Self-organizing maps (SOMs) represent a method widely used to obtain representative states of atmospheric circulation and link these states to local climate on one hand, and large-scale modes of circulation variability on the other. In the present Part II of our study, we focus on the temporal aspects of the link between SOM circulation types (CTs) and main building blocks of spatial-temporal variability represented by (i) three predefined idealized oscillatory variability modes and (ii) four leading modes of Euro-Atlantic circulation variability extracted by principal component analysis (PCA). The CTs respond to various changes in modes, including variations in their strength and preferred phase. However, compared to these responses, the detected sampling variability inherent to decadal-scale datasets generated from identical setting of modes is surprisingly high, showing that trends in the frequency of CTs of approximately ± 30% can occur without any change in the strength and phase of the underlying modes. This suggests that in order to achieve robust changes in CT frequencies, either an unrealistically large change in the underlying variability mode, which is inconsistent with reanalysis data, is required, or simultaneous contributions of two or more modes that superimpose one another are needed. Consequently, to attribute changes in CTs to variability modes, other methods (e.g., PCA) should be used in parallel to SOMs to avoid misinterpretations. Since the results obtained for the idealized modes agree with our findings for real-world circulation, we believe that the rather simplistic idealized modes may be used in future studies that would extend the research to non-linear aspects of teleconnectivity, including (but not limited to) non-stationary spatial patterns and non-linear combination of variability modes in generating synthetic datasets.