Northern Hemisphere winter storm track trends since 1959 derived from multiple reanalysis datasets

Abstract

In this study, a comprehensive comparison of Northern Hemisphere winter storm track trend since 1959 derived from multiple reanalysis datasets and rawinsonde observations has been conducted. In addition, trends in terms of variance and cyclone track statistics have been compared. Previous studies, based largely on the National Center for Environmental Prediction–National Center for Atmospheric Research Reanalysis (NNR), have suggested that both the Pacific and Atlantic storm tracks have significantly intensified between the 1950s and 1990s. Comparison with trends derived from rawinsonde observations suggest that the trends derived from NNR are significantly biased high, while those from the European Center for Medium Range Weather Forecasts 40-year Reanalysis and the Japanese 55-year Reanalysis are much less biased but still too high. Those from the two twentieth century reanalysis datasets are most consistent with observations but may exhibit slight biases of opposite signs. Between 1959 and 2010, Pacific storm track activity has likely increased by 10 % or more, while Atlantic storm track activity has likely increased by <10 %. Our analysis suggests that trends in Pacific and Atlantic basin wide storm track activity prior to the 1950s derived from the two twentieth century reanalysis datasets are unlikely to be reliable due to changes in density of surface observations. Nevertheless, these datasets may provide useful information on interannual variability, especially over the Atlantic.

Appendix: Trends derived from individual ensemble members of 20CR

In Sect. 6, we have examined the ensemble mean analysis from 20CR. Compo et al. (2011) suggested an alternative strategy of using the 56 individual ensemble members (instead of the ensemble mean analysis) from 20CR to derive storm track activity before taking the ensemble mean of the storm track activity to quantify its time evolution. This procedure has been carried out for pp and results are shown in Fig. 12d, e. As expected, results based on averaging pp computed from individual ensemble members are consistently larger than those based on the ensemble mean analysis. Over 1900–2010, no significant trend is found in both the Pacific and the Atlantic. Nevertheless, it is not clear how reliable these century long trends are.

A similar procedure was carried out by Wang et al. (2013) to derive trends of cyclone activity over the period 1871–2010. However, the fidelity of trends derived by this procedure has been questioned by Krueger et al. (2013, 2014) who suggested that storminess trends derived from the 20CR using this procedure is not consistent with those derived from pressure observations over the North Atlantic. Wang et al. (2014) argued that the storminess derived from 20CR shows high temporal correlation with those derived directly from observations throughout the entire period, and the correlation increases after erroneous observations have been removed.

By comparing storm surge data from tide gauges with predictions based on reanalysis wind and pressure data, Dangendorf et al. (2014) suggested that long term trends in storminess derived from 20CR may be useful over the North Sea (around 54°N, 8°E) all the way back to about 1910. Prior to that, predictions based on either the ensemble mean analysis or individual ensemble members exhibit significant biases. Examining the time series of pp from 20CR and ERA-20C from 1901–2010 over this region (not shown), the 31-year running correlation between the two time series remains above 0.97 all the way back to the beginning of the period, and the value of the spread in 20CR (Fig. 12f), even when normalized by the value of pp, remains small even back to 1910 compared to the spread over either the Atlantic or Pacific storm track prior to the 1950s, but the spread over that region is increasing steadily before 1910. In addition, the value of the ensemble averaged pp computed from individual ensemble members over this region is only slightly (<2 %) larger than those computed based on the ensemble mean analysis all the way back to 1900. These results suggest that the North Sea region is an area with high enough observation density that the reanalyses are strongly constrained by the available observations all the way back to the early 1900s. These results indicate that useful long term trends from 20CR may only be derivable from regions with consistently high observation density, and the utility of such trends probably needs to be assessed on a case by case basis. Given that the uncertainties in 20CR prior to 1950 over the Pacific and Atlantic basins are much larger than those over the North Sea region around the early 1900s, and the results of Dangendorf et al. (2014) suggesting that 20CR trends are not reliable over the North Sea region around that time, it is unlikely that the long term trends are reliable over the Pacific and Atlantic basins prior to 1950.