Bioavailable iron produced through benthic cycling in 1 glaciated Arctic fjords (Svalbard)

29 The Arctic has the highest warming rates world-wide. Glaciated fjord ecosystems, which are known hotspots of 30 carbon cycling and burial, are predicted to be extremely sensitive to this warming. Glaciers are important 31 sources of iron, an essential nutrient for phytoplankton, to high-latitude marine ecosystems. However, up to 32 95% of the glacially-sourced iron settles in sediments close to the glacial source. We found that only 0.6-12% of 33 the total glacially-sourced iron is potentially bioavailable. Our results also show that biogeochemical cycling in 34 fjord sediments converts the unreactive glacial iron into more reactive and bioavailable phases, leading to an 35 up to 9-fold increase in the amount of potentially bioavailable iron. Arctic fjord sediments therefore likely are 36 an important source of bioavailable iron. However, once glaciers retreat onto land, the flux of iron from 37 sediments into the water column is reduced, such that glacial retreat could exacerbate iron limitation in polar 38 oceans.

The parameters determined in AFeR extractions enable a detailed characterization of not only the iron reactivity, 130 but also the composition of the iron mineral pool. The heterogeneity parameter quantifies the heterogeneity of 131 reactivities in iron minerals extracted by ascorbate. This parameter can also be thought of as the diversity of 132 ascorbate-extractable iron minerals with different reactivities present. The highest heterogeneity was found for 133 particulates from the plume at the head of Kongsfjorden (1.83±0.6, Figure 2d), indicating heterogeneity in the 134 mineral composition with a range of corresponding reactivities. The lowest heterogeneity was found for 135 particulates from meltwater rivers in Kongsfjorden (0.99±0.1, Figure 2d), indicating that all iron that could be 136 extracted by ascorbate had a relatively similar reactivity and likely homogeneous iron mineral composition. 137 These results indicate that glacial iron transported by proglacial rivers gets sorted or even chemically or physically 138 modified, such that a uniform type of reactive iron mineral is supplied to Kongsfjorden. The initial rate parameter 139 determined in AFeR extractions is the best measure of biological availability and iron reducibility 26 , as it takes 140 into account the amount (M(0)) and reactivity (apparent rate) of FeR. Particulates collected from icebergs had the 141 lowest initial rate (reducibility) of all glacial sources in Kongsfjorden (1.1*10 -3 ±0.8*10 -3 µmol gdw -1 s -1 , Figure 2b). 142 The highest reducibility of glacial source material was found in the Kongsfjorden glacial plume, which had an 143 800% higher initial rate compared to the average of the icebergs (8.8*10 -3 ±2.9*10 -3 µmol gdw -1 s -1 , Figure 2, Table  144 S1). This highlights that glacial meltwater emanating as the plume in front of Kronebreen, contains FeR that was 145 produced by subglacial weathering. On the other hand, icebergs contain iron which has aged and become less 146 reactive while transported in glacial ice until delivered to the fjord through iceberg calving 18 . Even the highest 147 reducibility of the Kongsfjorden glacial source samples was ten times lower than the previously highest reported 148 values of Kongsfjorden sediment 26 (Figure 2). 149 Particulates collected in the plume of the meltwater river at the head of Dicksonfjorden (Dicksonelva, Figure 1) 150 had only ~20% the amount of FeR that was found in the Kongsfjorden plumes and meltwater rivers. The 151 amount of FeR in Dicksonelva particulates (5.95 µmol gdw -1 , Table S1, Figure 2a (Table S1), showing that the FeR in the Dicksonelva sample has a broader range of reactivities 157 compared to the meltwater rivers in Kongsfjorden. Dicksonelva is very different from the meltwater rivers in 158 7 transport and/or production of FeR. Previous studies have concluded that sediment transport in meltwater 160 rivers will transform minerals into more reactive phases due to increased weathering 58 . This does not seem to 161 hold true for Dicksonelva. 162 The differences in FeR amount and reducibility that we found for particulates from glacial sources in 163 Kongsfjorden and Dicksonfjorden contain a paucity of FeR, independent of glacial regime or source type. Still, 164 there were differences in the reactivity, heterogeneity, and amount of FeR delivered by the different types of 165 glacial sources, which add to predicted effects of glacial retreat with the potential to impact biogeochemical 166 cycles such as the linked iron and carbon cycles within the downstream fjord sediments 43,44,46 . 167 The reactivity and spatial distribution of FeR in Kongsfjorden sediment. The amount and reducibility of FeR at 168 the fjord head (KF1; Figure 1) was the lowest (Figure 3, Table S3) of all surface sediment samples within the 169 Kongsfjorden transects. In fact, the amount and reducibility of FeR at KF1 are similar to the average of 170 Kongsfjorden glacial sources and implies that there is little processing of iron upon sedimentation at the head 171 of the fjord (Figure 2a and b). However, the amount and reducibility of FeR in surface sediment in Kongsfjorden 172 increased by 9-fold and 19-fold, respectively, at the station furthest away from the fjord head (KFa7; Figure 2a  173 and b, Figure 3, Table S3). A similar increase in FeR amount and reducibility over distance was found in the 174 northern transect (KF1 to KFb5) of Kongsfjorden ( Figure 3). These increases are exponential as seen from the 175 linear increase in the semi-log-plot ( Figure 3) and an R 2 of 0.96 and 0.94 for the transects going towards KFa7 176 and KFb5, respectively, when fitting an exponential model through the data (Table S4). Further, time-course 177 extractions using a microbial pure culture (MFeR 26 ) showed even more pronounced differences in the 178 reducibility of FeR at the surface of station KFa7 and KFb5 compared to all the sources ( Figure 3, Table S5). 179 These increases are either produced by preferential transport of the smallest, most reactive particles or by 180 processing of the iron upon sedimentation. 181 AFeR extractions showed that FeR in the transects became more heterogeneous over the first few km distance, 182 likely due to the glacial sources containing FeR with different reactivities and authigenic production of reactive 183 Fe within the sediment. Further out in the fjord, FeR became more homogeneous again, reaching values even 184 lower than at KF1, indicating the presence of a uniform pool of highly reactive iron (Table S3, Figure S4). The 185 increase in amount and reducibility as well as the changes in the heterogeneity of FeR with increasing distance 186 from the fjord head imply that significant processing of iron occurred after sedimentation at stations further 187 away from the fjord head, likely through microbial dissimilatory iron reduction or interactions with sulfide 30,58 .
It seems as if a homogenous pool of highly reactive FeR is produced in the surface sediments as distance from 189 the glacial source increases (Figure 3, S5-7). 190 These results are consistent with 57 Fe Mössbauer spectroscopy, which showed that the relative abundance of 191 hematite in KF1 was 21.1±1.3%, similar to the Kronebreen plume. The abundance of hematite in the surface 192 sediment of KFa7 (10.3±2.1%) was only half that of KF1 ( Figure S1, Table S2). This distribution of iron minerals 193 with different crystallinities could be caused by the transport of the finest and most reactive particles to the 194 more distant stations, which would also explain the higher iron reactivity that we measured in AFeR 195 extractions. However, the reducibility of reactive iron in KF7 surface sediment is higher than any value 196 measured in the plume, and also notably higher (by 860%) than the average of all glacial sources (Figure 2b, 197 Table S1, S3). Taken together, these data indicate that the abundant iron mineral species were more 198 dominated by less crystalline, more reactive iron phases further from the head of the fjord and that they might 199 be authigenic. 200 The impact of contrasting catchment geology on the spatial distribution of FeR. Lilliehöökfjorden possesses 201 differing catchment geology than Kongsfjorden, yet the same increase in FeR amount and reducibility over 202 distance from the fjord head was found, reaching a maximum of 89 µmol gdw -1 at LF8, which is an increase of 203 390% within the 23 km transect. (Figure 3). The reducibility also increased by 430% in our Lilliehöökfjorden 204 transect ( Figure 3, Table S3). The Lilliehöökfjorden FeR pool develops in a manner similar to the two transects 205 in Kongsfjorden where a diverse pool of FeR becomes progressively more uniform in composition with distance 206 from the fjord head ( Figure S4). The MFeR extractions detected a similar increasing trend in FeR amount and 207 reducibility in Lilliehöökfjorden ( Figure 3, Table S5). The trend of increasing FeR amount and reducibility is 208 interrupted where Möllerfjorden and Lilliehöökfjorden merge (between LF6 and LF7; Figure 1, 3), with 209 Möllerfjorden likely supplying less reactive iron to the sediments. Also changes in pore water Fe and Mn were 210 found where these two fjords merge, with maximum dissolved Fe(II) (dFe(II)) concentrations decreasing and 211 maximum dissolved Mn (dMn) concentrations increasing at station LF6, compared to the stations closer to the 212 fjord head ( Figure S8). 213 No hematite could be identified by 57 Fe Mössbauer spectroscopy in Lilliehöökfjorden samples and the iron 214 mineral composition was different from Kongsfjorden as expected from the contrasting bedrock and sediment 215 color ( Figure 1, Figure S9, Table S2). Collected spectra were similar for LF1 and LF5, with a higher proportion of 216 away from the fjord head and mirrors the trend found in AFeR extractions. Consequently, the oxidation of Fe (II)  219 to Fe(III), by biotic or abiotic processes 31 , appears to be important for the production of FeR in 220 Lilliehöökfjorden. The results from Mössbauer spectroscopy helped to support findings from the AFeR 221 extractions but Mössbauer spectroscopy alone did not capture this distinct change in the amount and reactivity 222 of FeR over distance from the glacier. This highlights the value of direct quantification of FeR amount and 223 reactivity in AFeR and MFeR extractions. 224 In Kongsfjorden and Lilliehöökfjorden the amount and reducibility of FeR increased by up to 50 and 166%, 225 respectively, per km of distance from the fjord head independent of catchment geology (Figure 3 and S4). This 226 pattern of increasing FeR amount with distance from the fjord head was also observed in two fjords in 227 southwestern Svalbard 58 . Van Mijenfjorden and Van Kuelenfjorden in southern Spitsbergen, Svalbard drain 228 different bedrock assemblages and reinforce the widespread nature of these FeR patterns in fjord sediments. 229 The increases we observe in Kongsfjorden are statistically significant over the entire length of the transects 230 (Table S4). For the Lilliehöökforden transect, the flattening off after station LF6, causes the increase in the 231 amount of FeR over distance to have low significance and the increase in reactivity over distance to have no 232 significance. If only the data until LF5 are included in the analysis, the increase in the amount and reactivity 233 become statistically highly significant (Table S4). This again supports our hypothesis that Möllerfjorden outputs 234 impact the Lilliehöökforden transect and represents the sensitivity of fjord sediments to nearby marine-235 terminating glaciers. In conclusion, the increases of reactive Fe in fjord sediments toward the fjord mouth, 236 irrespective of catchment of geology, reveals that there is a gradual transformation of unreactive glacially-237 derived iron minerals towards higher reactivity and bioavailability. 238 FeR production through benthic cycling in fjord sediments. We propose that the main driving force 239 transforming the unreactive glacially-derived iron into FeR is benthic cycling through an interplay of abiotic and The strong gradient in FeR from fjord head to mouth is controlled by steep gradients in hydrology, biology, and 252 geochemistry due to inputs of glacial material at the fjord head and the marine influence at the fjord mouth 61-253 63 . At the fjord head, high sedimentation rates of detrital material and low primary productivity within a thin 254 photic zone 20,64 , lead to sediment with low TOC amount and high C:N ratio of up to 70 (Table S6, Figure 7). 255 Towards the fjord mouth TOC contents gradually increased, while C:N ratios decreased and approached a more Greenlandic fjords 73 . The C:N ratio is a measure of the quality of the organic matter and how readily it can be 264 respired by benthic microbial communities. Therefore, sediments close to glaciers can sustain only moderate 265 activity of Fe(III)-or sulfate-reduction due to the low amount and refractory characteristics of the organic 266 carbon (Figure 7). Further from the glacier, the sedimentation rate of inorganic detrital material decreases and 267 primary productivity in the water column increases, producing sediment with a higher TOC amount and lower 268 C:N 62,74,75 . This creates favorable conditions for Fe-cycling, as the organic carbon can support higher rates of 269 microbial Fe(III) reduction and sulfate reduction, both leading to the production of dFe(II). 270 Based on these results for TOC and C:N, we expect SRR to increase concurrent with the increase in TOC and the 271 decrease in C:N. However, depth-integrated rates of SRR in Kongsfjorden and Lilliehöökfjorden at first increase, 272 but then decrease further out in the fjord (Figure 7). This is likely caused by the consecutive increase in FeR 273 along the transect, enabling Fe-reducers to compete favorably with sulfate-reducers (for more detailed 274 discussion of SRR in relation to TOC and C:N, see supplemental information). Besides the increased activity of 275 benthic iron cycling, the lower sedimentation rates in the outer part of fjords 62,74 lead to a more abundant and 276 active benthic fauna 76 , further intensifying benthic cycling 77,78 , and increased time for iron to be repeatedly 277 cycled before it gets buried deeper in the sediment (Figure 8a). Sedimentation gradients in Arctic fjords caused by particle transport in freshwater lenses could explain the 279 observed FeR gradients by carrying the finest and most reactive grains furthest 20 . However, we did not find 280 evidence for long-distance transport of the finest and most reactive glacial iron in generating the observed 281 gradient of FeR. Over 95% of the grain size distributions from surface sediment samples recovered along the 282 transects are characterized by silt and clay (< 63 µm or 4 ) ( Figure S10). We found no systematic relationship 283 between the percent of fine-grained material and the distance from the fjord head ( Figure S10). Our results 284 corroborate other observations that the majority of the suspended material in fjords supplied from marine or 285 land-terminating glaciers does not reach further than 7 km from the source and that flocculation causes also 286 suspended colloidal and nano-particulate material to quickly settle from the water column 20,79 . For more 287 detailed interpretation of the grain size analysis, see supporting information. Thus, the increase of FeR over 288 distance cannot be explained as a function of the transport of small particles containing the most reactive Fe. 289 The increase in reactivity through benthic iron cycling, also called "rejuvenation", has also been shown to be an 290 important pathway for bioavailable iron production in continental margin sediments 77,80,81 . The steep gradients 291 in FeR seem to be unique for glaciated fjord systems and we conclude that in glaciated arctic fjords the 292 increasing intensity of benthic iron cycling, due to increased amount of labile organic carbon, and time before 293 burial produce the observed gradients of FeR from fjord head to mouth. 294

295
The impact of glacial retreat on FeR distribution and Fe-export to the water column. The general pattern of 296 increasing amount and reducibility of FeR with distance from the fjord head is also observed in Dicksonfjorden, 297 which is fed only by land-terminating glaciers (Figure 1). The amount of FeR was only about half of what was 298 found in Kongsfjorden and Lilliehöökfjorden at similar distances from the fjord head ( Figure 3a). However, the 299 reactivity of FeR was higher in Dicksonfjorden, such that the reducibility (initial rate) was within the same range 300 as for Kongsfjorden and Lilliehöökfjorden (Figure 3b and S4, Table S3). We propose that, similar to 301 Kongsfjorden and Lillihöökfjorden, benthic cycling is responsible for the increase in the amount and reducibility 302 of FeR from head to mouth in Dicksonfjorden. In contrast to Lilliehöökfjorden and Kongsfjorden, where the 303 amount of FeR (M(0)) peaked at the sediment surface, the maximum concentration of FeR was never found at 304 the sediment surface in Dicksonfjorden ( Figure 5, Figure S4 and S11-S17, Table S3). At station DF1, the amount 305 of FeR did not change significantly over sediment depth and at station DF3 and DF5 the maximum amount of 306 FeR was found at 3-4 and 6-8 cm sediment depth with 38.3 and 70.7 µmol gdw -1 , respectively (Figure 5, Figure  307 of glacial regime. However, the specific depth-distribution of FeR that we found in Dicksonfjorden might impact 309 the potential for FeR release to the water column. 310 The subsurface peaks of FeR in Dicksonfjorden are likely caused by deeper penetration of oxidants (such as 311 oxygen, nitrate or Mn(IV)-oxides; Table S7) and provide further evidence that the FeR is authigenic and not a 312 function of the fine grained and reactive material getting transported furthest. The presence of oxidants is 313 evident from the absence of dissimilatory sulfate reduction, dFe(II), and dMn just above the depth where the 314 maximum amount of FeR was found at station DF3 and DF5. This indicates that dFe(II) and dMn were oxidized 315 within the top 3-5 cm of the sediment and could not reach the sediment surface at these stations. At station 316 DF1, low SRR (<1 nmol cm -3 d -1 ) and dMn were found within the upper 4 cm of the sediment, but no dFe(II) was 317 detected ( Figure 6). Again, Dicksonfjorden is in contrast to Kongsfjorden and Lilliehöökfjorden where sediment 318 sulfate reduction was active and dFe(II) and/or dMn, could be detected within the upper 2 cm of the sediment 319 at all stations ( Figure 5 and 6). The deeper penetration of oxidants in Dicksonfjorden sediments is likely caused 320 by the generally lower primary productivity in fjords with land-terminating glaciers as they lack glacial 321 upwelling, which is known to entrain nutrient-rich bottom water and transport it up to the photic zone where it 322 supports primary productivity 44-47 . The smaller increase in TOC is likely due to diminished primary productivity 323 in the Dicksonfjorden water column and leads to a smaller increase in TOC content of the sediment with 324 distance from the head of the fjord compared to Kongsfjorden and Lilliehöökfjorden (Figure 7a, Table S6). The 325 lower TOC content also led to depth-integrated SRR that stayed low over the entire transect (Figure 7b). 326 Consequently, the sediment microbial community is less active, oxidants penetrate deeper into the sediment 327 and prevent dFe(II) from reaching the sediment surface to fuel authigenic Fe(III) production or from diffusing 328 into the overlying water column . 329 The production of authigenic, reactive Fe(III) at the sediment-water interface 82,83 , and the diffusion of dFe(II) 330 across the sediment-water interface 84 , have been shown to be important factors for Fe-transfer into the water 331 column. We propose that there is a decreased potential for Fe-flux to the water column in Dicksonfjorden 332 compared to Kongsfjorden or Lilliehöökfjorden because authigenic FeR is produced at several cm sediment 333 depth and dFe(II) did not reach the sediment surface (Figure 8). The deepening of the iron cycle is an additional 334 negative feedback mechanism on primary productivity in high-latitude marine systems when glaciers retreat 335 onto land. While Kongsfjorden and Lilliehöökfjorden are potentially important sources of FeR to the water 336 column, we conclude that when glaciers retreat onto land, benthic iron cycling is restricted to deeper sediment 337 layers and reduces the source strength of FeR or dFe(II) from the sediment to the overlying water column and, ultimately, the open ocean ( Figure 8b). As the iron and carbon cycles are intimately linked 44,46 , not only by 339 primary production but also by carbon remineralization, glacial retreat may impact both the biological carbon 340 pump and the function of sediments as carbon sinks 11 . 341

342
To improve our understanding of iron cycling in the ocean and production of essential bioavailable iron for 343 primary production, it is fundamental to know the sources and fate of iron along the continental margins. We 344 show that the amount and reactivity of FeR in glacially-derived material is low. While fjords were previously 345 expected to reduce glacial iron delivery to the ocean 18 , we show that fjord sediments are a biogeochemically 346 active interface in which glacially-sourced, unreactive iron is transformed into potentially bioavailable FeR 347 through benthic cycling. Our results show that sediments at the fjord mouth contain bioavailable FeR that 348 could be a source of iron to the marine shelf and open ocean environments, thereby promoting primary 349 productivity. Moreover, the study highlights the impact of glacial retreat on biogeochemical processes in fjord 350 sediments that may reduce their ability to serve as a source of iron for primary production in the Arctic Ocean. 351  (Table S8, Figure 1). Sediment was retrieved with a Haps corer 85 and sub-sampled aboard the ship using 359 2.8 cm (for SRR measurements) or 6 cm (for pore water and solid-phase geochemistry) diameter acrylic coring 360 tubes. Sediment was stored at 4°C until further processing within 2 days after sampling. 361

Material and Methods
Glacial source material was sampled in Kongsfjorden in June and July 2017, and July 2018. In total we collected 362 7 pieces from individual icebergs with embedded sediment (Figure S16), 4 samples of glacial plume water in 363 front of the KB/KV calving front, and 6 samples of meltwater from rivers along the southern and northern shore 364 of Kongsfjorden (Table S9, Figure 1). The material from the meltwater rivers was collected directly at their 365 mouth before entering the fjord. Material from the Dicksonelva plume at the head of Dicksonfjorden was The distances of the stations relative to the main glacial source was determined by geospatial analysis using 368 qGIS (v. 3.10). We used the imagery seen in Figure 1 to measure the distance from the glacier terminus to the 369 GPS determined sample point. The imagery was collected ~1 month after our samples were collected and 370 represented the glacial terminus at the time of sample collection. 371 Processing and subsampling of sediment cores. The 6-cm wide subcores were sliced in an anoxic glove bag 372 under N2 atmosphere ( < 0.5% atmospheric O2 concentration, checked with a trace-range optical oxygen sensor 373 TROXROB10 connected to a Firesting O2 -meter, Pyroscience). The cores were processed outside the laboratory 374 at ambient temperature (4-8 °C) using the technique of Keimowitz et al. 86 with slight modifications as 375 described in detail by Michaud et al. 30 . All plasticware used for subsampling was made anoxic by placing the 376 plasticware and an oxygen scrubber (AnaeroGen, ThermoFischer) in a heat-sealed gas-tight plastic bag (Escal 377 Neo, high gas barrier bag, Mitsubishi Gas Chemical Co., Inc.) for at least 24 h. The sediment cores were sliced 378 into 1-3 cm sections down to a depth of 13 cm. After each section was homogenized, subsamples of sediment 379 were taken for (i) Fe extractions, (ii) determination of porosity, water amount, TOC and TN, and (iii) pore water 380 geochemistry. The subsamples for Fe extractions, porosity, water amount, TOC and TN were immediately 381 frozen at -20°C. After closing the centrifuge tubes inside the glove bag under N2 atmosphere, the pore water 382 samples were centrifuged for 15 min at 4000 rpm outside the glove bag. The tubes were immediately returned 383 to the glove bag after centrifugation, and the supernatant was filtered by centrifugation (5 minutes, 14000 384 rpm) in spin filters (0.45 µm nylon membrane, Norgen Biotek). For dissolved Fe and Mn analysis an aliquot of 385 the filtrate was acidified (HCl, 1 M final concentration) and the remaining was used for sulfate quantification. 386 All pore water samples were stored at 4°C in the dark until analysis. 387 388 Processing and subsampling of glacial source material: Particulate material was extracted from plume 389 and river-water by centrifugation (15 min., 3000xg). Samples of sediment-loaded icebergs were first rinsed with 390 milliQ water on the exposed surfaces, then molten inside a clean plastic bag before centrifugation. In all cases, 391 the pellets were collected and frozen at -20°C until analysis. 392 393 Pore water chemistry: Dissolved Fe(II) and Mn in the pore water were measured spectrophotometrically by 394 the ferrozine assay 87 and the formaldoxime assay 88,89 , respectively. Both assays were performed in 96-well 395 plates and the absorbance was measured at 562 nm for Fe(II) and 450 nm for Mn with a plate reader 396 (FLUOstarOmega, BMG Labtech). The formaldoxime assay was adapted, according to Otte 90 , to exclude interference from the high Fe 2+ amount in the pore water 91 . Sulfate concentration in pore water was 398 quantified on 1:100 diluted samples using suppressed ion chromatography (Dionex). Some of the pore water 399 chemistry data is already published 26 . For which stations this is the case is stated in Table S8. 400 Sulfate reduction rate measurements: Sulfate reduction rates (SRR, nmol cm -3 d -1 ) were determined by 401 injecting 35 SO4 2into intact, 20 -25 cm long , 2.8 cm diameter sediment cores 92 . Fifty kBq of carrier-free 35 S-402 SO4 2was injected at 1 cm depth intervals through ports sealed with polyurethane-based elastic sealant 403 (Sikaflex ® -11FC + , Sika) 93 . After 10 to 14 hours of incubation at near in-situ temperature (2°C), the cores were 404 sliced in 1 cm sections, which were added immediately to 10 ml of 10% zinc acetate and homogenized by 405 vortexing. The zinc acetate-fixed samples were stored at -20°C until analysis. The cold chromium method 93 was 406 used to separate radiolabeled total reduced inorganic sulfur (TRIS) from the sample and the evolved H2S was 407 trapped as Zn 35 S in 5 mL of 5% zinc acetate solution. Scintillation counting was used to analyze the radioactivity 408 in the sulfate and TRIS pools and sulfate reduction rates were calculated according to Jørgensen 92 . To 409 determine the water amount and porosity of the sediment, required for calculation of SRR, the weight loss of a 410 known volume of sediment after drying to constant weight at 105°C was determined. Some of the SRR data is 411 already published in a recently accepted manuscript 98 . For which stations this is the case is stated in Table S8