Unravelling the tree cover dynamics over the last 20,000 years on the Northern Hemisphere

Abstract. Over the last 20,000 years, Northern Hemisphere vegetation underwent major shifts in response to orbital changes, rising CO₂, and ice sheet retreat. Using the large-scale pollen-based tree cover reconstruction by Schild et al. (2025), we evaluate the performance of the MPI-ESM Earth System Model in simulating tree cover dynamics from the Last Glacial Maximum to the present. Although the model reproduces the broad increase in tree cover during deglaciation and decrease throughout the Holocene, it fails to simulate the mid-Holocene maximum observed in the reconstructions. The model does capture the shift from energy-limited conditions during deglaciation to water-limited conditions in the early to mid-Holocene, and then back to energy-limited conditions in the late Holocene, but regional discrepancies remain substantial. MPI-ESM simulates too much forest in sparsely forested areas and too little forest in densely forested areas, particularly in mid- and high-latitude regions. Statistical analyses indicate that summer temperature dominates simulated high-latitude forests, while precipitation is critical in most other regions, contrasting with reconstructions that highlight cold-season temperature in temperate and boreal forests. Areas of model-data agreement show largely linear responses to climate drivers, whereas regions of disagreement exhibit non-linear dynamics to the temperature of the warmest month and over-sensitivity of the plant-physiological CO₂ response. Employing an emulator with a bias-corrected climate reduces the mismatch in the forest steppe transition zones, but does not lead to an overall improvement of the model-data agreement. In particular, the mismatch in the boreal region remains unresolved, suggesting structural limitations in the model. Improving dynamic vegetation models for simulating climatic transitions in both, past and future contexts, requires integrating realistic soil and permafrost processes, dynamic biome thresholds and disturbance regimes. Trait-based approaches could lead to better representation of the vegetation response to climate changes.