Introduction

The amount of radiation absorbed or reflected from sediment at any wavelength on the electromagnetic spectrum depends on its composition. In the visible-light spectrum (around 400–800 nm), this difference in absorptivity and reflectivity is perceived as a difference in sediment color. For instance, minerals with low reflectivity will appear optically dark, and minerals with high overall reflectivity will appear optically bright. Similarly, minerals with higher reflectivity in longer wavelengths will appear reddish. Sediment reflectivity is determined by the average reflectivity of its constituent components. We can therefore measure sediment reflectivity in an attempt to reconstruct sediment composition, evaluate mineral assemblages, or help define distinct sedimentary facies (Debret et al. 2011; Wei et al. 2014). Reflectively-based analyses have the advantage of being rapid, non-destructive, and obtainable at a very high spatial resolution.

Several methods have been used to provide objective, user independent measurements of color for use in sedimentary and stratigraphic analysis. Quantification is necessary as color is not itself a sediment property, but rather the human perception of sediment reflectance across the visible-light spectrum. The two most widely used are the Munsell color chart and CIELab L*a*b* referential (Schuttenhelm 1989; Nagao and Nakashima 1992; Nederbragt et al. 2006; Debret et al. 2011). Both are formulated as a cylindrical coordinate system with the vertical axis representing color lightness or intensity. The two horizontal axes represent different aspects of sediment color: ‘hue’ and ‘chroma’in the Munsell color space and ‘green–red’ and ‘blue–yellow’ in the CIELab color space. Both color systems wrap the color space such that the long and short wavelength limits of the visible color spectrum (violet and red) are joined. These color spaces have been used to define notable sedimentary units (Ericson et al. 1961) and calibrated to evaluate the relative abundance of constituent components (Roth and Reijmer 2005; Debret et al. 2006, 2011). These methods are, however, poorly adapted for quantitative analysis of sediment composition as the color spaces are defined for the optical perception of color instead of the underlying reflectance values (Debret et al. 2011).

Sediment reflectance spectra, which map out the sediment reflectance across the visible light spectrum, are better suited for quantitative compositional analyses or the identification of specific compounds (whether mineral or chemical). Raw spectra can be used, but more commonly the first derivative of the spectrum is taken with respect to wavelength to increase its sensitivity, known as a first derivative spectra (Debret et al. 2011). Raw spectra have been used for evaluating the effect of differing water content (Balsam et al. 1997) while first derivative spectra peaks have been used to identify different minerals (e.g. Goethite: 445 and 525 nm, Hematite: 555–575 nm) and other compounds (e.g. chlorophyll-\(\alpha \): 675 nm; Debret et al. (2011)). Spectral brightness across different wavelengths can also be used as independent variables for factorial analysis, and the results interpreted in terms of one or more mineral abundances (Damuth and Balsam 2003; Ortiz et al. 2009).

Quantitative spectrophotometric analysis of sediment composition, the direct reconstruction of sediment chemical or mineral composition using reflectance data, always requires external geochemical calibration, and remains limited to detecting a small number of constituents (Debret et al. 2011). Nevertheless, changes in sediment reflectance (often measured in terms of ‘color’), through its relation to underlying sediment properties, can provide a high resolution record of local climatic or environmental variability. Sediment reflectance or color has been used to reconstruct summer temperatures in South America (von Gunten et al. 2009; Elbert etal. 2015), evaluate precipitation changes from the Asian monsoon (Ji et al. 2005; Zhang et al. 2007), reveal wet-dry cyclicity in Far East Russia (Wei et al. 2014), track relative sea level changes in the tropics (Roth and Reijmer 2005), and reconstruct patterns of oceanic sediment transport in the high northern and southern latitudes (Helmke et al. 2002; Wu et al. 2019). Existing studies highlight the wide range of environmental and climatic variables which have been reconstructed using sediment optical properties, but also emphasize that a detailed understanding of the local system is crucial for its use as a proxy (Balsam et al. 1999; Debret et al. 2011; Wei et al. 2014).

A significant correlation between proxy and climatic data, commonly defined as a p-value<0.1 (Mann et al. 2008), is not sufficient for a proxy to have hindcasting value beyond the calibration time-period. For the standard linear regression approach to proxy reconstruction, two additional assumptions must be made: linearity and stationarity (National Research Council 2006). Linearity refers to the statistical relationship between a given proxy and the climatic variable it is calibrated to, which must approach a first-order polynomial. Stationarity, similar to Lyell’s “uniformity principle” (Lyell 1830; Camardi 1999) refers to the assumption that the statistical relationship between a given proxy and a given climatic variable remains constant throughout the entire period of interest (National Research Council 2006). In practice, the relationship between the proxy and climatic variable (leading to the correlation) must exhibit only small changes through time, or else the predictive power of the proxy will be lost and any reconstructions will be meaningless. In an extreme scenario, environmental changes could cause the relationship between proxy and climatic variable to become inverted through time, leading to the proxy hindcasting paleoclimate opposite to reality.

Prior literature shows that changes in sediment reflectance or color can be controlled by a wide range of climatic and environmental conditions. Here, we test two hypotheses related to the link between sediment reflectance and climate at Lago Argentino, Argentinian Patagonia. Our large dataset of varved lake cores across Lago Argentino’s 100 km long lake basin provides an ideal opportunity to evaluate the use of pixel intensity in a complex lake environment and gain insight into the climatic drivers of sedimentation in a relatively data-poor environment. First, we hypothesize that sediment pixel intensity, an indirect measure of reflectance and sediment properties derived from 20 \(\mu \)m resolution digital core scans, can be used to reconstruct past temperature, precipitation, or wind speed changes. As a part of this process, we evaluate on whether the assumptions of linearity and stationarity can be made at Lago Argentino. Secondly, we hypothesize that the correlation between sediment pixel intensity and climatic variables can provide insight into the climatic drivers of sedimentation at Lago Argentino.

Site description

Lago Argentino is the largest of several ultra-oligotrophic, ice-marginal, freshwater lakes located on the eastern flank of the Southern Patagonian Icefield (SPI) at 50 \(^\circ \)2’S, 72 \(^\circ \)4’W (Fig. 1). Lago Argentino has a surface area of around 1500 km\(^2\) and is up to 600 m deep (Sugiyama et al. 2016; Magnani et al. 2019). While precipitation is as high as several metres of water equivalent per year on the SPI itself, most of the landscape surrounding Lago Argentino is semiarid and receives less than 500 mm of precipitation per year (Garreaud et al. 2012; Lenaerts et al. 2014).

Fig. 1
figure 1

Location of Lago Argentino and other key locations. The central globe (a) shows the location of Patagonia, and the left hand panel (b) shows the location of Lago Argentino within Patagonia. The right hand panel (c) shows the position of the 12 cores used in this study (7A, 27A, and 28A for comparison with XRF data; 11A, 12A, 13A, 20A, 22A, 27A, 30A, 38A, 41A, and 42A for comparison with climatic data). NPI = Northern Patagonian Icefield; SPI = Southern Patagonian Icefield. Imagery \(\copyright \) Google Earth

Full size image

Lago Argentino is located within a tectonically and volcanically active region, and three active volcanoes are located within 100 km of the lake. Bedrock geology varies from metasedimentary and volcanic rocks in the Andean fold and thrust belt to the west of Lago Argentino, to a forearc basin sedimentary sequence and plateau basalts to the east of the fold and thrust belt (Coutand et al. 1999). Six glaciers feed into Lago Argentino, of which three (Upsala, Perito Moreno and Spegazzini) calve directly into the lake and three (Onelli, Mayo and Ameghino) calve into smaller peripheral lakes (Fig. 1). Upsala glacier drains around 400 km\(^3\) of ice, approximately three quarters of the ice volume in Lago Argentino’s catchment (Carrivick et al. 2016; Millan et al. 2019), and calves into the longest and deepest fjord (brazo) in the north-western branch of Lago Argentino. Two major rivers are located at the east of the main basin of Lago Argentino, one of which flows into the lake (Río La Leona), while the other drains the lake to the Atlantic Ocean (Río Santa Cruz).

The monthly average surface air temperature at Lago Argentino varies from a January (austral summer) maximum of around 10 \(^\circ \)C to a July (austral winter) minimum of around −2 \(^\circ \)C, as measured at El Calafate airport located immediately adjacent to the lake and 15 km east of El Calafate. The wind speed follows a similar cycle, with a monthly maximum of 14.2 m.s\(^{-1}\) (in January) and minimum of 6.8 m.s\(^{-1}\) (in June; source: El Calafate airport). Local winds are dominantly westerlies, with 70% of measured wind directions in the range 230–320 \(^\circ \) (SW–NW). High wind speeds keep the majority of Lago Argentino ice free even during the winter, when temperatures are below freezing. Local precipitation at El Calafate, on Lago Argentino’s southern shore (Fig. 1) reaches a minimum of 11.5 mm month−1 in November, and a maximum of 28.5 mm month−1 in May. The melt rates and flow velocities of the glaciers draining into Lago Argentino also vary seasonally, with a maximum in the summer (Mouginot and Rignot 2015; Minowa et al. 2021).

Materials and methods

Core collection and imaging

We collected 47 sediment cores from Lago Argentino in August–September 2019, with sediment recovery of up to 6.4 m at shallow water (<300 m depth) sites and 5.5 m at deep water (>300 m depth) sites. The total core length recovered was \(\sim \)108 m. We conducted the coring in the austral winter due to lower wind speeds. We collected cores from half the locations using a Kullenberg piston-coring system with 250–450 kg head weight, and from the other half using a smaller gravity surface coring device. We identified coring sites through a preceding seismic reflection survey (Magnani et al. 2019), and collected cores from all of Lago Argentino’s main depositional environments, as well as a proximal-distal transect away from the glacier fronts. The cores captured recently deposited sediment at each site, as confirmed by subsequent \(^{137}\)Cs dating (Van Wyk de Vries et al. 2022).

We performed initial whole-core scans of magnetic susceptibility and gamma-beam density, and split the cores lengthwise into two halves at the CSD Facility. We then conducted near-UV to near-IR spectrophotometric and magnetic scans on each split core face. We described the key sediment properties, lithological units, and sediment structures of the 47 cores in logging software PSICAT (Reed 2007) which are described in detail in Van Wyk de Vries et al. (2022).

We selected a subset of 10 gravity cores from our whole dataset for further analysis. The subset forms a proximal-distal transect away from Glaciar Upsala, the largest glacier calving into Lago Argentino. We selected the 10 cores for their well-preserved core tops and presence of intact, alternating bright and dark laminations (Van Wyk de Vries et al. 2022).

We used a Cr source ITRAX X-ray fluorescence (XRF) scanner to investigate the elemental composition of these cores. We ran a 5 mm resolution whole-core scan on three cores from the west (ice-proximal), centre, and east (ice-distal) of Lago Argentino. We normalized XRF data for individual elements by the total counts-per-second, but did not attempt external calibration to absolute concentrations. We used a GeoTek Geoscan V line scanner to capture 20 \(\mu \)m resolution sediment pixel intensity (PxI) core scans. We performed core scanning immediately following splitting to avoid optical property changes related to water loss or oxidation. The Geoscan V line scanner captures image data sequentially down the length of the entire core, avoiding stitching effects or spherical distortions associated with standard area-imaging cameras. We scanned each core with two focal ratios, F8 and F11, to ensure non-saturation while imaging both bright and dark sediment. The core subset used in this study is predominantly composed of bright-colored sediment, so we used the F8 focal ratio images.

Extraction of annual core color timeseries

Lago Argentino’s sediment is predominantly composed of alternating bright and dark laminae. These laminae are varves, which has been confirmed by their sedimentary structure and \(^{137}\)Cs dating (Van Wyk de Vries et al. 2022). The varves are represented as an alternating sequence of seasonal light (winter) and dark (summer) layers, with dark layers exhibiting a coarser overall grain size and a higher Ca/K ratio (Van Wyk de Vries et al. 2022). Investigation of smear slides also shows a higher abundance of mafic grains in dark layers, although the extremely fine grain size (\(<1 \mu \)m) complicates a direct optical assessment of mineral abundances. Manual counting of a large number of varves is challenging to reproduce and has a limited assessment of uncertainties (Sprowl 1993; Tian et al. 2005), we therefore used a semi-automated method to overcome these limitations. We counted the number and thickness of varves in our cores using sliding-window autocorrelation of grayscale digital core scans (CountMYvarves; see (Van Wyk de Vries et al. 2021)). CountMYvarves conducts multiple simultaneous lamina counts, which allows for uncertainties to be propagated into the resulting age-depth model and sedimentation rate curve.

We used the boundaries calculated from semi-automated varve counting to extract the 3-band (red, green, and blue; RGB) digital core scan image for each individual varve. We created four additional image bands, the grayscale, red/green, red/blue, and green/blue PxI (Fig. 2).

Fig. 2
figure 2

Extraction of color depth-series from raw digital core scans. a and b show a whole core scan and \(\sim \)10 laminae scan subsubsection. c shows a single varve split out into the four different bands, and d shows the whole-core PxI depth-series for each band

Full size image

Comparison between sediment pixel intensities and elemental composition

We correlated sediment grayscale and RGB pixel intensities with XRF results to investigate the relationship between digital core-scan color and sediment elemental composition. We extracted the mean pixel intensities for the seven image bands at the same location as each 5 mm resolution XRF measurement. We averaged the pixel intensities in a 5 by 5 mm sampling area corresponding to each XRF measurement. We repeat this process for each of the three cores analysed. We then calculated a correlation matrix between the down-core variation in 8 elements measured by the ITRAX XRF scanner (Si, Al, K, Ca, Fe, Ti, Rb and Sr) and the four mean PxIs (grayscale, red, green, and blue) and three PxI ratios (red/green, red/blue, and green/blue). The correlation coefficient (r-value) between any two variables A and B is calculated as:

$$\begin{aligned} r(A,B) = \frac{1}{N-1} \sum _{i=1}^N \left( \frac{A_i-\mu _A}{\sigma _A} \right) \left( \frac{B_i-\mu _B}{\sigma _B} \right) , \end{aligned}$$
(1)

with N being the size of each sample, \(\mu _A\) and \(\mu _A\) being the averages of each variable, and \(\sigma _A\) and \(\sigma _A\) being their standard deviations. Due to the risk of false correlations, we calculate a new significance threshold Sig by dividing the threshold for 95% statistical significance by the number of pairwise correlations computed n:

$$\begin{aligned} Sig = \frac{0.05}{n} = \frac{0.05}{8*9} = 0.0007. \end{aligned}$$
(2)

We calculated a corresponding correlation p-value matrix, and considered correlations with p-values lower than 0.0007 to be significant.

Correlation between sediment pixel intensities and climatic variables

We evaluated the correlation between sediment PxIs and three climatic variables: precipitation, temperature, and wind speed. We accounted for two separate temporal uncertainties in our age-depth model: possible missing time from disrupted varves at the sediment-water interface (i.e. the first identifiable varve might not correspond to 2019, the year the cores were collected), and uncertainties from varve counting (\(\sim \) 10%; Van Wyk de Vries et al. (2021)). We computed an ensemble of 1000 possible age-depth models for each core using Monte Carlo sampling of the uncertainty derived from the semi-automated varve counting and plausible numbers of missing varves at the sediment-water interface (typically \(\sim \)0 to 15 varves).

We obtained temperature and precipitation covering the period 1930–2020 from the CRU-TS 4.05 monthly time series for the half-degree grid cell centered on 50.25\(^\circ \)S, 72.75\(^\circ \)W (Harris et al. 2020). This dataset combines records from several meteorological stations around Lago Argentino (Ibarzabal et al. 1996) into a single homogeneous record. We obtained daily wind speed data from El Calafate airport covering the period 1963–2017. We resampled the precipitation, temperature, and wind records into three separate time series: a mean annual record, a mean austral summer December–January–February (DJF) record and a mean austral winter June–July–August (JJA) record. We used equation (2) to perform a cross-correlation between individual PxI time series and climate time series. We accounted for serial autocorrelation in both the climatic and pixel intensity timeseries by calculating the resulting reduction in degrees of freedom. The effective number of degrees of freedom \(N_{\text {eff}}\) is calculated as:

$$\begin{aligned} N_{\text {eff}} = N\frac{1-\phi _A\phi _B}{1+\phi _A\phi _B} \end{aligned}$$
(3)

with N being the number of independent samples and \(\phi _A\) and \(\phi _B\) being the lag-1 autocorrelation coefficients of each timeseries (Hu et al. 2017). We calculated a correlation coefficient for each individual PxI timeseries of the 1000 age-depth model ensemble. We also perform a false discovery rate test (Benjamini and Hochberg 1995; Hu et al. 2017) to evaluate the effect of test multiplicity on our significance thresholds and verify that the correlations are not statistical artifacts. Due to the additional presence of temporal uncertainties, we modify the standard false discovery rate test with a bootstrap sampling procedure. The correlations between sediment PxIs and climatic variables remain statistically significant even when this test is applied (see supplementary section S2).

Results

Compositional significance of pixel intensities

In ice-distal core 7A, grayscale, green, and blue PxIs show little to no correlation with XRF-derived sediment composition. Red PxI shows a strong positive correlation with Al, Ti, Fe, and Rb (r-value > 0.35), a weaker positive correlation with K, and a strong anticorrelation with Ca, Si, and Sr (r-value<\(-\)0.35). The red/green, red/blue, and green/blue PxIs show strong positive correlations with Si, Ca, and Sr, a weaker positive correlation with K, and a strong anticorrelation with Al, Ti and Fe (Fig. 3).

Fig. 3
figure 3

Correlations between the four PxI depth-series and selected elements from the XRF scanner. Significant correlation coefficients are shown in bold, while non-significant correlations are shown in italic

Full size image

In core 27A, from the western margin of the main-lake basin, grayscale PxI shows a weak anticorrelation with all elements. Blue PxI exhibits no strong correlation with any element. Red PxI shows a strong positive correlation with Al, Si, Ti, Ca, and Fe (r-value > 0.35) and a weak positive correlation with all other elements (Fig. 3). Green PxI shows a strong positive correlation with Sr, a weak positive correlation with Ca, a strong anticorrelation with K, and a weak anticorrelation with other elements. All band ratio PxIs are strongly anticorrelated with Al, K, Ti, and Fe (r-value<\(-\)0.35). The Green/blue PxI is also strongly correlated with Ca and Sr (r-value > 0.35).

In ice-proximal core 28A, blue PxI shows little to no correlation with any element. Grayscale PxI and all three band ratio PxIs are very strongly anticorrelated with Al, Si, K, Ca, Ti, and Fe (r-value<\(-\)0.7). The green PxI exhibits the same correlations to a slightly lesser degree (r-value of  0.5). The red PxI shows the opposite pattern, with a very strong positive correlation with Al, Si, K, Ca, Ti, and Fe (r-value > 0.8). Grayscale, green, red/green, and red/blue PxIs show a strong correlation with Sr (r-value > 0.35), while the red PxI shows a strong anticorrelation with Sr (r-value<\(-\)0.35), and the green/blue PxI shows a strong correlation with both Rb and Sr (r-value > 0.35).

Overall, the ice-proximal core PxI shows a strong (positive or negative) correlation with all major elements studied. The two cores from western and eastern margins of the main-lake basin exhibit more complex patterns, with an overall positive correlation between the red PxI and Al, Ti, K, Fe, and Rb. In addition, all three band ratios PxIs are strongly anticorrelated with Al, Ti, and Fe (Fig. 3). The green/blue ratio PxI exhibits a strong positive correlation with Ca and Sr in both cores. The two cores also exhibit notable differences, with Si being strongly anticorrelated with red PxI in the eastern main-lake basin (core 7A) and positively correlated to the west (core 27A). Similarly, Si is strongly correlated with red/blue and green/blue PxIs in core 7A, but shows no correlation with these PxIs for core 27A. Ca and Sr and both strongly anticorrelated with red PxI for core 7A, but are weakly correlated with red PxI for core 27A.

Correlation with climatic variables

We use the correlations between sediment PxI and XRF compositional data to narrow our analysis down to the most informative PxI bands: red, red/green, red/blue, and green/blue PxIs.

Temperature

Fig. 4
figure 4

Box plot of correlation coefficients r-values; a, d, g, p-values b, e, h, and bar plot of percentage of correlations significant at 95% level c, f, i for the correlation between red PxI and annual, austral summer (DJF), and austral winter (JJA) temperatures. The x-axis of each plot represents a transect of gravity cores from ice-proximal to ice-distal (Fig. 1). Each individual grey box shows the interquartile range of each distribution, with the whiskers showing the 95th confidence intervals and the central line representing the median. The distribution of values primarily reflects the temporal uncertainty in the varves and resulting PxI timeseries. We highlight series with>20% significant values in orange and >50% significant values in red

Full size image

Mean annual temperature is anticorrelated with red PxI in the three most ice-proximal cores (22A, 30A, 38A; 30–70% of p-values <0.05) and the westernmost core in the main-lake basin (27A; 35% of p-values <0.05). All three band ratio PxIs are anticorrelated with mean annual temperature in the main-lake basin, with 60–70% of p-values <0.05 for the easternmost core 11A (Fig. 4). JJA temperatures exhibit a weak positive correlation with red PxI in the main-lake basin, and no correlation with red PxI in the brazos (ice-proximal fjords) or with the band ratio PxIs in any part of the lake. DJF temperatures are anticorrelated with red PxI in the three most ice-proximal cores (22A, 30A, 38A; 30–75% of p-values <0.05) and westernmost core in the main-lake basin (27A; 60% of p-values <0.05). Band ratio PxIs show a weak positive correlation with DJF temperature in the brazos, and a strong anticorrelation with DJF temperature in the main-lake basin (30–40% and 80–85% of p-values <0.05 for cores 12A and 11A respectively).

Overall, temperature is most correlated, whether positively or negatively, with sediment PxI in the summer (Fig. 4). Red and all three band ratio PxIs show different signals, with red PxI most anticorrelated with temperature in the most ice-proximal cores and western margin of the main-lake basin, and band ratio PxIs most anticorrelated with temperature in the centre and east of the main-lake basin.

Precipitation

Fig. 5
figure 5

Box plot of correlation coefficients r-values; a, d, g, p-values b, e, h, and bar plot of percentage of correlations significant at 95% level c, f, i for the correlation between red PxI and annual, austral summer (DJF), and austral winter (JJA) precipitation

Full size image

Mean annual, austral winter (JJA), and austral summer (DJF) precipitation show very little or no correlation with red or band ratio PxIs in any region of Lago Argentino, with no correlation having more than  40% of significant values. Only the easternmost core 11A, red, red/green, red/blue, and green/blue PxIs exhibit a weak positive correlation with precipitation, having 25–40% of significant positive correlations with DJF precipitation (Fig. 5).

Wind speed

Fig. 6
figure 6

Box plot of correlation coefficients r-values; a, d, g, p-values b, e, h, and bar plot of percentage of correlations significant at 95% level c, f, i for the correlation between red PxI and annual, austral summer (DJF), and austral winter (JJA) wind speeds

Full size image

DJF and mean annual wind speed are strongly negatively correlated with red and band ratio PxIs in the brazos (60–90% of p-values <0.05) and positively correlated with band ratio PxIs at the western margin of the main-lake basin (core 27A; 20–40% of p-values <0.05). DJF wind speed is also strongly positively correlated with red PxI in easternmost core 11A ( 90% of p-values<0.05). In brazos cores 30A and 38A, JJA wind speed shows a similar overall pattern with weaker correlations, exhibiting a moderate anticorrelation with red PxI (15–40% of p-values <0.05 respectively; Fig. 6). JJA wind speeds show no correlation with sediment PxI in the main-lake basin. All band ratio PxIs are strongly negatively correlated with wind speed for most ice-proximal core 22A (60–95% of p-values <0.05) and positively correlated with wind speed in the western margin of the main-lake basin (core 27A; 35–55% of p-values <0.05).

JJA and mean annual wind speed are most correlated with sediment PxI (Fig. 6). Red PxI and band ratio PxIs show different signals, with red PxI most negatively correlated with wind speed in the most ice-proximal cores and positively correlated with wind speed in the eastern margin of the main-lake basin, and band ratio PxIs most anticorrelated with wind speed at the most ice-proximal location and western margin of the main-lake basin.

In summary, several of PxI’s ensemble median correlations meet the p-value<0.1 significance threshold for inclusion as a proxy, as defined by Mann et al. (2008). For summer air temperature, this correlation significance threshold is met by 3 cores (out of 10) for red PxI, 2 cores for red/green and red/blue PxIs, and 4 cores for green/blue PxI. For summer wind speed, this correlation significance threshold is reached by 6 cores for red PxI, 3 cores for red/green PxI, and 1 core for red/blue and green/blue PxIs. No core reaches this significance threshold for precipitation. Based on this correlation significance alone, we might conclude that PxI is a suitable paleo-temperature and paleo-wind speed proxy in multiple cores. We discuss the implications and limitations of this further in the following section.

Discussion

Correlation between sediment pixel intensity, elemental composition, and climate

Sediment PxI exhibits significant correlations with both XRF-derived compositional data and climatic variables, linking changes in core composition to changes in temperature and/or wind speed. In this section, we first discuss and interpret the compositional significance of PxI, and the relationship between PxI, climatic variables, and sediment transport. In the second section, we discuss the advantages and disadvantages of using PxI as a paleoclimatic or paleoenvironmental proxy in Lago Argentino and more broadly through our two hypotheses: 1) sediment pixel intensity can be used as a proxy to reconstruct past climatic changes, and 2) the correlation between sediment PxI and climatic variables might help elucidate the details of lacustrine sedimentation.

We use the correlation between PxI and XRF data to evaluate the compositional significance of core PxI in Lago Argentino’s different depositional environments. Ice proximal core 28A shows a strong positive correlation between red PxI and all major elements (Al, Si, K, Ca, Fe), as well as a strong anticorrelation between green PxI, band ratio PxIs and major elements. Possible minerals enriched in these elements include various feldspars, pyroxenes, and clays which more strongly reflect long wavelength red light, and weakly reflect shorter wavelength green light. Core 27A from the western main-lake basin shows a strong positive correlation between red PxI and Si, Al, Ti, K, and Fe, and the weakest correlation for Ca, Sr, and Rb (Fig. 3). Possible minerals enriched in these elements include quartz, K-feldspar, various pyroxenes, ilmenite and other phases. Many Fe-rich minerals appear optically red (Schuttenhelm 1989; Debret et al. 2011). Core 7A from the eastern margin of the main-lake basin exhibits a strong correlation between red PxI and Al, Ti, K, Fe, and Rb, and a strong anticorrelation with Si, Ca, and Sr. This suggests a similar mineral assemblage to the western margin of the main-lake basin, but with the optically red phases more depleted in silicon. This could be caused by a greater proportion of non-silicate minerals, with ilmenite or haematite being two possible candidates. Analysis of lamina-scale stratigraphy in Lago Argentino ((Van Wyk de Vries et al. 2022); see particularly Figure  5b and c therein) reveals distinct characteristics between the bright and dark layers:

  • Optically bright, winter layers have a relatively low Ca/K ratio and a finer grain size distribution. These layers have less abundant 5–10 \(\mu \)m grains and are composed primarily of 0.5 \(\mu \)m grain size fraction particles.

  • Optically dark, summer layers have a higher Ca/K ratio and a coarser grain size distribution. These layers have more abundant 5–10 \(\mu \)m grains and a secondary 0.5 \(\mu \)m grain size fraction particles.

Similarly, our results show a strong positive correlation between sediment red band PxI and K and a weak correlation or anticorrelation between red band PxI and Ca.

We evaluate the possible climatic controls on sediment production, transport, and preservation to contextualize the relationship between PxI and climatic variables. We separate sediment production, transport, and preservation, as each affects the quantity and type of sediment recorded in a lake in different ways. At Lago Argentino, sediment is primarily produced through glacial erosion, with a minor contribution from fluvial erosion (Van Wyk de Vries et al. 2022). At a first order, glacial erosion rate is controlled by basal sliding speed, which is in turn related to the subglacial water pressure (Cuffey and Paterson 2010; Cook et al. 2020). Temperature can therefore influence glacial erosion by affecting surface melting and the quantity of water at the glacier bed. The magnitude of this effect is, however, challenging to quantify as the relationship between water production, glacier sliding velocity, and glacial erosion are complex (Cuffey and Paterson 2010; Cook et al. 2020).

Sediment transport occurs primarily through the movement of suspended sediment particles through Lago Argentino, with a lesser contribution from subglacial and fluvial sediment transport. Subglacial sediment transport is high when abundant meltwater reaches the glacier bed and the subglacial drainage system is channelized (related to glacier bed dynamics). Fluvial sediment transport is high when river discharge is high, related to high precipitation or rapid ice melt in the case of proglacial rivers. Lacustrine sediment transport is related to water circulation within the lake, the degree of mixing of the water, and the settling velocity of individual particles (Fischer et al. 1979; Spigel and Imberger 1980; Imberger and Patterson 1989; Sugiyama et al. 2016). Lake geometry and bathymetry also affect sediment transport, with constrictions such as Boca del Diablo (Fig. 1) inhibiting sediment through-flow. Lacustrine sediment transport might therefore be affected by wind induced surface currents, seiches, and deeper circulation cells (Imberger and Patterson 1989; Tylmann et al. 2013; Richter et al. 2016) and temperature induced lake stratification (Lewis 1983; Imberger and Patterson 1989). Sediment can also be temporarily rafted in icebergs. Iceberg production rate is related to glacier calving rate, while iceberg motion through the lake is driven by surface winds and subsurface currents.

Finally, sediment quantity and type can be influenced by the resuspension, erosion, or reworking of previously deposited sediment. Lago Argentino is an ultra-oligotrophic lake, with very few traces of bioturbation in any of the cores studied. Sediment can be resuspended at a lake bed from strong wind-induced currents (Kristensen et al. 1992; Bailey and Hamilton 1997; Carper and Bachmann 2011), although this process is limited to lakes at most a few tens of metres deep. The shallowest core analysed in this study was collected at a water depth of 99 m, and the deepest at a water depth of 464 m; all well below the wave base (Bengtsson et al. 2012). Any sediment deposited in Lago Argentino is therefore preserved in the sedimentary record.

Temperature can influence both sediment production and sediment transport at Lago Argentino. We expect the strongest correlative relationship between temperature and sediment PxI in different locations depending on the mechanism by which temperature affects sedimentation: changes in glacier melt or changes in lake-water stratification. If it affects sedimentation through changes in glacier surface melting, erosion, and subglacial sediment transport, we expect the strongest correlative relationship close to the glacier fronts. Conversely, if temperature affects sedimentation through changes in lake-water stratification (Van Wyk de Vries et al. 2022), we expect the strongest correlative relationship to be distributed through the main-lake basin. Our results show high correlations in both of these zones: DJF temperature shows a strong anticorrelation with red PxI in the ice-proximal cores 22A, 30A, and 38A and a strong anticorrelation with band ratio PxIs in main-lake basin cores 11A and 12A (Fig. 4). At Lago Argentino, wind speed does not affect sediment production or resuspension, but might affect sediment transport. Both red and band ratio PxIs are strongly anticorrelated with DJF wind speed in the ice proximal brazos. Red PxI is positively correlated with DJF wind speed in the main-lake basin, while band ratios are weakly anticorrelated or show no relationship. The inverse relationship between the ice-proximal and ice-distal regions of Lago Argentino suggests that higher than average winds promote distal-to-proximal lacustrine circulation and transport sediment from ice-proximal towards ice-distal regions of the lake.

Climate data can vary over small distances, particularly in geographically and topographically complex regions such a Lago Argentino where strong climatic gradients exist (e.g. Lenaerts et al. (2014)). No single meteorological station will be fully representative of the full climatic range of Lago Argentino, which is a limitation of our study. The available climatic data is limited to the southern shore of the main basin of Lago Argentino, with no in-site data available from Brazo Norte or Brazo Upsala (Fig. 1). Therefore, the wind data (from El Calafate airport) and CRU-TS 4.05 climatic data (Harris et al. 2020) are likely most accurate in the eastern main-lake basin and have gradually reduced accuracy in the brazos. We aim to mitigate this limitation by creating multi-month or annual averages of the meteorological variables, which is expected to reduce cross-lake discrepancies relative to higher frequency data. Nevertheless, more detailed and denser meteorological monitoring networks, particularly in the NW of Lago Argentino or located on the lake itself, would provide valuable information for future iterations of this work.

Sediment pixel intensity as a paleoclimatic proxy

Sediment PxI is significantly correlated with summer temperature and wind speed in multiple cores. In this section, we explore whether the relationship between PxI and temperature or wind speed might be used as a paleoclimatic proxy to reconstruct past climatic conditions beyond the observational period. For PxI to be useful as a paleoclimatic or paleoenvironmental proxy, it is necessary to use its relationship with the modern instrumental period to calibrate a predictive model and reconstruct temperature over the period pre-dating instrumental measurements (Gornitz 2009; McShane and Wyner 2011; Tingley et al. 2012). PxI’s ensemble median correlations with air temperature and windspeed meet the p-value<0.1 significance threshold for inclusion as a proxy (Mann et al. 2008). This correlation significance threshold is met by 3 cores for red PxI and 4 cores for green/blue PxI for summer air temperature, and 6 cores for red PxI and 3 cores for red/green PxI for summer wind speed. Precipitation does not reach this significane threshold at any time of the year. While this correlation significance suggests that PxI may be a suitable paleo-temperature and paleo-wind speed proxy, a significant correlation between proxy and climatic data is not the only condition required for it to have predictive value beyond the study period (National Research Council 2006).

For any paleoclimatic reconstruction to be made using the present-day relationship between the proxy (PxI) and climatic variable, their relationship must remain consistent through time. Demonstrating this at Lago Argentino is challenging, as it exhibits strong spatial variability and a non-unique relationship between the proxy and climatic variables of interest. In particular the distance between any given location and the glacier fronts has evolved through time due to glacier advance and retreat (Strelin et al. 2014; Kaplan et al. 2016) and the local conditions have fluctuated on centennial timescales due to large-scale climatic systems (Van Wyk de Vries et al. 2023). The strong spatial variability in Lago Argentino is best illustrated by the correlation between DJF wind speed and red PxI in cores 41A and 42A, located within 3 km of each other in a region where the lake is at its narrowest. Both cores’ red PxI exhibit highly significant (p-values<0.01) correlations with DJF wind speed, but with opposing signs: a strong negative correlation at core 41A (median r-value of \(-\)0.45) and a strong positive correlation at core 42A (median r-value of 0.5). With opposing relationships to climatic drivers within a 3 km length-scale (for a  100 km long lake), even a small spatial shift in sedimentation patterns across Lago Argentino would strongly affect the relationship between sediment PxI and climate at any given location. Any down-core shift in the relationship between sediment PxI and climate would contradict the assumption of stationarity, and prevent the meaningful use of present-day climate-PxI correlations to reconstruct past climate.

The relationship between PxI and climatic variables is also non-unique, as our data shows a statistically significant relationship with both temperature and wind speed. Summer temperature and wind speed themselves do not co-vary (r-value = 0.06; p-value = 0.67), and therefore cannot both be accurately reconstructed using a single proxy. Furthermore, if the contribution of temperature and wind speed has not remained steady through time, PxI might record these climatic drivers to different degrees through time. For instance, the effect of wind could have dominated over temperature during periods when the mean temperature at Lago Argentino was colder than present. Such a case would also invalidate the assumption of stationarity. We cannot assess the relative effect of temperature and wind speed on PxI beyond the instrumental period, which limits the potential of PxI to predict paleo-temperatures and paleo-wind speeds.

PxI is not significantly related to precipitation at Lago Argentino, so cannot be used as a paleo-precipitation proxy. The use of PxI as a paleo-temperature or paleo-wind speed appears statistically plausible based on its correlation with climatic variables, but the non-uniqueness and strong spatial variability of this relationship suggest that stationarity cannot be assumed at Lago Argentino. We therefore conclude that PxI is also not a suitable proxy for temperature or wind speed at Lago Argentino. This does not mean that it is not a useful measure since we also demonstrate that PxI does reflect changes in certain climatological variables (here wind speed and temperature) over the instrumental period. Based on this, quantitative paleoclimatic reconstruction may be possible at other lakes where linearity and stationarity can be established. In complex lakes such as Lago Argentino where it cannot be used as a proxy, it can serve the alternative purpose of better understanding the climatic drivers of sedimentation over the instrumental period—information that is valuable for understanding lake dynamics even if it cannot be temporally extrapolated to the entire sediment record.

While it is not a suitable proxy for any climatic variable, correlation (or lack thereof) between sediment PxI and temperature, precipitation, and wind speed does reveal differing controls on sedimentation in the summer and winter, and across Lago Argentino. Mean annual, DJF, and JJA precipitation show no relationship with PxI, showing that precipitation does not impact sedimentation within Lago Argentino in a way that translates into a PxI signal. This suggests that precipitation does not have a major influence on the very slow settling of fine grains reflected in these cores, but it may still influence overall sedimentation in other ways—for instance through the deposition of coarser flood deposits near the lake margins not sampled in our coring campaign. Mean annual and DJF temperature and wind speed are significantly correlated with red and band ratio PxIs, suggesting that these do exert a control on local sedimentation. The seasonality and sign of these correlations can provide insight into the nature of these controls: DJF temperature anticorrelates with red PxI across the entire basin, while DJF wind speed is anticorrelated in the ice proximal brazos and correlated in the ice-distal main-lake basin. JJA temperature shows no relationship with PxI, while JJA wind speed is weakly anticorrelated in the ice proximal brazos and correlated in the ice-distal main-lake basin. The effect of temperature is therefore more seasonal (limited to the summer) and more similar across the whole lake, while the effect of wind differs in the ice-proximal to ice-distal regions of the lake and operates throughout the year. We note that choosing only one single core from our dataset would in a number of cases miss these relationships between climate and sedimentation, or lead to different conclusions if interpreted alone. Caution must be applied in studies using only a single lake core or a small number of cores from the same region of a lake, particularly in large and complex lakes such as Lago Argentino. Future studies should ensure that a sufficient number of cores are collected to capture the spatial patterns in sedimentation across the lake basins, and where possible target lakes with simpler geometries and depositional environments than Lago Argentino. Patagonian lakes with a single branch and with input dominated by a single glacier, such as Lago Viedma or Lago Grey, may provide more favourable environments for a linear and stationary relationship between climate and sediment properties.

Fig. 7
figure 7

Summary diagram of key climatic controls on sedimentation in Lago Argentino in the austral summer (DJF)

Full size image

The greater influx of optically dark sediment close to the glacier fronts during summer is enhanced by higher temperatures or suppressed by lower temperatures. In the main basin of the lake, settling of these same optically dark grains is promoted during times of high temperature by a strong thermal stratification and limited vertical water column mixing, while low temperatures suppress this stratification (Fig. 7). Dark grains are generally coarser (5–10 \(\mu \)m grain size) and have a Stokes settling timescale of several months, relative to optically bright grains in the \(\sim \)0.5 \(\mu \)m grain size fraction with a settling timescale of decades (Van Wyk de Vries et al. 2022). Vertical stratification and reduced mixing of the water columns enhances the settling of both grain size fractions, but the differential settling timescales results in a summer enrichment in coarse dark grains ( Van Wyk de Vries et al. (2022); see Appendix 1 for further details) In addition, higher than average summer wind speeds drive enhanced horizontal mixing and increased migration of sediment-rich waters towards ice-distal regions of the lake, thus reducing settling of optically bright grains close to the ice front and increasing it in the main-lake basin. The effect of higher than average wind speeds is also present in the winter, although to a lesser extent. Detailed, basin-wide measurements of lake circulation and suspended sediment concentrations (Sugiyama et al. 2016, 2021) yield similar information about the drivers of sedimentation across the lake basin. Such measurements are, however, expensive and logistically challenging to collect, and are available at few locations. We show here that correlation between sediment properties and local climatic variables can yield similar information about the large-scale climatic drivers of sedimentation.

Conclusions

We evaluated whether the optical properties of a large dataset of cores from Lago Argentino, southern Patagonia, provide a suitable paleoclimate proxy, and what they show about the dominant climatic drivers of sedimentation. We used 20 micron resolution linescan images and semi-automated varve counts to extract annual resolution sediment pixel intensity data. We then compared this pixel intensity data to both x-ray fluorescence compositional data of these same lake cores, and local temperature, precipitation, and wind speed data. We showed that sediment pixel intensities are significantly correlated with XRF compositional data, with red, red to green ratio, red to blue ratio, and green to blue ratio pixel intensities exhibiting the strongest relationship with available major elements. Pixel intensities have significant correlations with austral summer (DJF) temperatures and wind speeds, but not with precipitation. We established that several cores reach the correlation significance thresholds for use as paleo-temperature or paleo-wind speed proxies, but that the assumption of stationarity cannot be made at Lago Argentino. Despite being unsuitable as paleoclimatic proxies at Lago Argentino, our data reveals how climate affects sedimentation in this location: 1) precipitation does not affect sedimentation; 2) high summer temperatures promote ice melt and thermal stratification, increasing the settling of coarser, optically dark grains; 3) high wind speeds, particularly in the summer, increase the lateral flux of sediment-laden waters from ice-proximal to ice-distal regions of Lago Argentino, decreasing the settling of optically bright, fine grains in the brazos and increasing it in the main-lake basin. In addition, the significant correlations suggest that PxI may be useful as a paleoclimatic proxy in other, simpler lake systems. Our data highlights that caution must be applied when searching for paleoenvironmental proxies in complex environments, and that records unsuitable for use as proxies might still yield valuable information about their sedimentary environment.