Development of new global lake brGDGT-temperature calibrations: advances, applications, challenges, and recommendations

Branched glycerol dialkyl glycerol tetraethers (brGDGTs) are a group of temperature-sensitive membrane lipids found in bacteria that have been widely used in palaeo-temperature reconstruction. Despite recent advances in analytical methods, calibration datasets and statistical modelling approaches, one of the current challenges in Quaternary science remains in determining the most appropriate calibration model for reconstructing past changes in climate. We address this challenge by expanding existing calibration datasets, and by evaluating calibration models constructed using a range of statistical modelling approaches. We further evaluate model performance by applying the calibrations to published downcore records from contrasting environments and across different Quaternary timescales.

Our study expands existing calibrations and includes new data from Antarctic lakes, providing greater confidence and applicability across a wider range of global environments. Results show robust brGDGT-temperature relationships on a global scale within the temperature range of approximately −2 °C to +31 °C covered in this study, with the random forest (RF) models performing the best (highest R2cv and lowest RMSEP) to estimate mean temperature of Months Above Freezing (MAF) and Mean Summer (air) Temperature (MST). Examination of uncertainties suggests the best models are accurately modelling all the features of the brGDGT-temperature relationships.

To evaluate model performance downcore we apply and recommend a suite of exploratory statistical analyses to help identify core-samples that have unusual, no-analogue compositions, and use measures of correlation and concordance to summarise the similarity in trends and absolute values among reconstructions as a tool to suggest which reconstructions may be more reliable and where to use caution. Our results demonstrate that, although cross-validated calibration R2cv and RMSEP may indicate good model performance for the calibration data, a thorough assessment is required to assess reconstruction reliability when a model is applied downcore at a specific site. Our findings highlight the complexities and caveats of different methods for global temperature calibrations. The implications of our work are also relevant to other calibration studies in Quaternary science.