The effect of global warming on winter cold extremes remains uncertain. While rising global temperatures are expected to reduce the frequency and severity of cold events, the impact on temperature variability introduces complexity. Changes in variability, encompassing variance, skewness, and kurtosis, can influence the rate of change in temperature extremes independently of shifts in the mean temperature. Recent high-impact cold events have fueled debate about the extent of changes beyond simple mean shifts. Previous research has indicated a general decrease in daily winter temperature variance over North America. This decrease, linked to Arctic amplification (stronger warming in the Arctic than at lower latitudes), reduces the north-south temperature gradient. Consequently, cold Arctic air displaced southward during cold outbreaks warms faster than air from the south during warm periods. This reduction in variance suggests a decline in cold extreme frequency and intensity. However, several impactful cold outbreaks in recent decades challenge this view, leading to speculation that Arctic warming and sea ice loss could increase cold extreme frequency through changes in atmospheric circulation, specifically the stratospheric polar vortex and jet stream. The lack of consensus stems from discrepancies between model projections and observational studies, including questions about the ability of climate models to accurately capture midlatitude circulation responses. To resolve these discrepancies, this study investigates changes in winter cold extremes and temperature variability using both observations and climate model simulations, examining the full shape of winter temperature distributions and the contributions of mean shifts, variance changes, and higher-order moments.

Existing literature presents a mixed picture regarding the impact of global warming on winter cold extremes. Some studies report a decrease in winter temperature variability over North America, suggesting a reduction in cold extremes. This decrease is attributed to Arctic amplification and the resulting changes in the north-south temperature gradient. Models generally support this finding, showing a robust decrease in sub-seasonal winter temperature variance over northern mid-latitudes. This reduction in variance implies a faster decline in cold extremes than would be expected from mean temperature shifts alone. Observational data, particularly in recent decades, also supports the decrease in variance, especially over North America. However, contrasting evidence arises from high-impact cold air outbreaks in recent years. These events have spurred research into the role of Arctic warming and sea ice loss in influencing mid-latitude circulation patterns. Some studies suggest that changes in the stratospheric polar vortex and jet stream, driven by Arctic warming, could lead to more frequent and persistent cold extremes. The debate further centers on the ability of climate models to accurately represent these midlatitude circulation responses, raising concerns about the reliability of model-based projections of cold extremes.

This study uses daily near-surface temperature data from ERA5 reanalysis (1980-2022) for North America, supplemented by data from other reanalysis products and gridded observations to ensure robustness. Trends in winter (December-January-February) mean and extreme temperatures are analyzed using quantile regression, focusing on the 2nd and 98th percentiles to represent cold and warm extremes. To account for differing degrees of global warming across models, trends are normalized by the global annual mean temperature trend. Large ensemble historical simulations from seven climate models are compared with the reanalysis data. The multi-model mean is used to identify robust model features. The uncertainty due to internal variability is assessed using a bootstrapping approach for ERA5 and by analyzing the ensemble spread in the model simulations. The shape of the temperature distributions is analyzed by calculating trends for percentiles from 2nd to 98th and decomposing these trends into components linked to mean, variance, and higher moments. This decomposition helps to quantify the contribution of each component to the amplified warming of cold extremes. Pattern-based detection and attribution methods are employed to determine whether the observed changes in mean, variance, and skewness are detectable in observations and can be attributed to human influence. This involves regressing the spatial patterns of observed trends onto the corresponding model trend patterns (fingerprints) and analyzing the resulting scaling factors and uncertainties. Further attribution is made using single-forcing experiments from three of the models that separate the effects of greenhouse gases, anthropogenic aerosols, and natural forcings. Finally, the role of sea ice loss is explored using simulations from the PAMIP project, comparing the response to both SST warming and sea ice loss with the response to sea ice loss alone.

The analysis reveals that winter cold extremes over North America have warmed substantially faster than the winter mean temperature since 1980. Averaged over the United States and southern Canada, the coldest days (2nd percentile) warmed approximately 2.2 times faster than the winter mean and 3.2 times faster than the global annual mean trend. This amplified warming is observed in both ERA5 reanalysis and multi-model mean simulations. Examining the full shape of the temperature distributions reveals that, north of 45°N, colder days warmed faster than the mean and warmer days warmed slower, reflecting a decrease in variance. South of 45°N, only the coldest days showed amplified warming, suggesting additional changes in higher moments (e.g., skewness). The increase in skewness is confirmed by direct calculation of skewness trends. This increase is most noticeable south of 45°N, where changes in variance play a less significant role. The pattern-based detection and attribution analysis demonstrates that the observed changes in temperature variability are detectable and can be attributed to human influence. Notably, changes in variability have a higher signal-to-noise ratio than changes in the winter mean temperature, which is likely due to the robust nature of Arctic amplification in observed trends. The analysis using single-forcing experiments shows that greenhouse gas forcing is the primary driver of the observed trends. Experiments focusing on the role of sea ice loss indicate that Arctic warming from sea ice loss significantly contributes to the amplified warming of cold extremes, particularly in the reduction of variance and changes in higher-order moments south of 45°N. The results suggest that changes in temperature gradients, rather than reduced snow cover, are the dominant mechanism behind the higher-order moment changes.

The findings challenge the notion that Arctic warming will necessarily lead to more frequent or intense cold extremes over North America. The observed and modeled trends consistently show a significant amplified warming of cold extremes, primarily driven by human-induced changes in temperature variability. The discrepancy with some previous studies is likely due to the focus on shorter-term variability rather than long-term trends in extreme temperatures. This study highlights the importance of analyzing changes in the full shape of temperature distributions, including higher-order moments, to understand the impact of global warming on temperature extremes. The higher signal-to-noise ratio for variability changes compared to mean temperature changes suggests that the impact of Arctic amplification on variability is particularly robust and easily detectable. While winter cold extremes will continue to occur, their frequency and intensity are likely to decrease due to the combined effects of rising mean temperatures and changes in temperature variability. The consistency between observations and model simulations strengthens confidence in the projected decrease in the frequency and intensity of future cold extremes.

This study demonstrates that the amplified warming of North American winter cold extremes since 1980 is not solely attributable to shifts in the mean temperature or changes in variance. Changes in higher moments of the temperature distributions, particularly skewness, play a significant role, especially in more southern regions. The observed trends are robustly captured by climate models and are attributable to human influence, primarily through greenhouse gas emissions. The results highlight the importance of considering the full shape of temperature distributions when assessing the impact of climate change on extreme events. Future research could investigate the regional differences in the mechanisms driving these higher-order moment changes and further explore the implications for regional climate risk assessment.

The study primarily focuses on trends since 1980, limiting the analysis of long-term historical changes. The detection and attribution analysis relies on the assumption that climate models accurately represent the observed internal variability. The use of multiple reanalysis datasets and climate models helps mitigate potential biases but does not eliminate the possibility that some uncertainties remain. Future work incorporating additional observational data and a wider range of climate models could improve the robustness of the findings.