Snow in the Great Lakes: Past and Future


Executive Summary

Snow (including snow storms, snow cover, snow depth, and snow density) is a complicated variable in the Great Lakes region due to the influence of the lakes on local climate.  Great Lakes snow can be partitioned into two main categories - lake-effect snow and non-lake-effect snow.  The regions where lake-effect snow occurs are well defined, however, some are expanding in a warming climate.  Lake-effect regions receive greater amounts of snowfall because the lakes enhance precipitation in those areas.  Lake-effect snowfall only occurs when the lakes are not completely frozen over.  

There is a lot of evidence that snow is changing in the Great Lakes region, but the changes are not uniform. While snowstorms that impact the entire region are decreasing, lake-effect snowfall is increasing around Lakes Superior and Michigan.  Snow depths going into spring are decreasing as warming occurs, and earlier spring snowmelt is occurring. We have a situation where there is more snow during storms, but the faster melting means that snow cover is less in late winter and early spring. 

Projections of future climate in the Great lakes, and especially future snow, have a lot of uncertainty.  Global climate models are not a reliable source of information for lake-effect snowfall, and regional climate and weather models play a role in filling that gap.  High resolution models can be used to study the interactions between changing air and water temperatures, and their relation to lake ice cover, which are the primary variables for understanding how snow may change in the future.      

Lake ice cover plays a major role in the development and distribution of lake-effect snowfall.  When the lakes are completely frozen over there is essentially no lake-effect snowfall because the moisture supply for snow. evaporation from the lake, has been cut off.  Lake ice cover has been decreasing in recent years, which has been accompanied by increases in lake water temperature (due to increasing air temperatures).  Models run under these conditions predict increasing lake-effect snowfall and an expansion of the lake-effect zone, which is consistent with what has already been observed.

Jump to Section:

What influences snow formation?

Historical Climatology of Snow in the Great Lakes Region

The Future of Snow

What Influences Snow Formation?

Figure 1 Major winter weather storm tracks (Source: Illinois State Water Survey)

Snowfall in the Great Lakes region is primarily from synoptic (i.e., large-scale) weather systems moving across the region (Figure 1), lake-effects, or a combination of both.  When large air masses pass over the Great Lakes region they are modified by the presence of the lakes.1,2  Cold air from the north, such as storm Type 1 in Figure 1, enter the Great Lakes region as a low-pressure system and pick up moisture as they pass over the relatively warm surface waters of the lakes.  The additional moisture contributes to increased precipitation (snowfall when temperatures are cold enough) downwind (generally on the eastern side of the lakes).  Some snow events are initiated by synoptic systems and transition to lake-effect snow.

Lake-effect zones (Figure 2) can receive contributions of snowfall from large-scale snowstorms and lake-effects.  Large storms passing over the Great Lakes pick up moisture from the lakes and cause snowfall downwind, but as the air mass moves farther inland its moisture loss is not replenished and snowfall can diminish.  Unlike large-scale snowstorms that impact great portions of the region, lake-effect snowfall is localized.  For example, strong temperature differences between the lake and land create land-breezes that enhance local snow production.3  

Figure 2: Lake-effect snow regions in the Great Lakes (image source)

Lake-effect zones are located directly downwind of the lakes and in a fairly narrow band following the lakeshore.  Each lake has its own lake-effect zone and they do not receive equal amounts of snowfall.  Several factors are responsible for determining the amount of lake-effect snow at a given location.  The surface area, depth, and orientation of the lake plays a large role in determining the severity of lake-effect snowfall.4  The surface area of the lake combined with its orientation determines the extent and duration that the lake modifies a storm passing over it.  Lakes that are oriented parallel to the direction of the winds, primarily west-to-east, have the greatest spatial influence on passing air masses because the winds are over them for longer periods of time.  Lake surface area and depth are measures for the size of a lake, and larger lakes (i.e., Lake Superior) generally have a greater influence on local and regional weather.  Lake Superior has the greatest impact on local snowfall amounts with 100% more winter precipitation falling downwind compared to Lakes Erie and Ontario that only have precipitation increases of 15% from the lake-effects.4  The primary reason larger lakes have greater influence is because they hold their warmth from summertime longer into winter delaying ice formation, which maintains a moisture flux to the atmosphere that can support snowfall.

Historical Climatology of Snow in the Great Lakes Region

Wintertime Averages

Figure 3: "Average precipitation (mm) over the Great Lakes basin in winter: a) using all data and b) showing lake-induced changes. Heavy line on b) represents the 80-km lake-effect boundary. Dots indicate locations of precipitation sites. Values in parentheses are for Canada" (Source: Scott and Huff 1996)

The geographical distribution of winter precipitation in the Great Lakes is complex (Figure 3).  Depending on the location, vast differences can occur over relatively short distances.  Take for example the lake-effect zone of Lake Erie.  At its maximum, average winter precipitation is about 300 millimeters in western New York but decreases by almost 30% just over the New York/Pennsylvania border, and the contribution from lake-effects decreases by about 50%.  This is in contrast to areas upwind of the Great Lakes such as Wisconsin and Illinois and to a lesser extent the eastern half of Lower Michigan where snowfall is fairly uniform.    

Knowing where the lakes have the greatest influence is important for understanding how snowfall may change in the future.  Lake Superior's lake-effect is the greatest of all of the Great Lakes in both regional extent and magnitude.  Lake Michigan and to a lesser degree, Lake Huron, have fairly widespread regions of  influence but they are not as strong as Lake Superior.  For example, Lake Michigan's lake-effect zone covers roughly the western third of Lower Michigan but only causes increases in precipitation by about 40%.4

The timing of when snowstorms occur is important for both preparedness and planning.  Lake-effect zones and regions in the northern most part of the Great Lakes have snowstorms occurring as early as October, and the rest of the region is delayed until November.5 The remaining areas experience the most storms in January.  As winter progresses lake ice cover accumulates and acts as a barrier turning off the moisture supply mechanism for lake-effect snowfall.  Most of the region continues to experience snowstorms, primarily from large-scale atmospheric circulation, through April and as late as May in the far north.

Wet vs. Dry Snow

Terms like "wet" and "dry" are important descriptors of snow because five inches of dry snow has very different impacts on say, infrastructure, compared to five inches of wet snow.  Wet and dry snows are defined by the density of the snow, which is determined by the amount of equivalent liquid water it holds called the snow-to-liquid equivalent ratio (SLR).  A climatology of SLR values for the entire United States was put out by Baxter, Graves, and Moore in 20046for the purpose of improving the ratios that are used in weather forecasting.  They produced several maps that are useful for identifying snow characteristics over the Great Lakes region.

SLR values are used to calculate the depth of snowfall based on an individual forecast for liquid precipitation, so on their own they only tell the density of snow.  Table 1 relates SLR values with the relative density of the snow.7  The SLR value for a given location varies every snowstorm is different, but the maps in Figure 4 capture the general picture of how the statistical distribution of snow can be characterized.

Table 1: SLR values assigned to each characteristic of snowfall density as defined by Roebber et al. 2003.  Wet snow is "heavy" and dry snow is "light"
Snow Density Characteristic Range of SLR Values
Heavy 1:1 < ratio <  9:1
Average 9:1 ≤ ratio ≤ 15:1
Light ratio > 15:1

The Great Lakes region has some of the greatest variability of snow density compared to the United States as a whole.  SLR values in the Great Lakes range from about 8:1, which is heavy snow to 20:1, which is very light snow.  A large snowstorm passing over the region that produces roughly equal amounts of liquid precipitation will produce greater accumulations of snow in areas with high SLR values compared to low SLR values (ignoring factors such as drifting snow).  Lake effect snow is almost always less dense than synoptic-origin snowfall.  Snow in the southern Great Lakes region is typically heavier than in the north because colder temperatures in the north prevent the air from holding as much moisture, hence the snow is less dense.  Snow near the lakes is typically less dense, but there are greater amounts of it.  Only one-quarter of SLR values indicate high-density snows over the Great Lakes region, so snow is most often average to light density.

Figure 4: The 25th (left), 50th (center), and 75th (right) percentile SLR values during 1971-2000 (Source: Baxter, Graves, and Moore 2004)
Figure 5: The average SLR values in October and November (left), December, January, and February (center), and March and April (right)(Source: Baxter, Graves, and Moore 2004)

There is considerable variability in the seasonality of snow density particularly downwind of the Great Lakes (Figure 5).  When the lakes are not frozen over and their waters are warmer than the air above them convective updrafts can occur.  The additional moisture from the lakes in the updraft enhances snow crystal formation leading to more dense snowfalls.8 This is observed during early winter (October and November) near Lakes Erie and Ontario and is shown by lower SLR values near the lakes (12:1) than farther inland (13:1).  The ability of the lakes to increase snow densities is dampened once the waters are frozen over, which typically occurs throughout December, January, and February.9  However, not all lake-effect processes are terminated once the lakes freeze9 so the lake-effect zones can still receive considerable snowfall.  

Information about average winter weather combined with snow density information is useful for describing when there is greatest potential for damaging conditions.  Most snowstorms occur in December and January, which coincides when snow densities are lightest, so the greatest snow depths can be expected during these months.  Snow densities are typically lightest near the lakes but snow depths are greatest in the lake-effect zones.  Snowstorms in the southern portion of the Great Lakes during early winter pose a greater threat for dense snow, although typically snow depths are less.

Intense Snowstorms

Figure 6: The amount of snow (cm) expected in a snowstorm at least once every 5 years (Source: Changnon 2006)
Figure 7: The amount of snow (cm) expected in a snowstorm at least once in every 10 years (Source: Changnon 2006)

The lake-effect zones, which have been identified as regions receiving the most annual average snowfall, are also where the heaviest snowstorms are experienced.  A heavy snowstorm, as defined by (Changnon 2006), is an event when 15.2cm or more occurred in 1 or 2 days.  "Heavy" is not necessarily referring to a wet snow, rather, in this context it means a lot of snow.  Weather station data were used to identify where heavy snowstorms occurred in the United States between 1948-2001.10 The heaviest amount of snow that is expected to occur at least one time in a 5- and 10-year period are mapped (Figure 6 and Figure 7).

The structure of the lake-effect zones is not as apparent in the maps of heavy snowstorms, because there are less data points (compare the dots representing weather stations in Figures 6 and 7 to those in Figure 3a).  However, regions around the lakes (upwind and downwind) receive the heaviest snowstorms in the region.  Upwind locations receive slightly less intense snowstorms than downwind, but the lakes' influence is seen on all sides.

Localized differences emerge between the 5-year and 10-year snowstorm maps in the Great Lakes region.  For example, the range of variability across Lower Michigan in the 5-year snowstorm is greater than what is observed for the 10-year snowstorm.  This would suggest that the most extreme events (around 35 cm) in Lower Michigan occur less frequently (every 10-years instead of 5) but they are widespread.  There is little change in the spatial structure of snowstorms in the far north and eastern portions of the Great Lakes, and in general the 10-year storms have about 5 to 10 cm more snowfall than the 5-year events at a given location.

The Future of Snow

Is there a climate change signal in the historical observations?  

Figure 8:  Mean annual snowfall during two 30-year periods: 1961-1990 (left) and 1981-2010 (right) (Source: Historical Climate and Climate Trends in the Midwestern USA11)
Snowfall is both an indicator of climate change and it's surface properties (i.e., reflectivity) also impact local climate.  Snowfall observations are very important for describing how snowfall frequency and amounts are changing in the Great Lakes, but observations are not always reliable.  Several inconsistencies and potential biases were found in the U.S. Cooperative Observer Program (COOP) snowfall record.12 Kunkel et al. (2006) studied several weather stations, locations where observations are taken, and found that some stations very close to one another showed very different snowfall amounts.  Although it is possible for some differences to be attributed to natural climate variations, they found most are likely due to station inhomogeneities (i.e., moving a station to a new location that has different sunlight exposure).  To overcome data quality issues Kunkel et al. (2006) recommend careful assessment of station histories with surrounding stations to remove any stations that are unsuitable.  Kunkel was a co-author of the Historical Climate Sector Midwest Technical Input Report for the National Climate Assessment, which we use below to show how snow is changing in the Great Lakes region.        

Snowfall observations for the Great Lakes show a signal of climate change.  There is a general northward shift in the bands of snowfall amounts, and more snowfall has been observed in the north (Figure 8).  Upper Michigan had widespread increases in precipitation along with the northern tip of Lower Michigan.  Snow in Lake Superior's and Lake Michigan's lake-effect zones had upward trends.13 Lake Michigan's lake-effect zone pushes farther inland and the eastern part of Lower Michigan experienced less snowfall during the later period so the snowfall gradient between western and eastern Lower Michigan is increasing.  No trends were found in this study for the remaining lakes.14         

Additional studies report similar findings for changes in lake-effect snowfall, however, differences emerge in the statistical significance of changes as each study used a different set of quality control measures for their snowfall data.  Burnett et al. (2003)15show that there is a significant increasing trend in lake-effect snowfall and no change in non-lake-effect snowfall between 1931 and 2001.  Kunkel et al. (2009) imposed stricter data quality standards than the Burnett et al. study and showed significant increases to lake-effect snowfall only for Lakes Superior and Michigan between 1925 and 2007.  The remaining lakes had mixed results depending on the time frame selected for the analysis because there were only a few stations to choose from during the earlier part of the period. 

Another signal of change is the earlier timing of spring snow melt.16 The largest changes in the timing of snowmelt were found to start in late January and continue through April.  Shallower snow cover was observed during May and September, which contributes to the faster snowmelt.17

If there is a signal in the observations, is it consistent with basic theory?

Snowfall decreases in the south and increases in the lake-effect zones are consistent with observed decreases of synoptic snowsfalls and increases in lake-effect snowfalls.18 Increased lake-effect snowfall is commonly attributed to warmer air temperatures that cause warmer lake surface waters and less ice cover18,19  One study also suggested an increase to the snow liquid ratio in lake-effect zones over time which could explain increased snowfall there.19

Earlier spring snowmelt, a result of shallower snow depths, is related to a decrease in extratropical cyclones (synoptic-scale snowstorms) and warmer springtime temepratures.20    

Is there a climate change signal in the climate models?

There is better agreement amongst the models for projecions of wintertime precipitation compared to summertime precipitation.  The Midwest technical input report to the National Climate Assessment claims more precipitation can be expected during winter, but more rain and freezing-rain is likely instead of snow.21  The anticipation of more rain and freezing-rain is consistent with warming winter air temperatures. 

Individual model studies show how snow in the Great Lakes is influenced by lake ice cover.  Complete ice coverage results in major reductions of lake-effect snowfall along with colder temperatures and an overall more stable atmopshere over the lakes.22 Unfrozen lakes determine the spatial distribution of lake-effect snowfall compared to completely frozen lakes that have no impact on snow placement.23 When unfrozen surface waters are warmed, lake-effect snowfall amounts are increased downwind and the lake-effect region expands farther inland.24  The model results describe the exact changes that have already been observed for lake-effect snowfall near Lake Michigan and Superior.  Of the two modeled scenarios (complete ice coverage and ice-free lakes), observations point more toward decreasing ice coverage.[fn]Andresen, Jeff, Steve Hilberg, and Ken Kunkel. "National Climate Assessment Midwest Historical Climate." (2013).,25 Therefore, we can expect more precipitation in the lake-effect regions and beyond if ice cover continues to decline.  However, the form that precipitation takes will depend on the extent of warming.  Slight warming will increase snowfall amounts, but if the atmosphere is warmed beyond the freezing threshold more precipitation will fall as rain or freezing-rain.  

Are the projections reliable?

Climate model projections (GCM and downscaled) of precipitation, especially in the Great Lakes region, contain a lot of uncertainty.  Most of the uncertainty is related to how well weather is represented in the models for the Great Lakes region, but even the best GCMs have spatial resolutions too coarse to simulate lake-effects and other small-scale dynamics.  Many models do not even include the lakes, which is important for the interpretation of the projections and the description of their uncertainty.  In downscaled data the representation of the lakes is not necessarily improved so it is important to know how each model treats the lakes.  In addition, GCMs use simple snow models that do not capture the complexity of important snow processes that are highly sensitive in GCMs.26

Even though there are several reasons to be skeptical of future snow projections, the models can provide useful information.  For example, although the GCMs do not represent the Great Lakes well they are in fairly good agreement that air temperatures during all seasons will warm.21  There is a positive feedback between air temperatures, lake water temperatures, and ultimately the amount of winter lake ice cover.27 As the air temperature warms the lake waters are also warmed, and since lakes have a relatively high heat capacity (they are slower than the land surface to change temperature) their waters stay warmer later into fall.  The persistence of relatively warm waters into early winter delay the formation of ice, and the period that ice covers the lakes is shortened as springtime temperatures warm.  So, with atmospheric warming we see both a warming of lake waters and a decrease in lake ice cover.  This scenario was used to drive the high resolution weather forecasting model (Wright et a. 2013), which simulates lake-effect precipitation, and the result was an increase in the amount and spatial coverage of lake-effect snowfall.  So, even though the GCM projections are not the most reliable source of information about snowfall in the Great Lakes region, their information about regional changes can be used to inform localized simulations. 


There is a consistent story of how snow is changing in the Great Lakes, and the changes are supported by observations, theory, and modeling.  We place the most confidence in our observations of what has happened because that information is more certain than predictions about future snow.  Observations of snow depth can have their own sources of uncertainty, but we aimed to rely on information in the literature that used quality controlled snow data.

Observations from two 30-year periods (with 10 years of overlap) show how lake-effect precipitation, particularly that associated with Lakes Michigan and Superior, is increasing in both magnitude and spatial coverage.  Areas outside of the lake-effect zones have shown a decrease in snowfall, which is primarily attributed to a decrease of large-scale snowstorms.  The changes that have been observed are consistent with the modeling studies that we presented, so under the scenario of increased warming and decreased lake ice cover we can have greater confidence that those changes will continue to be magnified.  

We did not promote a set of snow projections (GCM or downscaled) to use for future planning because climate models have a high level of uncertainty associated with simulating precipitation and even more uncertainty for snow.  In addition, many models are not able to provide any information about lake-effect snow because their resolution is too coarse.  Equal caution must be taken for downscaled data to know what information (i.e., lake-effects) is simulated versus derived from observations.  If climate projections are needed for a particular planning application, we suggest evaluating all possible snow products to determine if some are better than others.  Models that do not simulate the Great Lakes should be flagged for a high amount of uncertainty.