Recent summertime North American weather regime trends in a very large seasonal model ensemble

Recurrent and persistent large-scale circulation patterns, known as weather regimes, are widely employed in operational medium-range and subseasonal prediction. However, they have been used less often in studies of long-term climate variability and change. Here, we use a recently defined year-round North American regime classification to identify trends in the summertime circulation from 1981 to 2024. We find large increases in the frequency, persistence and interannual variability of the Greenland High (GH) regime, which is similar to Greenland blocking and the negative summer North Atlantic Oscillation. Recent extremes include the summers of 2023, 2019 and 2016. A first-order Markov model shows that the increased GH frequency and interannual variability can arise from increased GH persistence.

The GH frequency trend resembles previously reported trends in summertime Greenland blocking, which are absent in uninitialised climate models but have been seldom analysed in initialised models. We therefore investigate whether the observed GH trends can be reproduced by SEAS5, ECMWF’s current operational seasonal prediction system. To do so, we construct a 10,000-member ensemble by randomly sampling a single member from the May initialisation each year from 1981 to 2024 and stitching them together to create 10,000 different time series.

Our results show that the very large SEAS5 ensemble fails to capture the observed trend in GH frequency because persistence trends are too weak. This occurs despite SEAS5 producing summers with more GH days and individual regimes more persistent than observed, so the issue is not simply an overall inability of the model to generate persistent regimes. Hence, the missing GH trends must arise from fundamental model deficiencies which develop on subseasonal timescales and are not rectified by initialisation. Our work adds to a growing body of literature showing the benefit of using seasonal model data to understand the development of climate model trend errors.