Methodology:

Chronology

A chronology for each record was established using Accelerator Mass Spectrometry (AMS) radiocarbon dating of 21 samples from the LP08 record and 15 samples from the LP16 record. Calibration of radiocarbon ages was undertaken in OXCAL v.4.4 using the SHCal20.14C Southern Hemisphere atmosphere calibration curve (SH20). Radiocarbon ages are reported as conventional radiocarbon years BP (14C years BP) ±1sigma; and calibrated ages as 2sigma; (95.4%) ranges, median and mean calendar years BP (cal a BP and cal ka BP, relative to 1950 CE), rounded to the nearest ten years. Age-depth models were developed using Bayesian age-depth modelling software (rBACON v.2.5). Modelled age data mean ages produced by the SH20M1H (Southern Hemisphere, SHcal20, radiocarbon calibration curve) in rBACON, where M1 indicates Model 1 and H indicates the inclusion of a hiatus in the model.

Geochemistry

Sediment cores were collected using a UWITEC-gravity corer, Livingston piston corer and a Russian corer from the deepest point (approx 3.5 m of water depth) in Lago Pato:

- LP08 record from 51.3003 S, 72.6786 W, 32 m a.s.l. is 600 cm long

- LP16 record from 51.3031 S, 72.6816 W, 33?34 m a.s.l. is 295 cm long

Contiguous downcore wet-sediment Energy Dispersive Spectrometry (EDS) X-ray fluorescence core scanning (XRF-CS) data was collected using an ITRAX XRF core scanner at Aberystwyth University fitted with a Molybdenum (Mo) anode X-ray tube (settings: 30 kV, 50 mA, count time 10 seconds, at 2 mm contiguous intervals and for LP08 Unit 6 (equivalent to mean ± 2-sigma: 4.5±7.0 years), at 200 μm intervals for LP08 Unit 1 (1.3±4.2 years), with LP08 basal Unit 1 scanned at 100 micrometers, and at 500 micrometers for LP16 Units 2-6 (9.6±17.4 years) and at 200 micrometers for LP16 Unit 1 200 micrometers (1.1±1.6 years). Data from finely laminated glaciolacustrine sediments in Units 1-2 were measured at or smoothed to 200 micrometers (from 100 micrometers interval data) before analysis.

Time series analysis

Log-n element/Ti ratio XRF-CS Z-scores were used for time series analysis (Fast Fourier Transform FFT, periodograms, Lomb-Scargle Power Spectrum, Wavelet Power Spectrum, Peak Identification) in MATLAB. Equally spaced (10-year and 100-year) time-intervals were generated using a Piecewise Cubic Hermite Interpolated Polynomial (PCHIP) function, which avoids spline artefacts and preserve the shape of the original XRF-CS data series (Grinsted et al., 2004; Trauth, 2015). Time series data were detrended (polynomial linear best fit) to remove the long-term linear trend. Second order polynomial Locally Weighted Scatterplot Smoothing (LOESS) 100-year smoothing (0.1 sampling interval with outliers removed) was also used to compare datasets to published data.

Data were analysed in MATLAB v. R2021a, R v. 4.1.0/Rstudio v. 1.4.171, using the R packages Vegan, Rioja, Tidyverse, ggplot2, Ggally v. 2.1.2. Code is available from: https://github.com/stever60/Lago_Pato

Data quality:

XRF-CS data from finely laminated glaciolacustrine sediments in Units 1-2 were smoothed to 200 micrometers, other data to 2mm, and equal spaced time-intervals (10-years and 100-years) for use in time series analysis were generated. Time series datasets were centred, standardised (Z-scores), detrended and interpolated using a 10-year or 100-year Piecewise Cubic Hermite Interpolating Polynomial (PCHIP) function, which avoids spline artefacts and preserve the shape of the original XRF-CS data series.