The Last Glacial Maximum (LGM), approximately 20,000 years ago, serves as a crucial test case for evaluating the accuracy of climate models in simulating climate states significantly different from the present. However, evaluating these models is complicated by the indirect and uncertain nature of past environmental variable reconstructions, such as sea surface temperature (SST). This study introduces a novel approach to assess LGM climate simulations by utilizing the fundamental macroecological principle that community similarity decreases with increasing thermal distance. This principle is applied to planktonic foraminifera, marine microorganisms whose distribution is strongly influenced by temperature. The study's central hypothesis is that if climate models accurately simulate LGM cooling, the similarity-decay pattern derived from combining fossil species assemblages with simulated temperatures should closely match the modern pattern. Deviations from this match indicate potential inaccuracies in the climate models' representation of LGM temperature gradients, particularly in their spatial patterns. Understanding the spatial patterns of glacial cooling is vital, as they are fundamental to climate dynamics and govern the distribution of habitats and ecosystems. Direct comparison of observational data and model output is challenging due to inherent ambiguities within both paleo-reconstructions and climate simulations. Paleo-reconstructions, particularly SST reconstructions derived from marine microfossils, are subject to uncertainties about depth and seasonality, while simulations face uncertainties about model setup, design, and structural aspects. The research aims to overcome these limitations by employing an ecological approach that does not rely on the often uncertain temperature reconstructions directly. The study uses a new, larger dataset of planktonic foraminifera assemblages and state-of-the-art climate model simulations to evaluate this new macroecological approach. The results have implications for refining climate models and improving predictions of future climate change, specifically regarding regional temperature variations, which are critically important for ecosystem health and human societies.

Previous studies have relied on data-model comparisons to evaluate climate model performance in simulating LGM climate. However, discrepancies between paleo-reconstructions (e.g., SST reconstructions based on marine microfossil assemblages) and simulations have made it difficult to attribute these differences to model skill or limitations in the proxy data. The indirect nature of proxy data, such as uncertainties in the depth and seasonality of the temperature signal and confounding nuisance variables, often complicates these comparisons. Early global reconstructions of LGM temperatures used transfer functions relating microfossil assemblage composition to seawater temperature. While newer geochemical methods are available, the transfer-function approach remains relevant. However, this study proposes a direct, ecologically grounded method that bypasses the reliance on explicitly reconstructed temperatures. This approach builds on the well-established ecological principle of decreasing community similarity with increasing environmental distance, specifically temperature. This pattern is observed in many taxa and ecosystems. Previous research has shown that temperature is a dominant driver of community assembly in planktonic foraminifera, a feature consistent with their low dispersal limitation and consistent response to glacial-interglacial cycles. The stability of the thermal niches of planktonic foraminifera suggests the similarity-decay pattern observed in core-top data should apply to LGM assemblages. This approach provides a direct comparison between observations and simulations, avoiding space-for-time substitutions inherent in proxy calibrations. This allows for assessment of temperature gradients in all directions and implicitly accounts for seasonal and vertical habitat tracking by foraminifera.

This study uses a new, expanded dataset of 2,085 LGM planktonic foraminifera morphospecies assemblages from 647 sites, representing a 50% increase in coverage compared to previous work. This dataset was combined with near-surface seawater temperature patterns from equilibrium simulations of LGM climate using state-of-the-art climate models. The modern similarity-decay pattern was characterized using core-top planktonic foraminifera assemblage data from the ForCenS compilations, complemented by LGM data which was extended by over 50% using the MARGO compilation and literature. The expanded dataset ensured a significant increase in coverage, particularly for the LGM. The Bray-Curtis dissimilarity was used to quantify assemblage similarity, reflecting differences in the relative proportions of taxa, accounting for sensitivity to changes in assemblage composition. The analysis compares assemblage similarity against environmental (temperature) distance (similarity-decay plots) using annual-mean temperature data at 50 m water depth. The study uses LGM temperatures derived from simulated temperature anomalies to minimize model bias. To address the potential for poor analogues (fossil samples dissimilar to modern samples), a cutoff was set for poor analogues. The study further investigates the LGM temperature patterns using a new approach to resolve ambiguities related to the seasonal and vertical origin of the proxy signal. This approach estimates LGM temperatures from assemblages by assessing the proportion of variance in assemblages explained by temperature at different seasons and depths. This resulted in temperature estimates based on annual seawater temperatures at 50 m water depth. The methodology includes a robust assessment of uncertainties in the temperature reconstruction through Monte Carlo simulation, accounting for spatial autocorrelation. The study also uses sensitivity tests to rule out other factors that could explain the discrepancies between simulated and observed patterns. For example, tests were conducted to evaluate the effects of chronological uncertainty, model bias (using pre-industrial control runs), and the role of poor analogues in influencing the reconstructions. Additionally, the study includes simulations with a weaker AMOC, achieved by prescribing freshwater flux into the North Atlantic, to test if this mechanism could reconcile the discrepancies between models and data. These simulations were performed with one model using identical LGM boundary conditions but with varying freshwater fluxes and durations.

The study reveals significant differences in biogeography between modern and LGM planktonic foraminifera, most pronounced in the North Atlantic where cold-water assemblages expanded to mid-latitudes at the expense of transitional fauna. The high compositional similarity (98%) between modern and LGM assemblages supports the assumption that similarity-decay patterns in core-top assemblages are also characteristic of LGM assemblages. While simulated temperatures initially indicated a similarity decay similar to the core-top data, discrepancies emerged at high similarities (>0.75), where simulated temperature differences exceeded modern limits. This discrepancy, present across all simulations, arises primarily from the North Atlantic and Nordic Seas, where models maintain large present-day temperature gradients. This pattern is robust against chronological uncertainty. Reconstructed LGM temperatures based on assemblages are 2.45 °C lower than today, with simulations showing similar global cooling (-2.15 °C). However, the spatial pattern differs significantly. The replacement of transitional assemblages with cold-water assemblages in the North Atlantic correlates with substantial cooling (~7.3 °C lower than today). In contrast, the simulated cooling is spatially uniform, leading to discrepancies of up to 4.9 °C in the subpolar North Atlantic. This indicates a significant mismatch between simulated and reconstructed LGM temperature change in the North Atlantic. This mismatch resembles the fingerprint of AMOC changes, suggesting a weaker AMOC during the LGM. Additional simulations with a weaker AMOC, achieved through freshwater forcing, improved the data-model agreement by producing a colder subpolar North Atlantic and resulting in similarity-decay patterns closer to the core-top pattern. These simulations, representing different freshwater flux scenarios (magnitude, location, duration), demonstrate that a reduced AMOC is a physically plausible explanation for the observed discrepancies. The findings rule out alternative explanations such as species adaptation or changes in thermal niches due to the high similarity between modern and LGM assemblages. The analysis robustly indicates distinct regional patterns in LGM temperature change, with the largest cooling observed in the northern North Atlantic likely related to a reduced AMOC. The significant spatial heterogeneity of climate change during the LGM, underscored by the plankton biogeography, emphasizes the need for climate model evaluation beyond globally averaged statistics.

The findings address the research question by demonstrating that existing climate models inadequately represent the spatial patterns of temperature change during the LGM, particularly in the North Atlantic Ocean. The discrepancies between simulated and observed temperature gradients highlight a critical limitation in the models' ability to accurately reproduce regional climate dynamics. The significant cooling in the subpolar North Atlantic, inferred from the planktonic foraminifera data but not fully captured by the standard LGM simulations, points to the importance of correctly representing the Atlantic Meridional Overturning Circulation (AMOC). The study's novel approach, grounded in macroecological principles, provides a robust and independent method for evaluating climate simulations. This method allows for direct comparison of observations and simulations, reducing reliance on often uncertain temperature reconstructions. The results underscore the importance of considering the spatial heterogeneity of climate change, as regionally specific changes in temperature have significant impacts on ecosystems and human societies. The improved agreement between the model simulations with weaker AMOC and the observed foraminifera data indicates that the AMOC is likely a key process regulating North Atlantic temperatures during glacial periods. The study demonstrates that incorporating improved representations of AMOC dynamics in climate models is crucial for accurately simulating past and future climate change.

This study demonstrates the utility of macroecology in evaluating past climate simulations by providing a robust, independent method for assessing the accuracy of climate models. The findings highlight significant discrepancies between current climate models and the observed patterns of LGM temperature change, especially in the subpolar North Atlantic, emphasizing the importance of accurately simulating the AMOC. The spatial heterogeneity of LGM climate change underscores the need for future model development focusing on regionally resolved climate patterns, which are essential for accurately understanding the impacts on ecosystems and societies. Future research should focus on incorporating improved representations of AMOC dynamics and other relevant oceanographic processes into climate models, using this macroecological framework to guide and validate these improvements. This could include transient climate simulations to better understand the dynamics of AMOC change. Further research could also extend this method to other time periods and geographic locations to broaden our understanding of past climate variability.

While this study utilizes a significantly expanded dataset, the spatial distribution of samples, especially for the LGM, could still influence the results. The study focuses on equilibrium simulations of the LGM and might not fully capture the complexities of transient climate changes. The analysis uses a single proxy (planktonic foraminifera) and may not fully capture the nuances of past oceanography. Although the study addresses uncertainties related to the seasonal and vertical habitat preferences of the foraminifera, limitations remain in completely resolving these aspects. Further, the choice of employing a particular metric (Bray-Curtis dissimilarity) for community similarity might have implications; the use of other metrics could lead to slightly different results. The study's findings could benefit from comparisons with other independent proxy data, such as those related to deep-ocean circulation. While the experiments to simulate reduced AMOC provided a better fit, the specific mechanism of freshwater forcing might need further scrutiny.