Proxy System Biases partially resolve long-standing paleoclimate data-model discrepancies in Tropical East Africa

Paleoclimate data-model comparison is a vital tool for evaluating climate model performance and constraining predictions of climate change. Paleoclimate records provide an important out-of-sample target to test models in climates with different boundary conditions and improve model physics. General Circulation Models (GCMs) from the Paleoclimate Model Intercomparison Project (PMIP) underestimate temperature changes in tropical Africa during key periods in the past, calling into question model predictions of future climate change in this region. We compare a new suite of 8 temperature records from tropical African lakes derived from glycerol dialkyl glycerol tetraethers (GDGTs) to simulations of the Last Glacial Maximum (LGM, 21 kya) and mid-Holocene (MH, 6 kya) from PMIP3 and PMIP4, and find that there is significant proxy/model temperature disagreement during both periods; GCMs are over 2 °C cooler than proxies during the MH, and 2 °C warmer than proxies during the LGM. This could arise from issues with GCMs or the GDGT records. We use a proxy system model (PSM) to investigate multiple potential sources of proxy uncertainty, including non-stationarity in the relationship between air and lake temperature, seasonality, and calibration error. We find that proxy system error can account for significant portions of this discrepancy during the MH (up to 80%) assuming the maximum possible error for each potential source, but not during the LGM. Our findings highlight potential areas for improvement in interpretation of proxy records by explicitly modeling and quantifying uncertainty in reconstructed temperatures.