Investigating the occurrence of blizzard events over the contiguous United States using observations and climate projections

Over previous decades, the United States has been plagued by severe winter storms or blizzards, which caused millions of dollars in damages. However, the historical trend of blizzard events and the possible impacts of future global climate change on blizzard occurrences remain unclear. In this study, we analyzed historical blizzard occurrences using the observed storm event database, which shows that the Northern Plains, such as North Dakota, South Dakota, and Minnesota, had the most blizzard activities over the past 25 years. No significant trend in blizzard occurrence is found in those regions. When considering blizzards as compound events of strong wind and extreme snowfall, we estimated blizzard occurrences based on wind speed and snowfall in climate datasets, including Automated Surface Observing Systems wind speed, national gridded snowfall analysis, ERA5 reanalysis and historical simulations of 19 models in Coupled Model Intercomparison Project phase 6 (CMIP6), which show a good agreement with the observations with respect to the climatology of blizzard occurrence. In the near-term and long-term future under two emission scenarios, CMIP6 projections suggest decreases in both strong wind and extreme snowfall events, eventually leading to a significantly reduced frequency of compound events. Significant decreases in blizzard occurrence are found in the Northern Plains and Upper Midwest, suggesting potentially reducing the risk of winter hazards over those regions in a warming climate.


Introduction
Winter storms cause significant disruption and damage to different socioeconomic sectors, such as agriculture (including livestock) and forestry, transportation and utility infrastructure, commercial operations, building safety, and human health (Changnon 2007, Coleman and Schwartz 2017, Janoski et al 2018).For instance, severe winter storms in Illinois produce more total damage than any other form of short-term severe weather including hail, lightning, and tornadoes (Changnon 1969).Blizzards, a severe form of winter storms, are usually a combination of heavy snowfall and strong winds that cause low visibility and extreme windchill (Wild 1997).According to the National Weather Service (NWS) Storm Data instruction, a blizzard is characterized as a winter storm with sustained winds at 35 mph or greater and visibility lower than 1 /4 mile for three consecutive hours or longer (NWS 2021).Due to the combined impacts of heavy snow, strong wind, and the consequent low visibility, blizzards often lead to significant structural damages and economic losses.As a weather-related disaster, blizzards in the past few decades have raised concerns regarding the impacts of climate change on the spatio-temporal variability of strong-wind and heavy-snowfall events.
Winter storms in North America are usually associated with the development of extratropical cyclones.As one of the most dangerous types of winter storms, blizzards are typically produced by a lowpressure system, which commonly develops over two locations east of the Rockies: eastern Colorado and the plains of Alberta (Atkinson 2010).For typical weather conditions that produce blizzards on the eastern side of the Rockies, a deep area of low pressure is centered over the Northern Plains with a cold front extending southward, while a high-pressure system is found northwest of the low with an extremely cold arctic air mass.Strong pressure gradients between the two pressure systems produce extremely strong winds.Meanwhile, moisture wraps around the low, and produces snowfall on the low's northwest side.The combined strong surface winds and snowfalls greatly reduce visibility, and create blizzard events (Ahrens and Samson 2010).
As the global mean surface temperature keeps increasing, total precipitation and heavy precipitation intensity are expected to increase, but snowfall will account for a smaller proportion of precipitation compared to rain (O'Gorman 2014).As a result, snowfall is projected to decrease across much of the Northern Hemisphere during the twentyfirst century (Krasting et al 2013, Danco et al 2016).Meanwhile, observations have shown a general decrease in global average surface wind speed since the 1980s, which is known as 'global terrestrial stilling' , and a reversal around 2010 (Zeng et al 2019).Climate projects show decreased annual mean wind speed across the Northern Hemisphere mid-latitudes (Karnauskas et al 2018).Given the complexity of the variability of snowfall and wind, it is important to investigate the historical and future trends of blizzard events.Schwartz and Schmidlin (2002) 'first time' presented the climatology of blizzards for the conterminous US, and identified the occurrence and location of blizzards using NWS storm data from 1959/60 to 1999/2000.Their analysis was updated 14 years later using longer records of storm data, which covers 55 winter seasons from 1959/60 to 2013/14 (Coleman and Schwartz 2017).They found that a relatively high occurrence rate of blizzard events is over the central plains, upper Midwest, and eastern Rockies.There are a few studies with a focus on future strong-wind and heavy-snowfall events only in regions like the eastern US and northern Europe (Lehtonen et al 2016, Janoski et al 2018).High-resolution (∼50 km) global model simulations show a significant decrease in the frequency of the high-snowfall and extreme-wind events over major areas of the eastern US (Janoski et al 2018).However, as a 'hotspot' of blizzards (Schwartz andSchmidlin 2002, Coleman andSchwartz 2017), regions like the central plains, upper Midwest, and eastern Rockies have not been studied, and how the compound events will change in those regions under a warming climate is still unclear.
To fill the knowledge gap discussed above, we will use climate models from state-of-the-art climate models, reanalysis data, and the NOAA storm events (SEs) database to investigate the spatio-temporal variability of blizzard events in the conterminous US.We will focus on two research questions: (1) Can we quantify blizzards as a compound event of strong wind and heavy snow?(2) If yes, how will the frequency of blizzards change in the future?

Observations of historical blizzard events
We used the SEs dataset provided by NWS NOAA to quantify the historical blizzard events over the contiguous US.The SE database documents the occurrence of storms and other extreme weather events (such as hail, tornado, flash flood, etc) that would have caused loss of life, injuries, significant property damages, and other necessary disruptions.Among all the types of events, blizzard records are available from 1996 to the present.Compared to previous studies (Schwartz and Schmidlin 2002, Coleman and Schwartz 2017), the current SE dataset provides the records for a shorter period, but it can effectively avoid inconsistency issues related to blizzard reporting parameters, detection capabilities, and spatial precision, which have been discussed in Coleman and Schwartz (2017).In the earlier storm reports, there was a lack of geographic preciseness of blizzard occurrence.For instance, the reported locations of blizzards were often relegated to general regions such as 'northeastern Illinois' or 'western Kansas.'Since January 1996, the digital database has been consistently recording the events by specific counties and forecast zones according to Coleman and Schwartz (2017).Each blizzard record provides the year, date, time, and the NWS forecast zone where the blizzard is observed.Based on the records from 1996 to 2020, we create the climatology of blizzard occurrence over the contiguous US.We consider SE-based blizzards as observations, which will be used to evaluate the estimated blizzard occurrence that is derived from daily wind speed and snowfall in reanalysis and climate models.

Identification of blizzard events using climate data
To evaluate the approach of using compound extremes to represent blizzard events, we use observed snowfall and wind speed from weather/climate data networks.It should be noted that snowfall and wind speed observations usually do not coexist in the same network.For instance, although many blizzard events are reported through Automated Surface Observing System (ASOS) sites, most of the ASOS sites only provide wind speed observations but no snowfall.Therefore, we use two sets of observational datasets: daily wind speeds from 1577 ASOS sites over the CONUS (figure S1), and daily snowfall over the ASOS sites extracted from National Gridded Snowfall Analysis from National Operational Hydrologic Remote Sensing Center (NOHRSC).NOHRSC provides 24 hour accumulated snowfall at a spatial resolution of 1 km from 2003 to the present, which is generated by assimilating observations from Cooperative observers, Community Collaborative Rain, Hail and Snow Network observers, and NWS spotter reports into a 24 hour background analysis based on Stage IV qualitative precipitation estimates (Carroll et al 2006).Due to the data coverage of NOHRSC, only the records from 2004 to 2022 are used.
The historical and future blizzard occurrences are also identified as compound events using daily wind speed and snowfall from the ERA5 reanalysis and Coupled Model Intercomparison Project (CMIP) phase 6 (CMIP6).ERA5 is the fifth generation ECMWF atmospheric reanalysis of the global climate covering the period from January 1950 to the present (Hersbach et al 2018).It provides hourly estimates of many atmospheric, land, and oceanic climate variables at a high spatial resolution (∼31 km).Compared to its predecessor, ERA-Interim, ERA5 has made improvements in the representation of model processes and the observation systems, including the assimilation of the NCEP stage IV quantitative precipitation estimates over the US (Hersbach et al 2020).Previous studies have evaluated the precipitation-phase partitioning, snowfall, and wind speed of ERA5, which exhibit a good performance compared to observations (Molina et al 2021) or satellite retrievals (Xiong et al 2022).We calculate daily average wind speed and total snowfall based on its hourly products from 1979 to 2020.
CMIP6 provides simulations of the past, current, and future climate from multiple climate models.Daily wind speed and snowfall data are obtained from the historical and future simulations of 19 climate models (table 1).Future climate projections are generated based on two Shared Socioeconomic Pathways (SSPs): SSP245 and SSP585, which represent the medium and high ends of the range of plausible future forcing pathways, respectively.Although some of the models provide more than one ensemble for some of the simulations (historical, SSP245, or SSP585), to keep the consistency among all the models, only the first available ensemble (e.g.r1i1p1f1) is used for each simulation from each model.For the projected changes in blizzard occurrence, we calculated the blizzard occurrence in the near-term future (2030-2059) and long-term future (2070-2099).The projected changes are based on the difference between the future periods and the historical period 1980-2009.
All the results from CMIP6 are shown as the median among the climate models, which are regridded into the same spatial resolution (0.5 • latitude × 0.5 • longitude) as ERA5.
Although the definition of blizzards by NWS contains the criteria for both wind speed and visibility of blizzards, climate datasets (e.g.observations, reanalysis, or model simulations) do not provide quantification of visibility.Therefore, the identification of blizzards from ERA5 and CMIP6 models is based on daily wind speed and snowfall.We consider blizzards as a compound event of strong wind and heavy snow.The identification of blizzard events was based on two criteria: (1) daily snowfall is greater than 10 mm of water equivalent, and (2) daily average wind speed is greater than 7 m s −1 .Days that satisfy both criteria would be considered blizzard events.We then compare the estimated blizzard occurrence from ASOS/NOHRSC, ERA5, and CMIP6 with the historical records from the SE dataset.
It would be noted that the thresholds (10 mm and 7 m s −1 ) used for daily snowfall and wind speed are arbitrary values.Previous studies used different definitions of heavy-snow or strong-wind events.For instance, Janoski et al (2018) use the criteria that a 2 day period should have snowfall greater than or equal to 25 mm water equivalent for 'heavy snow' , and 6 hourly average wind speed greater than the 98th percentile of 2 day maximum 6 hourly wind speed for 'strong wind' .Lehtonen et al (2016) use the criteria of daily average wind speed between 2.1 m s −1 and 5.6 m s −1 , total precipitation greater than 6.4 mm, and average temperature between −3.4 • C and 1.1 • C to characterize risky days for heavy snow loading.Other studies that investigate daily snow events and wind events also adopt different thresholds, such as greater than 5 mm d −1 for intense snowfall (Lin and Chen 2022), or greater than the 75th percentile for strong wind (Shen et al 2021).Different thresholds have been tested in our analysis (not shown).The number of detected blizzards can be sensitive to the thresholds, but the spatial pattern of blizzard climatology stays consistent.

Comparison between SE and the derived blizzard occurrence
The SE observations provide detailed information on each recorded blizzard, including the time when the storm starts and ends.Although most events do not last over 24 h according to the records, a few blizzards last for 2 or 3 days.Therefore, the recorded events are site-and event-specific.In other words, if a 3 day blizzard occurs over five different NWS forecast zones, there is one blizzard record for each forecast zone.For this reason, the quantification of compound events of strong wind and heavy snow is also event-specific.If a grid satisfies the criteria of compound events for two or more consecutive days, only one event is assigned to that grid.To make the SE observations directly comparable to the derived blizzard occurrence, the climatology of blizzard occurrence from the SE records at the forecast-zone scale needs to be interpolated into the same grid scale (e.g.0.5 • latitude × 0.5 • longitude) as the climate data.First, because the size of forecast zones varies, we normalize the value of observed blizzard occurrence at each forecast zone.The normalization follows the approach used by Coleman and Schwartz (2017), in which the number of blizzards at each zone center is the average of observed blizzards occurring within a 28.196 km radius of the zone center.For instance, if there are multiple zones within the searching radius, the value of blizzard occurrence at the target zone will be the average among all the zones.Second, based on the normalized values at all zone centers, we use the Kriging interpolation approach to create a gridded map of the climatology of blizzard occurrence during the period 1996-2020.

Results
According to the SE database, there are a total of 12 946 blizzard events documented over the contiguous US between 1996 and 2020.If considering different climate regions, blizzard activities are the highest in the Northern Rockies and Plains (Montana, Wyoming, North Dakota, South Dakota, Nebraska) and the Upper Midwest (Minnesota, Iowa, Wisconsin, Michigan) (figure 1(a)).These climate regions accounted for 10 232 blizzards in the past 25 years, with North Dakota being the highest contributor accumulating 2187 of those events.Seasonally, the Northern Rockies and Plains subregion has the most events (1667 total blizzards) in December, while the Upper Midwest peaks in January (1248 total blizzards) (figure 1(c)).Meanwhile, we examine the decadal changes in the blizzard occurrence (figure 1(b)).Overall, there is an evident increase in blizzard occurrence in the central US, including North Dakota, South Dakota, Minnesota, Nebraska, Kansas, Iowa, Missouri, and Texas.Some regions in the Northeast also show a slight increase, such as New York and Maine.Meanwhile, there is a reduction in blizzards in the west and mountainous regions, including Oregon, California, Idaho, Montana, and Colorado.If focusing on the states with an evident increase in blizzard occurrence between those two decades (figure 1(d)), we see relatively 'mild' years at the beginning of the century (e.g. from 2001 to 2006) even in those states with the most blizzard activities (figure 1(a)).Given the fact that blizzards were also very active in the late 1990s, the decadal variations only suggest the change between the 2000s and 2010, but do not necessarily indicate the trends of blizzard occurrence at a climate scale.Figure 2 shows the climatology of blizzard occurrence based on the three datasets.The SE dataset shows an evidently high frequency of blizzard events in the northern Great Plains, particularly in eastern North Dakota, north-eastern South Dakota, and north-western Minnesota, where there are at least two events per year in 1996-2020.This spatial pattern shows a good agreement with the blizzard climatology of the contiguous US in 1959-2014 (Coleman and Schwartz 2017), even though the datasets cover different periods.The 'hotspot' of blizzard events is also called the 'blizzard zones' (Schwartz andSchmidlin 2002, Coleman andSchwartz 2017).States outside the 'blizzard zones' , such as Iowa, Montana, and Wyoming, also exhibit frequent blizzard events (between 0.5-2 events per year).According to ASOS and NOHRSC observations (figure 2(b)), there is a high frequency of compound extremes in the blizzard zones, but also some sites in the Midwest and Northeast.It should be noted that wind speed is highly variable with significant spatial and temporal heterogeneity.Therefore, wind speed would be considerably smoothed out at a grid scale (e.g. 30 km or 50 km in our gridded datasets).In other words, if using a threshold of 7 m s −1 to identify extremes at a grid scale of 50 km, a higher threshold would be needed when applied to individual stations.Figure S2 shows the frequency of compound extremes derived from ASOS/NOHRSC using different thresholds.With a higher threshold of wind speed, the Northern Plains region appears to be the 'hotspot' of compound extreme events, exhibiting a good agreement with the SE dataset.For instance, northeastern Minnesota is not considered as a blizzard zone with frequent occurrence compared to southwestern Minnesota according to the SE dataset.This spatial variation is captured by ASOS/NOHRSC with a higher wind speed threshold.In general, the comparison between the SE dataset and ASOS/NOHRSC results demonstrates that the compound extremes of strong wind and heavy snowfall can represent blizzard events.
The identified compound events of strong wind and heavy snowfall in ERA5 and CMIP6 (figures 2(c) and (d)) generally capture the observed spatial distribution of blizzard occurrence, suggesting that the criteria of wind speed and snowfall can properly portray the blizzard activities.However, both ERA5 and CMIP6 underestimate the frequency in the 'blizzard zones' , and overestimate the frequency over the Great Lakes areas, especially in CMIP6.This discrepancy can be associated with the scale mismatch between the station observations and gridded data and the model uncertainties.Blizzard occurrence in SE is based on averaged records among the forecast zones (or observation sites) that fall within the 0.5 • latitude × 0.5 • longitude grid.If only a few zones (or sites) show high occurrence, the grid average will be smoothed out.On the other hand, blizzard occurrence in ERA5 and CMIP6 is based on wind speed and snowfall at the grid scale.
Compared to the CMIP6 models, ERA5 is a reanalysis assimilating observations and has a higher spatial resolution, so it can capture more spatial details of blizzard occurrence and shows a better consistency with SE, such as in Montana and Wyoming.ERA5 also shows blizzard activities over the mountainous regions of the western US, such as the Oregon Cascade, Arizona, and New Mexico, which are not shown in SE, but found in some CMIP6 models (such as CESM2, EC-Earth3, and MPI-ESM1-2-HR, figure S3).The discrepancy can possibly be related to uncertainties of the modeling data in simulating wind speed and snowfall, or unreported events in SE.
If examining individual CMIP6 models, we find that 4 out of 19 models (CMCC-CM2-RS5, CMCC-ESM2, KIOST-ESM, and MRI-ESM2-0) fail to capture the 'hotspot' of blizzard activities over the northern Great Plains (figure S3), mainly due to its deficiency in simulating strong-wind events (figure S4).The two CMCC models show significantly fewer strong-wind and compound-event days, and KIOST-ESM and MRI-ESM2-0 do not show consistent spatial patterns compared to other models.Therefore, those four models are not used in calculating the multimodel ensemble median in the following future projection analysis.
Figure 3 shows the projected changes based on the CMIP6 projections.In the near-term future, there will be a significant reduction in blizzard activities in the areas that show high frequency in the past, including the Northern Plains and Great Lakes region.Significantly reduced blizzard occurrence is also found over the same region in the long-term future but with a stronger reduction in magnitude (figure S5).Decreased blizzard frequency is also found over the Northeast.To understand the cause of the decrease in blizzard occurrence in the future, we analyze the changes in days with strong wind and heavy snow separately.The strong-wind criterion plays a major role in characterizing blizzard events.Historical data from 1980-2009 shows strong winds According to future projections, there will be a significant decrease in strong-wind days over the Northern Plains and Great Lakes region (figures 3(e) and (f)).Iowa will potentially experience the greatest decrease (more than 10 d per year) in strong wind days, while the Dakotas, Nebraska, Kansas, Minneapolis, Missouri, Illinois, Indiana, Michigan, and Ohio will experience a significant decrease in windy days as well.Figures 3(g)-(i) shows the changes in heavysnow events.Historically, the western and northeast regions experience more heavy-snow days per year than other regions across the US.Under a warming climate, there will be a significant decrease in extreme snowy days across the US.Therefore, both the decreases in strong-wind days and heavy-snow days contribute to decreased blizzard occurrences in the future.
Meanwhile, we compare the projected changes under two different scenarios (SSP245 versus SSP585).Both scenarios suggest a significant decrease in blizzard activities in the central US, but the decrease in heavy-snow days is slightly stronger in SSP585 but the decrease in strong-wind is slightly stronger in SSP245 in the near-term future.The long-term future projections also show significantly reduced blizzard occurrences in the Northern Plains and Northeast, with a stronger decrease in the SSP8585 scenario (figure S5).
When focusing on the blizzard zones, there is a gradual decrease in blizzard occurrence in both future scenarios (figure 4), in which SSP585 shows a stronger decrease in the second half of this century, which can also be confirmed by the spatial changes in figures 3 and S5.During the historical period, both ERA5 and CMIP6 show a slightly decreasing trend but not statistically significant.ERA5 shows interdecadal variability of blizzard occurrence, in which the low occurrence during the 2000s well agrees with the SE-based results (figures 1(b) and (d)).Such interdecadal variability is also evident in the CMIP6 multi-model ensemble median but with a small magnitude.

Discussions and summary
Using the SE database, ERA5 reanalysis, and CMIP6 output, this study examines the spatial and temporal variations of blizzard events in the contiguous United States.According to NWS (National Weather Service (NWS) 2021), a blizzard event is defined as a winter storm that satisfies three criteria: (1) sustained winds or frequent gusts 30 knots (35 mph) or greater, (2) falling and/or blowing snow reducing visibility frequently to less than 1 /4 mile, and (3) the above conditions for three consecutive hours or longer.Because visibility is difficult to quantify using reanalysis or climate model data, blizzard events are determined by wind speed and snowfall in our analysis.It should be noted that the calculation is based on daily wind speed and snowfall because those variables are not available at the hourly scale in all the models.Therefore, the time requirement (e.g. at least three consecutive hours) is not taken into account.Despite those limitations, the model-based results exhibit a good agreement with the SE observations, and capture the blizzard 'hotspots' in the Dakotas and Nebraska (figure 1). the spatial consistency, the authors acknowledge the uncertainties related to the thresholds used for compound extreme identification.Future work can be focused on climate downscaling with high spatial and temporal resolutions to better characterize regional or local blizzard events.
CMIP6 projections show a significant decrease in blizzard occurrence over the central US, which is attributed to the decreases in heavy snowfall events and strong wind events in the future.From a largescale perspective, the mean snowfall over the middle latitudes will decrease in a warming climate because snowfall statistics are strongly dependent on the climatological temperature (O'Gorman 2014).The shift in temperature distribution and reduced cold days will limit the snowfall events within a narrow temperature range, leading to less occurrence of snowfall days or extreme snowfall events (O'Gorman 2014).A decrease in land surface wind speed has been observed since the 1980s (Vautard et al 2010), which is followed by a reversal of stilling around 2010 (Zeng et al 2019).However, causes for the different trends remain controversial, such as the impacts of surface roughness change and the decadal-scale atmosphereocean oscillation.Future climate change shows a significant decrease in average wind speed across the northern mid-latitudes (Miao et al 2023).The projected decrease in wind speed can be associated with the reduced meridional temperature gradient as a result of the polar amplification of global warming and so reduced storm track intensity (Karnauskas et al 2018), which potentially leads to decreased strong wind events.When blizzards are treated as compound events of heavy snowfall and strong wind, those mechanisms discussed above would jointly lead to less occurrence of blizzards in the future.From the mesoscale perspective, blizzards are usually associated with extratropical cyclones.Analysis based on The North American Regional Climate Change Assessment Program shows a significant reduction in the frequency of winter extratropical cyclones (Eichler 2020).This is also consistent with the projected decrease in storm track during winter over North America (Chang et al 2012, Chang 2013), which is associated with the reduced baroclinicity due to arctic amplification and the consequently weakened meridional temperature gradient (Wang et al 2017).
Meanwhile, it should be noted that this study is only focused on the blizzard occurrence, because it is difficult to derive other blizzard metrics based on the observational dataset, like other studies (Schwartz andSchmidlin 2002, Coleman andSchwartz 2017).The intensity of blizzard events is beyond our scope.Some studies suggest that future extreme snowfall events will get more intense, especially in Northern America (Quante et al 2021).The intensification of extreme snowfall can be a result of increasing winter precipitation in this region and temperatures still remaining below freezing despite the warming climate (Diffenbaugh et al 2013).Meanwhile, as an important weather system that produces winter storms, the intensity of extratropical cyclones is also projected to increase in some regions like the central and western US (Eichler 2020).Therefore, even though the frequency of blizzard events is projected to decrease, some individual events may get more impactful due to the intensification of extreme snowfall.
This study extends previous work in blizzard climatology (Schwartz and Schmidlin 2002, Coleman and Schwartz 2017) by combing the observed SEs and climate model output and evaluating the future changes in blizzard occurrence using a compoundevent framework.The reduced occurrence of blizzard events may suggest certain benefits of global climate change in winter severe weather risks in the US.However, caution should be taken when applying our results to regional climate assessment and planning.As discussed previously, climate downscaling would better represent SEs at a local scale despite its inherent uncertainties (Deser 2020), and the intensity of individual SEs needs further investigation.and National Weather Service (NWS) for providing the storm events data.The authors are grateful to the reviewers whose insightful comments helped improve manuscript.We would also like to thank Thomas Kauzlarich for his valuable contributions regarding the ASOS and NOHRSC observations.The authors have no conflicts of interest.

Figure 1 .
Figure 1.Number of reported blizzards over the contiguous US based on the SE blizzard records from 1996 to 2020 (a), the changes between the two decades: the 2000s (2001-2010) versus the 2010s (2011-2020) (b), the seasonality of blizzards over two subregions based on the 25 year records (c), and the time series of reported blizzards in three states with a strong increase in Minnesota, North Dakota, and South Dakota (d).The red boundaries outline the two climate regions (Northern Rockies and Plains, and Upper Midwest) defined by the NOAA U.S. Climate Regions.

Figure 2 .
Figure 2. Climatology of blizzard occurrence (episodes per year) derived from (a) SE, (b) ASOS/NOHRSC, (c) ERA5, and (d) CMIP6.Both SE and ERA5 cover the period 1996-2020.Due to the availability, the period 2004-2020 is used in the ASOS/NOHRSC calculation, and the period 1996-2014 is used in the CMIP6 calculation.

Figure 3 .
Figure 3. Occurrence of the compound events (episodes per year) during the historical period 1980-2009 (a), and its projected changes in the future 2030-2059 under the SSP245 (b) and SSP585 (c) scenarios.Panels (d)-(f) and (g)-(i) are for the occurrence (days per year) of heavy snow and strong wind, respectively.Stippling indicates that more than 11 out of 15 models agree on the sign of the change.The red box in panel (a) shows the blizzard zones (40 • -49 • N, 93 • -104 • W) that are focused on for the time series of blizzard occurrence in figure 4.

Figure 4 .
Figure 4. Time series of blizzard occurrence (episodes) over the blizzard zones (defined in figure 3) from 1980 to 2100 based on ERA5, and ensemble medians of CMIP6 historical, SSP245 and SSP585 simulations with the 25th and 75th percentiles as shaded areas.Time series are smoothed by a 5 year running average here.The calculation of CMIP6 historical running average in years 2013 and 2014 uses SSP245 projections in years 2015 and 2016.

Table 1 .
Information on the 19 CMIP6 models used in this study.