| 15 Jul 2026
European seasonal temperature change since the Last Glacial Maximum from climate models
Abstract. The Last Glacial Maximum (LGM, ~21 ka BP) is a key period for evaluating General Circulation Models (GCMs) used for future climate projections. Determining whether regional cooling reflects a uniform decrease in monthly temperatures throughout the year (i.e., no change in temperature seasonality) or a change in temperature seasonality, would have significant implications for many paleotemperature reconstructions, including those based on paleo-glacier equilibrium lines. In this study, we provide an assessment of GCMs as well as iLOVECLIM Earth system model of intermediate complexity downscaled over Europe in terms of temperature seasonality and of its changes at the LGM, compared to the pre-industrial era. Although models show a large dispersion, some common features emerge. The average temperature seasonality decreases from PI to LGM over southern Europe, while it increases further north. Models simulate a slight and consistent increase in seasonality changes along the Atlantic coast, but a larger and more variable increase over eastern Europe. We identify variations in LGM MTCO (mean temperature of the coldest month) as the primary drivers of both temperature seasonality changes and inter-model dispersion in across the GCM and iLOVECLIM outputs, with the largest disagreements occurring in northeastern Europe, over and near the Fennoscandian ice sheet. Motivated by the fact that the iLOVECLIM model produces some features largely different from the GCMs, we perform a series of sensitivity experiments with iLOVECLIM, to understand its different sensitivity to LGM boundary conditions and forcings. These experiments study the sensitivity of the simulated temperature seasonality amplitude to changes in greenhouse gas concentration, thermohaline circulation, albedo and topography of the Fennoscandian ice sheet and vegetation cover. We show that none of these processes, including glacial forcings (inter-model ice sheet extents), explain the observed seasonality change mismatch between iLOVECLM and GCMs. A significant reduction in this mismatch is achieved only by modifying the vertical parametrisation of radiative profiles in iLOVECLIM, suggesting that the lack of an explicit vertical representation of radiative profiles in iLOVECLIM may bias the simulated temperature seasonality changes at the LGM relative to PI. Pollen-based reconstructions are generally consistent with model results. However, the regions that display the largest inter-model differences are also not covered by this type of reconstruction, which is restricted to areas outside the ice sheets. European temperature seasonality since the LGM therefore remains a key yet poorly constrained characteristic of LGM climates, calling for more single-model sensitivity experiments to improve our understanding of past and future temperature seasonality changes.
This manuscript investigates changes in European temperature seasonality during the Last Glacial Maximum (LGM) relative to the pre-industrial period (PI), combining PMIP3/PMIP4 GCM simulations with iLOVECLIM and a series of sensitivity experiments. The scientific question is important, particularly because the focus on MTWA and MTCO provides useful information beyond mean annual temperature changes. The pronounced inter-model disagreement around the Fennoscandian Ice Sheet, largely associated with winter temperature responses, is an interesting result, and the iLOVECLIM sensitivity experiments provide a valuable attempt to explore the underlying processes.
The manuscript presents useful results, but some of the mechanistic interpretations appear stronger than the current experiments can support. In particular, the distinction between identifying processes that affect seasonality and attributing the original inter-model differences to those processes needs to be made more carefully. Several aspects of the inter-model spread and the robustness of the conclusions also require further analysis or clarification.
I think the manuscript has good potential for publication after substantial revision.
Major Comments:
1. The manuscript discusses both inter-model differences across the ensemble and the distinctive response of iLOVECLIM relative to the GCMs. These questions are closely related, but not fully equivalent. The sensitivity experiments provide useful insight into processes that may contribute to the iLOVECLIM–GCM discrepancy, but it is less clear whether the same processes explain the spread among the GCMs.
I therefore suggest making this distinction clearer in the Introduction, Section 4, and Conclusion, and framing the iLOVECLIM experiments primarily as a mechanistic exploration of its distinctive response.
2. The main focus of the manuscript is the LGM–PI change in temperature seasonality. However, Fig. 5 mainly compares the inter-model spread in absolute LGM MTWA and MTCO. If the aim is to show that the disagreement in seasonality change is mainly associated with the winter response, it would be more direct to compare the inter-model spread in LGM–PI MTWA changes (ΔMTWA) and LGM–PI MTCO changes (ΔMTCO).
I suggest adding this comparison and, if the conclusion remains robust, using it to support the interpretation that MTCO is the main contributor to the inter-model differences in seasonality change.
3. The PMIP simulations use different LGM ice-sheet reconstructions, while the largest disagreement in seasonality and MTCO occurs over and around the Fennoscandian Ice Sheet. The inter-model spread in this region may therefore reflect not only differences in model physics, but also differences in ice-sheet extent, elevation, and topographic representation.
Since the ensemble includes configurations of the same model using different ice-sheet reconstructions (e.g., iLOVECLIM), these may provide a useful opportunity to assess the sensitivity to ice-sheet boundary conditions. If a quantitative separation is difficult, it would still be helpful to acknowledge that both boundary conditions and model physics may contribute to the reported spread.
4. In Section 4, the authors conclude that differences in ice-sheet representation are unlikely to explain the iLOVECLIM–GCM seasonality mismatch, noting that even the extreme FlatIS experiment produces only a relatively moderate seasonality response. However, the Conclusion states that ice-sheet-forced albedo changes explain a large part of the simulated seasonality changes and may contribute to persistent inter-model discrepancies. These statements seem difficult to reconcile. The authors should clarify whether ice-sheet effects are considered important for the mean LGM–PI seasonality response, for the inter-model disagreement, or both, and revise the conclusions accordingly.
5.The IRN experiment substantially reduces the difference between iLOVECLIM and the GCM multi-model mean, suggesting that the treatment of the vertical radiative profile may contribute to the distinctive iLOVECLIM winter response. However, a reduced model difference after modifying one process does not necessarily demonstrate that this process caused the original discrepancy, as other mechanisms or compensating effects may also be involved. I would therefore recommend that the authors reconsider the causal framing in the Abstract, Discussion, and Conclusion to better reflect what can be supported by the current experiments.
6. The authors note that CESM2-1 behaves as an outlier in terms of magnitude but retain it in the ensemble analysis. While this is reasonable, I wonder to what extent this outlier influences the inter-model standard deviation in Fig. 5, particularly the large MTCO spread. It would be helpful to clarify the robustness of this result.
7. The GCM outputs are remapped to 0.25° using nearest-neighbor interpolation, whereas iLOVECLIM is dynamically downscaled to 0.25°. These procedures are fundamentally different, as remapping does not add new spatial information. I suggest distinguishing them more clearly throughout the manuscript and clarifying whether this difference could affect the comparison, particularly around the FIS where topography is important.
8. The integration lengths and equilibration status of the sensitivity experiments are not sufficiently clear. This is particularly relevant for FWF and BRINES, given the longer adjustment timescales of the ocean circulation. It would be helpful to clarify the integration and averaging periods and whether these experiments had reached a sufficiently equilibrated state.
Minor comments:
1. PMIP simulations are averaged over 50 years whereas iLOVECLIM is averaged over 100 years. Please explain the reason for this difference and also check the Fig. 4 caption, which states that the last 50 years are used for each model.
2. Please clarify whether spatial averages are area weighted, as unweighted averaging would overrepresent high-latitude grid cells.
3. For the FWF experiment, please specify the exact spatial region over which the 0.2 Sv freshwater forcing is applied, its duration, and the resulting AMOC state.
4. In the Fig. 3 caption, “35°N–30°N” appears to be a typo, as the Methods describe the study region as 35–70°N. Please check.
5. Expressions such as “MTCO decreases by +5 to +15°C” contain an apparent inconsistency between the sign and the wording. Please revise throughout.
6. Please standardise terminology and spelling throughout, including “Fennoscandien/Fennoscandian,” iLOVECLIM, experiment names, and “Mean Model/multi-model mean.”
7. Some information in Table 1 should be checked carefully. For example, “ICE-6C_C” for MPI-ESM1-2 may be a typo for “ICE-6G_C.” The PMIP/CMIP designation for HadCM3 should also be verified.
8. Please standardise “near-neighbour” to “nearest neighbour” in the Methods.
9. L59, iLOVECLM
10. L225, peloclimate
11. L921, Citation year for Harmand et al.: it is cited as 2025 in the text but listed as 2015 in the references.
The manuscript would benefit from careful proofreading, as several typographical and citation inconsistencies remain.