Detecting Climate Change

in Canadian Ice Data

 

 

 

 

 

 

 

 

By

Carl Drews

 

 

 

jambo@dimensional.com

http://www.highestlake.com/

 

 

 

 

Final Research Paper

For

METO 1160-C

Introduction to Meteorology

USDA Graduate School

 

 

Instructor:

Dr. Raymond Motha

Deputy Chief Meteorologist, World Agricultural Outlook Board

Formerly at the University of Nebraska

 

 

Copyright September 2003

 

 

 

 

Dedicated to Christine, Isabel, Simon, and Graham,

who slept peacefully upstairs while I wrote this.

 

 

 

Abstract

 

The Canadian Ice Service maintains an archive of ice thickness measurements collected over the period 1947-1998.  The ice measuring stations are located at various lakes, rivers, and oceans throughout Canada.  The author analyzed these measurements for the purpose of detecting long-term climate trends.  A Java program calculates the ice centimeter-days for each winter season.  The ice archive shows that the climate of Canada became warmer during the study period, and that this warming trend has accelerated in recent years.

 

Introduction

 

The science of climatology uses various methods to recover historic records of temperatures in a region.  Some of these methods involve measurements taken by humans and recorded for later use.  The Canadian Ice Service has a web site at http://www.cis.ec.gc.ca/index.html that provides an on-line archive ice thickness measurements taken at numerous bodies of water in Canada.  I discovered this archive on the Internet and decided to analyze it as a research project for a course in meteorology.

 

At this writing it is well accepted in the scientific community that the earth’s climate is warming up.  The cause of the warming is a matter of debate; some scientists point to human activity as a major cause of global warming, while others suggest that the global increase in temperature is primarily natural.  I did not expect to overturn the general scientific consensus by this project, nor to prove or disprove the human link.  The purpose was to examine the data and see what it would reveal about global warming.

 

Climate influences both the length of a winter season and its severity (how cold it gets).  Knowing this, I wanted to use a metric that would reflect both the number of winter days with ice cover, and the thickness of the ice itself.  I decided to create and use the metric ice centimeter-days by multiplying the thickness by the number of days with that thickness.  Ice centimeter-days are similar to degree-days.

 

Data Discovery

 

The ice thickness archive can be downloaded from the Canadian Ice Service’s web site at http://www.cis.ec.gc.ca/index.html.  The archive itself is a zip file 300K in size, which decompresses to a text file of 1.74 megabytes.  There are 50,308 total records in eight columns, as follows:

 

Station Identifer

Station Name

Ice Thickness

Relevant Date

Snow Depth

Method of Observations

Surface Code

Water Features

Table 1

 

Dates in the archive range from September 22, 1947 to May 8, 1998.  The five most recent years are missing because the Canadian Ice Service recently instituted a new measuring program to replace the old one.  The ice thickness ranges from 0 to 345 centimeters (measured at station IC1 Isachsen (old ice) on May 31, 1968).  345 centimeters is 11 feet, 4 inches of sea ice.  A sample of the data is shown below:

 

Table 2

 

The Canadian Ice Service web site describes the measuring process as follows:

 

Measurements are taken approximately at the same location every year on a weekly basis starting after freeze-up when the ice is safe to walk on, and continuing until break-up or when the ice becomes unsafe. The location is selected close to shore, but over a depth of water which will exceed the maximum ice thickness. Ice thickness is measured to the nearest centimeter using either a special auger kit or a hot wire ice thickness gauge. The depth of snow on the ice at the location of ice thickness measurement is also measured and reported to the nearest centimeter. Measurements after 1982 include additional information (coded values as per code for additional information at bottom) such as character of ice surface, water features and method of observation.

 

A hot wire ice thickness gauge uses a length of copper wire inserted vertically into the ice for the winter.  Each week an ice technician runs a current through the wire with a battery, causing the ice around the wire to melt enough so that it slides freely up and down.  The technician pulls up the wire until a submerged washer hits the lower surface, at which point the technician can read off the ice thickness.

 

The archive contains 195 ice thickness stations.  These stations are distributed throughout Canada, but only a few of them contain records for the entire range of years.  The measurements are usually taken weekly.

 

Since the average daily temperature roughly follows a sinusoidal pattern throughout the year, I expected the ice thickness also to follow sinusoidal pattern.  I expected the ice thickness to increase gradually in the fall, peak in the winter, and gradually thin out in the spring.  This is not what I discovered.

 

The overall pattern for ice thickness stations throughout a long season is to increase gradually, maintain a constant level, then drop suddenly to zero.  The ice breakup pattern is always sudden, more sudden than the gradual increase in thickness during the fall.  For example, the graph below shows station YNI Nitchequon during the winter of 1971-1972.

 

Figure 1

 

Stations YVP Kuujjuaq 1974-1975 and YBK Baker Lake 1994-1995 are other good examples of the same pattern, showing gradual buildup and sudden breakup.

 

The sudden breakup in the spring is somewhat puzzling for standing bodies of water, but is not a problem for a long-term climate study.  I remember ice-skating as a teenager on lakes and ponds in New Jersey; in the spring the ice would melt not by getting thinner, but by separating into a vertical crystalline structure that creaked beneath our skates.  We called this “rotten ice”, and it was not nearly as strong as the thinner hard ice that first appeared in the fall.  The process of ice melting into vertical columns is called “candling”, and it usually precedes a sudden breakup.

 

Some Canadian ice seasons are not so predictable.  For example, station Q18 P18 (Portneuf) has only five thickness measurements for the winter 1983-1984.  The reading intervals are irregular, ranging from six to eight days.  The graph actually shows a mid-winter dip, with thicker ice at the beginning and end of the winter period than in the middle.  The methodology will have to handle this situation as well as the apparently more “well-behaved” sequences.  Although the dominant curve shape is known, the analysis must not assume any particular curve shape for an ice winter.

 

Figure 2

 

There are 50 records in the archive that have a blank for the thickness measurement (possibly because of heavy snow).  These records must be eliminated.

 

I decided to ignore the snow measurements.  There are complex interactions when heavy snow falls on ice.  The snow could blow off, or gradually settle and become incorporated into the ice cover through partial melting and re-freezing.  A heavy snow cover also presses down on the ice, causing lake water to seep up through small cracks and soak the lower layers of snow.  Without more detailed information, the snow measurements are not reliable with respect to ice thickness.

 

Methodology

 

The heart of the ice thickness analysis is a Java application program called “Canadice”.  Canadice calculates the ice thickness centimeter-days for each station for each ice winter, and writes this information to a text file for later analysis.

 

Canadice first separates the station measurements into annual seasons by determining the “midsummer date” that separates the winter seasons.  Summer is determined by the longest time span between ice thickness measurements.  By convention, the numbered year of the ice season contains the month of January; for example, the season November 1967 – May 1968 is reported as the ice year 1968.

 

The centimeter-days of an ice season are the integral of a bar chart representing the weekly thickness measurements.  Canadice uses trapezoidal interpolation, as shown in the figure below:

 

Figure 3

 

The total centimeter-days for this winter is the sum of the areas of the red trapezoids.

 

Canadice does not attempt to extrapolate beyond the first or the last measurements.  A few stations have only one measurement per year; for these we assume that the ice thickness lasted one full day.  This assumption is not valid in mixed comparison with seasons that include many measurements, but it does allow us to apply the same methodology to those stations.

 

After calculating the centimeter-days for each season, Canadice uses least squares to fit a line to the annual totals.  If the slope of the line is negative, then we can conclude that the ice cover is getting thinner over time.  This result suggests that the climate is warming up.  Any stations with a positive slope show that the ice cover is getting thicker over time, which suggests that the climate of Canada is cooling down.

 

Figure 4

 

Canadice calculates a linear fit to the data for the entire year span 1947-1998, but also separates the data by decade and by first and second half.  These separations allow us to look for any acceleration or deceleration in the overall rate of change.

 

Some stations do not contain enough years to fit a valid line for the period.  Canadice only reports trend lines for stations that contain 70% or more of the years in the time period.  This test assures that we are basing our conclusions on a full data set.

 

Any historical data set recorded by humans includes the possibility of bad data; that is, inconsistencies in the process of collecting and recording the measurements.  These inconsistencies can introduce a bias to the data that may affect the final results.  I reviewed the final results by hand to find any data points that looked “suspicious”; that is, outside the natural variation in the data points.  I examined the original measurements for these outliers and usually found that the thickness record for those particular years was incomplete.  The figure below shows station WZC Moosonee with the outlying point circled in red.

 

Figure 5

 

The outlying year is 1964.  For most years the ice record for WZC Moosonee goes into March or April.  But for ice year 1964 the archive contains only four records for December 1963, and those measurements show the ice getting thicker as usual.  Therefore we must conclude that we are missing the months January-March 1964, and so we eliminate that data point from further analysis.

 

Canadice rejects the following data points based on a hand review of the original data:

 

Station Identifier	Station Name	Year Eliminated
LT1	Alert	1989
WZC	Moosonee	1964
YBK	Baker Lake	1993
“	“	1996
		1998
YCB	Cambridge Bay	1959
		1997
		1998
YEV	Inuvik	1994
		1997
YFB	Iqaluit	1998
YIC	Isachsen	1948
YLT	Alert	1951
		1953
		1989
		1998
YMD	Mould Bay	1950
		1954
YRB	Resolute	1952
YUR	Gladman Point	1977
		1980
YVQ	Norman Wells	1960
		1991
YYR	Goose Bay	1959
YZS	Coral Harbour	1998

Table 3

 

Many more station-years out of the entire set could have been rejected by hand review; these particular station-years were candidates for the third data set (described below).

 

Results

 

The Canadice Java application produces three sets of data:

 

1.  The ice centimeter-days calculated for each station and year (3,456 rows of data).

 

2.  For each ice measuring station (195 rows):

            a.  the slope of the line that best fits the data points

            b.  the year-intercept of that line (where ice coverage would be zero)

            c.  the R-squared goodness of fit

            d.  the number of seasons measured

            e.  the average of those values across all stations

The second set is calculated without removing any outliers.

 

3.  For each ice measuring station (195 rows), Canadice removes outliers and only calculates results for stations that have at least 70% of the years in the range.

            a.  the slope of the best-fit line for 1948-1959

            b.  the slope for 1960-1969

            c.  1970-1979

            d.  1980-1989

            e.  1990-1998

            f.  1948-1973 (the first half of the entire range)

            g.  1973-1998 (second half)

            h.  1948-1998 (the entire range of years)

            i.  the average of those values across all stations

 

These three data sets may be useful for further climate analysis.  I encourage other researchers to use these data sets, so long as proper citation is given for the source of the processed data.  The data files are too large to include with this paper.  Please click on the link below to download the data sets as a single text file.

Get iceResults.txt

 

Second Data Set

 

The second set of results contains all measuring stations, without removing any outliers or requiring a minimum number of years.  This “raw” analysis produces the following average results:

 

Table 4

 

The slope is negative, indicating that average ice coverage is decreasing by 53 centimeter-days per year.  This result implies that the climate of Canada is warming up, as originally expected.  The year-intercept is technically the year at which we expect ice coverage in Canada to drop to zero.  This result is alarmingly soon because some stations have a positive slope, and therefore their year-intercept is in the past.  The average year-intercept is a mathematical result only and should not be used for any realistic prediction.

 

Third Data Set

 

There are 13 ice measuring stations that have 70% of the years in the entire range 1948-1998.  These stations are the most useful for long-term climate analysis.  After eliminating incomplete seasons, Canadice produces the following results for the slope:

 

Station Identifier	Station Name	1948-1959	1960-1969	1970-1979	1980-1989	1990-1998	1948-1973	1973-1998	1948-1998
LT1	Alert	-	877.1442	-181.864	-520.743	911.8142	-	-210.936	-103.937
WEU	Eureka	-82.648	1387.385	-824.307	-249.17	-868.809	744.1279	-268.605	43.30362
YBK	Baker Lake	-	393.5701	-169.506	1308.604	-	-	-115.273	-237.849
YCB	Cambridge Bay	-	196.8975	-488.347	709.8625	-1265.93	-	232.4336	-23.3449
YFB	Iqaluit	-	604.2655	-218.888	-78.6223	-325.141	-	-230.425	42.74262
YLT	Alert	-	721.2699	-353.482	-2.51528	1258.088	558.4761	-185.069	-43.0544
YMD	Mould Bay	-	291.7903	-480.76	195.823	-1829.91	149.9252	-181.532	-61.1656
YRB	Resolute	898.2383	730.3307	748.1971	-84.5247	-1092.2	615.2269	-180.133	46.09642
YUX	Hall Beach	-	-552.818	-296.219	-241.022	-1509.17	-	-484.104	-270.038
YVQ	Norman Wells	-	450.3597	-34.1241	-475.111	-729.295	-	-125.465	-57.1456
YYR	Goose Bay	-	-87.2758	123.9442	219.4774	-	-	31.22169	20.47411
YZF	Yellowknife	-	-449.983	-209	173.8819	-172.733	-	164.2509	-50.8068
YZS	Coral Harbour	-	-504.143	-179.021	329.5014	-	-	-236.472	-134.157
Average	Average	407.7952	312.2149	-197.183	98.88016	-562.329	516.939	-137.701	-63.7601

Table 5

 

Note that the average slope of the “cleaned-up” third data set (-63.7601) is fairly close to the average slope of the “raw” second data set (-53.061).  This result provides further confirmation that the ice cover of Canadian lakes, rivers, and oceans is decreasing.

 

Table 5 above contains a number of interesting rows and columns.  The following figure shows the average slope of the best-fit lines by decade (the bottom row).

 

Figure 6

 

Although there are anomalies, the overall trend is obvious.  The climate of Canada got warmer from 1948 to 1998.  The rate of warming increased in the most recent decades.

 

A curious result is that there are four stations that report a small overall increase in ice coverage during the study period (Eureka, Iqaluit, Resolute, and Goose Bay).  Although this result could be an artifact of human bias and under-reporting during the earlier part of the study period, without more detailed information we can’t be sure.  Graphing the right-hand column of Table 5 above allows us to compare the overall trends of the 13 final stations:

 

Figure 7

 

Again we see that the overall picture shows a reduction in the ice coverage.

 

It would be interesting to know why a few measuring stations are apparently bucking the overall trend.  Is there a smaller region of Canada that is getting colder, while the rest of the country is getting warmer?  To find out, we display in color the station trends on a map that shows all of the Canadian measuring stations.  Stations not in the final 13 are shown in black.  Warming stations are shown in red, and cooling stations are shown in blue.

Figure 8

 

The four cooling stations are all located along Canada’s eastern coastline.  This region faces out onto Baffin Bay and Davis Strait, a much deeper section of ocean that functions as an enormous thermal reservoir.  The cold Labrador Current also runs through this area, and during the winter the polar easterlies bring cold air from the Greenland ice cap to northern Canada’s eastern coast.  In winter the polar easterlies are aided by the counterclockwise circulation of the Icelandic low-pressure system, which also conveys cold air across Greenland’s thermal reservoir to Canada’s northeastern regions.  All these geographic features act to reduce any global warming that is felt throughout the rest of the country.

 

It would be nice to obtain better ice station coverage across Canada, especially in the southern and western regions.  Nevertheless, these 13 stations demonstrate overall climate warming and reveal local regions where the climate is actually cooling.  This result should be confirmed independently before using it as a basis for public policy decisions.

 

Conclusion

 

The Canadian Ice Service maintains and publishes an extensive archive of ice thickness measurements taken at various stations throughout the country.  If proper care is taken to remove incomplete data, the archive indicates that winter ice coverage on bodies of water is decreasing over time, and that the climate of Canada is warming up.  This warming trend has accelerated during the decade of the 1990s.  The author recommends that the measuring program continue in some form, since the archive provides valuable data on Canada’s climate.

 

One region of Canada is running against the overall warming trend, and that region is the northeast coastline adjacent to Baffin Bay, Davis Strait, and the Labrador Sea.  The climate of this region is actually cooling slightly.  The likely cause for this relative stability is the thermal reservoirs of the deep ocean in Baffin Bay and the Greenland Ice Cap.

About the Author

 

Carl Drews grew up in northern New Jersey, where he spent the winters ice-skating on the frozen ponds of the Great Swamp National Wildlife Refuge.  He attended Stanford University in Palo Alto, California, and graduated in 1982 with a Bachelor’s degree in Electrical Engineering.  Mr. Drews has lived in Boulder, Colorado for 20 years, writing software professionally and exploring Colorado’s mountains.