As global warming is proceeding due to rising greenhouse gas concentrations, the Earth system moves towards climate states that challenge adaptation. Past Earth system states are offering possible modelling systems for the global warming of the coming decades. These include the climate of the mid-Pliocene (similar to 3 Ma), the last interglacial (similar to 129-116 ka) and the mid-Holocene (similar to 6 ka). The simulations for these past warm periods are the key experiments in the Paleoclimate Model Intercomparison Project (PMIP) phase 4, contributing to phase 6 of the Coupled Model Intercomparison Project (CMIP6). Paleoclimate modelling has long been regarded as a robust out-of-sample test bed of the climate models used to project future climate changes. Here, we document the model setup for PMIP4 experiments with EC-Earth3-LR and present the large-scale features from the simulations for the mid-Holocene, the last interglacial and the mid-Pliocene. Using the pre-industrial climate as a reference state, we show global temperature changes, large-scale Hadley circulation and Walker circulation, polar warming, global monsoons and the climate variability modes - El Nino-Southern Oscillation (ENSO), the Pacific Decadal Oscillation (PDO) and the Atlantic Multidecadal Oscillation (AMO). EC-Earth3-LR simulates reasonable climate responses during past warm periods, as shown in the other PMIP4-CMIP6 model ensemble. The systematic comparison of these climate changes in past three warm periods in an individual model demonstrates the model's ability to capture the climate response under different climate forcings, providing potential implications for confidence in future projections with the EC-Earth model.

Palaeoclimate simulations improve our understanding of the climate, inform us about the performance of climate models in a different climate scenario, and help to identify robust features of the climate system. Here, we analyse Arctic warming in an ensemble of 16 simulations of the mid-Pliocene Warm Period (mPWP), derived from the Pliocene Model Intercomparison Project Phase 2 (PlioMIP2). The PlioMIP2 ensemble simulates Arctic (60-90 degrees N) annual mean surface air temperature (SAT) increases of 3.7 to 11.6 degrees C compared to the pre-industrial period, with a multimodel mean (MMM) increase of 7.2 degrees C. The Arctic warming amplification ratio relative to global SAT anomalies in the ensemble ranges from 1.8 to 3.1 (MMM is 2.3). Sea ice extent anomalies range from -3.0 to -10.4 x 10(6) km(2), with a MMM anomaly of -5.6 x 10 6 km(2), which constitutes a decrease of 53 % compared to the pre-industrial period. The majority (11 out of 16) of models simulate summer seaice-free conditions (<= 1 x 10(6) km(2)) in their mPWP simulation. The ensemble tends to underestimate SAT in the Arctic when compared to available reconstructions, although the degree of underestimation varies strongly between the simulations. The simulations with the highest Arctic SAT anomalies tend to match the proxy dataset in its current form better. The ensemble shows some agreement with reconstructions of sea ice, particularly with regard to seasonal sea ice. Large uncertainties limit the confidence that can be placed in the findings and the compatibility of the different proxy datasets. We show that while reducing uncertainties in the reconstructions could decrease the SAT data-model discord substantially, further improvements are likely to be found in enhanced boundary conditions or model physics. Lastly, we compare the Arctic warming in the mPWP to projections of future Arctic warming and find that the PlioMIP2 ensemble simulates greater Arctic amplification than CMIP5 future climate simulations and an increase instead of a decrease in Atlantic Meridional Overturning Circulation (AMOC) strength compared to pre-industrial period. The results highlight the importance of slow feedbacks in equilibrium climate simulations, and that caution must be taken when using simulations of the mPWP as an analogue for future climate change.