Marlene Klockmann

A model-data comparison is an effective method for evaluating both proxy-based climate reconstructions and simulated model outputs. The MIS3 period is marked by a wide range of proxy records that indicate millennial-scale climate variability globally. In this study, we use paleoclimate proxy data to assess the performance of four climate models: CCSM4, HadCM3, MPI-ESM, and COSMOS. We compare model results of Dansgaard–Oeschger (D-O)-like oscillations with recent North Atlantic sea surface temperature reconstructions from 10–50 ka. Additionally, we evaluate the models based on Antarctic warming during Greenland Stadials, the timing of Antarctic temperature changes relative to Greenland, the duration and amplitude of D-O events, and the position of the Intertropical Convergence Zone. All four models suggest that D-O event temperature anomalies are primarily driven by changes in the Atlantic Meridional Overturning Circulation (AMOC) and North Atlantic sea ice extent.

For climate forecasts it is crucial to initialise the ocean state from observations because they rely on the memory of the ocean. If, however, the initialised ocean state is far away from the model’s own preferred mean state, predictive skill will suffer due to model drift. We are testing whether an ocean grid with variable resolution - designed to represent sparse and well-observed regions with appropriate resolution - has advantages over an ordinary grid with uniform resolution. The locally high resolution could lead to an improved mean ocean state through a better representation of mesoscale processes. The observation-informed grid will allow for high-resolution data assimilation in well-observed areas, which will potentially lead to improved initial conditions and predictive skill. We developed such a grid for the ocean component of the coupled ICON model designed for seamless predictions (ICON-XPP). The grid resolution varies from 40 to 10km, depending on the observation density in the EN4 database from 1960 to 2023. The local refinement in well-observed areas leads to a better representation of ocean features such as fronts and western boundary currents. We assess the effect of these improvements on the mean climate state by comparing to a reference simulation with a uniform 20km ocean resolution.

Simulating and reproducing the past Atlantic meridional overturning circulation (AMOC) with comprehensive climate models are essential to test the ability of models to simulate different climates. At the Last Glacial Maximum (LGM), reconstructions show a shoaling of the AMOC compared to modern climate. However, almost all state-of-the-art climate models simulate a deeper LGM AMOC. Here, we conduct a multi-model analysis using outputs from all PMIP phases (PMIP2 to PMIP4) to consistently explore the causes of this paleodata-model mismatch. The analysis focuses on the role of sea-surface temperature biases in the piControl simulation as well as changes in ocean temperature, salinity and density in each oceanic basin. We further compare the deepwater formation regions in each model and explore potential implications on the interpretation of paleodata-model comparison.

Greenland ice core records feature Dansgaard–Oeschger (D-O) events, which are abrupt warming episodes followed by gradual cooling during ice age climate. The three climate models used in this study (CCSM4, MPI-ESM, and HadCM3) show spontaneous self-sustained D-O-like oscillations (albeit with differences in amplitude, duration and shape) in a remarkably similar, narrow window of carbon dioxide (CO2) concentration, roughly 185-230 ppm. This range matches atmospheric CO2 during Marine Isotopic Stage 3 (MIS 3: between 27.8 – 59.4 thousand of years BP, hereafter ka), a period when D-O events were most frequent. Insights from the three climate models point to NA sea-ice coverage as a key ingredient behind D-O type oscillations, which acts as a tipping point. No other climate property (NA salinity, Atlantic Meridional Overturning Circulation, Global mean Ocean temperature and Global mean temperature) is found to directly determine whether D-O type behaviour can occur in all three models.

Greenland ice core records feature Dansgaard–Oeschger (D‐O) events, which are abrupt warming episodes followed by gradual cooling during ice age climate. The three climate models used in this study (CCSM4, MPI‐ESM, and HadCM3) show spontaneous self‐sustained D‐O‐like oscillations (albeit with differences in amplitude, duration, and shape) in a remarkably similar, narrow window of carbon dioxide ( CO ₂ ) concentration, roughly 185–230 ppm. This range matches atmospheric CO ₂ during Marine Isotopic Stage 3 (MIS 3: between 27.8 and 59.4 thousand of years BP, hereafter ka), a period when D‐O events were most frequent. Insights from the three climate models point to North Atlantic (NA) sea‐ice coverage as a key ingredient behind D‐O type oscillations, which acts as a “tipping element.” Other climate state properties such as Mean Atlantic Meridional Overturning Circulation strength, global mean temperature and salinity gradient in the Atlantic Ocean do not determine whether D‐O type behavior can occur in all three models.

We present a new framework for the reconstruction of climate indices based on proxy data such as tree rings. The framework is based on the supervised learning method Gaussian Process Regression (GPR) and aims at preserving the amplitude of past climate variability. It can adequately handle noise-contaminated proxies and variable proxy availability over time. To this end, the GPR is formulated in a modified input space, termed here embedding space. We test the new framework for the reconstruction of the Atlantic multi-decadal variability (AMV) in a controlled environment with pseudo-proxies derived from coupled climate-model simulations. In this test environment, the GPR outperforms benchmark reconstructions based on multi-linear principal component regression. On AMV-relevant timescales, i.e. multi-decadal, the GPR is able to reconstruct the true amplitude of variability even if the proxies contain a realistic non-climatic noise signal and become sparser back in time. Thus, we conclude that the embedded GPR framework is a highly promising tool for climate-index reconstructions.

At unprecedented resolution we investigate the nature of Dansgaard-Oeschger events in the Fram Strait, the gateway between the Nordic Seas and the Arctic Ocean. The new reconstructions of biomarkers and sea ice variability, stable isotopes and IRD will be seen in context of sea ice conditions, ocean hydrography and climate of the Nordic Seas as seen in multi-model output from three transient glacial GCM simulations (NorESM, CESM, MPI-ESM) and high-resolution reconstructions from an eastern Nordic Seas transect (from the Faeroe-Shetland Channel, via the Norwegian Sea to the Fram Strait). The combined results show that ocean-atmosphere-sea ice processes and dynamics during the transition from H4 to GI8 are strongly coupled. Both model results and reconstructions suggest subsurface ocean warming and polynya events in the southern- and northernmost Nordic Seas during the cold stadial. For a short time during the stadial to interstadial transition, a corridor of open water and hence sea ice-free conditions existed from the southern Nordic Seas all the way to the Fram Strait. The breakup of the sea ice cover is likely caused by the overshoot of AMOC during the transition and the associated enhanced ocean heat transport into the Nordic Seas. After the transition, winter sea ice grows back in the Fram Strait during the interstadial state, but the Southern Nordic Seas remain ice-free.

The last deglaciation was accompanied by a gradual warming with superimposed abrupt climate changes. In transient simulations of the last deglaciation with the comprehensive Max Planck Institute Earth System Model (MPI-ESM) we show that the timing and occurrence of abrupt climate changes are highly dependent on the utilized ice-sheet boundary condition. Simulations with different ice-sheet reconstructions show that the variability of North Atlantic surface temperatures are dominated by the timing and amplitude of meltwater fluxes from ice sheets, as derived from reconstructions. While some abrupt climate events (e.g. the Younger Dryas) only occur under certain boundary conditions in the transient simulations, other climate events such as the Bølling Allerød warming (about 14.7-14.2 ka BP) cannot be simulated with any of the applied and widely used reconstructions. However, in a sensitivity experiment with changing ice sheets but no addition of meltwater into the ocean, the North Atlantic experiences a warming during the time of the Bølling-Allerød. This warming is associated with a reorganization of the ocean circulation and deep-water formation sites. Prior to this reorganization, during the glacial and early part of the deglaciation, a rather zonal jet stream maintains a strong subpolar gyre in the North Atlantic. In addition, salty and dense water masses form in the Arctic. Until about 16.5 ka BP the Arctic freshens significantly and the surface elevation over the Laurentide ice sheet reduces. The latter leads to a shift in the atmospheric circulation at around 14.2 ka BP. The resulting changes in wind stress strongly reduce the eastward extent of the North Atlantic subpolar gyre. Here, we examine the physical mechanisms behind the reorganization and explore additional simulations with fixed deglacial key parameters (e.g. CO2, insolation, ice sheets) to identify the key drivers of the climate changes during the early deglaciation and Bølling Allerød.

Three different climate field reconstruction (CFR) methods are employed to reconstruct spatially resolved North Atlantic–European (NAE) and Northern Hemisphere (NH) summer temperatures over the past millennium from proxy records. These are tested in the framework of pseudoproxy experiments derived from two climate simulations with comprehensive Earth system models. Two of these methods are traditional multivariate linear methods (principal component regression, PCR, and canonical correlation analysis, CCA), whereas the third method (bidirectional long short-term memory neural network, Bi-LSTM) belongs to the category of machine-learning methods. In contrast to PCR and CCA, Bi-LSTM does not need to assume a linear and temporally stable relationship between the underlying proxy network and the target climate field. In addition, Bi-LSTM naturally incorporates information about the serial correlation of the time series. Our working hypothesis is that the Bi-LSTM method will achieve a better reconstruction of the amplitude of past temperature variability. In all tests, the calibration period was set to the observational period, while the validation period was set to the pre-industrial centuries. All three methods tested herein achieve reasonable reconstruction performance on both spatial and temporal scales, with the exception of an overestimation of the interannual variance by PCR, which may be due to overfitting resulting from the rather short length of the calibration period and the large number of predictors. Generally, the reconstruction skill is higher in regions with denser proxy coverage, but it is also reasonably high in proxy-free areas due to climate teleconnections. All three CFR methodologies generally tend to more strongly underestimate the variability of spatially averaged temperature indices as more noise is introduced into the pseudoproxies. The Bi-LSTM method tested in our experiments using a limited calibration dataset shows relatively worse reconstruction skills compared to PCR and CCA, and therefore our working hypothesis that a more complex machine-learning method would provide better reconstructions for temperature fields was not confirmed. In this particular application with pseudoproxies, the implied link between proxies and climate fields is probably close to linear. However, a certain degree of reconstruction performance achieved by the nonlinear LSTM method shows that skill can be achieved even when using small samples with limited datasets, which indicates that Bi-LSTM can be a tool for exploring the suitability of nonlinear CFRs, especially in small data regimes.