• Allen, M. R., , and W. J. Ingram, 2002: Constraints on future changes in climate and the hydrologic cycle. Nature, 419, 224232, doi:10.1038/nature01092.

    • Search Google Scholar
    • Export Citation
  • Barnett, T. P., , J. C. Adam, , and D. P. Lettenmaier, 2005: Potential impacts of a warming climate on water availability in snow-dominated regions. Nature, 438, 303309, doi:10.1038/nature04141.

    • Search Google Scholar
    • Export Citation
  • Cattiaux, J., , H. Douville, , and Y. Peings, 2013: European temperatures in CMIP5: Origins of present-day biases and future uncertainties. Climate Dyn., 41, 28892907, doi:10.1007/s00382-013-1731-y.

    • Search Google Scholar
    • Export Citation
  • Changnon, S. A., , and D. Changnon, 2006: A spatial and temporal analysis of damaging snowstorms in the United States. Nat. Hazards, 37, 373389, doi:10.1007/s11069-005-6581-4.

    • Search Google Scholar
    • Export Citation
  • Collins, M., and Coauthors, 2013: Long-term climate change: Projections, commitments and irreversibility. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 1029–1136.

  • Eisenberg, D., , and K. E. Warner, 2005: Effects of snowfalls on motor vehicle collisions, injuries, and fatalities. Amer. J. Public Health, 95, 120124, doi:10.2105/AJPH.2004.048926.

    • Search Google Scholar
    • Export Citation
  • Franklin, B. A., , K. Bonzheim, , S. Gordon, , and G. Timmis, 1996: Snow shoveling: A trigger for acute myocardial infarction and sudden coronary death. Amer. J. Cardiol., 77, 855858, doi:10.1016/S0002-9149(97)89181-3.

    • Search Google Scholar
    • Export Citation
  • Harris, I., , P. D. Jones, , T. J. Osborn, , and D. H. Lister, 2014: Updated high-resolution grids of monthly climatic observations—The CRU TS3.10 Dataset. Int. J. Climatol., 34, 623642, doi:10.1002/joc.3711.

    • Search Google Scholar
    • Export Citation
  • Hartmann, D. L., and Coauthors, 2013: Observations: Atmosphere and surface. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 159–254.

  • IPCC, 2013: Summary for policymakers. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 3–29.

  • Kapnick, S. B., , and T. L. Delworth, 2013: Controls of global snow under a changed climate. J. Climate, 26, 55375562, doi:10.1175/JCLI-D-12-00528.1.

    • Search Google Scholar
    • Export Citation
  • Krasting, J. P., , A. J. Broccoli, , K. W. Dixon, , and J. R. Lanzante, 2013: Future changes in Northern Hemisphere snowfall. J. Climate, 26, 78137828, doi:10.1175/JCLI-D-12-00832.1.

    • Search Google Scholar
    • Export Citation
  • Mehran, A., , A. AghaKouchak, , and T. J. Phillips, 2014: Evaluation of CMIP5 continental precipitation simulations relative to satellite-based gauge-adjusted observations. J. Geophys. Res. Atmos., 119, 16951707, doi:10.1002/2013JD021152.

    • Search Google Scholar
    • Export Citation
  • Notaro, M., , D. Lorenz, , C. Hoving, , and M. Schummer, 2014: Twenty-first-century projections of snowfall and winter severity across central-eastern North America. J. Climate, 27, 65266550, doi:10.1175/JCLI-D-13-00520.1.

    • Search Google Scholar
    • Export Citation
  • Notaro, M., , V. Bennington, , and S. Vavrus, 2015: Dynamically downscaled projections of lake-effect snow in the Great Lakes basin. J. Climate, 28, 16611684, doi:10.1175/JCLI-D-14-00467.1.

    • Search Google Scholar
    • Export Citation
  • O’Gorman, P., 2014: Contrasting responses of mean and extreme snowfall to climate change. Nature, 512, 416418, doi:10.1038/nature13625.

    • Search Google Scholar
    • Export Citation
  • Riahi, K., and Coauthors, 2011: RCP 8.5—A scenario of comparatively high greenhouse gas emissions. Climatic Change, 109, 3357, doi:10.1007/s10584-011-0149-y.

    • Search Google Scholar
    • Export Citation
  • Sheffield, J., and Coauthors, 2013: North American climate in CMIP5 experiments. Part I: Evaluation of historical simulations of continental and regional climatology. J. Climate, 26, 92099245, doi:10.1175/JCLI-D-12-00592.1.

    • Search Google Scholar
    • Export Citation
  • Su, F., , X. Duan, , D. Chen, , Z. Hao, , and L. Cuo, 2013: Evaluation of the global climate models in the CMIP5 over the Tibetan Plateau. J. Climate, 26, 31873208, doi:10.1175/JCLI-D-12-00321.1.

    • Search Google Scholar
    • Export Citation
  • Sun, Y., , S. Solomon, , A. Dai, , and R. W. Portmann, 2007: How often will it rain? J. Climate, 20, 48014818, doi:10.1175/JCLI4263.1.

  • Taylor, K. E., , R. J. Stouffer, , and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485498, doi:10.1175/BAMS-D-11-00094.1.

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., 2011: Changes in precipitation with climate change. Climate Res., 47, 123138, doi:10.3354/cr00953.

  • Vavrus, S., 2007: The role of terrestrial snow cover in the climate system. Climate Dyn., 29, 7388, doi:10.1007/s00382-007-0226-0.

  • View in gallery

    Percent change in frequency of measurable daily snowfall events (≥0.25 cm) in all seasons simulated by the CMIP5 multimodel ensemble for the periods of (a) 2021–50 and (c) 2071–2100. The changes are expressed relative to a reference period of 1971–2000 [see Eq. (1)]. Grid boxes that had less than five events in the reference period are shaded gray. Line contours display the model-mean annual temperature over the reference period at intervals of 10°C. In addition, the percent of models projecting a decrease in frequency of measurable events, excluding any models that had zero events in both the historical and future time periods, is given for the periods (b) 2021–50 and (d) 2071–2100.

  • View in gallery

    As in Fig. 1, but for daily snowfall events ≥ 10 cm.

  • View in gallery

    (a),(b) As in Figs. 1a and 1c, but for daily snowfall events ≥ 25 cm.

  • View in gallery

    (a),(b) As in Figs. 2a and 2c, but for midwinter (January–February) events and the midwinter temperature only.

  • View in gallery

    Regions used for regional analysis of projected daily snowfall changes. Regions 1–20 are from Krasting et al. (2013), while regions 21–23 are new regions in interior North America defined for this study.

  • View in gallery

    Histograms depicting the change in frequency of daily snowfall events (January–February) as a function of intensity for the periods of (a) 2021–50 and (b) 2071–2100 for central and southern Europe. The ordinate is displayed on a nonlinear scale in order to facilitate the representation of values that can span several orders of magnitude. The histogram bars represent the mean difference in frequency among the ensemble means of all the models for that particular intensity bin [see Eq. (2)], and each white circle represents the median difference in frequency among the ensemble means. Within a bin, the narrow black tick marks represent the difference in frequency of each multimodel ensemble member for that bin, while the upper and lower whiskers display the maximum and minimum difference, respectively, among all the members. All grid cells containing more than 50% water are masked out, and all of the frequency differences are divided by the total number of land grid points in the region. Average daily snowfall is displayed on each histogram for the historical and future period.

  • View in gallery

    As in Fig. 6, but for the (a) New England and (b) Japan regions, for 2071–2100.

  • View in gallery

    As in Fig. 6, but for the north-central United States for (a) January–February and (b) March–April for 2021–50.

  • View in gallery

    As in Fig. 6, but for 2071–2100 for southern Québec and New Brunswick for (a) January–February and (b) November–December.

  • View in gallery

    As in Fig. 6, but for northwestern Siberia for (a),(b) November–December and (c),(d) September–October for (left) 2021–50 and (right) 2071–2100.

  • View in gallery

    As in Fig. 6, but for (a),(b) northeastern Greenland for September–October and (c),(d) southeastern Greenland for May–June for (left) 2021–50 and (right) 2071–2100.

  • View in gallery

    As in Fig. 6, but modified to show relationships between changes in snowfall events and temperature biases for (a) southern Québec and New Brunswick, (b) the north-central United States, and (c) northern Scandinavia for January–February for the period of 2071–2100. Within each intensity bin, the individual ensemble members are divided into five groups based on their bias in simulated temperature over the period of 1971–2000 for the specified region. The five colored bars in a bin represent the mean difference in frequency of daily snowfall events computed over each of the temperature bias groups: Darker and lighter blue bars average over members with cold biases of >5°C and 2°–5°C, respectively; darker and lighter red bars average over members with warm biases of >5°C and 2°–5°C, respectively; and gray bars average over members with biases smaller than 2°C. The numbers next to the colored boxes at the top of each graph display the total count of members belonging to the corresponding temperature bias group. For reference, the wider white bars represent the mean difference in frequency among the ensemble means of all the models for that particular intensity bin (like the cyan bars in Figs. 611), and each white circle represents the median difference in frequency among the ensemble means.

  • View in gallery

    As in Fig. 12, but for (a) northeastern Siberia for March–April, (b) northeastern Québec for November–December, and (c) southeastern Greenland for May–June for 2071–2100.

  • View in gallery

    As in Fig. 12, but for (a) southern Québec and New Brunswick for March–April, (b) the U.S. Mid-Atlantic for January–February, and (c) northwestern Siberia for January–February for 2071–2100.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 3 3 3
PDF Downloads 2 2 2

Effects of a Warming Climate on Daily Snowfall Events in the Northern Hemisphere

View More View Less
  • 1 Department of Environmental Sciences, Rutgers University, New Brunswick, New Jersey
  • 2 Department of Atmospheric and Oceanic Sciences, University of California, Los Angeles, Los Angeles, California
  • 3 Department of Environmental Sciences, Rutgers University, New Brunswick, New Jersey
  • 4 Department of Environmental Sciences, and Institute for Earth, Ocean, and Atmospheric Sciences, Rutgers University, New Brunswick, New Jersey
© Get Permissions
Full access

Abstract

Using simulations performed with 24 coupled atmosphere–ocean global climate models from phase 5 of the Coupled Model Intercomparison Project (CMIP5), projections of Northern Hemisphere daily snowfall events under the RCP8.5 emissions scenario are analyzed for the periods of 2021–50 and 2071–2100 and compared to the historical period of 1971–2000. The overall frequency of daily snowfall events is simulated to decrease across much of the Northern Hemisphere, except at the highest latitudes such as northern Canada, northern Siberia, and Greenland. Seasonal redistributions of daily snowfall event frequency and average daily snowfall are also projected to occur in some regions. For example, large portions of the Northern Hemisphere, including much of Canada, Tibet, northern Scandinavia, northern Siberia, and Greenland, are projected to experience increases in average daily snowfall and event frequency in midwinter. But in warmer months, the regions with increased snowfall become fewer in number and are limited to northern Canada, northern Siberia, and Greenland. These simulations also show changes in the frequency distribution of daily snowfall event intensity, including an increase in heavier snowfall events even in some regions where the overall snowfall decreases. The projected changes in daily snowfall event frequency exhibit some dependence on the temperature biases of the individual models in certain regions and times of the year, with colder models typically toward the positive end of the distribution of event frequency changes and warmer models toward the negative end, particularly in regions near the transition zone between increasing and decreasing snowfall.

Current affiliation: School of Meteorology, University of Oklahoma, Norman, Oklahoma.

Corresponding author address: James F. Danco, Department of Environmental Sciences, Rutgers University, 14 College Farm Road, New Brunswick, NJ 08901. E-mail: jdanco@ou.edu

Abstract

Using simulations performed with 24 coupled atmosphere–ocean global climate models from phase 5 of the Coupled Model Intercomparison Project (CMIP5), projections of Northern Hemisphere daily snowfall events under the RCP8.5 emissions scenario are analyzed for the periods of 2021–50 and 2071–2100 and compared to the historical period of 1971–2000. The overall frequency of daily snowfall events is simulated to decrease across much of the Northern Hemisphere, except at the highest latitudes such as northern Canada, northern Siberia, and Greenland. Seasonal redistributions of daily snowfall event frequency and average daily snowfall are also projected to occur in some regions. For example, large portions of the Northern Hemisphere, including much of Canada, Tibet, northern Scandinavia, northern Siberia, and Greenland, are projected to experience increases in average daily snowfall and event frequency in midwinter. But in warmer months, the regions with increased snowfall become fewer in number and are limited to northern Canada, northern Siberia, and Greenland. These simulations also show changes in the frequency distribution of daily snowfall event intensity, including an increase in heavier snowfall events even in some regions where the overall snowfall decreases. The projected changes in daily snowfall event frequency exhibit some dependence on the temperature biases of the individual models in certain regions and times of the year, with colder models typically toward the positive end of the distribution of event frequency changes and warmer models toward the negative end, particularly in regions near the transition zone between increasing and decreasing snowfall.

Current affiliation: School of Meteorology, University of Oklahoma, Norman, Oklahoma.

Corresponding author address: James F. Danco, Department of Environmental Sciences, Rutgers University, 14 College Farm Road, New Brunswick, NJ 08901. E-mail: jdanco@ou.edu

1. Introduction

Snow is an important aspect of weather and climate with physical, ecological, and societal impacts. The presence of snow cover increases Earth’s surface albedo, which has a major cooling effect on the climate, and it has a dramatic influence on the hydrological cycle in the midlatitudes and in the Arctic through spring melting (Vavrus 2007). Accumulated snow is a source of ground and surface water supply via natural (e.g., lakes and rivers) and engineered (e.g., reservoirs and aqueducts) storage and distribution systems. Over one-sixth of the world’s population relies on melting snow and ice for their water supply (Barnett et al. 2005). Snowfall, which can be defined as the accumulation of snow during a given time period, greatly disrupts transportation and causes automobile accidents (Eisenberg and Warner 2005), and it has been linked with increases in heart attacks (Franklin et al. 1996). Adverse economic effects include canceled events, roof and building damage, snow removal costs, and damaging floods from rapid melting; in fact, during 1949–2000, major snowstorms caused a total of $21.6 billion in property losses in the United States (Changnon and Changnon 2006).

Global mean surface temperature has increased during the past half-century, primarily due to the emission of greenhouse gases as a result of human activities, and this trend is likely to continue and perhaps accelerate in the coming decades (IPCC 2013). If there were no changes in the amount of precipitation, rising temperatures would produce less snowfall almost everywhere around the globe because a higher fraction of the total precipitation would fall as rain rather than snow. However, the expected response is not this simple because it is virtually certain that global precipitation will increase with increased global temperature, at a likely rate of 1% to 3% per °C of warming, with precipitation increases likely occurring in high latitudes and currently moist midlatitude regions under the RCP8.5 greenhouse gas forcing scenario (Collins et al. 2013). This is primarily due to exponential increases in atmospheric moisture content, resulting from exponential increases in the water-holding capacity of the atmosphere by ~7% per °C of warming, and to increased transport of water vapor from the tropics (Sun et al. 2007; Collins et al. 2013). In the winter months of December–February (DJF), model simulations from phase 5 of the Coupled Model Intercomparison Project (CMIP5) indicate especially strong increases in precipitation in middle and high latitudes compared to other months of the year. Mean DJF precipitation is projected to increase by at least 10% throughout much of the Northern Hemisphere where snow falls and by greater than 50% across far northern Canada and northern Asia by 2081–2100 under the RCP8.5 forcing scenario (Collins et al. 2013; see Fig. 1 in their box 12.1). Extreme precipitation events in particular are exhibiting strong changes. Hartmann et al. (2013) found that there have likely been increases in either the frequency or intensity of heavy precipitation in North America and Europe since the middle of the twentieth century, and in Asia increases are being observed in more regions than decreases. Furthermore, Collins et al. (2013) concluded that a global shift to more intense precipitation events and fewer weak events is likely to continue as temperatures increase, especially over most midlatitude landmasses and wet tropical regions. The total effect of global warming on snowfall is a delicate balance between these competing effects of increased temperature and increased precipitation.

A number of recent studies have examined projected future changes in snowfall and the mechanisms associated with them. Kapnick and Delworth (2013) analyzed the response of snowfall to an idealized increase in carbon dioxide concentration in two versions of a GFDL climate model that differ primarily in resolution. They found that snowfall is simulated to decrease over most continental locations, but that increases in snowfall are simulated in high-latitude and high-elevation regions where the contribution of increased precipitation dominates the reduced fraction of precipitation that falls in the form of snow. Krasting et al. (2013) examined trends in annual, seasonal, and monthly Northern Hemisphere snowfall in a multimodel ensemble of CMIP5 simulations of projected future climate under the RCP4.5 emissions scenario. Similarly, the study discovered that total annual snowfall is projected to decrease over areas in which winter temperatures are relatively mild and even a modest amount of warming will decrease the amount of precipitation that falls as snow. At higher latitudes, where the climate is typically cold enough to remain below freezing and still support snow even with moderate warming, they found that the simulated increase in winter precipitation is sufficient to offset the decrease in snow fraction that results from rising temperatures. This includes large regions of Eurasia and North America in the midwinter but is more limited to the northernmost regions in warmer months. The study also found that the −10°C isotherm in the simulated late twentieth-century climate approximately separates the regions of increasing and decreasing snowfall (Krasting et al. 2013).

Other studies have focused on higher intensity snowfall events. Rather than using model output directly, Notaro et al. (2014) processed climate model output using a statistical downscaling approach to examine projected trends in snowfall and snow depth over central and eastern North America. As in the studies cited previously, they found that annual snowfall is expected to decrease across this region. However, their analysis also indicated that daily snowfall events over much of this region are projected to become less frequent but more intense. O’Gorman (2014) found that warming from CMIP5 model simulations causes widespread decreases in annual mean snowfall across the middle latitudes from 1981–2000 to 2081–2100, with increases confined to extreme northern Canada, Greenland, and northern Siberia. In contrast, extreme snowfall events exhibit a much more muted response. For example, in the multimodel median, mean snowfall decreases by 65% for monthly climatological temperatures just below freezing while the 99.99th percentile of daily snowfall only decreases by 8% (O’Gorman 2014).

This study will specifically examine how the frequency distribution of daily snowfall events in the Northern Hemisphere will be affected by increasing temperatures, thus extending the work of Krasting et al. (2013) that examined snowfall on monthly and annual time scales. The approach will be somewhat different from that of O’Gorman (2014) by focusing on spatial variations in the response of daily snowfall. As in both of these previous studies, a key issue will be the interplay between changes in frequency distribution of precipitation amount and changes in the fraction of precipitation falling as snow.

Section 2 of this paper includes a description of the methodology, including the model output and observed data used in the analysis. Section 3 discusses the projected future changes in daily snowfall and how well they match the findings from previous studies. Section 4 examines the extent to which daily snowfall projections may be affected by the temperature biases of the models. Section 5 summarizes the findings of the study while identifying questions that could be the subject of future research.

2. Data and methodology

This analysis uses output from an ensemble of climate model simulations coordinated under CMIP5, promoted by the World Climate Research Programme (Taylor et al. 2012). The CMIP5 output was downloaded using the Program for Climate Model Diagnosis and Intercomparison web portal (http://pcmdi9.llnl.gov/), and two experimental designs are considered. The first is a historical simulation of twentieth-century climate and the second is a twenty-first-century climate simulation using the RCP8.5 forcing scenario, which increases greenhouse gas emissions by a factor of about 3 over the course of the twenty-first century (Riahi et al. 2011). A total of 24 coupled climate models and 37 ensemble members had daily snowfall data available (in terms of mass per unit area per unit time) and thus are used in this study. Table 1 provides a list of the models used, the number of ensemble members from each model, and each model’s original resolution. Since the models were run at a wide range of spatial resolutions, their output is interpolated to a common 1° latitude by 1° longitude grid to facilitate comparison among the models. This resolution is chosen to approximate the highest resolution among the models. All grid boxes that contain more than 50% water (based on the GFDL-ESM2M land fractions that are interpolated down to 1° × 1° resolution) are discarded because the impacts of snowfall are confined mainly to land regions. As noted by Krasting et al. (2013), the CMIP5 models determine precipitation type using a variety of methods. The snowfall output from the models is used regardless of any differences in methods of determining precipitation type. Since the model output provides no information about snow density, snowfall is determined from its water equivalent by assuming a uniform 10:1 snow-to-liquid ratio, as in Krasting et al. (2013).

Table 1.

List of CMIP5 models included in the analysis. The number of ensemble members used and the horizontal resolution are indicated for each model.

Table 1.

Since this study uses climate model simulations from CMIP5, it is important that any biases in the CMIP5 models are identified and considered when examining their simulations of Northern Hemisphere snowfall. Krasting et al. (2013) compared CMIP5 snowfall output in the Northern Hemisphere to estimates of observed snowfall and found that a positive snowfall bias exists in the multimodel ensemble over the western half of North America and much of Eurasia, except for central Europe where snowfall is underestimated. However, the models do reasonably capture the patterns of relative maxima and minima of snowfall.

A general cold bias exhibited by CMIP5 models, particularly in cold regions and months, can at least partially explain the positive bias in snowfall in most of the Northern Hemisphere. For example, an evaluation of European temperatures and their projected changes from the CMIP5 ensemble found that, on average, these models have a cold bias in winter, especially in northern Europe (Cattiaux et al. 2013). Another study by Su et al. (2013) compared CMIP5 model output to ground observations in the eastern Tibetan Plateau for the period of 1961–2005 and found that the majority of models have cold biases averaging 1.1°–2.5°C for the months of December–May, and less than 1°C for June–October. A substantial wintertime cold bias has also been shown to exist among CMIP5 models in the very high latitudes of North America (Sheffield et al. 2013).

CMIP5 simulations suffer from biases in precipitation as well. An analysis of the precipitation biases of 17 CMIP5 models over North America for 1979–2005 found that the models have a mean positive bias of 12% in DJF, with an overestimation of precipitation in more humid and cooler regions and an underestimation in drier regions (Sheffield et al. 2013). Another study found that most CMIP5 models have a positive bias in precipitation over regions of complex topography, such as western North and South America and southern Africa and Asia, and a negative bias over arid regions (Mehran et al. 2014).

To determine the temperature biases associated with each model and how such biases may impact the response of daily snowfall, historical land-only monthly temperature data from the Climatic Research Unit (CRU) TS v. 3.23 dataset over the period of 1971–2000 are also used in this study (Harris et al. 2014). They were downloaded from the CRU website (https://crudata.uea.ac.uk/cru/data/hrg/) and have a native resolution of 0.5° latitude × 0.5° longitude.

3. Projected future changes in daily snowfall

a. Spatial patterns

We begin by examining projected future changes in daily snowfall event frequency for events exceeding a given threshold amount. The historical time period used is 1971–2000, and it is compared to two future time periods: 2021–50 and 2071–2100. To enable each model to have the same weight regardless of the number of available ensemble members from that model, an ensemble mean event frequency for each model at each grid cell is calculated by averaging the frequency of daily snowfall events at or above a given threshold across the ensemble members. The model-mean percent change in frequency between two time periods is then computed as
e1
where is the number of models and the superscripts indicate the time period. This calculation is done for all months of the year (“annual”) and for a midwinter period (January–February) chosen to be two months in length so that the sample size is large enough to get reasonably robust results.

A map of the projected percent change in the annual number of measurable daily snowfall events, defined here as events ≥ 0.25 cm, from 1971–2000 to 2021–50 is shown in Fig. 1a. The analysis indicates that measurable snowfall events decrease in frequency in most regions of the Northern Hemisphere where snow occurs, with the percent decrease in frequency especially large across the southern United States and Mexico, much of central and southern Europe, and central and southern Asia, where decreases of more than 30% are projected. The only areas with little change or a slight increase in measurable snowfall events are in some very high-latitude regions, including far northern Canada, northeastern Greenland, and northeastern Siberia. This projected response would be expected with a warmer planet as higher temperatures cause more precipitation events to become rain instead of snow, except at the very coldest regions where temperatures average so far below freezing that they can rise substantially and still be cold enough for snow, and thus the increase in precipitation becomes the dominant factor. Figure 1a indicates that this transition between increasing and decreasing changes in measurable snowfall events by 2021–50 generally occurs near the −10°C isotherm in the historical period in northeastern Siberia and far northern Canada and near the −20°C isotherm in Greenland. The majority of models project an increase in events in far northern Canada, northeastern Greenland, and northeastern Siberia (Fig. 1b). In most of the rest of the Northern Hemisphere, there is nearly unanimous agreement among all the models that measurable snowfall events will decrease (Fig. 1b). By 2071–2100 (Fig. 1c), the changes discussed above become even more pronounced as the climate continues to warm. Thus the frequency of measurable daily snowfall events once again decreases throughout almost the entire Northern Hemisphere, but now the drop is even greater. The only regions where these events are projected to increase are the same very high-latitude areas as the earlier time period, and the transition between increasing and decreasing events occurs near the same climatological temperatures as in the earlier time period. The map of the percentage of models projecting a decrease by 2071–2100 (Fig. 1d) is very similar to Fig. 1b, but the models have 95%–100% agreement in even more areas than in the earlier time period.

Fig. 1.
Fig. 1.

Percent change in frequency of measurable daily snowfall events (≥0.25 cm) in all seasons simulated by the CMIP5 multimodel ensemble for the periods of (a) 2021–50 and (c) 2071–2100. The changes are expressed relative to a reference period of 1971–2000 [see Eq. (1)]. Grid boxes that had less than five events in the reference period are shaded gray. Line contours display the model-mean annual temperature over the reference period at intervals of 10°C. In addition, the percent of models projecting a decrease in frequency of measurable events, excluding any models that had zero events in both the historical and future time periods, is given for the periods (b) 2021–50 and (d) 2071–2100.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

Examining changes in the frequency of events for higher thresholds yields substantially different results. For example, the geographic boundary between positive and negative changes in the frequency of annual events ≥ 10 cm day−1 from 1971–2000 to 2021–50 is much farther south, and it generally coincides with an annual average temperature over 1971–2000 of about 0°C (Fig. 2a). Many more places are projected to have a higher number of these events, including much of Canada, Greenland, northern Asia, and the Tibetan Plateau region. The likely reason for this disparity in response between measurable events and larger events is that warming increases absolute humidity and leads to higher sea surface temperatures in moisture source regions, which leads to a disproportionate increase in heavier precipitation events (Allen and Ingram 2002; Sun et al. 2007; Trenberth 2011; Collins et al. 2013) and thus more heavy snowfall events during the instances when it is below freezing. The dependence of changes in frequency of ≥10 cm day−1 events on temperature results in disparities among regions of similar latitude. Regions with more continental climates are more likely to have projected increases in ≥10 cm day−1 events while regions with stronger marine influence such as Europe and the west coast of North America (even as far north as southern Alaska) have projected decreases. The effects of elevation are also clearly seen in the Tibetan Plateau region of central Asia, where the frequency of these events increases despite decreases in places at similar latitudes to the east and west. Since the elevation here is very high, and the marine influence is minimal, temperatures are able to remain cold enough for snow even with projected future warming. In the case of ≥10 cm day−1 events, there are few areas with nearly unanimous agreement among the models in the projected sign of change, and there is considerable disagreement in many areas (Fig. 2b).

Fig. 2.
Fig. 2.

As in Fig. 1, but for daily snowfall events ≥ 10 cm.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

By 2071–2100, the frequency of events ≥10 cm day−1 is projected to decrease even further in the United States, Europe, and central Asia outside of Tibet, while the opposite occurs in high-latitude or high-elevation regions, including northern and central Canada, Greenland, northern Asia, and Tibet (Fig. 2c). In some of the highest latitudes, percent increases even approach and exceed 100%. Thus the trends in the earlier time period are amplified in the later time period during which higher temperatures are simulated. Interestingly, however, the geographic boundary between increasing and decreasing frequency is not projected to retreat poleward very much. This is likely because the competing effects of warming temperatures and rising precipitation remain nearly the same close to the boundary, so while both of these variables increase, they each continue to have an equal but opposite effect on snowfall. As a result, they offset each other, and the net result is little movement of the boundary. There is considerably better model agreement in the sign of change of >10 cm day−1 events for 2071–2100 compared to 2021–50, with 95%–100% of models projecting an increase or decrease in many regions (Fig. 2d). The patterns seen in Figs. 2b and 2d hold true for even larger intensity thresholds and in January–February only (not shown).

Raising the intensity threshold further to 25 cm day−1 results in the geographic boundary between increasing and decreasing frequency shifting even farther equatorward (Fig. 3a), including large portions of Asia and North America in projected increases (extending as far south as the northern United States). Note that the fewer number of these events results in a larger range for the percent change metric and a noisier spatial pattern. The boundary between increasing and decreasing events at this intensity occurs at even warmer average annual temperatures over 1971–2000, generally around 5°C. As with the ≥ 10 cm day−1 events, by 2071–2100, the geographic boundary between increasing and decreasing frequency experiences little movement, and areas with a projected decrease in the earlier time period are projected to experience even fewer ≥ 25 cm day−1 events, while regions with an increase are projected to experience even more ≥ 25 cm day−1 events (Fig. 3b).

Fig. 3.
Fig. 3.

(a),(b) As in Figs. 1a and 1c, but for daily snowfall events ≥ 25 cm.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

Focusing next on midwinter (i.e., January–February), the frequency of larger snowfall events is generally projected to increase in more regions than it does when all months are considered. For instance, by 2021–50, the frequency of January–February days with snowfall ≥ 10 cm is simulated to rise in not only Canada, Greenland, and much of northern and central Asia, but also in parts of northern Europe and the northern United States (Fig. 4a). High latitudes are projected to experience a substantial increase in these events, greater than 100%, which is much higher than their projected annual increase. Because midwinter months are colder, more regions are able to stay cold enough for snow that might be too warm in other months of the year, and since a greater fraction of precipitation is still able to stay as snow, the increase in precipitation becomes an even more dominant factor over warming temperatures. The boundary between increasing and decreasing events coincides with an average January–February temperature over 1971–2000 of about −10°C. By 2071–2100, the geographic boundary between increasing and decreasing frequency of midwinter events ≥ 10 cm day−1 stays in roughly the same position as it was during the earlier time period (Fig. 4b), similar to what was seen with annual events, but the magnitude of the changes goes up considerably.

Fig. 4.
Fig. 4.

(a),(b) As in Figs. 2a and 2c, but for midwinter (January–February) events and the midwinter temperature only.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

To conduct a more regional analysis, percent changes in the frequency of measurable and heavy daily snowfall events for the entire year are computed over each of the 20 Northern Hemisphere regions used in the study by Krasting et al. (2013), plus three additional regions in interior North America (Table 2; see Fig. 5 for a geographic representation of the regions). As might be expected from Fig. 1, every region is projected to experience a decrease in the annual number of measurable snowfall events by 2021–50, with the exception of northeastern Greenland where a small increase is projected. As was also evident in the maps, regions with a projected increase in events are more common for heavier events than for all measurable events, especially for events ≥ 25 cm day−1. Colder regions are also more likely to show projected increases in snowfall events (note that regions are sorted from warmest to coldest model-mean annual climatological temperature in Table 2). However, the dependence on climatological temperature is less clear for large snowfall events than for all measurable events, as some regions with projected decreases in the frequency of large events are colder than regions with projected increases in these events. This is likely at least partly because regions with similar annual mean temperatures can have very different snowfall climatologies if they have maritime versus continental climates, such as British Columbia versus southern Québec and New Brunswick (Table 2). Furthermore, the spread between the 25th and 75th percentiles among the ensemble members is much greater for large events than for all measurable events, indicating a larger uncertainty in the effects of a warming climate on high-intensity snowfall events. This is especially true over Europe and central and eastern Asia, where the sign of the projected change in extreme events varies considerably among the ensemble members (Table 2).

Table 2.

Percent change in the frequency of daily snowfall events at or above specified thresholds from all seasons. The changes are expressed relative to the period 1971–2000 and are organized by future time period and region. Regions are sorted top to bottom from warmest to coldest model-mean annual climatological temperature over 1971–2000. The model-mean percent frequency changes in snowfall (as described in the text) are shown. Adjacent numbers in parentheses represent the 25th and 75th percentile frequency changes among the 37 individual ensemble members. Undefined values (—) occurred if there were no days with snowfall in the historical period for many models.

Table 2.
Fig. 5.
Fig. 5.

Regions used for regional analysis of projected daily snowfall changes. Regions 1–20 are from Krasting et al. (2013), while regions 21–23 are new regions in interior North America defined for this study.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

The time of year also has a large influence on the change in daily snowfall events. Compared with events throughout the year, simulated percent changes in the number of January–February events are considerably more positive in colder regions and considerably less negative in warmer regions (Table 3). For example, while measurable events during the entire year decrease in every region but one, in midwinter they are projected to stay the same or increase by 2021–50 in the three northeastern Canadian regions, southern Canada, northern Scandinavia, Tibet, both regions in northern Siberia, and both regions in Greenland. Once again, regions with a colder January–February climatological temperature are much more likely to have projected increases in snowfall events during these months, but there are some regions that have smaller projected increases (or even projected decreases) in heavy midwinter events compared to regions that are warmer. The spread in model projections is greater for large midwinter events as well (Table 3).

Table 3.

As in Table 2, but for midwinter (January–February) daily snowfall events and regions sorted by model-mean midwinter climatological temperature.

Table 3.

b. Daily snowfall histograms

Histograms depicting daily snowfall changes are constructed for each of the 23 regions from the previous subsection. The same historical and future time periods are used as well. Daily snowfall events at each grid point in the region are categorized into 12 bins, with the lowest including events < 0.25 cm, defined as a trace (T), which is approximately the cutoff between measurable and nonmeasurable snowfall. The second bin ranges from T to 5 cm and the next nine have a uniform width of 5 cm. The last bin includes daily snowfall events greater than 50 cm. The model-mean frequency difference between two time periods is then computed for each bin and for each region as
e2
where is the ensemble mean event frequency for each model at each grid cell , is the number of models, is the number of land grid cells in the region, and the superscripts indicate the time period. The frequency distribution is normalized by the total number of land grid points in the region in order to effectively compare regions of different sizes. The model–median differences using the ensemble means for each model are also computed. Additionally, the maximum, minimum, and individual differences among all 37 separate ensemble members are calculated and shown in the histograms. These difference histograms are computed for each of six 2-month intervals (January–February through November–December), and for each future period (2021–50 and 2071–2100). The July–August interval is only analyzed for the two Greenland regions, because in all other regions these months are too warm for substantial amounts of snow to fall. Since it is not feasible to show histograms for all 23 regions and all 2-month intervals, selected histograms are analyzed to illustrate some examples of each type of snowfall response.

In some regions that are examined, the climate is warm enough that an increase in temperature will cause a large decrease in daily snowfall for all bins in all months of the year, due to less precipitation falling as snow and more as rain. The only bin that increases in such regions is 0–T, as the frequency of days with no snow must go up when the overall frequency of daily snowfall events goes down. Regions exhibiting this behavior include the southeastern United States, the U.S. Mid-Atlantic, the south-central United States, the U.S. West Coast, British Columbia, southern Alaska, and central and southern Europe. A difference histogram for the last of these regions for January–February for 2021–50 serves as an example (Fig. 6a). As is clearly shown in this figure, a sharp decline in daily snowfall event frequency is projected by most of the models in this region by 2021–50, and all 37 ensemble members simulate that the frequency of days with zero snowfall will go up and the frequency of days with T–5 cm will go down. The average daily snowfall is projected to decrease substantially as well.

Fig. 6.
Fig. 6.

Histograms depicting the change in frequency of daily snowfall events (January–February) as a function of intensity for the periods of (a) 2021–50 and (b) 2071–2100 for central and southern Europe. The ordinate is displayed on a nonlinear scale in order to facilitate the representation of values that can span several orders of magnitude. The histogram bars represent the mean difference in frequency among the ensemble means of all the models for that particular intensity bin [see Eq. (2)], and each white circle represents the median difference in frequency among the ensemble means. Within a bin, the narrow black tick marks represent the difference in frequency of each multimodel ensemble member for that bin, while the upper and lower whiskers display the maximum and minimum difference, respectively, among all the members. All grid cells containing more than 50% water are masked out, and all of the frequency differences are divided by the total number of land grid points in the region. Average daily snowfall is displayed on each histogram for the historical and future period.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

By 2071–2100, this downward trend in daily event frequency is larger, with mean midwinter daily snowfall in central and southern Europe reduced by more than half (Fig. 6b). This dramatic decrease in daily snowfall in these regions is due to relatively high winter temperatures that make the climate marginal for snow. Thus the higher temperatures that models simulate in the future periods would allow for more precipitation that would have fallen as snow to instead fall as rain. While increases in precipitation are likely in at least some of these regions (Sun et al. 2007; Collins et al. 2013), this is not enough to offset the effects of increasing temperatures. Decreases in snowfall are projected to be even more dramatic in these regions in the warmer months such as the fall transition months (November–December) and the spring transition months (March–April), during which the fraction of precipitation that falls as snow is even more sensitive to rising temperatures (not shown). For example, by the later time period, average daily snowfall in these transition months in the U.S. mid-Atlantic is projected to be less than a third of what it was in 1971–2000.

In other regions that are slightly farther north and colder, including the New England, Japan, Caucasus Mountains, and Baltic Sea regions, average snowfall is still simulated to decrease in all months of the year, but in January–February the frequency of larger snowfall events stays about the same or even slightly increases. Figure 7a shows a clear example of this pattern, with the frequency of small and moderate midwinter snow events in New England dropping by 2071–2100 but the occurrence of large events ≥ 30 cm day−1 slightly increasing in the ensemble mean. Midwinter snowfall is also projected to decrease in Japan by 2071–2100, but the frequency of events ≥ 25 cm day−1 goes up slightly (Fig. 7b). As discussed earlier, this is likely due to higher atmospheric moisture and higher sea surface temperatures causing a greater number of intense precipitation events (Allen and Ingram 2002; Sun et al. 2007; Trenberth 2011; Collins et al. 2013), including heavy snowfall events when temperatures are cold enough.

Fig. 7.
Fig. 7.

As in Fig. 6, but for the (a) New England and (b) Japan regions, for 2071–2100.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

The north-central United States, southern Canada, southern Québec and New Brunswick, Labrador, eastern Hudson Bay, Tibet, and northern Scandinavia are regions where the models project an overall increase or little change in daily January–February snowfall by 2021–50. For example, an increase or little difference in snowfall event frequency in midwinter is projected for all bins except for the lightest snowfall events (T–5 cm day−1) in the north-central United States, with no change in average midwinter daily snowfall (Fig. 8a). In these places, it is cold enough that some warming is not enough to greatly decrease the fraction of precipitation that falls as snow, and the increase in precipitation is just as important or the more dominant factor. In the warmer transition months of November–December and March–April, a decrease or little change in overall snowfall is projected, but the large event frequency still increases (e.g., Fig. 8b). This is comparable to what is projected to happen in regions slightly farther south in the midwinter (e.g., Figs. 7a,b) and is likely occurring for similar reasons. These projections also demonstrate that increases in midwinter snowfall in these regions are likely to be offset by decreases in snowfall during warmer months, resulting in a shortening of the overall snowfall season, as was found by Krasting et al. (2013). By 2071–2100, the regions projected to experience an increase in average daily snowfall in midwinter no longer include the north-central United States, southern Québec and New Brunswick, and northern Scandinavia, because the temperatures are finally warm enough to cause a substantial drop in the fraction of precipitation that falls as snow. But this does not stop the more intense events from increasing. For instance, in southern Québec and New Brunswick, average daily snowfall in January–February drops by 2071–2100, but the frequency of events ≥ 20 cm day−1 still increases (Fig. 9a). And, while average daily snowfall in November–December plummets by over 40% from what it was in 1971–2000, daily events ≥ 35 cm slightly increase (Fig. 9b).

Fig. 8.
Fig. 8.

As in Fig. 6, but for the north-central United States for (a) January–February and (b) March–April for 2021–50.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

Fig. 9.
Fig. 9.

As in Fig. 6, but for 2071–2100 for southern Québec and New Brunswick for (a) January–February and (b) November–December.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

For places as far north as northeastern Québec, northwestern Siberia, and northeastern Siberia, where the fraction of precipitation that falls as snow is even less sensitive to warming, the frequency of daily snowfall events is simulated to rise in the midwinter months as well as the transition months due to the dominating effect of higher atmospheric moisture and thus greater precipitation. However, even in climates this cold, the warm months of September–October and May–June receive less snowfall overall. Once again though, in some cases these months are still projected to have an increase in large events. As an example, daily snowfall events in northwestern Siberia are projected to increase in November–December by 2021–50 for all measurable bins, and average daily snowfall goes up (Fig. 10a), but in September–October the frequency of small events decreases greatly and events ≥ 10 cm day−1 increase (Fig. 10c). This is yet another example of a shortening of the snowfall season that is likely to occur in many regions, even where snowfall in colder months may increase. By 2071–2100, a similar pattern is projected, but increases in snowfall during November–April and decreases during May–October are even more dramatic (e.g., Figs. 10b,d), because as temperatures continue to rise, their effects on snowfall should continue to strengthen.

Fig. 10.
Fig. 10.

As in Fig. 6, but for northwestern Siberia for (a),(b) November–December and (c),(d) September–October for (left) 2021–50 and (right) 2071–2100.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

The two Greenland regions are the only ones among those examined where daily snowfall is simulated to increase by 2021–50 for all 2-month intervals and nearly all bins, with the exception of July–August. For example, the frequency of events increases for northeastern Greenland in September–October for all measurable bins, and the average daily snowfall increases as well (Fig. 11a). By 2071–2100, average daily snowfall is projected to increase even more in northeastern Greenland in September–October (Fig. 11b). In cold months the increase is even more dramatic; for instance, in northeastern Greenland average midwinter daily snowfall in the later time frame is about 30% higher than what it was in the earlier time frame (not shown). In southeastern Greenland, however, some subtle negative effects of increased temperatures on snowfall are evident where average daily snowfall is projected to slightly decline from 2021–50 to 2071–2100 in the warmer months of May–October (e.g., Figs. 11c,d). Overall, these model projections for Greenland are not surprising because it is a region where increases in winter precipitation are certainly expected (Collins et al. 2013), but it is sufficiently cold that increases in average temperature, even by several degrees, should not be enough to lessen the fraction of precipitation that falls as snow, except in the summer months and possibly May–June and September–October in the southern part of the island. Northeastern Greenland is also the only region examined where warming temperatures are not projected to decrease the length of the snowfall season by 2071–2100, as average daily snowfall increases in all 2-month intervals except July–August. But lighter snowfall events do decrease in southeastern Greenland and in the warmest months of the year in northeastern Greenland, which is more evidence of increasing temperatures causing fewer light events and more heavy events, as has been seen in other regions.

Fig. 11.
Fig. 11.

As in Fig. 6, but for (a),(b) northeastern Greenland for September–October and (c),(d) southeastern Greenland for May–June for (left) 2021–50 and (right) 2071–2100.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

4. Model temperature biases

Since temperature has a major effect on snowfall frequency and intensity, when examining daily snowfall projections from the CMIP5 models it may be useful to look at the model temperature biases in order to determine whether or not such biases influence the results and to see if they can partly explain intermodel spread in the projections. To accomplish this, CRU monthly temperature data from 1971–2000 (described in section 2) for each of the 23 regions were obtained and compared to the simulated temperatures for the same time period from the historical simulations of all 37 ensemble members used in this study. Regional-mean temperature biases were computed by subtracting the CRU mean temperature over this 30-yr period from the mean temperature for each ensemble member for the same region. This produced temperature biases for each ensemble member, 2-month interval, and region, which were used to divide the ensemble members into five distinct bias groups (see description in the Fig. 12 caption). For each intensity bin, the individual ensemble members within each temperature bias group were used to compute the mean difference in frequency of daily snowfall events for that group, which are plotted on their respective histograms as colored bars. Overall, more CMIP5 models exhibit a cold bias than a warm bias in the coldest regions examined in this study, including northeastern Québec, northern Siberia, northern Scandinavia, and Greenland, especially in the midwinter months. This finding is supported by regional studies which came to similar conclusions (Cattiaux et al. 2013; Sheffield et al. 2013).

Fig. 12.
Fig. 12.

As in Fig. 6, but modified to show relationships between changes in snowfall events and temperature biases for (a) southern Québec and New Brunswick, (b) the north-central United States, and (c) northern Scandinavia for January–February for the period of 2071–2100. Within each intensity bin, the individual ensemble members are divided into five groups based on their bias in simulated temperature over the period of 1971–2000 for the specified region. The five colored bars in a bin represent the mean difference in frequency of daily snowfall events computed over each of the temperature bias groups: Darker and lighter blue bars average over members with cold biases of >5°C and 2°–5°C, respectively; darker and lighter red bars average over members with warm biases of >5°C and 2°–5°C, respectively; and gray bars average over members with biases smaller than 2°C. The numbers next to the colored boxes at the top of each graph display the total count of members belonging to the corresponding temperature bias group. For reference, the wider white bars represent the mean difference in frequency among the ensemble means of all the models for that particular intensity bin (like the cyan bars in Figs. 611), and each white circle represents the median difference in frequency among the ensemble means.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

In many Northern Hemisphere regions examined, temperature biases appear to influence model snowfall projections in January–February, with warm ensemble members usually residing toward the negative end of the distribution of projected frequency changes and the cold members toward the positive end. This pattern is especially strong in regions near the boundary of increasing and decreasing average daily snowfall during these months, including the northern United States, central and southern Canada, central Asia, and northern Europe (see section 3b), which is likely because these are places where many precipitation events occur with near-freezing temperatures, and thus even small differences in temperature can influence precipitation type. For example, in Fig. 12a, the mean changes in midwinter daily snowfall event frequency for members in the southern Québec and New Brunswick region with cold temperature biases of 2°C or greater (blue bars) are more positive or less negative than warmer members (the opposite occurs for days with no snow, with cold members showing a decrease and warm members showing an increase). Figure 12b displays a similar scenario for the north-central United States, with the means over warm-biased ensemble members (red bars) falling below the overall mean snowfall change (white bars), and vice versa for the cold-biased members, for snowfall events up to 30 cm day−1. In northern Scandinavia, the members suffering from a severe cold bias project a large decrease in days with no measurable snowfall and an increase in days with T–5 cm and 5–10 cm, and vice versa for members with little bias or a warm bias. In the latter two examples, the response of heavy snowfall events (an increase) appears less sensitive to the regional-mean climatological temperatures of the individual ensemble members (Figs. 12b,c). This pattern of strong dependence on model temperature bias in lighter bins but less dependence in heavier bins occurs relatively frequently in many regions and times of the year (not shown).

This apparent influence of temperature bias also exists in other months besides midwinter in some regions. For example, in northeastern Siberia in March–April, the mean of the ensemble members with a severe warm bias projects an increase in days with zero snowfall and a discernable decrease in days with snowfall up to 20 cm day−1, a pattern that is opposite to that of the colder members (Fig. 13a). In northeastern Québec in November–December, the members with a cold bias project a large decrease in days with zero snowfall and an increase in snowfall events up to 20 cm day−1; the opposite is true for warm-biased members (Fig. 13b). From the results in the previous section, these regions in northern Canada and northern Siberia are all located near the boundary between increasing and decreasing average daily snowfall in the months of November–December and March–April, once again demonstrating that this dependence of snowfall projection on model temperature bias is strongest near this transition zone. In southeastern Greenland, near the boundary between increasing and decreasing average daily snowfall in even warmer months of the year, the ensemble members with a severe cold bias project increased snowfall events of nearly all intensities in May–June, while members with little bias project fewer events (Fig. 13c). In all of these examples, the temperature biases are primarily affecting the results by influencing the fraction of precipitation falling as snow, with cold biases making the regions less sensitive to temperature changes and warm biases having the opposite effect. From Fig. 12 it is also evident that the mean frequency change of the warm-biased ensemble members follows an “up–down–up” pattern in many cases, even falling near the middle or high end of the model range for large events, which is consistent with findings in the previous section that a warmer climate often results in fewer light snowfall events but as many or more heavy events in somewhat milder regions and/or months.

Fig. 13.
Fig. 13.

As in Fig. 12, but for (a) northeastern Siberia for March–April, (b) northeastern Québec for November–December, and (c) southeastern Greenland for May–June for 2071–2100.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

While certain regions/months near the transition zone of increasing and decreasing snowfall can show a clear influence of the temperature biases on the results, the influence of the biases in other regions/months are much less conclusive. For example, in southern Québec and New Brunswick, while there is a clear relationship between temperature bias and snowfall projection in January–February (see Fig. 12a), in the warmer spring months of March–April there is no clear relationship (Fig. 14a). For events of most intensities, the members with a cold bias and members with a warm bias exhibit mean frequency change values very close to one another, and the mean for warm members actually lies slightly above the overall mean and median (Fig. 14a). Even in January–February, one can see examples of regions without a clear pattern, such as the U.S. mid-Atlantic (Fig. 14b) and northwestern Siberia (Fig. 14c). The lack of influence of temperature biases in such regions/months likely occurs because the climate is sufficiently warm that most simulations project a decrease in snowfall (e.g., Figs. 14a,b), or sufficiently cold that increases in snowfall are projected (e.g., Fig. 14c), regardless of their climatological temperature biases.

Fig. 14.
Fig. 14.

As in Fig. 12, but for (a) southern Québec and New Brunswick for March–April, (b) the U.S. Mid-Atlantic for January–February, and (c) northwestern Siberia for January–February for 2071–2100.

Citation: Journal of Climate 29, 17; 10.1175/JCLI-D-15-0687.1

In most of the histograms presented in this study, there is considerable model spread in the projected frequency changes of daily snowfall events. The findings of this section suggest that some of the spread, particularly for regions close to the boundary between increasing and decreasing snowfall, is related to differences in the simulated climatological temperature. Modeling efforts aimed at reducing biases in climatological temperature are thus critical for generating more accurate and robust projections of the daily snowfall distribution. That many high-latitude regions exhibit a cold bias, and that many cold-biased simulations project a larger snowfall increase, implies that models may be overestimating projected snowfall increases in these regions. A large amount of model spread, however, cannot be explained by climatological temperatures. The remaining spread is likely due to numerous factors, including differences in the amount of projected warming, projected precipitation changes, precipitation biases in the current climate, and differences in the algorithms that determine precipitation type (Krasting et al. 2013). Additional research is necessary to disentangle the contributions of these factors to the model spread and reduce their impact in future generations of climate models.

5. Discussion and conclusions

The primary focus of this study was to examine projected changes in the frequency of daily snowfall events of varying intensities over the course of the twenty-first century in the CMIP5 climate model simulations under the RCP8.5 emissions scenario. Based on the results from these simulations, a large decrease in daily snowfall events is projected in all times of the year for all intensities in the warmest regions, generally the lower latitudes or areas where there is a strong marine influence such as the southern United States, central and southern Europe, and the west coast of the United States. In these places, temperatures are marginal for snowfall in the present climate, so future warming due to climate change will cause precipitation that would have fallen as snow to instead fall as rain. In contrast, in regions that are sufficiently cold for future temperature increases to have little effect on the fraction of precipitation falling as snow, which mainly include northeastern Québec, northern Siberia, and Greenland, the frequency of daily snowfall events is projected to increase in a large portion of the year. This is likely because higher temperatures should allow for more precipitation over high latitudes and moist midlatitude regions, due to increased atmospheric water vapor content and increased moisture transport (Allen and Ingram 2002; Sun et al. 2007; Trenberth 2011; Collins et al. 2013). Krasting et al. (2013) found similar decreases in mean annual snowfall in the midlatitude regions of the Northern Hemisphere, but increases over Greenland and northern Siberia, over the course of the twenty-first century in CMIP5 multimodel ensemble projections. The results here are also consistent with a study by Kapnick and Delworth (2013), which found that using a high-emissions scenario similar to RCP8.5, the GFDL CM2.5 and CM2.1 models simulate decreased mean annual snowfall everywhere in the Northern Hemisphere by the late twenty-first century, except for extreme northern Canada, Greenland, and northern Siberia. Furthermore, their study used a simple regression model to demonstrate that these changes are largely due to the competing effects of temperature and precipitation, as the effects of temperature changes are shown to decrease snowfall almost universally throughout the Northern Hemisphere by reducing the percent of total precipitation that falls as snow, while precipitation contributes to increased snowfall in the high latitudes and high-elevation regions (Kapnick and Delworth 2013).

Time of year is also shown to have a considerable effect on the results for each region. Average daily snowfall is modeled to increase in the midwinter months of January–February in a relatively large portion of the Northern Hemisphere, including much of Canada, Tibet, northern Scandinavia, northern Siberia, and Greenland, matching the regions where Krasting et al. (2013) found that mean midwinter snowfall is simulated to increase. However, when looking at the warmer transition months of November–December and March–April, the region of increased snowfall shrinks to include only northeastern Québec, northern Siberia, and Greenland. In the even warmer months of September–October and May–June, nearly all Northern Hemisphere regions are projected to have a decrease in daily snowfall, with the only exception being Greenland, and not a single region experiences increased daily snowfall in July–August. Temperatures in the warmer months are more likely to be marginal for snowfall, even at very high latitudes, which means the fraction of precipitation falling as snow is more prone to decrease as temperatures rise from climate change. Krasting et al. (2013) found similar decreases in snowfall during the autumn and spring seasons throughout much of the Northern Hemisphere except for the very high latitudes, resulting in a shortening of the snow season and offsetting the increases in midwinter snowfall in some regions.

Another common pattern of response that was found for many mid- to high-latitude regions at certain times of year is the frequency of large snowfall events staying the same or increasing in the future, even while mean daily snowfall decreases. The probable reason for more intense events is that a warmer climate will cause higher sea surface temperatures in moisture source regions, and more abundant atmospheric moisture (Collins et al. 2013). Thus even though temperatures may not be cold enough for snow as frequently as in the past, during the instances when it still is cold enough, the storms will produce more precipitation, maintaining or even increasing the frequency of heavy snowfall events. Other studies that have examined projections of intense snowfall events have discovered similar results. Notaro et al. (2014) found that heavy daily snowfall events are projected to increase in frequency across the U.S. Great Plains, upper Midwest, and Great Lakes regions, even while the total number of daily snowfall events decreases. Furthermore, while O’Gorman (2014) discovered that CMIP5 models simulate widespread decreases in mean snowfall in the midlatitudes as a result of a warming climate, the changes in snowfall extremes are much weaker and even exhibit a slight increase in parts of central and southern Asia and southern Canada. O’Gorman (2014) also hypothesized that heavy snowfall events respond differently to climate change than mean snowfall because they tend to occur in a narrow range of temperatures near an optimal value that is insensitive to climate warming (roughly −2°C in the simulations and observations). Thus he found that in a warming climate, extreme snowfall events can continue to increase at higher climatological temperatures (as high as −9°C) than can mean snowfall (−14°C).

The difference between projected daily snowfall by 2021–50 versus 2071–2100 follows a clear pattern as well. For the northernmost regions, including northeastern Québec, northern Siberia, and Greenland, where daily snowfall frequency is modeled to increase in the earlier time period, this trend is projected to continue further in the later time period, with an even greater increase projected. Similarly, in places where daily snowfall decreases substantially in the earlier time frame, such as the U.S. West Coast and central and southern Europe, it decreases further in the years beyond that. Furthermore, model agreement on the sign of snowfall event frequency change increases substantially in many regions from 2021–50 to 2071–2100. This reinforcement of the early period tendencies in the later period suggests that the continuing increase in temperature during the twenty-first century is the primary driver of the changes in snowfall intensity and frequency. It is important to note that despite this tendency for stronger responses later in the century, the geographic boundary between increased and decreased snowfall events does not appear to shift very much between the two time periods. This is likely because close to the boundary, the two competing effects of warming temperatures and increasing precipitation continue to get stronger, but because they each have equal and opposite effects on snowfall, they still cancel each other out.

Another important conclusion is that model temperature biases have some influence on the daily snowfall projections from the individual models analyzed in this study, with the ensemble members suffering from a warm bias generally falling on the lower end of the projected frequency changes and the cold ensemble members on the higher end. This influence is especially strong in regions near the transition zone of projected increasing and decreasing snowfall, because this is likely where temperatures are near freezing when precipitation is occurring and thus where projected snowfall can be most affected by slight temperature differences between the models. Warmer and colder regions/months may be less sensitive to temperature biases because the temperatures are so far above or below freezing, respectively, that a difference of even a few degrees does not have a significant effect on snowfall. Projections of light snowfall events also appear to be much more affected by model temperature bias than projections of heavy snowfall events, which is likely because the frequency of heavy snowfall events is less influenced by temperature and more affected by other factors such as precipitation. To reduce these model biases of temperature, as well as precipitation and snowfall, the application of a statistical downscaling method to the model projections may be useful, as was found by Notaro et al. (2014).

This study has made a number of assumptions that could impact its results, and further research should be done to address the magnitude of such impacts, their influence on the results, and how they can be reduced. A major assumption made in this study is the uniform 10:1 snow-to-liquid ratio when converting snowfall mass flux to snowfall depth. As noted by Krasting et al. (2013), research has shown that this ratio can differ widely, and climate models have a very difficult time simulating the causes of these variations, such as temperature and vertical velocity in the snow growth regions. This could have an impact on the seasonal comparisons done in this study, as this may cause snowfall projections to be biased positive in warmer months and regions where the ratio is often lower than 10:1 and biased negative in colder months and regions where the ratio is usually higher. Furthermore, it is assumed here that the 10:1 snow-to-liquid ratio will not change in the future, which may not be the case. More research is needed for models to improve their representation of snow formation processes that affect the snow-to-liquid ratio.

The CMIP5 models suffer from a number of other limitations as well. They have relatively coarse horizontal resolution, causing many topographic features to be inadequately represented, and they do not represent the mesoscale processes that can have a large influence on snowfall in certain regions, such as lake-effect snow belts (Notaro et al. 2015). Increasing model resolution may make their snowfall projections more accurate in such regions. In addition, coarse-resolution models may be underestimating the frequency of snowfall extremes that occur on relatively small spatial scales. Most of the models also diagnose precipitation type by simple algorithms which are based on lower tropospheric temperature, and future research should analyze whether snowfall simulations would be improved by using explicit mixed-phase microphysical schemes that keep independent water and ice budgets (Krasting et al. 2013). The effect that snow has on the temperature profile of the atmosphere is significant as well, and this feedback may not be addressed adequately by the models (Krasting et al. 2013). Additionally, many of the models suffer from substantial temperature biases (Cattiaux et al. 2013; Sheffield et al. 2013; Su et al. 2013) and precipitation biases (Sheffield et al. 2013; Mehran et al. 2014) that can influence the projected changes in snowfall. Efforts to reduce such biases would likely reduce the intermodel spread in the snowfall difference histograms, resulting in more robust projections. Any precipitation biases that may exist in the models were not addressed in this study, but further research should attempt to find the extent and influence of such biases on simulated future changes in snowfall.

Acknowledgments

This work was funded in part by grants from the NOAA Climate Program Office (NA09OAR4310109), the Office of Science (BER), U.S. Department of Energy (DE-SC0005467), and the National Science Foundation (ATM-0902735). The efforts of A.M.D. were also supported by a fellowship from the Environmental Protection Agency Science to Achieve Results (EPA STAR) program and by the Regional and Global Climate Modeling Program of the Office of Science of the U.S. Department of Energy. We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP5, and we thank the climate modelling groups (listed in Table 1 of this paper) for producing and making available their model output. For CMIP5, the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. CRU temperature observations were provided by the Climatic Research Unit, University of East Anglia, from their Web site at https://crudata.uea.ac.uk/cru/data/hrg/. The writers also acknowledge the helpful comments and suggestions by three anonymous reviewers.

REFERENCES

  • Allen, M. R., , and W. J. Ingram, 2002: Constraints on future changes in climate and the hydrologic cycle. Nature, 419, 224232, doi:10.1038/nature01092.

    • Search Google Scholar
    • Export Citation
  • Barnett, T. P., , J. C. Adam, , and D. P. Lettenmaier, 2005: Potential impacts of a warming climate on water availability in snow-dominated regions. Nature, 438, 303309, doi:10.1038/nature04141.

    • Search Google Scholar
    • Export Citation
  • Cattiaux, J., , H. Douville, , and Y. Peings, 2013: European temperatures in CMIP5: Origins of present-day biases and future uncertainties. Climate Dyn., 41, 28892907, doi:10.1007/s00382-013-1731-y.

    • Search Google Scholar
    • Export Citation
  • Changnon, S. A., , and D. Changnon, 2006: A spatial and temporal analysis of damaging snowstorms in the United States. Nat. Hazards, 37, 373389, doi:10.1007/s11069-005-6581-4.

    • Search Google Scholar
    • Export Citation
  • Collins, M., and Coauthors, 2013: Long-term climate change: Projections, commitments and irreversibility. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 1029–1136.

  • Eisenberg, D., , and K. E. Warner, 2005: Effects of snowfalls on motor vehicle collisions, injuries, and fatalities. Amer. J. Public Health, 95, 120124, doi:10.2105/AJPH.2004.048926.

    • Search Google Scholar
    • Export Citation
  • Franklin, B. A., , K. Bonzheim, , S. Gordon, , and G. Timmis, 1996: Snow shoveling: A trigger for acute myocardial infarction and sudden coronary death. Amer. J. Cardiol., 77, 855858, doi:10.1016/S0002-9149(97)89181-3.

    • Search Google Scholar
    • Export Citation
  • Harris, I., , P. D. Jones, , T. J. Osborn, , and D. H. Lister, 2014: Updated high-resolution grids of monthly climatic observations—The CRU TS3.10 Dataset. Int. J. Climatol., 34, 623642, doi:10.1002/joc.3711.

    • Search Google Scholar
    • Export Citation
  • Hartmann, D. L., and Coauthors, 2013: Observations: Atmosphere and surface. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 159–254.

  • IPCC, 2013: Summary for policymakers. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 3–29.

  • Kapnick, S. B., , and T. L. Delworth, 2013: Controls of global snow under a changed climate. J. Climate, 26, 55375562, doi:10.1175/JCLI-D-12-00528.1.

    • Search Google Scholar
    • Export Citation
  • Krasting, J. P., , A. J. Broccoli, , K. W. Dixon, , and J. R. Lanzante, 2013: Future changes in Northern Hemisphere snowfall. J. Climate, 26, 78137828, doi:10.1175/JCLI-D-12-00832.1.

    • Search Google Scholar
    • Export Citation
  • Mehran, A., , A. AghaKouchak, , and T. J. Phillips, 2014: Evaluation of CMIP5 continental precipitation simulations relative to satellite-based gauge-adjusted observations. J. Geophys. Res. Atmos., 119, 16951707, doi:10.1002/2013JD021152.

    • Search Google Scholar
    • Export Citation
  • Notaro, M., , D. Lorenz, , C. Hoving, , and M. Schummer, 2014: Twenty-first-century projections of snowfall and winter severity across central-eastern North America. J. Climate, 27, 65266550, doi:10.1175/JCLI-D-13-00520.1.

    • Search Google Scholar
    • Export Citation
  • Notaro, M., , V. Bennington, , and S. Vavrus, 2015: Dynamically downscaled projections of lake-effect snow in the Great Lakes basin. J. Climate, 28, 16611684, doi:10.1175/JCLI-D-14-00467.1.

    • Search Google Scholar
    • Export Citation
  • O’Gorman, P., 2014: Contrasting responses of mean and extreme snowfall to climate change. Nature, 512, 416418, doi:10.1038/nature13625.

    • Search Google Scholar
    • Export Citation
  • Riahi, K., and Coauthors, 2011: RCP 8.5—A scenario of comparatively high greenhouse gas emissions. Climatic Change, 109, 3357, doi:10.1007/s10584-011-0149-y.

    • Search Google Scholar
    • Export Citation
  • Sheffield, J., and Coauthors, 2013: North American climate in CMIP5 experiments. Part I: Evaluation of historical simulations of continental and regional climatology. J. Climate, 26, 92099245, doi:10.1175/JCLI-D-12-00592.1.

    • Search Google Scholar
    • Export Citation
  • Su, F., , X. Duan, , D. Chen, , Z. Hao, , and L. Cuo, 2013: Evaluation of the global climate models in the CMIP5 over the Tibetan Plateau. J. Climate, 26, 31873208, doi:10.1175/JCLI-D-12-00321.1.

    • Search Google Scholar
    • Export Citation
  • Sun, Y., , S. Solomon, , A. Dai, , and R. W. Portmann, 2007: How often will it rain? J. Climate, 20, 48014818, doi:10.1175/JCLI4263.1.

  • Taylor, K. E., , R. J. Stouffer, , and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485498, doi:10.1175/BAMS-D-11-00094.1.

    • Search Google Scholar
    • Export Citation
  • Trenberth, K. E., 2011: Changes in precipitation with climate change. Climate Res., 47, 123138, doi:10.3354/cr00953.

  • Vavrus, S., 2007: The role of terrestrial snow cover in the climate system. Climate Dyn., 29, 7388, doi:10.1007/s00382-007-0226-0.

Save