Past climate reconstructions are essential for understanding climate mechanisms and drivers. Our focus is on the South American continent over the past 2000 years. We offer a new reconstruction that particularly utilizes data from speleothems, previously absent from continent-wide reconstructions. We use paleoclimate data assimilation, a reconstruction method that combines information from climate archives and climate simulations.

To look at climate over the past 12 000 years, we reconstruct spatial temperature using natural climate archives and information from model simulations. Our results show mild global mean warmth around 6000 years ago, which differs somewhat from past reconstructions. Undiagnosed seasonal biases in the data could explain some of the observed temperature change, but this still would not explain the large difference between many reconstructions and climate models over this period.

We simulate the past 1000 years of glacier length variability using a simple glacier model and an ensemble of global climate model simulations. Glaciers with long response times are more likely to record global climate changes caused by events like volcanic eruptions and greenhouse gas emissions, while glaciers with short response times are more likely to record natural variability. This difference stems from differences in the frequency spectra of natural and forced temperature variability.

The El Niño–Southern Oscillation (ENSO) and Southern Annular Mode (SAM) are two important modes of climate variability, strongly influencing climate across the tropics and Southern Hemisphere mid- to high latitudes. This study sheds light on their relationship over the past millennium, combining evidence from palaeoclimate proxy archives and climate models. We show that their indices were mostly negatively correlated with fluctuations likely driven by internal variability in the climate system.

Antarctic temperature changes over the past millennia have been reconstructed from isotope records in ice cores in several studies. However, the link between both variables is complex. Here, we investigate the extent to which this affects the robustness of temperature reconstructions using pseudoproxy and data assimilation experiments. We show that the reconstruction skill is limited, especially at the regional scale, due to a weak and nonstationary covariance between δ¹⁸O and temperature.