Sensitivity of moving PCA in detecting spatial changes in teleconnections

Principal Component Analysis (PCA) is a well-known and widely used technique in climatology to identify modes of low-frequency variability of atmospheric circulation. PCA is also used to examine changes in mode patterns over time. We present temporal variability of modes using moving PCA of winter (DJF) monthly mean 500 hPa height anomalies for 20 to 50-year long moving periods with one year step. We compare modes identified for the ensemble mean and ensemble members of the 20CRv2, CERA20, ERA5 and NCEP/NCAR reanalyses. It shows that PCA is sensitive to the period analysed and results can vary considerably if the period is shifted by even one year. Changes in the pattern of a mode are usually interpreted as real changes in atmospheric circulation. However, these changes can be influenced by many aspects, e.g., data quality, length of the analysed period, or settings of PCA. The sudden changes from period to period are more common whit shorter moving periods, while longer periods give more stable results. Some of these sudden changes appear to be more or less random, while others occur in all ensemble members or even in all reanalyses used in the same period. This suggests that comparisons of modes from just a few periods, or even just two periods, may be affected by the presence of a sudden change. Comparing modes from periods that are shifted by only one year, for example, may then provide different results and hence a different interpretation of the shift in the centres of variability.