the ocean, depending on what flux gate window is defined 18 . But variation also arises from a process perspective in how fluxes are scaled. For Si(OH)4, a ten-fold difference in the two available flux estimates from the Greenland Ice Sheet occurs despite using similar freshwater and intermediate salinity Si(OH)4 concentrations 5,17 .

Applying linear regression to 79NG
A linear regression of any concentration against salinity should derive the approximate freshwater concentration inclusive of non-conservative effects. However, even in the absence of nonconservative chemical effects, Arctic glacier fjords can exhibit v-shaped nutrient distributions across the salinity gradient because of the multi-dimensional nature of saline water inflow at depth and modified water outflow closer to the surface. In the immediate vicinity of a glacier outflow, a transient increase in all macronutrient concentrations with salinity can therefore be observed followed by a steady decline in surface waters due to biological drawdown. The 0-salinity intercept of a linear regression is therefore sensitive to the range of salinity data selected.
Supplementary Table 3: Linear regression for all nutrient data at trace metal clean stations S1-13 (Supplementary Figure 4 and 5). The 0 salinity intercepts for all macronutrients downstream of 79NG are negative. Whilst this suggests a minor role for freshwater outflow on the scale of the region sampled, it is not particularly informative concerning the freshwater concentration as it mainly reflects the v-shaped macronutrient distribution seen in other Greenland fjord systems when considering data across a salinity gradient from ~10-35 18 . For dFe, conversely, the intercept is positive (7.3 ± 0.7 nM for the region, or 5.0 ± 0.8 nM for stations S1-6) suggesting a freshwater concentration of ~5-7 nM. This is derived from saline data with S > 24, so the intercept is informative concerning the dFe remaining after non-conservative loss and not the freshwater concentration before this loss (which would invariably be much higher) 11,13,14 .

General Additive Model (GAM)
GAMs are more useful than linear regressions because they are able to optimize a non-linear fit to multiple parameters simultaneously. Some nutrients clearly display similar spatial trends that can be explained by the same key factors. A Redundancy Analysis clearly shows that nitrate (NO3), PO4, and Si(OH)4 cluster together, as do the dissolved metals Fe, Co and Mn ( Figure 6). This suggests that similar factors explain their distributions. Relatively good (R 2 >0.75) fits were obtained for all nutrients with a GAM generated by the interaction between salinity and the distance to glacier terminus demonstrating that these variables were able to explain most of the variance in nutrient datasets and these fits were used to obtain estimated 0 salinity concentrations at the glacier terminus ( Figure 7).
Supplementary In all cases, a GAM fit was very good and the predicted intercepts were very similar to that predicted from linear regression. In the case of dFe, a positive intercept (3.13 ± 0.96 nM) is consistent with the dFe enrichment expected from fresh water and is, as per linear regression, a value derived after the majority of non-conservative loses have occurred. The negative predictions for macronutrients again verify that there must be a change in gradient for all macronutrients within the subglacial cavity. The differences between linear regression and GAMs in terms of the predicted intercept can be explained by considering that the GAM fit includes variables other than salinity (i.e. distance to glacier terminus), although salinity remains a major factor in explaining the variance in all nutrient datasets (as is shown by RDA).

Contrasting concentrations of macronutrients in mAIW and AIW
To add robustness to our discussion we test how changes in the collection of stations used to assess Atlantic Intermediate Water (AIW) and modified AIW (mAIW) properties would affect our interpretation. For trace metal data, all stations on the shelf are included within the main text. For macronutrients, there is additional data from the large volume CTD that expands the data available.
Here we define mAIW (27.00-27.73 kg/m 3 ) and AIW (>27.73 kg/m 3 ) using the same density definition throughout to different subsets of all the large CTD cruise data. The standard deviation of measurements is generally lower using the more extensive combined datasets, although the means remain similar.
From an oceanographic perspective concerning the detection of any changes attributable to local input from the 79NG, the most meaningful definitions of mAIW are those determined immediately adjacent to the ice-tongue as this determines the properties of outflow before any extensive nutrient drawdown, or dilution of mAIW can occur. The most meaningful definition of AIW is that determined at the deepest stations outside the fjord and upstream of the glacier outflow as this precludes any local processes in addition to those occurring underneath the 79NG ice-tongue (e.g. benthic inputs across the shelf) 19 which could affect the properties of AIW as it flows along the fjord prior to entering underneath the subglacial cavity.
Supplementary Comparing the mAIW and AIW concentrations of macronutrients suggests that there is possibly a decline in NO3 concentration between AIW and mAIW within Nioghalvfjerdsfjorden, but whether or not this is the case is sensitive to the definition of mAIW and only significant (p < 0.05) when using the broader definition of mAIW. This suggests that NO3 loss is not specifically related to processes occurring within the cavity and likely reflects biological drawdown of NO3 beyond the ice-tongue at stations downstream of S1 (Supplementary Figure 1D). For PO4, the changes between mAIW and AIW are more evident than for NO3 as a statistically significant loss (p < 0.05) of PO4 is determined when comparing AIW with mAIW outflow by any definition. This may therefore partially reflect a process occurring under the ice cavity as the difference is already evident at the ice-tongue. Some low PO4 concentrations are evident at S1 when considering the section S1 to S3 and this is similar to observations in Sermilik Fjord (East Greenland) 8 . For Si(OH)4, there is no significant difference between any of the defined water masses (ANOVA, p > 0.9).