Global temperature calibrations based on 3-hydroxy fatty acids in lacustrine settings

Sai Ke , Pierre Sabatier, Christelle Anquetil, Arnaud Huguet

Lakes are important archives of palaeoclimate records on the continent and decoding them could improve the parameterization of climate models to better predict the future. However, the limited temperature proxies currently applicable in lake environments make high-resolution climate records scarce. 3-hydroxy fatty acids (3-OH FAs) are bacterial lipids which have recently been proposed as temperature proxies in terrestrial and aquatic settings, with only regional studies in lakes. Here, the aim was to investigate the applicability of 3-OH FAs as temperature proxies in lakes at the global scale using an extended dataset (n=241) consisting of previously published data (n=113) and 128 newly analyzed lake surface sediments from all over the globe. The distribution of 3-OH FAs in lakes was shown to be generally homogeneous and distinct from the one observed in global soil and marine datasets, suggesting a major contribution of in situ production for these lipids at the global scale. Among the lakes, modest changes in 3-OH FA signatures were observed, likely reflecting variations in lipid sources and/or microbial diversity, with general independence of the 3-OH FA signal to climatic forcing, supporting the applicability of these compounds across lakes with diverse geographic settings. Only weak linear relationships between 3-OH FA relative abundances and temperature were observed. Three machine-learning models (multiple linear regression, random forest and TabPFN) were tested and revealed moderate to strong correlations with temperature. The application of the TabPFN model to our global lacustrine dataset especially generated a strong calibration to the mean temperature of Months Above Freezing (MAF), with an R2 of 0.75 and RMSE of 3.2 °C. The newly developed calibrations were applied to the only 3-OH FA lake core record available to date. The global model based on the TabPFN algorithm provided reliable reconstructions of mean summer temperature. Meanwhile, strong regional effects were found in the models reconstructing mean annual air temperatures and MAF temperatures, leading to the development of local 3-OH FA temperature calibrations based on the same machine-learning algorithms. The latter provided similar trends with 3-OH FA-based (RAN13 index) and brGDGT-based (MBT’5Me index) regional records. Altogether, these results demonstrate the potential of 3-OH FAs as temperature proxies in global lakes through the use of refined statistical models.