The Effects of Glacial Cover on Riverine Silicon Isotope Compositions in Chilean Patagonia

Proglacial rivers have been shown to have distinctive silicon (Si) isotope compositions, providing new insights into the mechanisms controlling Si cycling in the subglacial environment and suggesting terrestrial Si isotope exports may have varied between glacial and interglacial periods. However, Si isotope data are currently limited to a small number of glacial systems in the northern hemisphere, and it is unclear how compositions might vary across a spectrum of glacial influence. Using Chilean Patagonia as a unique natural laboratory, we present Si isotope compositions of 0.45 μm filtered (fSi), 0.02 μm filtered (DSi), and reactive amorphous (ASi) fractions from 40 river catchments with variable glacial cover and explore the key controls on Si cycling. The 0.45 μm filtered glacier-fed river samples displayed isotopically light compositions and a positive linear correlation with upstream glacial cover. This relationship was controlled by the inclusion of an isotopically light colloidal-nanoparticulate (0.02–0.45 μm) silicate phase that was only present in glacier-fed rivers and dominated Si budgets in these catchments. This phase was predominately composed of feldspars and its lability in seawater is uncertain, representing a significant unknown in resolving glacial Si isotope exports from this region. When the colloidal-nanoparticulates were removed from solution by ultra-filtration, the resultant DSi isotope compositions of glacier-fed catchments were not isotopically distinct from some non-glacial rivers and exhibited no clear relationship with glacial cover. The colloidal-nanoparticulate concentration of other weathering-sensitive elements (Li, Mg, Ba, Sr) also showed a linear relationship with glacial cover, suggesting that their isotopic compositions could be affected in a similar manner. These findings highlight the benefit of size-fractionated sampling and the need for more research to understand the lability of colloidal-nanoparticulate species, especially in glacier-fed rivers. Finally, we explore the controls on river ASi isotope compositions and show how including these reactive particulate phases is critical to quantifying terrestrial Si isotope budgets, both in Patagonia and other global regions.


INTRODUCTION
The silicon (Si) cycle is intricately linked to the carbon cycle, with climatic controls on silicate weathering regulating the drawdown of atmospheric CO 2 on geological timescales (Urey, 1952;Berner et al., 1983;Berner and Caldeira, 1997). Additionally, Si is an essential nutrient for many photosynthetic organisms including diatoms, which account for up to 70% of primary productivity in the ocean, form the base of food-chains and maintain the efficiency of the biological carbon pump Treguer and De La Rocha, 2013). Rivers deliver roughly 80% of the terrestrial Si flux to the oceans, acting as the critical link between continental and marine sub-cycles and affecting the rate of siliceous primary productivity in the ocean Treguer and De La Rocha, 2013). As such, understanding the controls on riverine Si exports is critical for our understanding of climate feedback mechanisms, both during the past and with future change. Si isotope compositions are increasingly used to understand the Si cycle (Frings et al., 2016;Sutton et al., 2018) and as a valuable palaeo-oceanographic proxy (De La Rocha et al., 1998;Hendry and Robinson, 2012;Hendry and Brzezinski, 2014). However, there are still gaps in our knowledge, particularly relating to how changing glacial cover could impact the Si cycle and global Si isotope budgets Sutton et al., 2018).
Until recently, glaciers and ice sheets were not considered to be active components of the Si cycle. The dissolved Si (DSi) concentrations of glacier-fed rivers tend to be much lower than non-glacial rivers, which is generally interpreted to reflect suppressed chemical weathering of silicate minerals in near-freezing subglacial environments (Tranter et al., 2002;Anderson, 2007). However, research from the Greenland Ice Sheet has shown that glacial meltwaters also contain high concentrations of a labile amorphous Si phase (ASi) bound to riverine suspended sediments (Hawkings et al., 2017;Hatton et al., 2019a). Concentrations of ASi from Leverett Glacier were roughly 40 times higher than the DSi phase, suggesting that glacial meltwaters could represent a significant missing component of terrestrial Si exports to the ocean (Hawkings et al., 2017). These ASi phases have been shown to readily dissolve in seawater to form bioavailable DSi, which could fuel siliceous primary productivity in downstream ecosystems (Frings, 2017;Hawkings et al., 2017). Similar ASi components have also been reported from other proglacial rivers in the Arctic and sub-Arctic (Hatton et al., 2019b), although the controls on concentrations and glacial ASi formation mechanisms are still the subject of active debate (Blackburn et al., 2019;Hatton et al., 2019b).
Biogeochemical reactions involving Si can cause subtle but measurable shifts away from the natural abundance ratios of the three stable isotopes of Si ( 28 Si, 29 Si, and 30 Si; Ding et al., 2005). Variations in these ratios function as a tracer for processes operating within the Si cycle. Riverine Si isotope compositions integrate basin-wide signals and reflect a dynamic balance between upstream catchment characteristics, silicate weathering processes, variable primary productivity by diatoms and other siliceous organisms, and inputs from Si-rich plant phytoliths (Frings et al., 2016;Sutton et al., 2018). During the formation of secondary weathering products (e.g., clays) and biogenic Si, solid phase material preferentially incorporates the lighter 28 Si isotope, leaving the residual river water enriched in the heavier 30 Si isotope relative to the parent rock (Ziegler et al., 2005;Georg et al., 2006;Frings et al., 2016). Conversely, incongruent dissolution of solid Si phases leads to the preferential release of the lighter isotope back into the river water (Cardinal et al., 2010). Si isotope compositions have been measured for many of the world's largest rivers, where signals have been used to infer the rates and congruency of silicate weathering (Georg et al., 2006;Cardinal et al., 2010;Opfergelt and Delmelle, 2012;Opfergelt et al., 2013), the degree of biological uptake (Hughes et al., 2011;Panizzo et al., 2017;Sun et al., 2018), anthropogenic impacts on the Si cycle (Hughes et al., 2012;Cockerton et al., 2013;Delvaux et al., 2013), and the terrestrial Si isotope composition exported to the ocean (Opfergelt et al., 2013;Frings et al., 2016;Hawkings et al., 2018;Sun et al., 2018).
Glacial processes also appear to affect the Si isotope composition of downstream river waters. Proglacial rivers in Iceland and Greenland have been shown to have distinct, isotopically light DSi compared to non-glacial river systems (Georg et al., 2007;Opfergelt et al., 2013;Hawkings et al., 2018;Hatton et al., 2019a). Such light DSi isotope compositions are interpreted to reflect extensive dissolution of ASi and other secondary weathering products in undersaturated, high pH meltwaters at the base of glaciers Hatton et al., 2019a). These findings have brought into question the long-held assumption that terrestrial Si isotope exports to the ocean have remained stable over glacial-interglacial cycles, with implications for interpreting Si isotope paleoclimate proxies (Opfergelt et al., 2013;Hawkings et al., 2018). However, compiled Si isotope data from proglacial rivers across the Arctic and sub-Arctic show complex variability in Si isotope compositions exported from different glacial systems (Hatton et al., 2019b). At present, glacial Si isotope data are limited to glacier-fed rivers in high-latitude environments of the northern hemisphere, and it is unknown how compositions could vary in other regions and across a spectrum of glacial cover.
Furthermore, the Si isotope composition of ASi phases is generally overlooked when quantifying riverine Si isotope budgets. Estimates of riverine Si isotope compositions are usually based on 0.45 or 0.22 µm filtered Si and do not include reactive ASi associated with the suspended particulate material (SPM; > 0.45 µm) load. Since ASi phases are estimated to account for between ∼23 and 58% of reactive riverine Si exports (Conley, 1997;Dürr et al., 2011;Frings et al., 2014), excluding these phases likely represents an incomplete picture of the labile flux of Si isotopes and could hinder our understanding of the terrestrial Si cycle. In proglacial river systems, ASi phases have been shown to have isotopically light Si compositions, although the controls and potential variability are not fully understood (Hatton et al., 2019b). In particular, measurements of the isotopic composition of riverine ASi are currently only available for a handful of high-latitude glacier-fed rivers Hatton et al., 2019a,b) and have never been reported from non-glacial river systems.
Chilean Patagonia is a unique and under-studied natural laboratory to investigate Si cycling processes for several reasons. First, this region has a wide spectrum of glacial cover, with rivers spanning completely deglaciated land, catchments with small mountain top glaciers and rivers draining from the large glacial systems of the Northern and Southern Patagonian Ice Fields. Second, it is a near-pristine environment with very low population density and minimal agriculture. As such, Si cycling processes are not overprinted or affected by anthropogenic signals, and we can observe the natural controls on riverine isotopic compositions. Finally, Patagonia is rapidly deglaciating and experiencing dramatic landscape change, with evidence that the Patagonian Ice Fields shrank by roughly 30 km 2 per year between 1986 and 2011 (Davies and Glasser, 2012) and 1,226 new glacial lakes formed between 1986 and 2016 (Wilson et al., 2018). Therefore, our data acts as a valuable baseline to understand future change to Si cycling processes across this region.
Here, we present data from 40 rivers in Chilean Patagonia and explore the key controls on the isotopic composition of dissolved (<0.02 µm), colloidal-nanoparticulate (0.02-0.45 µm) and amorphous (>0.45 µm) Si phases. We discuss the relative lability of each component and highlight how size-fractionated analyses are needed to understand the controls on riverine isotope signals. We also explore the dissolved and colloidalnanoparticulate concentrations of several additional weatheringsensitive elements (Li, Sr, Ba, Mg) to infer if similar controls are likely to influence their size-partitioning and isotopic compositions. Finally, by combining the relative proportions, isotopic composition and potential lability of each Si phase, we test the hypothesis that glacier-fed rivers in Patagonia have distinct Si isotope compositions compared to non-glacial rivers.

Sampling Campaign
River water samples were collected from 40 rivers located between 42 and 48 • S in Chilean Patagonia during the austral summer in January 2017 (Figure 1). The pH, electrical FIGURE 1 | Map showing river sampling locations in Patagonia. Yellow dots show river sampling locations collected in January 2017. The red square and top inset show the location and catchment area of the non-glacial time-series location (PTR01-R4, Río Lloncochaigua) and the blue square and bottom inset show the location and catchment area of the proglacial time-series location (PFC01-SA, Río Huemules-Steffen Glacier). Upstream catchment areas are outlined in black and were modeled using Q-GIS and Whitebox GAT (Lindsay, 2016). conductivity and water temperature were measured using handheld sensors. Water samples were collected from fastflowing sections of the river and upstream of settlements. Aliquots for major ions, DSi and Si isotope analysis were filtered immediately through 0.45 µm pore-size Whatman GD/XP polypropylene (PP) syringe filters and stored in acid-cleaned high-density polyethene (HDPE) bottles. Trace metal samples were collected by sequentially filtering river water through 0.45 µm Whatman GD/XP (PP) and 0.02 µm Whatman Anotop 25 syringe filters. These pore sizes were chosen to investigate differences between operationally defined filterable (<0.45 µm), colloidal-nanoparticulate (0.02-0.45 µm) and truly dissolved (<0.02 µm) concentrations, following protocols defined by Shiller (2003) and used in multiple glacial river studies (Schroth et al., 2011;Hawkings et al., 2014;Raiswell et al., 2018). Samples for Si isotope analysis were subsequently refiltered through 0.02 µm Whatman Anotop 25 syringe filters to investigate differences between dissolved and colloidal-nanoparticulate Si isotope signatures. SPM samples for determination of ASi concentrations and isotopic composition were collected by filtering 1-5 L of river water through pre-weighed 47 mm diameter Millipore 0.45 µm polyethersulfone (PES) membrane filters using an acid-washed Nalgene filter tower. The exact volume of water that passed through the membrane was recorded to calculate SPM concentrations, and the filter papers were stored in the dark at 4 • C until analysis. The Si, iron (Fe), and SPM concentration data from these samples are also reported in Pryer et al. (2020).
To constrain temporal and seasonal variability in riverine Si isotope composition, a glacial and non-glacial end-member catchment were sampled repeatedly over multiple periods of 2 weeks to 5 months from January 2017 until November 2018, covering both the austral summer and winter at each location. The glacial site was the Río Huemules, draining from Steffen Glacier at the southern edge of the Northern Patagonian Icefield (NPI). The non-glacial site was the Río Lloncochaigua, located on the eastern edge of the Comau Fjord (Figure 1).

Geospatial Analysis
Geospatial analysis of this region was conducted using a combination of Q-GIS and Whitebox-GAT (Lindsay, 2016). Upstream catchment areas were calculated using the NASA Shuttle Radar Topography Mission (SRTM) 3 arc-second digital elevation model (DEM). The glacial cover was calculated as a percentage of each catchment area using the Randolph Glaciers Inventory (RGI Consortium, 2017) and is used as a proxy to understand the influence of variable glaciation across this region. We define glacier-fed systems as all catchments with >20% glacial cover, where the main tributary is a proglacial river. The significance of differences between catchments grouped by glacial cover was analyzed using unpaired, two-tail t-tests assuming unequal variance. Catchment lithology and lake area were calculated using datasets from Hartmann and Moosdorf (2012) and Messager et al. (2016), respectively.

Silicon and Major Ion Concentration Analysis
Truly dissolved Si concentrations (DSi, as silicic acid) were measured colorimetrically using a LaChat 8500 Series 2 flow injection analyser (FIA) (QuikChem Method 31-114-27-1-D) using an external calibration curve that spanned the range of sample concentrations, gravimetrically diluted from a 1,000 ppm Si ISO 17034-certified stock standard (TraceCERT ). Mean accuracy for DSi measurements was 2.6% and mean precision was ±1.2%. Colloidal-nanoparticulate Si (0.02-0.45 µm; CNSi) concentrations were calculated as the difference between 0.02 and 0.45 µm filtered concentrations measured by multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS), with errors of ±5.0%. Concentrations of ASi associated with SPM were measured using the 0.1 M Na 2 CO 3 extraction method adapted from DeMaster (1981). Aliquots of the extraction solutions were analyzed using the same method as for DSi to obtain ASi concentrations in µM, with propagated errors of ±7.5%. Major cation and anion concentrations were measured using a Thermo Scientific TM Dionex TM capillary ICS-5000 ion chromatograph. Mean accuracy was 0.2 ± 2.5% and precision for all ions was better than ±2.5%. Bicarbonate (HCO 3 -) concentrations were estimated using the mass charge deficit (Tranter et al., 2002;Hawkings et al., 2015), with a propagated error of ±3.3%. Major ion concentrations were seasalt-corrected, using values from Millero et al. (2008). Total dissolved solid (TDS) concentrations were calculated as the sum of all major ions and DSi, with a propagated error of ±6.0%.

Trace Metal Concentration Analysis
Dissolved (<0.02 µm) and filterable (<0.45 µm) trace metal concentrations were analyzed using a Thermo Scientific X-SERIES 2 quadrupole ICP-MS with a collision-reaction cell at the National Oceanography Centre, Southampton. Berylium, indium and rhenium were used as internal standards to correct for drift and matrix effects, and external calibration solutions (n = 7) matched the concentration range observed in samples. For trace metals presented in this study, precision and accuracy were always within ± 10% of a multi-element standard, prepared from a gravimetrically weighed mix of NIST-traceable SPEX CertiPrep standards. Colloidal-nanoparticulate (0.02-0.45 µm) concentrations were calculated as the difference between 0.45 µm filterable and dissolved size-fractions (Shiller, 2003).

Silicon Isotope Analysis
Si isotope analyses were carried out at the University of Bristol Isotope Group clean laboratories and mass spectrometer facility. River water samples filtered through 0.02 µm filters were measured for DSi isotope analyses (δ 30 Si DSi ) and samples filtered through 0.45 µm filters were measured for filterable Si compositions (δ 30 Si fSi ). Sample volumes ranged from 0.9 to 20 mL, to target 7.2 µg of Si. Aliquots of filtered river water with Si concentrations <150 µM were pre-concentrated by evaporating to ∼2 mL volume at 90 • C in acid-cleaned Teflon vials, whereas other samples were added directly to the column. To measure the ASi isotopic composition (δ 30 Si ASi ), 10-30 mg of dry SPM was weighed into Teflon vials, before adding 1 mL of 0.2 M NaOH for every 0.1 mg of ASi, using concentrations previously measured using the DeMaster (1981) method described in section 2.3. Samples were refluxed at 100 • C for an hour to extract the reactive ASi phase, neutralized with trace-metal grade 8M HNO 3 , diluted with Milli-Q and passed through 0.22 µm pore-size PES syringe filters to remove sediment particles from the solution. This extraction method compares well with the more established Na 2 CO 3 method, extracting equal or slightly lower concentrations of ASi and giving assurance that more refractory phases of Si were not incorporated into solution . Isotopic analysis of ASi could only be measured for a subset of samples (n = 11), as insufficient particulate material was collected from rivers with low SPM loads.
All samples were added to pre-cleaned BioRad exchange resin columns (AG50W-X12 200-400 mesh) using a method adapted from Georg et al. (2006) and eluted with Milli-Q to give a solution with a concentration of 2 ppm Si. A 1 mL aliquot of each sample was taken for each analysis, doped with 100 µL of 10 ppm magnesium (Mg) standard to correct for mass bias (Cardinal et al., 2003;Hendry and Robinson, 2012) and 100 µL of 0.1 M sulphuric acid to balance matrix differences between samples and standards (Hughes et al., 2011). Si and Mg isotope ratios were measured on a Thermo Scientific Neptune Plus TM MC-ICP-MS using a standard-sample bracketing protocol. International reference standard NBS-28 (RM8546) was used as the bracketing standard, and all isotope results were calculated using δ 30 Si notation (Equation 1), as the deviation of the 30 Si/ 28 Si ratio of each sample from NBS-28 in parts per mille (‰).  (1) Instrument blanks of 0.1 M distilled HCl were always <1% of the intensity of the 28 Si signal of samples, and procedural blanks were always below the limit of detection. Long-term reproducibility was quantified by repeat measurement of two reference standards; LMG08 and Diatomite. The mean measured δ 30 Si value for LMG08 was −3.45‰ (n = 70, σ = 0.04; δ 29 Si = −1.76‰) and the mean δ 30 Si value for Diatomite was +1.23‰ (n = 25, σ = 0.04; δ 29 Si = +0.66‰). The values of these standards are within error of published values that have been calibrated between multiple labs Hendry et al., 2011). A three-isotope plot of all standards and samples measured during this study fit along a straight line with a gradient of 0.517 ± 0.04 (r 2 = 1.0, p < 0.01, n = 526; Supplementary Figure 1), showing mass-dependant fractionation . All samples were measured at least in duplicate and a conservative error of ± 0.08‰ was assigned to all δ 30 Si values, equalling ± 2σ variability of δ 30 Si values from repeat measurements of internal standards and replicate samples. The isotopic composition of the colloidal-nanoparticulate Si phase (δ 30 Si CNSi , 0.02-0.45 µm) was calculated using a mass balance approach to account for the relative proportions and isotopic composition of 0.02 and 0.45 µm filtered Si fractions. Total riverine Si isotope budgets were calculated using a mass balance accounting for the relative proportions and isotope ratios of the DSi and ASi components (δ 30 Si DSi + ASi ), and the DSi and fSi components (δ 30 Si fSi + ASi ). The mean measured δ 30 Si ASi was used as an estimate for river systems where no data were available. Using the mean value had little impact on the total budget compared to analytical uncertainty, since most rivers where δ 30 Si ASi was not measured had very low SPM loads.

Size-Partitioning of Other Elements
The relative proportions of colloidal-nanoparticulate and dissolved phases were also investigated for other weatheringsensitive elements, including lithium (Li), magnesium (Mg), strontium (Sr), and barium (Ba). Similar to Si, all of these elements showed a significant positive relationship, with higher relative proportions of the colloidal-nanoparticulate phases with increasing glacial cover (r 2 = 0.78-0.86, p < 0.01, n = 26; Figures 5A-D). In the rivers with the highest glacial cover, over 70% of 0.45 µm filterable Li (fLi), 55% of filterable Mg (fMg), 35% of filterable Sr (fSr), and 80% of filterable Ba (fBa) was in the colloidal-nanoparticulate size-fraction.
The time-series data collected from the non-glacial river (Río Lloncochaigua; PTR01-R4) showed more variability of DSi concentrations than the glacial site, with values ranging from 21.5 to 58.3 µM (mean = 41.6 ± 9.2 µM, n = 78; Figure 6A), although with no obvious seasonal trends between summer and winter. Despite variable DSi concentrations, the δ 30 Si values of the river water were always within analytical error, with δ 30 Si fSi values ranging from +0.32 to +0.45‰ (mean = +0.37 ± 0.04‰, n = 6; Figure 6B). No differences were observed between δ 30 Si fSi and δ 30 Si DSi of the Río Lloncochaigua, with all values within analytical error.

Colloidal-Nanoparticulate Phases Control Filterable Silicon Isotope Compositions in Glacier-Fed Rivers
A key observation in our dataset was that glacier-fed rivers in Patagonia had low 0.45 µm filtered Si isotope compositions (δ 30 Si fSi ), with values that varied as a linear function of upstream glacial cover (Figure 2C). Similar light Si isotope compositions have been reported from proglacial rivers in Greenland, where signals were interpreted to reflect leaching of mineral surfaces and dissolution of secondary weathering products such as clays and ASi in undersaturated subglacial meltwaters Hatton et al., 2019a). However, in Patagonia we found significant differences in the Si concentrations and isotopic compositions of the 0.45 and 0.02 µm samples from glacierfed rivers, which highlighted the presence of an isotopically light colloidal-nanoparticulate silicate phase (CNSi). This phase was present in all rivers with >20% glacial cover, and relative proportions increased linearly with increasing glacial cover ( Figure 4A). As the CNSi dominated the 0.45 µm Si budget in many glacial rivers, this phase effectively controlled the δ 30 Si fSi and shifted values to isotopically light compositions ( Figure 4B). As such, the low δ 30 Si fSi observed in glacial rivers in Patagonia reflect variable amounts of CNSi, rather than the release of isotopically light DSi due to subglacial secondary weathering product dissolution as inferred in Greenland Hatton et al., 2019a). CNSi was also detected in three rivers with <20% glacial cover ( Figure 4A). These rivers all had very large catchment areas (>15,000 km 2 ) and significant glacial meltwater inputs from tributaries draining from the Patagonian Ice Fields, implying that glacial cover was not a good proxy for meltwater inputs in these systems. The Si isotope composition of these rivers also fell above the linear regression between δ 30 Si fSi and the proportions of CNSi defined by glacier-fed rivers with <20% glacial cover ( Figure 4B). The higher δ 30 Si fSi values of these three rivers were likely caused by the presence of large non-glacial lakes in their upstream catchment area, as lake waters typically have higher δ 30 Si values due to enhanced uptake of the light 28 Si isotope by diatoms, which leaves the water enriched in 30 Si (De La Rocha et al., 1997;Opfergelt et al., 2011). The fact that no CNSi was detected in any other river systems sampled across the region gives inference that these phases were formed as a glacial weathering product (Pryer et al., 2020).
The relatively narrow range of isotope values of the CNSi (δ 30 Si CNSi ; −0.49 to −0.23‰; Figure 2D) suggests that compositions were similar across the sampling region, despite differences in glacial size and catchment properties. The  Hawkings et al., 2018;Hatton et al., 2019a). However, previous investigation of these CNSi phases indicated that they were predominately composed of alkali feldspars, without a significant ASi component, and were formed by glacial comminution of silicate bedrock (Pryer et al., 2020). The δ 30 Si CNSi values we observed in this region were significantly lower than the isotopic compositions reported for unweathered feldspar (mean = −0.15 ± 0.04‰ (2σ ); Georg et al., 2009), likely due to variability in feldspar composition, the fine-grained particle size, the incorporation of minor amounts of other mineral phases, and/or the possibility of isotopically light Si adsorbed to the iron-oxyhydroxide phases present in this size-fraction Opfergelt et al., 2009;Pryer et al., 2020). The potential lability of CNSi phases is currently unknown and represents a significant uncertainty in resolving bioavailable Si fluxes and Si isotope exports from glacier-fed rivers in Chilean Patagonia (Pryer et al., 2020). Traditionally feldspars and other aluminosilicate minerals are not considered to be reactive in seawater, and potential contributions to oceanic elemental cycles and isotope budgets are therefore overlooked. Assuming that the CNSi feldspars are unreactive in seawater, their inclusion could lead to significant misinterpretation of glacial river isotope compositions if δ 30 Si fSi values are interpreted to represent truly dissolved signatures. However, recent research has suggested that dissolution of feldspars in seawater may be more significant than previously appreciated, with dissolution rates an order of magnitude higher than in freshwater (Gruber et al., 2019). Indeed, dissolution of the CNSi from glacierfed rivers in Patagonia may be even more pronounced given the sub-micron particle size, associated large surface area and the potential for distal transport away from the coast. Furthermore, rock-crushing experiments have demonstrated how submicron size feldspar particles are formed due to physical grinding and that these phases are highly unstable in solution and rapidly dissolve to form DSi (Holdren and Berner, 1979). As such, we speculate that CNSi phases could potentially contribute to the labile Si concentrations and isotope compositions exported from glacial rivers, with implications for resolving marine elemental budgets and understanding how variable glaciation could impact the Si cycle. However, more research about the solubility, transformations and fate of these phases is needed to fully resolve their potential impact on downstream ecosystems.

Implications of Riverine Colloidal-Nanoparticulate Phases
At present, it is unclear whether CNSi phases are ubiquitous in proglacial rivers in other glaciated regions, or how environmental or geochemical processes control their formation and abundance. Samples for DSi analysis are typically filtered through either 0.45 or 0.22 µm filter membranes and concentrations are commonly measured using colorimetric techniques, which only measure truly dissolved Si in the form of silicic acid. As such, CNSi phases can go undetected unless using samples that have been size-fractionated (e.g., 0.45 and 0.02 µm filtered; Shiller, 2003) and measured by mass spectrometry techniques that ionize all Si present in the solution (e.g., ICP-MS). CNSi was only detected in Patagonian rivers because both colorimetric and ICP methods were used, giving Si concentrations that differed by almost an order of magnitude (Pryer et al., 2020). These large differences between methods were not reported from proglacial rivers in Greenland Hatton et al., 2019a), indicating that CNSi phases were negligible or absent in these river systems. We speculate that the presence of CNSi phases could be affected by a combination of factors including variable filtering techniques, transport through proglacial lakes, the surrounding geology or riverine pH conditions, which could influence the concentrations of fine-grained species in the filtrate (Horowitz et al., 1992(Horowitz et al., , 1996, the grain-size distribution of eroded bedrock (Crompton and Flowers, 2016) or the solubility of finely-ground feldspars (Yuan et al., 2019). High concentrations of CNSi have been reported in glacier-fed rivers in Alaska (mean = 61.1 ± 67.6 µM; Schroth et al., 2011), showing that these phases are present in other glaciated regions. We emphasize that CNSi may have been overlooked in other glacier-fed rivers and should be considered when interpreting Si isotope compositions.
The size-partitioning of several other weathering-sensitive elements including Li, Sr, Mg, and Ba (Figures 5A-D) indicate that these size-fractionated trends may not be unique to Si isotopes. All of these elements had a significant colloidalnanoparticulate component in glacier-fed rivers, and relative proportions varied as a linear function of glacial cover, suggesting associations with the CNSi or similar glaciogenic controls. For all of these elements, primary mineral phases and secondary weathering products are generally isotopically different from the composition of dissolved species (Peucker-Ehrenbrink et al., 2010;Wimpenny et al., 2010Wimpenny et al., , 2011Pogge von Strandmann et al., 2019;Gou et al., 2020). As such, it is likely that isotopic trends would show similar offsets between 0.45 µm filtered and dissolved (0.02 µm) compositions as observed for Si. This could potentially obscure the true controls on riverine isotope values and affect terrestrial isotope budgets. These results highlight that size-fractionated filtering is needed to quantify colloidal-nanoparticulate concentrations and isotopic compositions, especially in glacier-fed river systems where these phases may dominate.
Furthermore, there are currently no standardized sampling protocols for analysing "dissolved" stable isotopic composition in rivers, with 0.45 µm or 0.2/0.22 µm pore-sizes and a wide variety of filter membrane types commonly used. We postulate that filtering protocols may dramatically affect filterable concentrations and isotopic composition when a significant colloidal-nanoparticulate fraction is present. These filtering artifacts have been shown to affect the concentration of colloidalnanoparticulate species by up to an order of magnitude in nonglacial rivers (Horowitz et al., 1992), but the implications for glacier-fed riverine concentrations or for isotope signatures have not been tested before. Quantifying these filtering artifacts and the size-partitioning of different chemical species in glacier-fed rivers may provide valuable insight and help to disentangle the controls on riverine isotope signatures.

Controls on the Isotopic Composition of Dissolved Silicon in Patagonian Rivers
The dissolved Si isotope composition (δ 30 Si DSi ) of all rivers with a CNSi component shifted to significantly heavier values compared to corresponding δ 30 Si fSi . These offsets were well outside analytical error, reinforcing the importance of accounting for contributions from colloidal-nanoparticulate species. Glacierfed river δ 30 Si DSi values were still lower than the mean for nonglacial rivers in this region, likely due to suppressed diatom primary productivity, limited secondary mineral formation and enhanced dissolution of isotopically light ASi phases in turbid, low temperature and undersaturated glacial meltwaters Hatton et al., 2019b). However, the mean δ 30 Si DSi composition from glacier-fed rivers in Patagonia (+0.33 ± 0.15‰) was higher than the mean for glacial rivers in Iceland (+0.17 ± 0.18‰, n = 9; Opfergelt et al., 2013), and much higher than mean values from Leverett Glacier in Greenland (−0.25 ± 0.12‰, n = 16; Hawkings et al., 2018) or Qeqertarsuaq (Disko Island) (−0.01 ± 0.29‰, n = 4; Hatton et al., 2019b). These differences in δ 30 Si DSi are potentially due to the sampling of Patagonian rivers downstream of glacial outlets and after transport through large proglacial lakes, with variable non-glacial inputs, secondary weathering processes, and/or because samples from other regions were not ultra-filtered through 0.02 µm and could potentially contain CNSi phases. Recent research has found that riverine δ 30 Si DSi values can also be highly variable within glacial regions, with δ 30 Si DSi values in Icelandic proglacial rivers ranging from −0.58 to +0.78‰ (Hatton et al., 2019b). The controls on the variability of glacial meltwater δ 30 Si DSi signatures are not fully understood at present, with values likely to be influenced by the geology, catchment size, hydrology, and water residence times in subglacial environments, as well as physical and chemical weathering conditions controlling the dissolution and precipitation of primary and secondary weathering products Hatton et al., 2019a,b).
Non-glacial river samples showed a wide range of δ 30 Si DSi values, reflecting the heterogeneity of Si cycling processes across the region. The δ 30 Si fSi and δ 30 Si DSi values were always within error as there was no detectable CNSi present in rivers without glacial meltwater inputs. The highest δ 30 Si DSi values (+0.95 to +1.25‰) were from non-glacial rivers fed by large lake systems, likely due to enhanced diatom productivity (De La Rocha et al., 1997;Opfergelt et al., 2011). However, the lake-related high δ 30 Si DSi values were similar to the global mean riverine δ 30 Si DSi of +1.25‰ (Frings et al., 2016) and significantly lower than other large lake values such as Lake Baikal in Russia (mean = +2.30 ± 0.38‰; Panizzo et al., 2017). As lakes in Patagonia are generally oligotrophic and primary production is limited by inputs of nitrogen and phosphorus (Baigún et al., 2006), unrestricted diatom growth and further 30 Si enrichment of DSi may be curtailed by nutrient availability in this region. The lightest non-glacial δ 30 Si DSi values were found in tea-colored, organicrich rivers, which drained from catchments with wetlands and peat-bogs. All rivers of this type had a very narrow range of δ 30 Si DSi values from +0.14 to +0.21‰ and relatively low DSi concentrations of 31 to 40 µM. Such low δ 30 Si DSi and low DSi concentrations are similar to findings from "blackwater" tributaries of the Congo and Amazon rivers and are interpreted to reflect enhanced dissolution of isotopically light clays by organic acids (Cardinal et al., 2010;Hughes et al., 2013).
The relationships observed between δ 30 Si DSi and other geochemical variables highlight how chemical weathering plays a key role in driving riverine Si isotope compositions. The positive linear correlations between δ 30 Si DSi and concentrations of DSi (Figure 3A), total dissolved solids (TDS) (Figure 3B) and other ions such as bicarbonate (Supplementary Figure 2) indicate that progressive chemical weathering of the bedrock creates high δ 30 Si DSi values, similar to trends reported from other global regions such as the Congo (Cardinal et al., 2010), the Himalaya (Fontorbe et al., 2013), and Iceland (Opfergelt et al., 2013). The weak negative correlation observed between δ 30 Si DSi values and DSi/Na+K ratios ( Figure 3D) demonstrates how the removal of DSi relative to Na and K causes preferential loss of the 28 Si isotope, indicating that secondary clay and biogenic Si formation may explain some of the variability in riverine Si isotope compositions (Cardinal et al., 2010;Cockerton et al., 2013;Sun et al., 2018). No relationship was observed between δ 30 Si DSi and indicators of silicate and carbonate weathering such as divalent:monovalent cation ratios (Tranter et al., 2002;Wadham et al., 2010; Figure 3C) or end-member mixing plots (Gaillardet et al., 2014 ; Supplementary Figure 4), which have been inferred to influence temporal trends in Si isotope compositions in single catchment systems (Hatton et al., 2019a). The lack of significant trends is likely due to the diverse environmental controls on Si cycling and weathering in Patagonia, combined with mixed catchment lithologies complicating signals.

Isotopic Composition of Amorphous Silicon Phases
Amorphous silicon (ASi) phases associated with suspended particulate material (SPM) have been shown to readily dissolve in seawater and represent an important component of reactive riverine Si exports (Conley, 1997;Treguer and De La Rocha, 2013;Frings et al., 2014;Hawkings et al., 2017). However, measurements of ASi isotope compositions are currently limited to high-latitude proglacial rivers Hatton et al., 2019a,b), potentially representing a significant gap in resolving terrestrial Si isotope budgets. ASi was present in all rivers where particulate samples were available for extraction, with concentrations that varied primarily as a function of the SPM load (Pryer et al., 2020) and equalled 1-100% of the respective DSi concentration ( Figure 2B). All δ 30 Si ASi values from Patagonian rivers were far lower than corresponding riverine δ 30 Si DSi showing that the lighter 28 Si isotope is preferentially incorporated into ASi phases ( Figure 2D) with mean 30 Si ASi -DSi values of −1.06 ± 0.42‰. These offsets are similar but more variable than empirically derived values for abiogenic ASi formation in freshwater (−0.98 ± 0.12‰; Zheng et al., 2019), likely due to the presence of biogenic ASi phases such as diatoms, which can have a variable 30 Si ASi -DSi of −1 to −2‰ (Frings et al., 2016). Coupled to this, the dissolution of ASi phases can impart 30 Si ASi -DSi of +0.25 to −2.5‰, depending on the initial isotopic composition of the ASi phase and the congruency of weathering (Frings et al., 2016). The range of ASi isotopic compositions in Patagonian rivers therefore likely reflects a dynamic balance between the composition of ASi material from biogenic and abiogenic sources, combined with variable dissolution of these phases controlled by factors such as temperature, pH and the saturation state of the river water (Ziegler et al., 2005;Stamm et al., 2019;Zheng et al., 2019).
The δ 30 Si ASi compositions in some non-glacial rivers in Patagonia were lower than the range of values reported for proglacial rivers (−0.86 to −0.05‰; Figure 2D; Hatton et al., 2019b). Such low δ 30 Si ASi values from non-glacial river systems are likely due to the presence of phytoliths, diatoms and other biogenic ASi phases, which have large fractionation factors and are consequently highly enriched in 28 Si (Opfergelt et al., 2008;Frings et al., 2016;Sutton et al., 2018). The δ 30 Si ASi values from glacier and volcano-fed rivers were notably higher than δ 30 Si ASi compositions in other river systems, likely due to the reduction or absence of biogenic ASi phases in turbid, light-limited waters ( Figure 2D). Additionally, dissolution of ASi phases in subglacial environments and glacier-fed rivers, where rates of physical erosion are high and DSi concentrations are low, may cause preferential loss of 28 Si from the ASi Hatton et al., 2019a), and drive residual ASi phases to heavier isotopic compositions. The δ 30 Si ASi values observed for glacierfed rivers in Patagonia are similar to data from two glacial systems in Greenland (−0.21 ± 0.06‰ and −0.47 ± 0.06‰; Hawkings et al., 2018;Hatton et al., 2019a), suggesting broad similarities in the formation mechanisms and composition of glacial ASi between regions.

Seasonal Variability in Riverine Si Isotope Compositions
For the two catchments where time-series and seasonal data were collected, both δ 30 Si fSi and δ 30 Si DSi values showed no significant variability (Figure 6). For the glacier-fed site (Río Huemules, 70% glacial cover), the consistent DSi concentrations, δ 30 Si fSi and δ 30 Si DSi compositions were likely due to the presence of a large proglacial lake homogenizing and modulating temporal variability (Figures 6A-C). This lack of variability was also observed in the ASi concentrations and δ 30 Si ASi values, suggesting no seasonal change in ASi composition. We also observed no large discharge events, such as glacial lake outburst floods, or obvious seasonal shifts in the hydrological regime of the glacier (i.e., channelized vs. distributed drainage pathways), which are the primary drivers of variability in δ 30 Si DSi values in proglacial rivers in Greenland Hatton et al., 2019a). As all glacier-fed rivers sampled in Patagonia drain through large proglacial lakes, we speculate that other glacial catchments will also show negligible temporal variability and our Si isotope compositions are likely to be broadly representative of mean annual values from these catchments.
At the non-glacial time-series site (Río Lloncochaigua, 1.2% glacial cover), consistent riverine Si isotopic compositions were observed, despite variability in DSi concentrations (Figures 6D,E). This lack of isotopic variability indicates that Si sources and biogeochemical cycling within this river catchment were stable across the sampling period. However, other nonglacial rivers in Patagonia may show greater temporal variability in δ 30 Si DSi values, especially in catchments with non-glacial lakes that likely experience seasonal shifts in siliceous productivity (Alleman et al., 2005;Hughes et al., 2011;Opfergelt et al., 2011). Seasonal variability in δ 30 Si DSi on the order of ± 1‰ has been reported from other global river systems, driven by factors such as snow-melt influencing groundwater flow, shifts in weathering intensity, variable inputs and dissolution of phytoliths from plants, and changes in water temperature that can affect the precipitation or solubility of secondary weathering products (Ding et al., 2004;Pokrovsky et al., 2013;Sun et al., 2018). As such, investigating seasonal controls on riverine Si cycling should be accounted for in future sampling campaigns to fully resolve the Si isotope composition of freshwater flowing into the coastal region of Chilean Patagonia.

Importance of Including ASi Phases in Labile Riverine δ 30 Si Export Budgets
We calculated a mass balance accounting for the relative proportions and isotopic compositions of the DSi and ASi components in Patagonian rivers to investigate whether including ASi is important for resolving the "labile" Si isotope composition (δ 30 Si DSi + ASi ) of rivers in this region. The potential contributions from CNSi were excluded in the following discussion, as their solubility and fate in seawater are currently uncertain (Pryer et al., 2020). All rivers with >20% glacial cover had significantly lower δ 30 Si DSi + ASi compared to δ 30 Si DSi , with differences that ranged from −0.10 to −0.30‰ (mean = −0.20 ± 0.07‰; Figure 7A). These values were well outside analytical error and show how including ASi components can significantly change estimates of the riverine Si isotope flux to downstream fjords. However, it is not only glacial rivers where the labile Si isotope composition was affected by including ASi. Other rivers with high SPM loads, such as those draining from active volcanoes (e.g., Río Ibanez) or with unconsolidated sediments (e.g., Río Chacabuco), also exhibited large offsets between δ 30 Si DSi + ASi and δ 30 Si DSi values, with differences ranging from −0.27 to −0.14‰. Conversely, including the ASi phase did not influence the labile Si isotope budget for non-glacial rivers with low SPM concentrations, where δ 30 Si DSi + ASi and δ 30 Si DSi compositions were within analytical uncertainty. Overall, we find that in all river systems where the ASi phases exceed 10% of the labile Si budget (i.e., DSi + ASi), the δ 30 Si DSi + ASi values were significantly different to δ 30 Si DSi compositions ( Figure 7B). Although there is variability around this trend caused by independent factors that can influence the relative proportions and isotopic compositions of ASi and DSi, this relationship provides a useful indication of river systems where ASi phases are likely to significantly affect estimates of Si isotope exports.
We extend our analysis to investigate whether including ASi phases could change the Si isotope exports from other global watersheds. An example is the Lena River, which has one of the world's largest catchment areas and provides ∼20% of the freshwater flux to the Arctic Ocean (Walker, 1998;Holmes et al., 2012). Previous research explored the δ 30 Si DSi values of the Lena river to quantify the Si isotope composition exported to the Arctic Ocean but did not discuss contributions from reactive ASi phases (Sun et al., 2018). Assuming that ASi concentrations are similar to rivers in Patagonia (mean = 1.12 wt. %; Pryer et al., 2020) and scaling by the mean SPM concentrations of the Lena River (40.5 mg L -1 ; Fedorova et al., 2015), we approximate that ASi could deliver an additional ∼16 µM of reactive Si from this catchment to downstream ecosystems. These assumptions are likely broadly valid due to the relatively narrow range of ASi (wt. %) values reported from global rivers (Frings et al., 2014;Hatton et al., 2019b;Pryer et al., 2020) and the strong linear relationship between SPM and ASi (µM) concentrations (Frings et al., 2014;Pryer et al., 2020). As such, ASi phases in the Lena River could represent ∼20% of the total labile Si flux from this catchment, relative to mean annual DSi values of 69.7 µM (Sun et al., 2018). Using the mean δ 30 Si ASi of −0.59‰ from this study, we calculate that including ASi in the Si isotope budget for the Lena river could shift δ 30 Si values by −0.27‰, from δ 30 Si DSi values of +0.86‰ (Sun et al., 2018) to δ 30 Si DSi + ASi of +0.59‰. We speculate that including ASi components will significantly change the terrestrial Si isotope flux to the ocean from many other rivers and should be considered in future work.

Do Glacier-Fed Rivers in Patagonia Have Distinct Si Isotope Compositions?
A key hypothesis of this research was that glacier-fed rivers in Patagonia have distinct Si isotope compositions compared to non-glacial rivers. This has been shown to be true in other regions such as Iceland (Georg et al., 2007;Opfergelt et al., 2013) and Greenland , and suggests that glacial meltwater inputs could influence the oceanic Si isotope inventory over glacial-interglacial periods, with implications for understanding Si isotope paleoclimate records . Initially, the δ 30 Si fSi data showed that glacier-fed rivers in Patagonia were isotopically light compared to non-glacial rivers and compositions varied as a linear function of upstream glacial cover ( Figure 2C). However, more detailed analyses showed that this trend was mainly controlled by variable proportions of an isotopically light CNSi phase, which shifted δ 30 Si fSi to lower values. Once the CNSi were removed by ultra-filtration, the δ 30 Si DSi values of glacier-fed rivers were still very low (mean = +0.33‰; Figure 2C) relative to the global riverine mean (+1.25‰; Frings et al., 2016), and were statistically different to the mean composition of rivers with <20% glacial cover (+0.54‰, t 24 = 3.3, p < 0.01). However, the δ 30 Si DSi values of glacier-fed rivers were no longer isotopically distinct from some non-glacial rivers in Patagonia and were often isotopically heavier than organic-rich rivers (mean = +0.17‰).
When ASi phases were included to give δ 30 Si DSi + ASi as estimates of the labile Si isotope budget, glacier-fed rivers still did not have significantly lower isotope compositions than some non-glacial river systems ( Figure 7A). Although all glacier-fed rivers in this region had low δ 30 Si DSi + ASi (range = −0.12 to +0.26‰), there was overlap with rivers with <20% glacial cover (range = +0.03 to +1.17‰). These results demonstrate that low δ 30 Si values are not necessarily indicative of glacial meltwater inputs into the Patagonian fjords. Ultimately, the reactivity of CNSi is a key uncertainty in quantifying riverine Si isotope exports from glacier-fed rivers in Patagonia. If the CNSi dissolves in seawater, the reactive Si isotope composition of all rivers with >50% glacial cover would be isotopically distinct compared to non-glacial river systems in this region. Additional research is therefore needed to understand the lability and fate of CNSi within downstream fjords to fully resolve reactive Si isotope exports from glacial rivers in Patagonia. Overall, these data show that glacier-fed rivers have light Si isotope compositions, similar to findings from other glaciated regions. However, glacial river signatures were not isotopically distinct from some non-glacial river systems, highlighting the complex controls on riverine Si isotope exports in Chilean Patagonia.

CONCLUSIONS
This study provides new insights into how variable glacial cover can affect Si cycling processes and riverine Si isotope compositions using a latitudinal transect of rivers in Chilean Patagonia. A negative linear trend between glacial cover and δ 30 Si fSi in glacier-fed rivers initially indicated that glacial meltwaters could have a distinctive, light Si isotope composition and that the relationship could represent broadly proportional mixing with non-glacial water. However, significant differences between 0.45 and 0.02 µm filtered Si concentrations and Si isotopic compositions of all glacier-fed rivers indicated that an isotopically light colloidal-nanoparticulate Si phase was present and often dominated Si budgets. This CNSi was primarily composed of feldspars and its lability in seawater is currently uncertain, representing a significant gap in resolving glacial Si exports from this region. Similar linear relationships between glacial cover and the colloidal-nanoparticulate concentrations of other weathering sensitive elements, such as Li, Sr, Mg, and Ba, highlight that this issue is not unique to Si and that the isotopic compositions of these elements could also show sizedependent trends in glacial rivers. Once the CNSi phases were removed by ultra-filtration, glacier-fed rivers still had light Si isotope compositions relative to the non-glacial mean, but did not show a significant trend with glacial cover and were no longer isotopically distinct from non-glacial rivers with the lowest δ 30 Si DSi values. The variability observed in δ 30 Si DSi compositions reflects the diverse landscape and biogeochemical weathering pathways, with productivity in lakes, interactions with organicrich waters and chemical weathering intensity exerting important controls. These data give new insights into the factors influencing riverine Si cycling and isotope compositions, providing a baseline to understand future change in Chilean Patagonia.
ASi isotope data from a range of non-glacial and glacial rivers showed that ASi phases were isotopically lighter than DSi and provided preliminary evidence that riverine turbidity affects ASi isotope compositions. Currently, the contribution of ASi phases to Si fluxes and corresponding Si isotope compositions are not routinely measured, and interpretations of riverine Si exports are based on DSi alone. Since ASi was found in all rivers sampled in Patagonia and can equal the flux of DSi, excluding this reactive component potentially gives an incomplete picture, with implications for quantifying terrestrial Si exports, understanding Si cycling and interpreting Si-based paleoclimate records. We postulate that including ASi phases is critical for resolving riverine Si export and isotope budgets both in Chilean Patagonia and other global rivers. Future research on glacial isotope compositions should account for the size-partitioning and potential reactivity of different phases to fully disentangle how terrestrial exports may have varied in the past and are likely to change in our warming world.

DATA AVAILABILITY STATEMENT
The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. All data used in this study are freely available to download (https:// zenodo.org/record/4114629#.X5AjjC2ZMkg).

AUTHOR CONTRIBUTIONS
The PISCES project was devised by JW, LR, and KH. Data analysis was conducted by HP, with assistance from JHat and JHaw. HP, JHaw, JW, LR, AK, MM, AU, AC, GD, and VH were involved with fieldwork in Patagonia. The manuscript was written by HP, with comments provided by all coauthors. All authors contributed to the article and approved the submitted version.

FUNDING
This work was funded as part of the NERC-CONICYT project PISCES (NE/P003133/1-PII20150106). HP was supported by the NERC GW4+ DTP and the Natural Environment Research Council (NE/L002434/1). KH and JHat were funded by the European Research Council ERC-ICY-LAB (#678371) and Royal Society grant (RGF-EA-181036). JW was supported by a Royal Society Wolfson Merit Award. JHaw was supported by a European Commission Horizon 2020 Marie Skłodowska-Curie Actions fellowship ICICLES (#793962). This is publication #182 of Huinay Scientific Field Station.