Simulations of the last two glacial periods are run using a computer model in which the atmosphere and ice sheets interact. The results show that the initial conditions used in the simulations are the primary reason for the difference in simulated North American ice sheet volume between each period. Thus, the climate leading up to the glacial maxima and other factors, such as vegetation, are important contributors to the differences in the ice sheets at the Last and Penultimate glacial maxima.

The year 1740 was the coldest in central Europe since at least 1421. New monthly global climate reconstructions, together with daily weather reconstructions, allow a detailed view of this climatic event. Following several severe cold spells in January and February, a persistent circulation pattern with blocking over the British Isles caused northerly flow towards western Europe during a large part of the year. It was one of the strongest, arguably unforced excursions in European temperature.

The Benguela Upwelling System is a region in the SE Atlantic Ocean of high biological productivity. It comprises several water masses such as the Benguela Current, South Atlantic Central Water, and Indian Ocean Agulhas waters. We analyzed planktonic foraminifera from IODP Sites U1575 and U1576 to characterize water masses and their interplay in the Pleistocene. This defined changes in the local thermocline, which were linked to long-term Benguela Niño- and Niña-like and deglaciation events.

Old observations are necessary to understand the atmosphere. When direct observations are not available, one can use indirect observations, such as tide gauges, which measure the sea level in port cities. The sea level rises when local air pressure decreases and when wind pushes water towards the coast. Several centuries-long tide gauge records are available. We show that these can be complementary to direct pressure observations for studying storms and anticyclones in the 19th century.

This study used 1802 drought and 1977 famine records from historical documents to reconstruct the spatial–temporal progression of the Chongzhen drought (1627–1644) in China and its impacts. We advance this research by reconstructing the annual spatial patterns and regional series of drought; demonstrating drought as the primary factor triggering famine; and identifying the transmission pathway of the drought's impacts and how social factors, especially human responses, regulated these impacts.

We present simulations of the mid-Holocene and pre-industrial climate of Australia using coarse- (2°) and finer-resolution (0.44°) climate models. These simulations are compared to bioclimatic representations of the palaeoclimate of the mid-Holocene. The finer-resolution simulations reduce the bias between the model and the bioclimatic results and highlight the improved value of using finer-resolution models to simulate the palaeoclimate.

In our work, we reconstruct the weather of the extremely cold and wet summer in 1816 using a weather forecasting model to obtain high-resolution, three-dimensional weather simulations. We refine our simulations with surface pressure and temperature observations, representing a novel approach for this period. Our results show that this approach yields detailed and accurate weather reconstructions, opening the door to analyzing past weather events and their impacts in detail.

Studying climate conditions near the Antarctic ice sheet (AIS) during Earth’s past warm periods informs us about how global warming may influence AIS ice loss. Using a global climate model, we investigate climate conditions near the AIS during the Last Interglacial (129 to 116 kyr ago), a period with warmer global temperatures and higher sea level than today. We identify the orbital and freshwater forcings that could cause ice loss and probe the mechanisms that lead to warmer climate conditions.

We develop the Multi Ensemble Machine Learning Model (MEMLM) for reconstructing palaeoenvironments from microfossil assemblages. The machine-learning approaches, which include random tree and natural language processing techniques, substantially outperform classical approaches under cross-validation, but they can fail when applied to reconstruct past environments. Statistical significance testing is found sufficient to identify these unreliable reconstructions.