In our experiment, the temperatures in Scandinavia were highest during the first 500 years of the experiment, after which they decrease almost linearly towards the preindustrial climate (Fig.  5 a, b and Table  1 ).

The latter anomaly is strongly influenced by the strong winter cooling over the Nordic Seas (Greenland-Icelandic-Norwegian Seas, see Fig.  4 a), of which a part is included in the calculation of the average depicted in Fig.  5 a.

In Fig.  12 , positive anomalies are noted that are simultaneous with the temperature depressions found in the Scandinavian time series around 2.5 and 1.7 kyr BP (Fig.  5 ).

The simulated timing of the thermal maximum between 9 and 8 kyr BP (Fig.  5 b) is consistent with most sites, although we find no sign of a delayed thermal optimum as reported for Northern Finland, but this lag is probably caused by local effects (e.g., vegetation development, topography) not resolved by our model.

Besides, the model simulates large cold events at decadal to centennial timescales (Fig.  5 ).

The simulated July temperature anomalies (Fig.  5 b) show a reasonable correspondence with the latter excursions, suggesting that the progressive increase in the variability of the THC strength could have taken place in the real world during the Holocene..

This is clearly evident in the temperatures in Scandinavia and North America (Figs.  5 , 6 and Table  1 ), in the different components of the ocean circulation in the Nordic Seas (Fig.  14 and Table  2 ) and in the sea-ice evolution (Table  2 ).

The enhanced variability is most clearly expressed in the Nordic Seas region, where it culminates in two decadal-scale events that are characterized by cooling in Northern Europe (−15°C in January temperature, Fig.  5 ), negative excursions in overturning strength (−1.5 Sv, Fig.  14 ) and positive anomalies in the sea-ice extent (Fig.  12 ).