Assessing the Increasing Trend in Northern Hemisphere Winter Storm Track Activity Using Surface Ship Observations and a Statistical Storm Track Model

Abstract

Abstract Recent studies, based largely on analyses of reanalysis datasets, suggest that the Northern Hemisphere winter storm track activity has increased significantly during the second half of the twentieth century. In this study, this increasing trend, in terms of filtered mean sea level pressure (MSLP) variance statistics, is assessed using surface ship observations and a statistical storm track model. MSLP observations made by ships, archived as part of the reanalysis project conducted by the National Centers for Environmental Prediction–National Center for Atmospheric Research, have been analyzed. Observational errors are estimated by comparing reports of nearly collocated observations. Consistent with previous studies, the observational errors of ship pressure observations are found to be very large during the late 1960s and early 1970s. Without correcting for observational errors, the storm track activity over the Pacific, computed based on ship observations, is found to be decreasing with time, while the upward trend in the Atlantic is much smaller than that found in the reanalysis data. Even after corrections have been made to account for secular changes in observational error statistics, the ship-based trend in the Pacific is still found to be much smaller than that found in the reanalysis, while over the Atlantic, the corrected ship-based trend is consistent with that found in the reanalysis. The robustness of the results is tested by application of data trimming based on the reanalysis products. Ship observations that are different from the reanalysis by more than a prescribed limit are removed before the statistics are computed. As the prescribed limit is reduced from 30 to 2.5 hPa, the ship-based storm track activity becomes increasingly consistent with that based on the reanalysis. However, even when the smallest limit is used, the trends computed from the ship observations are still smaller than those computed from the reanalysis, strongly suggesting that the trends in the reanalysis are biased high. Nevertheless, the results suggest that decadal-scale variability of the Atlantic storm track activity is not very sensitive to the trimming limit, while results for the Pacific storm track are not as robust. As an independent corroboration of the ship observation results, a statistical model is used to test whether the storm track trend found in the reanalysis is dynamically consistent with observed mean flow change. Five hundred winters of GCM simulations are used to construct a linear model based on canonical correlation analysis (CCA), using monthly mean distribution of MSLP anomalies as a predictor to hindcast monthly mean MSLP variance. The Atlantic storm track in the CCA model hindcast is well correlated with the storm track in the reanalysis on both interannual and decadal time scales, with the hindcast trend being 82% of that found in the reanalysis. Over the Pacific, the CCA hindcast does not perform as well, and the hindcast trend is only 32% of that found in the reanalysis. The results of this study suggest that the actual trend in Pacific storm track activity is probably only about 20%–60% of that found in the reanalysis, while over the Atlantic, the actual trend is likely to be about 70%–80% of that found in the reanalysis. Two new basinwide storm track indices, which should contain less bias in the secular trends, have been defined based mainly on ship observations.