Calibration and application of the branched GDGT temperature proxy on East African lake sediments

Branched glycerol dialkyl glycerol tetraethers (brGDGTs) are a novel proxy for mean annual air temperature (MAAT) and have the potential to be broadly applicable to climate reconstruction using lacustrine sediments. Several calibrations have been put forth relating brGDGT distributions to MAAT using a variety of linear regressions, including the methylation (MBT) and cyclization (CBT) indices of brGDGTs, the relative abundances of the major, non-cyclized brGDGTs (MbrGDGTs), and best subsets regression (BSR) of the fractional abundances of the nine most common brGDGTs. However, these calibrations have rarely been applied to lake sediment cores to reconstruct temperatures and test the applicability of this proxy as a paleothermometer.We present an expanded East African lakes surface sediment brGDGT dataset based upon 111 lakes and examine three methods of calibrating brGDGTs to MAAT. These methods include recalculations of the East African lake MBT/CBT calibration and MbrGDGTs calibrations, as well as a new stepwise forward selection (SFS) calibration that uses the four combined brGDGTs that explain the most variance in temperature in our calibration set. We apply these new calibrations as well as five previously published lacustrine brGDGT calibrations to the brGDGT distributions of our surface sediment dataset and a 48. kyr sediment core from Sacred Lake, Mt. Kenya, producing the first brGDGT temperature reconstruction available from a small tropical lake. We compare the reconstructed temperatures to previously published paleotemperature records from East Africa to help us assess the performance of the brGDGT calibrations. We find that the SFS calibration has a consistently low root mean squared error of prediction (RMSEP) over the entire range of MAAT, while the MBT/CBT and MbrGDGT calibrations have relatively large RMSEPs, particularly between lakes with similar temperatures but variable pH. This suggests that these techniques do not properly deconvolve the temperature and pH signals recorded in the distributions of the brGDGTs. We further find that only the SFS calibration produces a credible reconstructed temperature history from Sacred Lake when compared to other last glacial maximum paleotemperature estimates from East Africa. Thus, we advocate for the use of the SFS calibration when reconstructing paleotemperatures from brGDGTs in East Africa.