Centennial-scale variability of the Atlantic Meridional Overturning Circulation (AMOC) in the absence of external forcing has been identified in several climate models, but proposed mechanisms differ considerably. Therefore, better understanding of processes governing AMOC variability at these timescales is needed. Here, we analyze numerical simulations with PlaSim-LSG, an Earth System Model Intermediate Complexity (EMIC), which exhibits strong multicentennial oscillations of AMOC strength under constant pre-industrial boundary conditions. We identify a novel mechanism in which these oscillations are driven by salinity anomalies from the Arctic Ocean, which can be attributed to changes in high-latitude precipitation. We further corroborate our findings by conducting a set of millennial-length sensitivity experiments, and we interpret the mechanism by formulating a three-box model which qualitatively reproduces regular oscillations of the AMOC. While PlaSim-LSG lacks complexity compared to state-of-the-art models, our results reveal that precipitation minus evaporation (P–E) change in the Arctic is a physically plausible driver of centennial-scale AMOC variability. We discuss how this mechanism might be most relevant in climate states warmer than the present-day, raising questions about the state-dependence of multicentennial AMOC variability.

Centennial-scale variability of the Atlantic Meridional Overturning Circulation (AMOC) in the absence of external forcing has been identified in several climate models, but proposed mechanisms differ considerably. Therefore, better understanding of processes governing AMOC variability at these timescales is needed. Here, we analyze numerical simulations with PlaSim–LSG, an Earth System Model Intermediate Complexity (EMIC), which exhibit strong multicentennial oscillations of AMOC strength under constant pre-industrial boundary conditions. We identify a novel mechanism in which these oscillations are driven by salinity anomalies from the Arctic Ocean, which can be attributed to changes in high-latitude precipitation. We further corroborate our findings by conducting a set of millennial-length sensitivity experiments, and we interpret the mechanism by formulating a three-box model which qualitatively reproduces regular oscillations of the AMOC. While PlaSim–LSG lacks complexity compared to state-of-the-art models, our results reveal that precipitation minus evaporation (P-E) change in the Arctic is a physically plausible driver of centennial-scale AMOC variability. We discuss how this mechanism might be most relevant in climate states warmer than the present-day, raising questions about the state-dependence of multicentennial AMOC variability.

<p>It has been hypothesized that climate variability on centennial timescales – in the North Atlantic region and beyond – is linked to unforced variability of the Atlantic Meridional Overturning Circulation (AMOC). Because of the presence of external forcings, uncertainties in proxy reconstructions of the AMOC and the short observational record, coupled climate models represent a key tool in assessing low-frequency AMOC variability. However, sufficiently long pre-industrial control (piControl) simulations with state-of-the-art climate models have only become widely available during the past decade. While significant centennial-scale AMOC variability has been identified in several single-model studies, proposed physical mechanisms differ considerably.</p> <p>Here, we assess mechanisms of AMOC variability on centennial timescales in the CMIP6 multi-model piControl ensemble. We find that a relatively large number of models – 11 out of the 15 analyzed – exhibit a statistically significant mode of centennial-scale MOC variability in the Atlantic. We review previously proposed mechanisms for centennial-scale AMOC variability and test whether their key elements are present in the CMIP6 ensemble.</p> <p>We find that salinity exchanges between the Arctic and North Atlantic basins, which have previously been proposed as drivers of multi-centennial AMOC variability in two CMIP6 models (IPSL-CM6A-LR and EC-Earth3), can also be identified in other CMIP6 models using the same ocean component (NEMO). However, we find only a weak or no signature of this mechanism in models that do not include NEMO. Even among NEMO models, the amplitude and timescale of centennial-scale AMOC variability is model-dependent, and we assess the relative role of deep-water formation sites in shaping these differences. Because AMOC fluctuations are linked to surface temperature anomalies and related impacts over land, our results motivate the need for more paleoclimate evidence at sub-centennial resolution, which would help constrain the CMIP6 inter-model spread in centennial-scale AMOC variability.</p>

<p>The global hydrological cycle is of crucial importance for life on Earth. Hence, it is a focus of both future climate projections and paleoclimate modeling. The latter typically requires long integrations or large ensembles of simulations, and therefore models of reduced complexity are needed to reduce the computational cost. Here, we study the hydrological cycle of the the Planet Simulator (PlaSim) [1], a general circulation model (GCM) of intermediate complexity, which includes evaporation, precipitation, soil hydrology, and river advection.</p><p>Using published parameter configurations for T21 resolution [2, 3], PlaSim strongly underestimates precipitation in the mid-latitudes as well as global atmospheric water compared to ERA5 reanalysis data [4]. However, the tuning of PlaSim has been limited to optimizing atmospheric temperatures and net radiative fluxes so far [3].</p><p>Here, we present a different approach by tuning the model’s atmospheric energy balance and water budget simultaneously. We argue for the use of the globally averaged mean absolute error (MAE) for 2 m temperature, net radiation, and evaporation in the objective function. To select relevant model parameters, especially with respect to radiation and the hydrological cycle, we perform a sensitivity analysis and evaluate the feature importance using a Random Forest regressor. An optimal set of parameters is obtained via Bayesian optimization.</p><p>Using the optimized set of parameters, the mean absolute error of temperature and cloud cover is reduced on most model levels, and mid-latitude precipitation patterns are improved. In addition to annual zonal-mean patterns, we examine the agreement with the seasonal cycle and discuss regions in which the bias remains considerable, such as the monsoon region over the Pacific.</p><p>We discuss the robustness of this tuning with regards to resolution (T21, T31, and T42), and compare the atmosphere-only results to simulations with a mixed-layer ocean. Finally, we provide an outlook on the applicability of our parametrization to climate states other than present-day conditions.</p><p>[1] K. Fraedrich et al., <em>Meteorol. Z.</em> <strong>1</strong><strong>4</strong>, 299–304 (2005)<br>[2] F. Lunkeit et al., <em>Planet Simulator User’s Guide Version 16.0</em> (University of Hamburg, 2016)<br>[3] G. Lyu et al., <em>J. Adv. Model. Earth Sy</em><em>st</em><em>.</em> <strong>10</strong>, 207–222 (2018)<br>[4] H. Hersbach et al., <em>Q. J. R. Meteorol. Soc.</em><em> </em><strong>146</strong>, 1999–2049 (2020)</p>

<p>Centennial-scale climate variability in the North Atlantic is characterized by the absence of a clear external forcing. Hence, identifying mechanisms of internal variability at these timescales is crucial to understand low-frequency climate variations. For this task, long control simulations with coupled climate models represent a key tool.</p><p>Although significant spectral peaks in centennial variability in the Atlantic Meridional Overturning Circulation (AMOC) were found among some state-of-the-art models, CMIP6 models disagree on the amplitude, periodicity and even existence of centennial AMOC variability. This disagreement motivates the use of models of reduced complexity with idealized setups and perturbed physics ensembles to elucidate the mechanisms of AMOC variability at long timescales.</p><p>Here, we investigate multi-millennial piControl simulations of PlaSim-LSG, an earth system model intermediate complexity (EMIC). For a range of vertical oceanic diffusion parameters, PlaSim-LSG exhibits strong oscillations of AMOC strength, as well as of salinity and surface temperatures in the North Atlantic, with a period of about 270 years.</p><p>Lag correlation analysis shows that a positive feedback involving the interplay of surface salinity, freshwater flux and sea ice concentration in the Norwegian Sea and the Arctic Ocean is the key driver behind these oscillations. In contrast to previous studies with other models, interhemispheric coupling only plays a minor role. We discuss preliminary results of sensitivity experiments for testing the proposed mechanism, and compare our results with previously proposed mechanisms of AMOC oscillations in CMIP6 models.</p>