Climate Trends at Eureka in the Canadian High Arctic

Climate Trends at Eureka in the Canadian High Arctic G. Lesins*, T. J. Duck and J. R. Drummond Department of Physics and Atmospheric Science Dalhousie University, Halifax, Nova Scotia [Original manuscript received 9 April 2009; accepted 2 December 2009] ABSTRACT Weather observations made at Eureka, on Ellesmere Island in the Canadian High Arctic, have been archived since 1953. The time series, averages, and seasonal cycles of surface temperature, pressure, dew point, relative humidity, cloud cover, wind speed, and direction are presented for the period from 1954 to 2007. Also shown are the time series and averages for the 500 mb temperature, 900 to 500 mb thickness, 500 mb wind speed, and various boundary-layer stability parameters. Some of the main trends found are 1) an annual average surface warming of 3.2 C since 1972, with summer exhibiting the least warming, 2) a reduction in the frequency of strong anticyclonic events in the winter, 3) a reduction in surface wind speeds except in the summer, 4) a 1.0 C warming in the 500 mb temperature since 1961, with the greatest warming occurring in the spring and summer, and 5) a 10% increase in precipitable water all year round since 1961 but dominated by the spring, summer, and autumn seasons. The importance of open water in the Arctic Ocean for summer temperatures and humidity, of the North Atlantic Oscillation for winter interannual pressure variability, and of precipitable water for winter temperatures are highlighted in this climatology. RéSUMé [Traduit par la rédaction] Les observations météorologiques faites à Eureka, sur l île Ellesmere, dans l Arctique canadien septentrional, ont été archivées depuis 1953. Nous présentons les séries chronologiques, les moyennes et les cycles saisonniers de la température de surface, de la pression, du point de rosée, de l humidité relative, de la couverture nuageuse ainsi que de la vitesse et de la direction du vent pour la période allant de 1954 à 2007. Nous présentons également les séries chronologiques et les moyennes pour la température à 500 mb, l épaisseur 900-500 mb, la vitesse moyenne du vent à 500 mb et divers paramètres de stabilité dans la couche limite. Certaines des principales tendances que nous trouvons sont 1) un réchauffement annuel moyen à la surface de 3,2 C depuis 1972, l été étant le moment du plus faible réchauffement; 2) une réduction de la fréquence des forts événements anticycloniques en hiver; 3) une réduction de la vitesse moyenne des vents de surface, sauf en été; 4) une augmentation de 1,0 C de la température à 500 mb depuis 1961, le réchauffement le plus marqué se produisant au printemps et à l été; et 5) une augmentation de 10 % de l eau précipitable toute l année depuis 1961 mais plus particulièrement au printemps, à l été et à l automne. Cette climatologie met en évidence le rôle important de l eau libre dans l océan Arctique relativement aux températures et à l humidité en été, de l oscillation Nord-Atlantique relativement à la variabilité interannuelle de la pression ainsi que de l eau précipitable relativement aux températures en hiver. 1 Introduction Global climate models predict the Arctic to be particularly sensitive to the warming induced by the increase in greenhouse gases (IPCC, 2007). The measurements from the past 50 years show that the Arctic already has the most warming of all regions on the globe (Hansen et al., 1999; Hansen et al., 2006). A downward trend in late summer sea-ice areal extent has been measured by satellites for the last 30 years (Stroeve et al., 2008; Comiso et al., 2008). Against this backdrop the Canadian Network for the Detection of Atmospheric Change (CANDAC) project was initiated to study the atmospheric processes at Eureka with climate change science as an important theme (Lesins et al., 2009; Bourdages et al., 2009). Eureka is located on Ellesmere Island in Nunavut, Canada. Ellesmere Island is the northernmost of the islands making up the Canadian Arctic Archipelago. The Archipelago is surrounded by the Arctic Ocean to the north and west, the mainland of Canada to the south, and Greenland to the east. Eureka is at the southern end of the Station Creek and Black Top Creek basins on the western side of Fosheim Peninsula in northern Ellesmere Island. Black Top Ridge, which is about 600 m above mean sea level (MSL), is located some 15 to 20 km to the northeast while another ridge of similar height is found about 15 to 20 km to the northwest (Fig. 1). The weather station is designated by World Meteorological Organization (WMO) identification number 71917 and by the Environment Canada and Transport Canada code WEU. There are no permanent residents living in Eureka, although a variable number of research and operational staff *Corresponding author s e-mail: glen.lesins@dal.ca Canadian Meteorological and Oceanographic Society

60 / G. Lesins et al. Fig. 1 A topographic map (NRCan, 2008), centred near Eureka (marked with a red airplane symbol on the north shore of Slidre Fiord), shows that 600 m high ridges are located about 15 km to the northwest, northeast, and southwest. The width of the map is about 120 km and the distance across Slidre Fiord at Eureka is about 4 km. La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 61 rotate in and out of the facility. From 1993 to 2001 a laboratory about 15 km to the west-northwest of Eureka housed the Arctic Stratospheric Ozone Observatory. This site, now called the Polar Environment Atmospheric Research Laboratory (PEARL), was reopened in 2006 as part of the CANDAC project to house a suite of advanced instrumentation to monitor clouds, aerosols, precipitation, chemistry, and upper atmospheric dynamics. Eureka s climate is fairly typical for the Canadian Arctic Archipelago with a long, dark winter that promotes a strong surface-based temperature inversion which remains intact for most of the winter making it difficult for free tropospheric air to mix down to the ground. The transition to summer occurs with a rapid warming that is forced by the nearly continuous daylight conditions and the breakdown of the Arctic winter vortex. In this paper the hourly surface observations and data from the twice daily radiosonde flights will be used to compile a climatology of weather conditions and trends at Eureka. This is done to provide background information for the many research activities currently ongoing as part of CANDAC and the recently ended International Polar Year and to support a long-term monitoring station as part of the mission for the International Arctic Systems for Observing the Atmosphere (IASOA). The goal is to report on various trends in weather observations at Eureka and suggest possible relationships rather than providing a detailed analysis on the causes for the trends. The prospects of rapid warming in the Arctic demand that a thorough analysis of past weather conditions be considered and understood. In Section 2, the surface and upper air datasets are described. Section 3 considers the seasonal cycles and Section 4 discusses the trends in the climate data. Section 5 examines the special wintertime temperature and humidity conditions, and finally the conclusions are summarized in Section 6. 2 Datasets Hourly surface data records for Eureka in the Environment Canada archive begin on 1 January 1953 at 01:00 Local Standard Time (LST is in the Eastern Time Zone of North America) or 06:00 UTC. The weather observations are taken at latitude 79.98 N, longitude 85.93 W and at an altitude of 10.4 m above MSL on the coast along the north side of Slidre Fiord. There is no record of any substantial movement of the Stevenson screen, which is located 1.3 m above the ground, during the total archive period. In 1953, observations are available only at 01:00, 07:00, 13:00, and 19:00 LST. Observations at 01:00 LST were taken continuously from 1 January 1953 until 31 May 1992. Observations are available for each hour of the day from 1 April 1981 until 2 May 1987. From 1982 to the present, continuous observations are available for all hours except 00:00, 01:00, and 23:00 LST. Starting on 1 June 1992 surface data are also available at 23:00 LST. Although there is considerable variation in observing frequency starting in 1954 and continuing to the present day, hourly observations are available continuously at 04:00, 07:00, 10:00, 13:00, 16:00, 19:00, and 22:00 LST. This seven times per day record is 54 years long, extending from 1954 to 2007, and is used for all analyses of average surface weather time series presented in this paper to maintain consistency of the observation times. The dew point temperature record has more data gaps compared to the air temperature record. In particular, the cold season months from November to April are inconsistently represented before 1988. Hence, the surface moisture analysis is restricted to post-1987. The hourly surface measurements at Eureka also include the cloud fractional sky coverage and the cloud type for up to four levels as determined by a human observer. By choosing the hours 4:00, 7:00, 10:00, 13:00, 16:00, 19:00, and 22:00 LST the cloud dataset is complete from 1955 to 2004 with only a few missing reports. Since it is difficult to identify cloud amount and type consistently during the night, the cloud dataset is further restricted to cover only the months from April to September when there is at least sufficient twilight for the entire day. The upper-air data from radiosondes are available for the period 1961 to 2007 at Eureka with profiles of temperature, humidity, wind speed, and wind direction measured at 00:00 and 12:00 UTC (19:00 and 07:00 LST). Note that surface winds reported here are actually measured at the standard 10 m height above the ground. Both surface and upper air data were obtained from Environment Canada. 3 Seasonal cycles The average monthly temperatures are shown in Fig. 2 for the entire record. Based on the seasonality pattern we define for use in this work, the winter months as the three coldest months of the year: January, February, and March. Summer is defined as the three warmest months: June, July, and August. Spring consists of the rapidly warming months of April and May while autumn comprises the four months of September to December. Although it is unconventional to define the seasons in this way, it better reflects the actual character of the annual changes at Eureka. The spring transition occurs more rapidly than the autumn transition. The inclusion of December in the autumn season is debatable, but our choice is also based on the fact that the deep, stable and quiescent winter boundary layer is not fully established until January and that the properties of the boundary layer are more alike in January, February, and March than they are in December. It does mean that there are fewer measurements in spring than autumn, but the advantage is that the winter and summer seasons both have three months and are centred on the coldest and warmest times of the year, respectively. Figure 3 shows the mean sea level pressure for each month averaged over the entire time series from 1954 to 2007. The average for the entire period is 1014.8 mb. Above-average annual pressures occur from February to May while belowaverage pressure occurs from June to September. The maximum monthly mean pressure of 1021.1 mb occurs in April, whereas the minimum monthly mean pressure of 1009.4 mb occurs in July. Canadian Meteorological and Oceanographic Society

62 / G. Lesins et al. Fig. 2 Monthly average surface dry-bulb air temperature averaged over the entire record from 1954 to 2007. The solid horizontal line is the annual temperature averaged over the entire data period. The vertical bars, here and in the other seasonal figures, represent plus and minus one standard deviation of the monthly average data over the entire 54-year record. Monthly Averages (1954 2007) Fig. 3 The monthly variation in the mean sea level pressure is plotted averaged over the record 1954 to 2007. The solid horizontal line is the average pressure over the entire record. Figure 4 shows the annual cycle of the mean monthly dew point temperatures ranging from 41 C to +1 C with an annual average of 22.3 C. The near 0 C plateau in the dew point temperature is not surprising when considering the mixture of open water and sea ice in the Arctic Ocean that exists in the summer. There is a minimum plateau of approximately 40 C during the winter months, however, as discussed in Section 5, this appears to be an artifact. The annual variation of the surface wind speed by monthly averages is shown in Fig. 5a. June, July, and August have La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 63 Monthly Averages (1988 2007) Fig. 4 Monthly mean surface dew point air temperatures averaged from 1988 to 2007. The solid horizontal line is the average dew point temperature over the entire available record. SurfaceWindSpeed (km h 1 ) Wind Speed (km h 1 ) Fig. 5 a) Monthly mean surface wind speed averaged from 1954 to 2007. The thin horizontal line is the annual average over the entire time series. b) Histogram of the surface wind speeds binned over intervals of 3 km h 1 for winter (dark solid line) and summer (grey dotted line) for the period 1954 to 2007. The sum of the frequencies equals one for each season. much higher average winds than the colder months of October to April. The annual average wind speed over the entire record is 11.2 km h 1. Figure 5b plots the wind speed histogram showing that the most frequent wind speed during the winter is calm, whereas in the summer it ranges most commonly from 10 to 20 km h 1. This is in sharp contrast to the annual wind behaviour of many stations in southern Canada where the summer season experiences the lightest Canadian Meteorological and Oceanographic Society

64 / G. Lesins et al. surface winds (http://www.windatlas.ca). In the high Arctic, the strong surface-based temperature inversion during the winter inhibits the mixing of higher momentum air from above the boundary layer to the surface and as a result the wind speeds are lower. A reduction in the frequency of baroclinic disturbances during the winter could also contribute to weaker winds at Eureka. The surface wind direction measurements are available starting only in 1976. Figure 6 shows the histograms of wind direction frequency by season. A direction of 0 corresponds to calm winds while 360 indicates a wind from the north. The most frequent wind directions are southeast and west throughout the year, although north also occurs with some regularity in the summer. Slidre Fiord extends to the west and southeast of Eureka (Fig. 1), and the local topography channels the wind increasing the frequency of winds from these directions. Calm winds occur 39% of the time during the winter but only 4.3% of the time in the summer, demonstrating the effect of the strong surface temperature inversion and few storm systems during the winter. Figure 7 plots the monthly total cloud amounts averaged from 1954 to 2007. It should be noted that the cloud fraction is estimated by a human observer so there is a subjective element to this dataset. Furthermore, the skies are dark during the winter which imposes even greater challenges in determining cloud coverage. The summer and early autumn are cloudier than the rest of the year with an annual average of 0.5 sky coverage. Figure 8 shows the seasonally averaged temperature profile for the entire time series. Surface-based inversions occur during all seasons except summer. The tropopause height shows little seasonal variation near 300 mb. The lighter lines represent one standard deviation from the mean. The summer variation in upper temperature is considerably less than for other seasons. Note that the winter lower stratosphere has standard deviations of about 12 C which is likely due, in part, to sudden warming events at these heights and to variations in the position of the Arctic circumpolar vortex (Duck et al., 2000). Using the temperature and humidity profiles, the precipitable water of the column (vertically integrated water vapour amount) was calculated. The calculation was not extended above 400 mb to enforce a uniform upper limit to the precipitable water integration and to avoid using humidity measurements taken at very cold temperatures which may not be reliable. This does not pose any problems since the absolute amount of water vapour above 400 mb is usually insignificant compared to the lower troposphere. Figure 9 plots the monthly averages of precipitable water which ranges from about 1.8 mm in late winter to about 12 mm in July, with an annual average of 4.9 mm. 4 Atmospheric trends a Air Temperature Figure 10 shows the annual average dry bulb air temperatures and the 5-year running mean time series. The average temperature over the entire 54 year record is 19.1 C, with the Wind Direction ( ) Wind Direction ( ) Wind Direction ( ) Wind Direction ( ) Fig. 6 Seasonal histograms of the surface wind direction measured from degrees north and binned every 10 degrees for the period 1976 to 2007. A direction of 0 indicates calm conditions while a direction of 360 indicates a north wind. La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 65 Fig. 7 Monthly total cloud amount averaged over the entire time series from 1954 to 2004. The thin solid line is the average over the entire record. Fig. 8 The seasonal mean temperature profiles up to 100 mb from 1961 to 2007. The thin lines are one standard deviation away from the mean. warmest year, at 16.5 C, occurring in 1958 and the coldest year, at 21.6 C, occurring in 1972. The smoothed annual temperatures were on the decline during the first part of the record until the early 1970s, with a warming period from the early 1970s to the early 1980s, a brief cooling period to the mid-1980s and finally warming to the present day. The highest recorded temperature was 20.0 C on 22 July 2007 while the lowest recorded temperature was 54.6 C on 15 February 1979. Linear regression was used to determine the temperature trend since 1972 when the warming started as indicated by the Canadian Meteorological and Oceanographic Society

66 / G. Lesins et al. Fig. 9 Monthly mean precipitable water averaged from 1961 to 2007. The precipitable water is computed from radiosonde data, integrating the water vapour content from the surface to 400 mb. The thin horizontal line is the overall average. 5-year running mean. The warming trend is 0.88 C per decade with an error of 0.17 C per decade. The trend error is given by one standard deviation in the regressed slope calculation. The trend and its error are computed the same way for all other figures containing a trend line. When the magnitude of the trend error is comparable to or larger than the magnitude of the trend then the trend is not significant. Since 1972 the 5-year smoothed surface annual temperature has increased by about 3.2 C. This is consistent with the Goddard Institute for Space Studies (GISS) range of warming from 1.18 to 4.7 C for 11 stations poleward of 75 N for the same time period (http://data.giss.nasa.gov/gistemp/maps/). Figure 11 shows the 54-year time series of the average surface temperatures for each season. Although each season has its own trend, all have higher temperatures at the beginning and end of the record compared to portions in the middle of the series. Winter has a pronounced warm period similar to current winter temperatures lasting from 1976 to 1983 with the exception of 1979. The temperature scale is the same for all four subplots which helps to depict the weaker interannual variability in the summer. The summer season warming is the smallest, at 1.5 C, since 1972. This is expected, due in part to the damping effect on temperature fluctuations caused by the extensive area of mixed open water and sea ice that make up the surface of the Arctic Ocean during the summer. Much of any enhanced downward radiative forcing is used to melt sea ice and snow during the summer months thereby reducing the impact on surface air temperature. The autumn season exhibits the largest and most consistent warming of 4.5 C since 1972. Figure 12 plots the cumulative temperatures above 0 C for the entire year based on measurements taken 7 times per day. This was computed by summing all above-freezing temperatures for the entire year. The trend is very similar to the average summer temperature in Fig. 11c. This is a simple indicator for the surface ice and snow melting potential for each year which has been increasing most strongly since the 1990s. The increasing trend in the above-freezing temperatures since the 1970s is consistent with the decreasing trend in Arctic summer sea ice measured by satellite microwave sensors since 1978. It is unclear whether the warm summer period around 1960 is associated with a reduced amount of sea ice since this predates satellite microwave measurements. Based on ship observations, the period from 1958 to 1963 experienced below-average Arctic sea-ice extent (Walsh and Johnson, 1979). Figure 11d shows that the greatest warming since 1972 occurred during the autumn, which is related to the delay in the sea ice refreezing at this time of year. We consider two measures of the tropospheric air temperature: the 500 mb air temperature and the thickness from 900 to 500 mb. 900 mb is used as the base of the thickness calculation instead of the commonly used 1000 mb to avoid extrapolation issues when the surface pressure is less than 1000 mb. Also 900 mb is a mandatory level for reporting upper-air variables from a radiosonde which minimizes missing data and interpolation issues. Figure 13a plots the annual averages of the 500 mb temperature from 1961 to 2007 which has a mean of 34.7 C over the entire time series. There is a fairly consistent warming trend throughout the period and, when a linear fit is La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 67 Fig. 10 Time series of the annual average surface dry-bulb air temperature at Eureka from 1954 to 2007 marked with crosses. The dark solid line is the 5-year running mean. The thin horizontal line is the average over the entire time series. The dashed line shows the linear regressed trend from 1972 to 2007. The numerical values for the trend and trend error also apply to the period 1972 to 2007. Fig. 11 Time series of the surface dry-bulb air temperature from 1954 to 2007 for each season. The dark solid line is the 5-year running mean and the thin horizontal line is the seasonal average temperature over the entire time series. The range of the vertical scale is the same for all four subplots which helps to emphasize the reduced interannual variability during the summer. The dashed line shows the linear regressed trend from 1972 to 2007. The values for the trend and trend error also apply to the period 1972 to 2007. Canadian Meteorological and Oceanographic Society

68 / G. Lesins et al. Annual Cumulative Melting Temperatures Fig. 12 Time series of the annual sum of surface dry-bulb air temperatures greater than 0 C using the 7 standard hours in each day as described in the text. The dark solid line is the 5-year running mean. Thickness 900 500 mb (m) Cumulative Temperature (degc) Fig. 13 a) Time series of the annual average 500 mb temperature from 1961 to 2007. The heavy solid line is the 5-year running mean and the dashed grey line is the trend using linear regression. The thin horizontal line is the average 500 mb temperature for the entire time series. b) Time series of the annual average 900 to 500 mb thickness from 1961 to 2007. The heavy solid line is the 5-year running mean and the dashed grey line is the trend using linear regression. The thin horizontal line is the average for the entire time series. imposed, we find a warming trend of 0.22 C per decade or about 1.0 C in 46 years. This is in contrast to the surface temperature trend in Fig. 10 which started with a cooling period before reversing into a strong warming trend. The seasonal trends in Fig. 14 show that spring and autumn have experienced the greatest warming. The winter variation more closely matches the surface temperature time series possibly due to the importance of downward infrared radiative forcing in La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 69 Fig. 14 Time series of the 500 mb temperature from 1961 to 2007 for each season. The dark solid line is the 5-year running mean, the thin horizontal line is the seasonal average over the entire time series and the dashed grey line is the trend from linear regression. The range of the vertical scale is the same for all four subplots. determining the winter surface temperatures. The more consistent warming trend in the other three seasons dominates the annual trend in the 500 mb temperature. No temperature trend was apparent at 300 mb which is near the tropopause (not shown). To see whether the warming trend is apparent over the entire lower tropospheric column the annual averages of the 900 to 500 mb thickness are given in Fig. 13b. A warming trend is apparent with a thickness increase of 18 m over the 46-year period (or about +4.0 m per decade). The fluctuations and trends in thickness follow those of the 500 mb temperature which indicates that the lower tropospheric column varies quite uniformly. Viewed seasonally, the increasing thickness trend is most apparent in spring and autumn with much smaller warming trends in summer and winter (Fig. 15). This differs slightly from the 500 mb temperature analysis where the summer warming trend is comparable to spring and autumn. The seasonal trends are also distinct from the seasonal surface air temperature trends shown in Fig. 11. b Surface Pressure Figure 16 shows the time series of the seasonal averages of surface pressure. The winter season exhibits the largest interannual variability. In contrast with the temperature record, there are no significant multi-decadal trends for any of the seasons. Much of the winter pressure variability at Eureka can be explained by the phase of the North Atlantic Oscillation (NAO) as shown in Fig. 17a for the winters of 1954/55 to 2006/07. The NAO index is the normalized sea level pressure difference between Lisbon, Portugal and Stykkisholmur, Iceland for the winter months of December, January, February, and March, as defined by Hurrell et al. (2003). The strong negative correlation coefficient between the NAO index and surface pressure of -0.62 indicates that when the NAO phase is positive, i.e., when the Icelandic low is deeper than average, the geographic extent of the Icelandic low is large enough to lower the average pressure at Eureka. The correlation between the NAO index and the surface winter temperature is also shown. In contrast to the pressure, the negative correlation with temperature is not significant (Fig. 17b). The histograms in Fig. 18 are based on the hourly pressure measurements and are binned at a resolution of 1 mb. In Fig. 18a the percentage of time the winter pressure was equal to or greater than 1035 mb declined from 9.4% in the first half of the record to 4.8% in the second half, a reduction by almost a factor of 2. Figure 18c shows that in summer the frequency of lower than average pressure measurements from 995 to 1005 mb increased in the second half of the time series. Figure 19 shows the seasonal time series of the pressure standard deviations which are related to the histograms in Fig. 18. In the winter there has been a reduction in surface high pressure events which is apparent when comparing the first half of the time series from 1954 to 1980 with the second half of the time series from 1981 to 2007. This is related to the reduction in the winter pressure standard deviation as shown in Fig. 19a. Note from Fig. 19c that in summer the pressure Canadian Meteorological and Oceanographic Society

70 / G. Lesins et al. Fig. 15 Time series of the average 900 to 500 mb thickness from 1961 to 2007 for each season. The dark solid line is the 5-year running mean, the thin horizontal line is the seasonal average over the entire time series and the dashed grey line is the trend from linear regression. The range of the vertical scale is the same for all four subplots. Fig. 16 Time series of the mean sea level pressure from 1954 to 2007 for each season. The dark solid line is the 5-year running mean and the thin horizontal line is the seasonal average pressure over the entire time series. The vertical scale is the same for all four subplots which demonstrates the enhanced interannual variability in winter. No consistent trends are apparent. La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 71 Fig. 17 Scatter plot of a) mean winter mean sea level pressure and b) surface air temperature versus the NAO index for the winters 1954/55 to 2006/07 as defined in the text. Each cross indicates one winter and in this figure winter includes December to be consistent with the Hurrell NAO dataset. The solid lines are the least squares linear fits and the correlation coefficient is displayed. Fig. 18 Histograms by season of the mean sea level pressure divided into the first half of the time series from 1954 to 1980 in solid black and the second half from 1981 to 2007 in dotted grey. Note that the ranges for the abscissa and ordinate are the same for all four subplots. The pressure axis is binned at 1 mb intervals and the sum over all bins of the fractions for each season equals one. standard deviation increased from about 6.0 mb in the late 1950s to about 7.0 mb in the early 2000s, an upward trend of about 0.18 mb per decade. An increase in summer cyclonic activity would be the most likely explanation for this trend. Results from the summer months suggest a periodicity of about 9 11 years in the intensity of the pressure fluctuations Canadian Meteorological and Oceanographic Society

72 / G. Lesins et al. Fig. 19 Seasonal trends in the mean sea level pressure standard deviation are plotted for the period 1954 to 2007. The dark solid line is the 5-year running mean and the thin straight line is the linear regression. which is of unknown origin. The other seasons do not show significant trends in the standard deviation although a slight decrease has occurred during the winter from about 10.3 to 9.5 mb in the linear trend. c Water Vapour The seasonal time series and trends for the surface dew point temperature from 1988 to 2007 are shown in Fig. 20. Spring has the greatest interannual variability of the seasons partly as a result of being comprised of only two months. Autumn is the only season with a significant trend of +1.2 C per decade with the caveat that the time series is only 20 years long. This is likely related to the observation that autumn has had the greatest warming in the dry-bulb temperature since 1972 (Fig. 11d). The relative humidity can be computed from the dry-bulb and dew point air temperatures either with respect to liquid water or ice. Hence, the relative humidity record will have the same data gaps as the dew point temperature and so the time series also starts in 1988. Figure 21 shows the seasonal time series of the relative humidities with respect to liquid water saturation. In Figs 21a, 21b, and 21d the upper curves are the relative humidities with respect to ice saturation. The summer subplot is only with respect to liquid saturation because the air temperatures are usually above freezing. Trends are computed for relative humidity with respect to ice, except for summer, in which case humidity is calculated with respect to the liquid phase. All seasons show a weak downward trend since 1988 of about 1.4 to 1.6% per decade but this trend is only marginally outside the one standard deviation error. This weak downward trend since 1988 is driven mostly by the warming temperatures which dominate the increasing dew point temperatures in determining the net effect on relative humidity. Note how close the surface air is to ice saturation during the winter with the 1991 average actually supersaturated with respect to ice. This explains why ice crystals are commonly observed in the air near the ground during the winter months. A moistening trend can be seen in Fig. 22 which plots the annual averages of the precipitable water. By forcing a linear trend through the series, a 10% increase in water content occurred from 1961 to 2007. However, the 5-year running mean shows that there is considerable deviation from the linear trend with moister profiles in the 1980s and drier conditions in the late 1990s. The seasonal results are shown in Fig. 23 which shows that all seasons except winter have experienced a 10% increase in precipitable water from 1961 to 2007. As an absolute amount, the summer season has the greatest increase at 0.3 mm per decade since 1961. La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 73 Fig. 20 Time series of the surface dew point temperature from 1988 to 2007 for each season. The dark solid line is the 5-year running mean, the thin horizontal line is the seasonal average temperature over the entire time series and the heavier straight line is the trend from linear regression. The range of the vertical scale is the same for all four subplots. RelativeHumidity (%) RelativeHumidity (%) RelativeHumidity (%) RelativeHumidity (%) Fig. 21 The seasonal time series of the annual average relative humidity is plotted with crosses. The dark solid line is the 5-year running mean, the thin horizontal line is the seasonal average relative humidity over the entire time series and the heavier straight line is the trend from linear regression. For winter, spring, and autumn, the upper set of curves are the relative humidity with respect to ice while the lower set are with respect to water. For summer only plots, the relative humidity with respect to water is plotted since the air temperatures are normally above freezing. Canadian Meteorological and Oceanographic Society

74 / G. Lesins et al. Fig. 22 Time series of the annual average precipitable water from 1961 to 2007. The heavy solid line is the 5-year running mean and the dashed line is the trend using linear regression. The thin horizontal line is the average precipitable water for the whole time series. Fig. 23 Time series of the precipitable water from 1961 to 2007 for each season. The dark solid line is the 5-year running mean, the thin horizontal line is the seasonal average over the entire time series, and the dashed grey line is the trend from linear regression. Note that the range of the vertical scale is not the same for all four subplots. d Winds The seasonal time series of wind speeds are shown in Fig. 24. There has been a wind speed reduction in spring and autumn. However, at about 0.6 km h 1 per decade over the entire time series, the reduction has been greater in winter. It is interesting to note that the weakening winter wind speeds occurred in La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 75 SurfaceWindSpeed (km h 1 ) SurfaceWindSpeed (km h 1 ) SurfaceWindSpeed (km h 1 ) SurfaceWindSpeed (km h 1 ) Fig. 24 Time series of the average surface wind speed from 1954 to 2007 for each season. The dark solid line is the 5-year running mean and the thin horizontal line is the seasonal average over the entire time series. The sloped solid line is the trend based on linear regression from 1954 to 2007. Fig. 25 a) Time series of the daylight cirrus/cirrostratus frequency from 1955 to 2004 averaged from April to September inclusively when daylight conditions exist nearly 24 hours a day. The frequency is computed by summing the total number of hourly observations with cirrus divided by the total number of observing hours. The dark solid line is the 5-year running mean. b) Time series of the daylight cumulus/stratocumulus frequency from 1955 to 2004 averaged from April to September inclusively. c) Time series of the daylight clear sky frequency from 1955 to 2004 averaged from April to September inclusively. Canadian Meteorological and Oceanographic Society

76 / G. Lesins et al. Apr to Sept Annual & 5-Yr Running Mean Fig. 26 Time series of the total cloud amount from 1954 to 2004 averaged from April to September inclusively when daylight conditions exist nearly 24 hours a day. The heavy solid line is the 5-year running mean and the thin horizontal solid line is the mean over the analyzed period. spite of a slight weakening of the surface-based temperature inversion shown in Fig. 27b (Section 3f). The 500 mb wind speed has not changed significantly in any of the seasons although there is a large amount of interannual variability (not shown). e Cloud Observations The series of plots in Fig. 25 compute the annual cloud frequencies. This is computed by taking the ratio of the number of observations which report a particular cloud type, irrespective of which level it might be reported in, to the total number of observations. It does not use the cloud amount information since that observation can have uncertainties associated with multiple cloud levels, varying opacity, and subjective factors in assessing coverage. These restrictions in the dataset should help to reduce some of the subjective nature of the cloud observations. Figure 25a shows that cirrus and cirrostratus frequency has been increasing since the start of the record in 1955. The increase is steep until the mid-1960s and then becomes more gradual. The time series for the frequency of cumulus (including cumulus fractus and towering cumulus) and stratocumulus in Fig. 25b shows a similar trend to the cirrus frequency. After rapidly increasing during the first 10 years of the record, the occurrence of low clouds slowly continues to increase from about 50% in the early 1970s to just over 60% by 2004. With the increase in low and high cloud frequencies, it is not surprising to see large reductions in the occurrence of clear sky conditions (Fig. 25c) from near 20% in the late 1950s to about 7% from the late 1990s to 2004. The trend in clear sky frequency is not linear, it recovers to near 20% in the mid- 1980s then drops steeply in 1994 and has not shown any signs of recovering. Figure 26 is the time series from the total cloud amount data field, irrespective of cloud type and number of levels. This observation is complete from 1954 to 2007 and is computed for the same 7 hours used in the cloud frequencies. Figure 26 includes the months from April to September when at least some daylight is present throughout the day. The overall average is 0.6 sky coverage with little variation and no trend. This result might appear to be contrary to the cloud frequency results but it should be noted that the frequency does not account for the amount of visible cloud. The cloud frequency can increase at the same time as the cloud amount decreases if there are more days with cloud but the cloud covers less of the sky. Note that the cloud amount in tenths of the sky for individual cloud types cannot be plotted in a similar way to the total cloud amount because, effective in 1977, all Canadian stations reported individual cloud amounts only as scattered, broken or overcast. f Winter Surface-Based Temperature Inversion A strong surface-based temperature inversion is a consistent feature in the lower tropospheric profile during most of the non-summer months, especially during the winter. The lack of solar heating and the intense infrared cooling from the surface to space during clear and dry conditions allow the surface skin temperature and the air in close proximity to the surface to cool below the air temperature values found two kilometres higher (Curry, 1983). La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 77 Inversion Lapse Rate (degc/km) Fig. 27 Time series of various properties of the winter average surface-based dry-bulb temperature inversion. The crosses connected by the dotted line are the annual average, the heavy solid line is the 5-year running mean, and the dashed grey line is the trend from 1984 to 2007. The numerical trends and trend errors are for the period from 1984 to 2007. a) Inversion temperature defined as the maximum temperature in the troposphere for each profile, averaged over each winter. b) Inversion lapse rate defined as the difference between the inversion temperature and the minimum boundary-layer temperature divided by the height difference between these two temperatures (a positive value here indicates that temperature is increasing with height which is opposite to the sign convention typically used in meteorology). c) Height difference between the inversion temperature (warm) and the minimum boundary-layer temperature (cold). d) Minimum boundary-layer temperature, which typically occurs at the surface. To determine whether the inversion characteristics have changed since 1961, the time series for various parameters that define the inversion are plotted in Fig. 27. Before 1985 there were years with inconsistent reporting of altitudes in the lowest 1 to 2 km of the sounding. For this reason we restrict the discussion of any trends to post-1985. The inversion temperature, defined as the temperature at the warmest level in the sounding, increased by about 2 C over the last 22 years. The amount of warming of the inversion temperature (0.93 C per decade from Fig. 27a) is somewhat less than for the minimum boundary-layer temperature (1.47 C per decade from Fig. 27d) which usually occurs at the surface and is consistent with the warming recorded by the surface air temperature (Fig. 11a). Since the inversion thickness remains nearly the same (Fig. 27c), the inversion lapse rate has been decreasing which has resulted in a weakening of the stratification near the ground since 1985 (Fig. 27b). Note that the inversion lapse rate is given by the temperature difference between the inversion temperature and the minimum boundary-layer temperature divided by the altitude difference between these two levels. 5 Minimum winter dry bulb and dew point temperatures The correlation between the surface air temperature and the surface relative humidity is given as a scatterplot for the winter months from 1988 to 2007 in Fig. 28. Each point on the scatterplot is an hourly observation. The 47 C dew point line is a lower limit for almost the entire dataset with only two exceptions. Unfortunately this is an instrument issue in which the dew point temperature is recorded as missing when the dry bulb temperature is less than about 47 C. Figure 28 has 36,559 points plotted but is missing another 1112 points that are colder than -47 C because of the missing dew point temperatures. The coldest post-1987 dry-bulb temperature was - 52.2 C on 30 January 1989. Figure 28 shows that, as the air temperature drops below 35 C, the frequency of supersaturated air with respect to ice increases rapidly. By 45 C the air is always supersaturated with respect to ice. This is consistent with the nearly constant reporting of ice crystals by the weather observer during the winter months (Lesins et al., 2009) and suggests that the dehydration process proposed by Blanchet and Girard (1995) Canadian Meteorological and Oceanographic Society

78 / G. Lesins et al. Fig. 28 Scatterplot of winter hourly observations from 1988 to 2007 of the dry bulb surface temperature and the surface relative humidity with respect to liquid water. The line labelled Ice Sat is the saturation water vapour pressure line with respect to ice. The three lines labelled Tdew are lines of constant dew point temperature. The paucity of points for dew points less than 47 C is an artifact caused by the missing humidity measurements at very cold temperatures. is able to proceed. In the dehydration process water vapour is removed by deposition onto ice crystals which slowly precipitate out of the air column. It is interesting to consider what determines the minimum surface temperature attainable in the winter. Figure 29 is a scatterplot of the surface air temperature versus precipitable water based on individual concurrent measurements. Only times with clear sky conditions and calm surface winds are plotted to eliminate the warming effect of the additional downward infrared irradiance from clouds and precipitation and the warming effect of a downward flux of turbulent sensible heat caused by a surface wind. It shows that the air temperature is strongly correlated to the precipitable water and that water vapour exerts a strong greenhouse forcing even during the coldest time of the year. 6 Discussion and conclusions A cooling of about 2 C took place in the annual average surface temperatures from 1954 to the early 1970s. The period from 1972 to 1987 was colder than average with the exception of 1981. Since 1972, the surface annual average temperature has increased by 3.2 C with all but one year in the 2000s being above the complete dataset average. The warming during the last 20 years has mostly been driven by the winter and autumn seasons. The large warming that has taken place since the mid- 1980s is not a direct result of the positive feedback between surface albedo and air temperature because there is no sunlight in the winter. Indeed the warming is smallest during the summer when the sun exerts its greatest forcing. The winter is particularly sensitive to increasing downward radiative forcing at the surface because of the strong surface-based temperature inversion which prevents the warming from mixing easily to higher altitudes thereby amplifying the effect near the surface. This is in sharp contrast to tropical latitudes where the near moist adiabatic lapse rate allows surface heating to be transported throughout the entire tropospheric column. There is an interesting winter warming period from 1975 to 1978 evident in the surface temperature time series. The warming is also evident at 500 mb and in the 900 to 500 mb thickness. The boundary layer inversion remains at the same strength with both the surface and inversion temperatures increasing about the same amount. The one distinction between this warming and the most recent warming is that the 1970s winter warming was accompanied by an increase in precipitable water. Hence, one possible explanation is that meridional transport was enhanced for several years during the winter bringing with it warmer and moister air that then radiatively heated the surface. La Société canadienne de météorologie et d océanographie

Climate Trends at Eureka in the Canadian High Arctic / 79 Winter (1961 2007) Fig. 29 Scatter plot of winter dry-bulb surface temperature and the precipitable water using observations at radiosonde release times for the winter months from 1961 to 2007 when the skies were clear of any clouds or other weather obscurations and when surface winds were calm. The solid line is the linear regression. The winter season surface pressure has a strong negative correlation with the NAO index yielding a correlation coefficient of 0.62 but the negative correlation with surface air temperature is not significant. The second half of the surface hourly record shows a 50% reduction in the frequency of strong winter anticyclones with pressures greater than 1035 mb. This observation can be explained by one or both of the following: 1) a geographic shifting of strong anticyclones events away from Eureka, or 2) a weakening of the strength of the anticyclone centres. The annually averaged surface dew point temperatures and relative humidities show considerable noise that make any trends too difficult to discern. This is also due to the limited period of reliable measurements, covering only 20 consecutive years. Autumn shows the strongest indication of an upward trend in dew point temperatures since 1988. Winter dew point and relative humidity trends cannot be used because of missing reports when the dew point temperature dropped below 47 C. Relative humidities have decreased slightly but this is being driven mostly by surface warming. Both the air and dew point temperatures are more strongly constrained in summer due to the moderating influence of open water and sea ice in the nearby Arctic Ocean. The annual average surface wind speed has decreased about 20% since 1954 although most of this decrease took place in the first ten years of the record. The rest of the decrease took place during the late 1990s. The surface wind direction shows a strong preference to be from the west or the southeast which is forced mostly by the local topography. The winter surface-based temperature inversion has weakened somewhat since 1985. This is caused by stronger warming taking place at the surface than at the top of the inversion layer. Although this trend is currently small, if it continues and the static stability of the boundary layer continues to weaken then the turbulent vertical fluxes through the boundary layer can be expected to increase. Greater mixing will allow more free tropospheric air to reach the surface which will modify the vertical profiles of temperature, wind speed, water vapour, and aerosols. Before 1980 the static stability was significantly weaker and this was a period during which surface wind speeds were higher. Precipitable water has increased by about 10% since 1961 in all seasons except winter. The 500 mb temperature has increased by about 1 C since 1961 particularly in spring and autumn. The thickness between 900 mb and 500 mb has increased by about 18 m since 1961. No trends are discernible around 300 mb which is near the troposphere. A strong positive correlation was found between the surface air temperature and precipitable water during the winter months when the skies were clear and the surface winds were Canadian Meteorological and Oceanographic Society