1. Characteristics and Trends of North American Snowfall from a Comprehensive Gridded Data Set Daria Kluver MS Thesis Presentation Department of Geography University of Delaware April 10, 2007

2. Outline Introduction Brief review of Previous literature Aims of this study Data Verification Snowfall Climatology Trend Analysis Teleconnection Analysis Conclusions http://www.wunderground.com/blog/smadsen8486/archive.html?tstamp=200601 http://www.cfcl.com/~vlb/weblog/images/WinterTimeRoad.jpeg

3. Introduction The factors controlling each snowfall event are numerous, and sometimes last for only a few hours (Leathers et al. (1993)). In contrast, snow cover studies, concerned with the presence of the snow cover over at least several days, incorporate a low temperature persistence factor (Harrington et al., 1987). Because of these differences, snowfall may be more representative of the short-term meteorological events that produce it. There have recently been several snow cover, and snow water equivalent (SWE), studies yet few have assessed the trends, climatological aspects, and climate change indication capabilities of actual snowfall (IPCC, 2001, Leathers et al., 1993).

4. North American Snowfall Half-century snowfall trends show decreases in the Pacific North West and increases in the Ohio River Valley (Scott and Kaiser, 2003,2004) Great Lakes/Upper Mid-West and High Plains experienced increases in snowfall from 1945-1984 (Leathers et al., 1993). –Number and intensity of Alberta Clippers Studies on Lake-effect snowfall show increases (Burnett et al, 2003; Ellis and Leathers, 1996; and Leather and Ellis, 1996; Scott and Kaiser, 2003, 2004) In southern Canadian regions increases in frequency of snowfall events corresponds to increases in winter snow cover (Brown and Goodison, 1996). However, reduced spring snowfall events is associated with decreased snow cover duration. Snowfall’s human impacts and cost (Changnon, 1979; Schmidlin, 1993).

5. Teleconnections and snowfall Arctic Oscillation- leading empirical orthogonal function of wintertime monthly mean Northern Hemisphere sea level pressure. –Positive phase corresponds to low pressure over the polar region and high pressure at the midlatitudes. Oceanic storms in the Pacific are pushed to the North, so the western U.S is dryer, Alaska is wetter. East of the Rocky Mountains cold weather outbreaks are not as severe. www.cpc.noaa.gov

6. North Atlantic Oscillation (NAO)- fluctuation in sea level pressure between two centers of action (Azores high and Icelandic low). It is the leading wintertime mode of variability in the Atlantic basin. –Positive years have stronger than normal subtropical high pressure center and deeper than normal Icelandic low. This larger pressure difference leads to stronger storms crossing the Atlantic Ocean at a more northerly track. Associated with warm and wet winters in both Europe and the eastern United States. http://www.ldeo.columbia.edu/NAO/

7. El Nino Southern Oscillation (ENSO)- phenomenon in the equatorial Pacific that affects precipitation, pressure and wind patterns in the tropics. Its effect on the position of the mid-latitude jet stream influences U.S. storm tracks. –Warm phase strengthens the upper level ridge off western North America, producing warmer temperatures. While the more frequent/stronger Pacific storms in the Pacific Northwest during a cold phase produces cool and wet conditions. http://www.cpc.noaa.gov/products/analysis_monitoring/ensocycle/nawinter.html

8. Pacific Decadal Oscillation (PDO)- Pacific Ocean phenomenon, defined as a leading model of multi-decadal variability in SSTs in the extratropical North Pacific. –Warm (positive) phase is characterized by cool SSTs in the central North Pacific, a more intense Aleutian low, and warm SSTs along the West Coast of North America. There is correspondingly wet periods in the coastal Gulf of Alaska. http://jisao.washington.edu/pdo/

9. Pacific North American Index (PNA)- fluctuation of the mid-tropospheric mean flow resulting in an intensification or damping of the typical PNA pattern. –Positive years, the flow is more meridional (ridge over the Rocky Mountains, trough in eastern North America), and during negative years, the flow is more zonal. This changes temperature characteristics, and frequency of precipitation across North America. www.cpc.noaa.gov

10. Few snowfall studies No studies cover all of North America Few studies looking at teleconnection patterns Existing studies have spatial or temporal limitations.

11. Aims of study Determine the quality of a new gridded data set Construct a climatology of North American snowfall Calculate trends in various snowfall characteristics Identify correlations between snowfall and teleconnection patterns

12. 1  by 1  interpolated snowfall data (Dyer and Mote, 2006) from U.S. National Weather Service (NWS) cooperative stations and the Canadian daily surface observations The interpolation was completed using the Spheremap spatial interpolation program, (Willmott et. al, 1984; Shepard, 1968) quality controlled using criteria from Robinson (1989) The period of record is 1900-2000 with a daily resolution Grid values for each day include maximum snowfall, minimum snowfall, median snowfall, mean snowfall, standard deviation. In this study, mean snowfall values were chosen to approximate the daily value at each grid point. Data

13. Data Verification Results

14. Snow season is July 1 to June 30.

15. From a total of 2891 grid cells.

16. The first ~50 years of the record had a large increase in the number of reporting stations contributing to the interpolation 190019091919 19291939 1949 1959

17. The first season with reporting stations Last season with reporting stations Period of Record with Station Data. Calculated as last season – first season from 1900 to 1999.

18. We wanted to find a period of record with the best spatial coverage possible Grids with 100 seasons period of record. Grids with >= 90 seasons period of record. Grids with >=75 seasons period of record. Grids with >=50 seasons period of record. Grids with >=25 seasons period of record.

19. This becomes an issue when trends are calculated. Example:

21. Black out grids with 10 or more seasonal differences that are greater than or equal to 10% of the number of stations over the period 1949 through 1999. Gray grid cells indicate no station data in the cell for the period of record. Solution: determine a criteria for grid cells to be blacked out.

22. Summary of Data verification results Potential data problems –Number of stations vary greatly with time –Only 50% of grid cells have a reporting station within them at a given time. –Spatially, before the 1940s few regions have continuous coverage of grid cells containing stations within them. –Grid cell timeseries illustrates how changes in station density can effect trend analysis. Reasons for problems –Data has historically be available where there are people to record it –Number of stations depend on amount of government funding available –Earlier data that has been recorded on paper is still in the process of being digitized.

23. Data verification results cont. The reliability of this data set is maximized by selecting a time period with the most consistent data distribution possible –Reduces the amount of variability and bias due to uneven spatial coverage. –1949 to 1999 most consistent data distribution Some grid cells are deemed unreliable and left out of the trend analysis

24. Snowfall Climatology

26. Mean seasonal snowfall over the time period 1949 to 1999 Remember: Snow season is July 1 to June 30.

27. Seasonal coefficient of variation over the time period 1949 to 1999

28. Seasonal maximum snowfall over the time period 1949 to 1999

29. Seasonal minimum snowfall over the time period 1949 to 1999

30. Pacific Northwest mean snowfall over the time period 1949 to 1999 This shows that the 1° by 1° resolution can capture smaller features Northeast mean snowfall over the time period 1949 to 1999

31. Summary of Climatology The annual cycle of North American snowfall is well documented by this data set. –Summer-Few grid cells with any mean snowfall, with the exception of Northern Canada –Autumn-consistent snowfall moves south from Canada, first is the Rocky Mountains, then western U.S., northeastern U.S. and Great Lakes. Canada and Alaska grid cells also have minimum values above zero. –Winter-high mean values in southern Alaska, inter- mountain west, eastern Canada and Great Lakes region. –Spring-mean snowfall values decrease. Ephemeral snow line moves north. Data set’s fine scale resolution identifies smaller scale features in the regional maps.

32. Snowfall Trends 1949-1999

33. Least squares linear regressions are calculated between each variable and time. –Slope of the linear regression identifies temporal changes in the dependent variable Correlation coefficients are calculated but not shown. Of interest to this study are physically significant changes in snowfall over time and highlighting statistically significant grid cells with physically insignificant trends would be a distraction.

34. Total seasonal snowfall, slope of the linear regression for 1949 to 1999.

35. Number of seasonal snowfall events, slope of the linear regression for 1949 to 1999

36. Date of first seasonal snowfall, slope of the linear regression for 1949 to 1999 Date of last seasonal snowfall, slope of the linear regression for 1949 to 1999. Length of snowfall season, slope of the linear regression for 1949 to 1999.

37. March MaySeptemberOctober November DecemberFebruary January Monthly Trends: April

38. Summary of Trend Analysis Decreases in of up to -30 mm of snowfall per half- century in the Pacific Northwest Increases in total seasonal snowfall are seen in several areas across the continent –Alaska, the Great Plains, the Great Lakes, Northeast United States, and reliable grid cells in northern Canada of as much as 30 mm over the period a shorter snow season in Southern California and parts of the Rocky mountains, reduced at both ends of the annual cycle. Monthly trend maps identify the largest changes in monthly snowfall as occurring in the winter and spring months.

39. Teleconnection patterns

40. Data Teleconnection Pattern Data sourceRecord length Arctic Oscillation (AO)Climate Prediction Center’s Monitoring and Data Index page 1950 -2004 North Atlantic Oscillation (NAO) Climate Prediction Center’s Monitoring and Data Index page 1821-2000 Pacific North American (PNA) index Climate Prediction Center’s Monitoring and Data Index page 1950-2004 Pacific Decadal Oscillation (PDO) University of Washington website1900-2000 Southern Oscillation Index (SOI) University of East Anglia Climate Research Unit’s Data site 1866-2003.

41. In order to identify basic relationships between snowfall and teleconnection patterns, simple linear regressions as well as Pearson correlation coefficients are calculated. These calculations are done monthly for each of the 5 teleconnection patterns and snowfall. Only maps with strong, visible signals are shown.

42. Arctic Oscillation and monthly snowfall

43. North Atlantic Oscillation and monthly snowfall

44. Pacific Decadal Oscillation and monthly snowfall

45. Pacific North American index and monthly snowfall

46. Southern Oscillation Index and monthly snowfall

47. For each month a multilinear stepwise regression is calculated between monthly snowfall (dependent variable) and the AO, NAO, PDO, PNA, and SOI (independent variables). Time period used for this analysis is 1950 to 1999 due to availability of teleconnection data. A multiple linear regression model –is built with stepwise selection of independent variables. –This is done using the variance-covariance matrix of the monthly snowfall and monthly teleconnection data. –This method of analysis is useful in this situation because it allows the observational data (snowfall) to be characterized by several different variables (teleconnection data).

48. January FebruaryMarch April September OctoberNovemberDecember Multilinear Stepwise Regressions

49. Summary of Teleconnections AO- –general negative correlation between the eastern half of the United States and the AO are consistent with the previous literature. –During a positive AO phase, the reduction of cold air outbreaks east of the Rocky Mountains leads to higher temperatures. This could cause the precipitation to fall more frequently as rain rather than snow, decreasing the snowfall amounts in these regions during a positive AO. NAO- –Correlations most likely related to changes of the mid-tropospheric flow, such as an eastward displacement of the eastern trough (Bradbury et al.,2002) during its negative phase –which could explain the negative correlations with the northern Great Plains in September, and the Great Lakes/ Eastern United States in October. PDO- –striking areas of strong correlations that are fairly persistent throughout the snow season. –The negative correlations in the Pacific Northwest are consistent with previous findings of decreases in winter precipitation in the Pacific Northwest and of a strengthened Aleutian low during positive (warm) phases (Gedalof and Mantua, 2002).

50. PNA- –Similar to the PDO. –Negative correlations in the Pacific Northwest correspond to the more meridional characteristics of the Pacific North American pattern during the positive phase of the PNA index (the ridge over the western part of the continent is accentuated, resulting in changes in storm tracks and higher temperatures). –The positive correlations in the eastern United States are associated with the deepening of the eastern trough, which results in colder temperatures and changes in storm tracks (Leathers et al., 1991; Serreze et al., 1998; Bradbury et al., 2002; Notaro et al., 2006). SOI- –Largest correlations in the central United States, along with a small area of positive correlations in the West. Multilinear stepwise regressions –teleconnection patterns can account for up to 70 % of the variance in snowfall, principally in the Pacific Northwest.

51. Conclusions

52. The Dyer and Mote (2006) gridded snowfall product for North America is a unique and useful data set for climatological studies. However, a thorough understanding of the limitations of the data is necessary before using the gridded product. The Rocky Mountains, coastal Alaska, Great Lakes snowbelts and southeastern Canada have the largest snowfall accumulations on the continent. Smallest accumulations are found along the southern tier of the United States. Temporal trends in snowfall identified in this study agree with previous work. Large decreases in snowfall are seen in the Pacific Northwest, and increases in snowfall are seen in several areas across the continent Atmospheric teleconnections account for a substantial amount of variation in snowfall, especially the AO, PDO, and PNA.

53. Future research is warranted in many areas. –In order to improve predictions of North American snowfall, the relationship between teleconnection patterns and snowfall requires more exploration. –More detailed studies of regional teleconnection/snowfall relationships could enhance our understanding of the physical processes involved in the statistical relationships. –Finally, continued extension of the trend analysis into the past can aid in the attribution of climate change.