Canadian historical and homogeneous temperature datasets for

INTERNATIONAL JOURNAL OF CLIMATOLOGY
Int. J. Climatol. 19: 1375–1388 (1999)
CANADIAN HISTORICAL AND HOMOGENEOUS TEMPERATURE
DATASETS FOR CLIMATE CHANGE ANALYSES
LUCIE A. VINCENT* and D.W. GULLETT
Climate Research Branch, Atmospheric En6ironment Ser6ice, En6ironment Canada, 4905 Dufferin Street, Downs6iew,
Ont., Canada M3H 5T4
Recei6ed 6 May 1998
Re6ised 9 February 1999
Accepted 16 February 1999
ABSTRACT
The Canadian Historical Temperature Database (CHTD) was developed to meet the need for detailed individual
station datasets and to produce an improved historical climate change database. It contains datasets of monthly mean
maximum and minimum temperatures for 210 Canadian stations. Stations were selected on the basis of length of
record, data completeness, and spatial distribution across the country. Records from separate stations were sometimes
joined to temporally extend their series backward. Missing data gaps were estimated using highly correlated
neighbour stations. Relative homogeneity was assessed using a Canadian developed technique based on regression
models. Nonclimatic steps resulting from station or site relocations were identified and quantified. Data adjustments
were performed for large steps (greater than 0.6°C) with or without metadata and for medium steps (0.4 – 0.6°C) with
support. A bias in minimum temperatures was also identified and adjusted at principal stations located in eastern
Canada. The bias results from a change in observing procedure in 1961 throughout the country, and it typically
produces a decreasing step of about 0.6–0.8°C in the annual series in the eastern part of the country. Although also
detectable in western Canada, it appears to be small there and no bias adjustments were performed in the western part
of the country. Large temporal and spatial differences in data availability exist between areas north and south of
60°N latitude making national analyses possible for only the latest 50 years of data. Spatial presentation of the linear
trends before and after adjustments shows overall improvement to the regional and national trends in terms of spatial
consistency. The CHTD contains the best available monthly temperature data in Canada and these datasets are now
available to the climate change research community. Copyright © 1999 Royal Meteorological Society.
KEY WORDS: Canada;
homogeneity; trends; time series; temperature; database
1. INTRODUCTION
Recent intense media coverage of global climate change and related issues, along with heightened
scientific activity, and a perception of increasing extreme climate events, has created a sense of public
urgency for more detailed information on climate change in Canada. Reliable historical climate datasets
are essential for accurate trend and variability analysis, for verification of regional and global climate
models, for ground-truthing of remotely sensed data from space platforms, and ultimately for detection
and attribution of climate change. There remain, however, difficult technical problems for the analysis of
these datasets, brought about by observing site modifications or closures, changing observing procedures
and instruments, changing numbers of stations over time, and most recently, downsizing of traditional
climate networks along with increasing use of automated instrumentation.
High quality station datasets with good temporal and spatial resolution throughout the regions of
Canada are essential for reliable climate analyses at local, regional, and national scales. These national
datasets are also an important contribution to the global datasets needed for climate change detection at
the global scale. For historical analyses, large temporal and spatial disparities in data availability exist
* Correspondence to: Climate Research Branch, Atmospheric Environment Service, Environment Canada, 4905 Dufferin Street,
Downsview, Ont., Canada M3H 5T4; tel.: + 1-416-7394337; fax: +1-416-7395700; e-mail: [email protected]
CCC 0899–8418/99/121375 – 14$17.50
Copyright © 1999 Royal Meteorological Society
1376
L.A. VINCENT AND D.W. GULLETT
between areas north and south of 60°N latitude. These are directly attributed to the highly uneven
population distribution, the brief human history, and the geographic vastness and physical diversity of the
country. In northern Canada, the average station density is about one station per 100000 km2 of land
area, and the longest available period with an adequate number of stations is 1945–1995. South of 60°N,
it is about one station per 30000 km2, and the longest period is 1895–1995. Some sparse instrumental
datasets also exist prior to these dates in various regions of the country, but they are highly irregular in
both time and space and are not considered suitable for historical analyses.
An important task in creating representative climate datasets is the proper identification and adjustment
of nonclimatic inhomogeneities. In the late 1980s, Gullett et al. (1990) investigated an approach for
homogeneity assessment of temperature series based on the early work of Mitchell (1961). A new
homogeneity assessment procedure was subsequently developed by Vincent (1990), and used to assess the
homogeneity of maximum and minimum temperature series at over 350 Canadian stations (Gullett et al.,
1991). This led to the creation of the Historical Canadian Climate Database (HCCD) (Gullett et al., 1992)
which comprises of temperature data for 131 stations reliable for regional and national scale analyses. The
HCCD was widely used and reported on in a variety of scientific documents: the State of the Climate
Reports (Environment Canada, 1992, 1995), the Climatological Bulletin (Gullett et al., 1992; Skinner and
Gullett, 1993), Trends ’93 (Boden et al., 1994) and the Climate Trends and Variations Bulletin
(Environment Canada, 1998). Recently, the need for more stations and updated datasets led the authors
to the creation of a second version of the historical database.
The purpose of the present paper is to describe the new and improved Canadian Historical Temperature Database (CHTD). Long-term, complete and homogenised, monthly mean maximum and minimum
temperature datasets for 210 individual stations that are relatively evenly distributed across the country
have been assembled. Description of station selections, methodologies used for temporal extension of
datasets, estimation of missing data, and identification and adjustment of inhomogeneities are provided.
Results from the homogeneity assessment and the impact of data adjustments on the temperature trends
for the period 1945 – 1995 are also presented. These datasets are suitable for a wide variety of climate
change analyses in Canada and they are now available to the climate change community.
2. DATA
The National Climate Data Archive (NCDA) of Environment Canada (Atmospheric Environment
Service, 1987) contains quality controlled daily data for over 5000 Canadian stations covering different
periods of time. Only about half of these stations are presently in operation. For this project, daily
maximum and minimum temperatures were retrieved for about 2000 stations; 210 stations were selected
as base stations for inclusion in the CHTD and the remainder were used for estimation of missing values
and homogeneity testing of the base stations. Monthly mean maximum and minimum temperatures were
then calculated from the daily values. If more than three consecutive daily values or five random daily
values were missing in a month, the monthly mean was not calculated, but instead it was flagged as
missing. This procedure is currently used to archive monthly temperature values in the NCDA. Monthly
series of maximum and minimum temperature, containing missing data and without any form of
adjustment at this stage, were input as original datasets for this project. Further processing of these is
described later in the paper.
3. SELECTION OF STATIONS
Stations were selected to represent all major climate regimes of Canada (Phillips, 1990). The main
considerations were the following: spatial distribution of the stations over different climate regions and
across the country; availability of the longest possible records within the period 1895–1995; completeness
of the records (usually less than 5% missing data). The 131 stations of the HCCD (Gullett et al., 1992)
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
CANADIAN HOMOGENEOUS TEMPERATURE DATASETS
1377
were given initial priority since they had been previously selected on the basis of their strategic locations,
length of record, and availability of data in near-real-time (data immediately available to users for
monitoring purposes). In addition, Canada’s Reference Climate Stations (Environment Canada, 1996)
were also considered since they had been designated in compliance with World Meteorological Organization guidelines (World Meteorological Organization, 1986), and were identified specifically for their
suitability for climate change studies. The stations of the Global Climate Observing System (Environment
Canada, 1996) were also considered for inclusion in the CHTD. It is hoped that such internationally
recognised and supported sites would be less likely to succumb to administrative or political pressures to
close, relocate, or other changes that would modify their physical surroundings.
Consideration was also given to ‘principal’ versus ‘ordinary’ climate stations. Principal stations are
often airport stations with hourly observations, available in near-real-time, but usually having a short
period of record (about 50 years or less). Ordinary stations are often run by volunteers, have daily
observations and frequently have long periods of record (between 50 and 100 years), but they are not
available in near-real-time. Principal and ordinary climate stations have different observing procedures,
practices, and administrative regulations, and the impacts of most of these differences are evident in the
use of the datasets. Principal stations are also subject to a bias in their minimum temperatures, addressed
in detail in a subsequent section; ordinary stations are not. Finally, it is important for future studies that
all CHTD sites be currently operational and have reasonable prospects for continuation.
All of these considerations led to the selection of 210 stations. Figure 1 shows the spatial distribution
of the stations and the major Canadian political boundaries with their provincial/territorial names and
corresponding two-letter abbreviations that are referenced occasionally in the paper. A relatively uniform
and dense network of stations exists across southern Canada (south of 60°N) while across the north the
network is less dense but still relatively uniform. There are also temporal disparities between north and
Figure 1. Distribution of the 210 stations of the Canadian Historical Temperature Database. Provincial/territorial names are given
with corresponding two-letter abbreviations
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
1378
L.A. VINCENT AND D.W. GULLETT
south. With the exception of a few sites located in the western part of the Northwest Territories, no
long-term northern datasets are available extending back prior to about 1945. This paucity of northern
data has created and continues to create serious difficulties for large-scale national climate change
assessments in Canada. On the other hand, southern Canada contains a reasonable selection of sites
covering the full period 1895 – 1995. There are a few cases in which CHTD sites are close to one another
(only a few kilometre separation) in apparent violation of the concept of a uniform distribution, however,
this is intentional on the author’s part. Most are Canada Department of Agriculture (CDA) sites
(ordinary stations) with ‘sister’ sites located at nearby airports (principal stations), for example, Charlottetown CDA and Charlottetown A, PE. CDA sites are generally long-term sites located in rural
environments, while airport sites frequently have shorter records and are sometimes located in semi-urban
or ‘small town’ environments as defined in Peterson and Vose (1997). These two types of sites usually
have about 50 years of coincident data which is good for comparative studies. While the CDA sites have
the longest records, they are not usually available in near-real-time. The airport sites, on the other hand,
with their shorter record lengths are almost always available in near-real-time, which is essential for
monitoring purposes.
Station selections were carried out so as to minimise the urban effect. Although some Canadian airports
are located near cities (population \ 50000), a number are also near small towns (10000–50000 people)
and many are located in rural areas (population of B 10000). They are usually located outside the city
limits, well away from the city and therefore usually do not contain a significant urban warming
component. In the CHTD, about 10% of all stations are affiliated with cities, 10% with small towns and
80% are rural. Only one station, Montreal McGill, QC, is likely to exhibit significant urban warming. It
was included as a special case because of its 100+ year record at the same location with the same agency
throughout its history. Other major centres where a strong urban effect is likely are Toronto, ON and
Vancouver, BC, but these have not been included in the CHTD at this time.
For each of the 210 base stations, multiple neighbours, usually 2–4, are needed to fill missing data gaps
and for relative homogeneity assessment. Neighbour stations were originally selected within 50–100 km
of each base station to ensure that they shared the same climate regime. Sometimes, however, it was
necessary to search up to 300 km or more from a base station in order to find suitable neighbours. This
was especially true in northern Canada where on occasion it was necessary to go as far as 800 km.
Fortunately this was the exception rather than the rule. Neighbours were not used unless they were highly
correlated with the base (\ 0.70 on their annual values over 30 years or more). Across the far north,
correlations on annual values tend to be high, since the stable Arctic High is a semipermanent
atmospheric feature of this region. Coastal and mountain areas are subject to greater climatic variation,
and topography was a major consideration in the selection of neighbour stations. High altitude stations
were rarely used, and in the mountainous regions of British Columbia and Yukon Territory, base and
neighbours are located in the valleys.
4. TEMPORAL EXTENSION OF RECORDS
There are many reasons that make it necessary and desirable to ‘join’ historical climate records but the
primary one is the need for ‘single’ series spanning as much of the 1895–1995 period as possible.
Sometimes records from different stations are joined either because the individual records are too short
or there are large gaps within the single station record. At other times, one station may close and re-open
at a later date. Often in the late 1940s and early 1950s, rural climate stations were closed in Canada, as
elsewhere, and were simultaneously re-opened as airport stations some distance away. A classic example
is Moncton and Moncton A, NB, which moved to its airport site in 1939. More recently, in the late 1980s
and early 1990s, many sites fully or partially ceased human-based observations and began machine-based
observations. Fort Reliance, NT and Ennadai Lake, NT, are good examples of this. Both are isolated
human-based principal stations in continuous operation from the early 1940s until 1989 and 1979,
respectively, when they were closed and re-opened using automated observing systems. At any one station
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
CANADIAN HOMOGENEOUS TEMPERATURE DATASETS
1379
Figure 2. Number of stations in the Canadian Historical Temperature Database reporting during the period 1895 – 1995 in northern
Canada (north of 60°N) (long dash), in southern Canada (south of 60°N) (short dash), and for the entire country (solid line)
in the CHTD, no more than two or three segments are joined, and over half of the 210 stations have no
joins at all.
Certain conditions must be met before stations are joined. Potential candidates must be located within
a few kilometres of one another, be at similar elevations and exhibit similar climate characteristics. The
F-test is used to determine whether the variances before and after a potential join are statistically
different, and segments failing the test are not joined. Since the means of each segment, before and after
a join, are often different, an artificial step is frequently created by the join. This was not considered a
problem, since the step would be identified and adjusted during the homogeneity assessment procedure.
Joining of stations proved to be extremely useful for creating additional long-term datasets to assist with
missing data estimation and for facilitating long-term homogeneity assessment.
Figure 2 shows the number of stations included in the database over time. There are 47 stations
covering the full 1895 – 1995 period of time. The interval of most complete data coverage is between 1945
and 1995 when the largest number of stations was reporting. This was generally a period of administrative
and economic stability in climatological organisations resulting in the expansion of networks in Canada
and around the world. A secondary period of relative stability can be seen from about the mid-teens to
the mid-1940s, for southern Canada, while prior to this time there were relatively few stations reporting
anywhere in the country. Less confidence is placed in the analyses of the early Canadian data because
stations were fewer, station inspections were rare or nonexistent, and instrument siting conditions were
highly irregular and somewhat ‘creative’ to say the least. Such inconsistencies as the instrument enclosure
mounted on the wall of a house, or the enclosure in extremely poor physical condition were typical. Other
common problems in the early days included excessive nearby vegetation or buildings near the instruments, and irregular times of observation. The effects of such irregularities are often visible in the early
portions of the data series in the form of a sudden change in the mean, a change in variability, or the
occurrence of an unexplained isolated extreme value (outlier).
5. MISSING DATA ESTIMATION
Correlation between base and neighbouring stations on annual values and over the 30 years or more of
coincident data were at least greater than 0.70 and most often greater than 0.90. Usually no more than
5% of the total number of monthly values were missing, and most of the time these were scattered
throughout the dataset. Sometimes, a sequence of missing months or years occurred at either the
beginning or the end of the series. Missing values at the beginning are usually related to program
irregularities such as those mentioned in the previous section, while missing values at the end are most
often related to station automation and a resultant loss of data making its way into the archive. Hopefully
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
1380
L.A. VINCENT AND D.W. GULLETT
technical problems associated with automation, archiving, and retrieval of data will be resolved in the
near future. Modifications are under way and improvements have already been realised during the past
year.
The methodology used for monthly data estimation is based on the World Meteorological Organization
average difference method described by Thom (1966). The series of the monthly values (same month,
different years) of a number of neighbouring stations are used to estimate the missing monthly value at
the candidate station. The differences between monthly values of candidate and neighbour are averaged
over the coincident interval of 5 years before and 5 years after the missing value. This average is then used
as a correction factor to the corresponding monthly value at the neighbour to estimate the missing value
of the candidate. This procedure is repeated using each neighbour and the mean of all estimates is used
as a final estimate of the missing value.
6. HOMOGENEITY TESTING AND ADJUSTMENT PROCEDURE
A new technique has been recently developed for the identification and adjustment of inhomogeneities in
Canadian temperature series (Vincent and Gullett, 1997; Vincent, 1998). The technique compares the
series of a base station with those of a number of neighbours, and identifies patterns in the differences
between the series. In this project, two regression models are applied in succession, the first to determine
if the base is homogeneous, and the second to determine if there is a nonclimatic step. The dependent
variable is the annual temperature series at the base and the independent variables are the annual
temperatures of the neighbours. In the second model, there is an additional independent variable to
describe and measure a potential step. After the application of each model, the residuals are analysed to
assess the fit. The autocorrelation for residuals several lags apart are obtained, and the graph of the
autocorrelation rk against lag k is produced with an approximate 95% confidence interval. Coefficients
outside the confidence interval are significantly different from zero at the 5% level. If the autocorrelations
are not different from zero after the application of the first model, the base series is homogeneous for the
tested interval of time. On the other hand, if consecutive significant autocorrelations are identified at low
lags, the first model is rejected and the second is applied instead to identify a step. The series is then
divided at the step, and each segment is tested separately starting with the first model. In this way, the
procedure divides and subdivides the series until each segment is determined to be homogeneous.
The success of homogeneity testing is highly dependent on the reference series. In Canada, there is a
large number of stations that have very few (perhaps 2–4) neighbours to use for homogeneity testing. At
first, the models are applied in station pairs to be able to identify steps belonging to the base and those
associated with the neighbours. The graph of the difference between base and neighbour is also very
useful to visualise the inhomogeneities. Often, it is necessary to either correct inhomogeneities in
neighbour series before testing the base or to discard a neighbour that appears to have too many
problems. Good detective skills are sometimes required on the part of the analyst when starting the
process in a new area, and it is often necessary to re-test and re-assess some stations as new information
is revealed by the process. During the creation of the CHTD, however, it was always possible to finalise
each analysis with confidence.
Adjustments are applied to bring each homogeneous segment into agreement with the most recent
homogeneous part of the base series. Results from testing the technique on a large number of simulated
annual temperature series (Vincent, 1998) were used as guidance in formulating an adjustment decision
rule. It was shown that the ability of the technique to properly identify a step was greatly dependent on
the magnitude of the step and that historical support for the cause of the inhomogeneity was sometimes
helpful in determining whether adjustment should be applied. The following rule was therefore implemented: steps greater than 0.6°C, identified in the annual series, are always adjusted regardless of the
presence or lack of supporting station history; steps between 0.4 and 0.6°C are adjusted only with
supporting station history; and steps less than 0.4°C are not adjusted since they are not often adequately
identified by this or any other technique. Although the original testing is carried out on the annual
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
CANADIAN HOMOGENEOUS TEMPERATURE DATASETS
1381
datasets, correction factors are derived on monthly values. After the identification of the step in the
annual series, the second model is applied to the 12 monthly series (same month, different years)
separately to estimate the 12 correction factors at the identified step. Changes around the site can produce
inhomogeneities of varying magnitudes in the different seasons, and by adjusting the monthly values, it
was possible to avoid the effects of seasonality.
Although the value of the historical metadata information to the homogeneity and adjustment process
is well documented (Jones et al., 1986; Karl and Williams, 1987; Gullett et al., 1991), it was hoped to be
able to minimize the use of either the electronic or paper metadata files retained in the NCDA. In fact
the authors have found them, as in previous work, to be an important component of the process. It was
often necessary to seek out metadata support information from these files to help with the interpretation
and verification of the homogeneity results. Greater confidence was usually placed in the data adjustments
if they could be supported with solid historical information. After many years of working with these data,
the authors believe that the historical metadata information is and will continue to be an important
component of the homogeneity and adjustment process, and therefore, they strongly recommend that
national meteorological services take action to protect this irreplaceable source of information.
As a final step in the process, linear trends are calculated at all stations after adjustment and over
different time intervals, for interstation comparison purposes. The assumption is that temperature trends
for stations within a homogeneous area and calculated for matching time periods should be of similar
magnitude. As a result of this exercise, it was sometimes necessary to re-assess and re-adjust some
temperature series. However, the purpose of this last step was not to attempt to quality control the
homogenised datasets, but only to identify the stations in which temperature trends significantly differed
from their surroundings. Agreement between the trends from stations located in the same climate regime
once again enhanced confidence in the data adjustments.
Figure 3 illustrates a typical station homogeneity assessment and adjustment situation. The annual
mean minimum temperatures for Saint John, NB, are tested for homogeneity for the period 1895–1995.
The series shows a decreasing linear trend of 2.4°C/101 years before adjustment (Figure 3(a)). Four steps
are clearly identified by the procedure (Figure 3(b)). The step in 1961 results from the change of observing
window which produced a bias in the minimum temperature in 1961 at principal stations (described in the
following section); the step in 1950 is associated with a site relocation with a 13-m increase in elevation;
the step in 1947 is caused by the joining of two segments, Saint John and Saint John A in 1947; and
finally the step in 1915 is caused by another site relocation with an increase in elevation of 9 m in 1914.
All of these steps are large enough to be adjusted with confidence without metadata support, however,
supporting history information was readily available to determine their causes. The adjusted series for
Saint John, NB, shows a temperature increase of 0.2°C/101 years (Figure 3(c)), compared to a decrease
of 2.4°C/101 years in the unadjusted series. Increasing linear trends of about this magnitude are also
observed at neighbouring stations, and it is concluded that the adjusted series reflects more closely the
temperature variations of the area.
7. ADJUSTMENT FOR A BIAS IN MINIMUM TEMPERATURES
At principal stations throughout eastern Canada, from the Manitoba–Ontario border to the Atlantic
coast, the homogeneity procedure identified a step decrease in 1961 of about 0.6–0.8°C in the annual
mean minimum temperature series. This step decrease, known as a ‘bias in minimum temperatures’, has
been addressed by Burrows (1964), Bootsma (1976), Schaal and Dale (1977), Karl et al. (1986) and Lewis
(1996). This bias accentuates the cooling trend over eastern Canada for the past 50- and 100-year periods.
The bias resulted from a nation-wide change in observing procedure at principal stations in 1961. At that
time, it was decided to change the reporting of the daily maximum and minimum temperatures from the
00Z-00Z climatological day to the 06Z-06Z day, ostensibly because it was easier for the public to
understand. The 06Z hour corresponds more closely to the time when the actual minimum temperature
is more likely to occur (i.e. somewhere between midnight and 07:00 LST). The closer the actual time of
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
1382
L.A. VINCENT AND D.W. GULLETT
occurrence of the minimum temperature to the 06Z observation hour, the greater the likelihood that the
same minimum temperature will be recorded on two successive calendar days (Bootsma, 1976). Cameron
and Wilson (1996) assessed the frequency of occurrence of the bias and its magnitude through case studies
in which the daily minimums were calculated for both observation windows (00Z and 06Z) using original
synoptic data extracted from microfilm held in the NCDA. Preliminary results based on limited samples
Figure 3. (a) Annual mean minimum temperature for Saint John, NB, 1895 – 1995: the dashed line shows a decreasing linear trend
of 2.4°C/101 years; (b) difference between the annual mean minimum temperature and the reference series: the long dashed lines
indicate the mean of each segment; (c) adjusted annual mean minimum temperature for Saint John: the dashed line shows an
increasing linear trend of 0.2°C/101 years
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
CANADIAN HOMOGENEOUS TEMPERATURE DATASETS
1383
show about 40% of the daily minimum temperatures were affected for stations in Ontario and Atlantic
Canada, 34% for those on the Prairies and 22% in south central British Columbia. Annually, magnitudes
were typically from 0.6 to 0.8°C in the east and much smaller at 0.2°C in the west. The decrease in
frequency and magnitude westward is expected since the 06Z-06Z window aligns more closely with the
evening hours in local time in the west, than with the early morning hours, as in the east.
The homogeneity technique was successful most of the time in identifying these nonclimatic negative
steps in eastern Canada. In a few instances, no bias could be detected at some principal stations, such as
at Sable Island, NS, for example, where it is believed that the moderating maritime influence overwhelms
and offsets the cold temperature phenomenon. In western Canada, a significant bias in minimum
temperature was not identified using this technique: magnitudes of 0.2°C or less were sometimes detected
in the annual series, however, these small steps were below the lower limit for adjustment. The
adjustments were performed in the following way. Since the most recent period (1961–1995) is in error
(period in which sometimes the same minimum is recorded on two successive calendar days), the
correction factors were calculated and applied to bring the 1961–1995 period into agreement with the
second most recent homogeneous part of the series following the procedure described in section 6. The
minimum temperature bias problem warrants further investigation in the various regions of Canada, but
for the present time, the authors are satisfied that they have addressed the problem adequately for the
purpose of creating homogeneous monthly and annual temperature datasets.
8. TABULATED RESULTS OF HOMOGENEITY ASSESSMENT
Maximum and minimum temperature series for the 210 CHTD stations were assessed for homogeneity
and adjusted where necessary. Results of the homogeneity assessment were tabulated for each station in
the following way (not presented in this paper). Station name and archive number of each data
station/segment are provided along with the year of each identified step. Beginning and ending dates of
each adjusted interval, period to which adjustment is made, and magnitude of adjustment are also given.
Finally, possible factors or causes contributing to each inhomogeneity are listed if known. These factors
have been extracted from the station history metadata files when available and when sufficiently complete.
Because metadata searches tend to be extremely labor-intensive, it was usually considered unnecessary to
search the metadata files for supporting documentation when steps greater than 0.6°C were identified.
However, supporting information was often already known either from sources such as the climatological
station catalogue (Environment Canada, 1989), or from previously recorded work.
9. DISCUSSION
In southern Canada, a total of 172 stations or 344 maximum and minimum temperature series were
assessed, and it was found that only 36% of these were homogeneous (or, in other words, did not need
adjustment). In the north, with 38 stations or 76 series, 55% were found to be homogenous. The higher
percentage in the north is partly due to the shorter record length (usually 50 years or less). In southern
Canada, it was also found that the most recent 50-year period tended to be more homogeneous than the
earlier period. Few stations were homogeneous in both maximum and minimum temperatures and it was
observed that physical changes at the observing site do not necessarily result in identical inhomogeneities
in both series. The reason is that different local conditions and site characteristics often affect daytime
(maximum) and night-time (minimum) temperatures differently. Usually no more than two or three steps
were adjusted in any one record, and only a few stations had as many as four or five adjustments. By far
the most common reason for the occurrence of steps is station or instrument relocation with an
accompanying change in elevation and/or in siting characteristics. Station joins are in this category.
Frequently, a station join resulted in a significant inhomogeneity in either the maximum or minimum
series, but not necessarily in both. It was also common to find a small step of about 0.5°C or less in the
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
1384
L.A. VINCENT AND D.W. GULLETT
early 1990s which was strongly suspected to be related to station automation. Most of these steps are too
close to the end of the series to be adjusted with confidence at this time, and they should be re-examined
in future studies after an additional 3 – 5 years of data have been observed and archived.
In the case of isolated stations, such as those located in northern Canada, it is realised that
homogeneity assessment is not as precise as in the south; never-the-less, the authors were able to identify
and adjust for large steps in some northern stations with a reasonable degree of success. An example is
Resolute A, NT, a key station located in the geographic centre of the high Canadian Arctic. For this site
it was necessary to use Mould Bay, NT, about 700 km to the WNW and Cambridge Bay, NT, about the
same distance to the SSW, as neighbours. All three stations are in fact in the same large-scale climate
regime and in the same ecoclimatic region at the smaller regional scale. Annual temperature correlations
for these stations were high, greater than 0.70, and the authors were able to clearly identify separate steps
in the maximum and minimum series that are obviously related to site relocations. The various
deficiencies of the northern stations, with records of only 50 years or less, are somewhat offset by the fact
that they usually have fewer site relocations, their record length is short, and they are located in rural
environments. For these reasons, isolated stations are identified as homogeneous more often than their
southern counterparts, a fact that should be kept in mind when using these datasets.
Linear trends before and after adjustments were calculated for the annual maximum and minimum
temperatures for the period 1946 – 1995. They were plotted and contoured in an attempt to examine the
impact of the adjustments on various regions of the country. Spatial representation of the maximum and
minimum temperature trends are presented in Figures 4 and 5, respectively. It is obvious from the patterns
that the datasets before adjustments are in disagreement in time and space, especially in the case of the
minimums. The ‘before’ maps (Figures 4(a) and 5(a)) are more noisy, even when using the relatively
coarse isotherm interval of one degree. The much smoother field of the adjusted datasets (Figures 4(b)
and 5(b)) shows a marked improvement in spatial homogeneity over the original NCDA data.
These analyses also clearly show the broad-scale temperature warming–cooling patterns over the past
50 years in Canada in both the daytime (maximums) and night-time (minimums) temperatures. The
well-known pattern of warming in western Canada and cooling in the eastern regions, separated by a
broad north–south ‘neutral’ zone from the high Arctic Islands in the north, to the lower Great Lakes
region in the south, is reaffirmed by this latest analysis. But some significant differences from earlier
analyses (Skinner and Gullett, 1993) also exist, namely with regard to the cooling in the east. Most
previous analyses using either the NCDA or the HCCD data have shown a large area of cooling
throughout much of eastern Canada similar to that shown in Figure 5(a). With the newly homogenised
datasets, that include for the first time adjustments for the bias in minimum temperatures, the cooling is
much less pronounced in terms of its magnitude and in its spatial extent (Figure 5(b)). In the author’s
opinion, the present study provides a more realistic representation of the 50-year temperature trends in
Canada, and is a significant improvement over previous analyses by Canadian researchers or by others
using Canadian datasets. The present results continue to support the cooling over eastern Canada as seen
in both daytime and night-time annual mean temperature analyses. At the same time, however, the
authors also assert that in the case of the night-time cooling, the magnitude and spatial extent is
somewhat less than previously believed.
10. SUMMARY AND CONCLUSION
This paper summarises the creation of a new and improved homogeneous database, the CHTD,
containing monthly and annual, maximum and minimum temperatures for 210 stations across Canada.
These datasets were gap-filled and assessed for relative homogeneity against highly correlated neighbours.
They were adjusted when necessary to remove nonclimatic step inhomogeneities. The bias in minimum
temperatures at principal stations was also identified and adjusted. Linear trends were fit to the resulting
homogeneous time series and compared with neighbours as a final homogeneity check. Spatial analyses
of the trends before and after adjustments, over the most-recent 50-year period, show the impact of
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
CANADIAN HOMOGENEOUS TEMPERATURE DATASETS
1385
Figure 4. Linear trends for annual mean maximum temperatures, 1946 – 1995; (a) before and (b) after adjustments. The slashed area
indicates trends less than − 0.5°C/50 years, the white area trends – 0.5 to 0.5°C/50 years, the light grey area trends 0.5 – 1.5°C/50
years and the medium grey area trends greater than 1.5°C/50 years
adjustments on the temperature trends. Original and adjusted data are available from the database, and
there has never been any intention of replacing the data in the NCDA. Data flags are used when a missing
monthly value has been estimated, or a value has been adjusted as a result of homogeneity assessment.
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
1386
L.A. VINCENT AND D.W. GULLETT
Figure 5. Linear trends for annual mean minimum temperatures, 1946 – 1995; (a) before and (b) after adjustments. The crossed area
indicate trends less than − 1.5°C/50 years, the slashed area trends −1.5 to − 0.5°C/50 years, the white area trends − 0.5 to
0.5°C/50 years, the light grey area trends 0.5–1.5°C/50 years, the medium grey area trends 1.5 – 2.5°C/50 years and the dark grey
area trends greater than 2.5°C/50 years
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
CANADIAN HOMOGENEOUS TEMPERATURE DATASETS
1387
The CHTD is a specialised database designed for climate trends and variability analyses in Canada. It can
be used for spatial and temporal analyses at various scales, either by station or by gridded format. These
datasets will also be of considerable value for validation of regional climate model outputs at different
scales.
Large spatial and temporal differences exist in the stations north and south of 60°N latitude. In
northern Canada, with the exception of a few stations in the western Arctic, the longest possible period
of continuous homogeneous data is 1945 – 1995, and the average land area represented by each station is
about 100000 km2. In the south, the period 1895–1995 is reasonably well represented and each station
covers about 30000 km2. These disparities create serious difficulties for temporal and spatial analysis.
Unfortunately this paucity of data in the north cannot be alleviated. Stations there have always been few
and distant, and the situation has been worsening in recent years as fewer and fewer stations are being
supported as part of the national climate network. Due to the high costs associated with operating the
northern stations, the pressure to close them is extreme and a constant battle ensues between scientists
who use the data and network administrators whose resources are dwindling. In future work, attempts
should be made to extend the temperature series back in time using any data and information available
(e.g. historical, anecdotal, proxy, etc.).
The CHTD incorporates many significant improvements over previous temperature databases such as
the HCCD. There are more stations in this new database (210 vs. 131) ensuring better and more uniform
spatial representation. Far fewer segments (stations) were joined than previously and segments considered
for joining were tested for difference of variances before and after a potential join. As a result of station
joins, there are ten additional stations covering the 1895–1995 period, for a total of 47 stations with the
full 101 years of record. Neighbours were more carefully selected and, as often as possible, they were
chosen from within 50 – 100 km of the base to ensure climate representativeness. They were also subjected
to preliminary homogeneity assessment before their use as reference series. The data estimation routine
for base and neighbour stations was improved with the estimating window using only five years on either
side of the missing datum instead of the entire record as in previous work. Finally, the CHTD contains
long-term, gap-filled and homogenised datasets up to and including 1995 while the HCCD was completed
to only 1989.
Numerous improvements were introduced to the homogeneity assessment procedure and in its
application to the datasets. Results from rigorous testing on simulated temperature series made possible
the formulation of a decision rule for adjustment, thereby ensuring greater objectivity in the application
of the technique. Fewer adjustments were made to the datasets than previously. Only steps greater then
0.6°C were adjusted without supporting metadata documentation and steps of 0.4–0.6°C were only
adjusted when sufficient metadata information was available. Correction factors were derived on the
monthly series instead of using the annual offset which is another significant improvement over previous
work. The CHTD contains minimum temperature datasets that are adjusted for the bias at principal
stations throughout eastern Canada. This has resulted in minimum temperature trends that are lesssteeply negative than previously reported. Finally a major improvement is the trend interstation
comparison among base and neighbour stations after adjustments, as a means of verifying and assuring
homogeneity. The authors have learned from this exercise that these comparisons are essential to the
process of constructing the national database and this step must not be overlooked.
A great deal of research and meticulous data evaluation has gone into the creation of this new database.
The knowledge and experience gained over many years of working with these datasets has proven
invaluable to the process. The CHTD is the best Canadian monthly and annual temperature data
assembled to date. Information on the availability of these datasets can be obtained by contacting the lead
author at the address shown.
ACKNOWLEDGEMENTS
The authors would like to thank Ross Brown and Xuebin Zhang of the Atmospheric Environment Service
for their valuable comments and review of this manuscript. Additional comments and suggestions from
two anonymous reviewers were also very much appreciated.
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)
1388
L.A. VINCENT AND D.W. GULLETT
REFERENCES
Atmospheric Environment Service 1987. Documentation for the Digital Archi6e of Canadian Climate Data (Surface) Identified by
Element, Internal Report, Climatological Services Division, Downsview, Ontario, Canada, 21 pp.
Boden, T.A., Kaiser, D.P., Sepanski, R.J. and Stoss, F.W. (eds) 1994. Trends ’93: A Compendium of Data on Global Change,
ORNL/CDIAC-65, pp. 800–828.
Bootsma, A. 1976. ‘A note on minimum temperature and the climatological day at first order stations’, Atmos., 14(1), 53 – 55.
Burrows, W.R. 1964. ‘Differences in temperature data from ordinary climatological stations arising from once daily readings as
compared to twice daily readings’, in Tech. Circ. 498, Dept. of Transport, 12 pp.
Cameron, T. and Wilson, H. 1996. Minimum Temperature Bias Introduced by a Redefinition of the Climatological Day, Internal
Report, 34 pp.
Environment Canada 1989. Climatological Station Catalogue des Stations Climatologiques, Climatological Services Division,
Downsview, Canada.
Environment Canada 1992. ‘The state of Canada’s climate: temperature change in Canada 1895 – 1991’, in State of the En6ironment
Report No. 92 -2, State of the Environment Directorate, Environment Canada, Ottawa, Canada, 36 pp.
Environment Canada 1995. The State of Canada’s Climate: Monitoring Variability and Change, State of the Environment Report No.
95-1, State of the Environment Directorate, Environment Canada, Ottawa, Canada, 52 pp.
Environment Canada 1998. Climate Trends and Variations Bulletin, Available on the internet at: ec.gc.ca/ccrm/bulletin, Climate
Research Branch, Environment Canada.
Environment Canada 1996. Climate Network Rationalization, Internal Report, National Weather Services Directorate, Downsview,
Canada, 50 pp.
Gullett, D.W., Skinner, W.R. and Vincent, L. 1992. ‘Development of an historical Canadian climate database for temperature and
other climate elements’, Climatol. Bull., 26(2), 125–131.
Gullett, D.W., Vincent, L. and Malone, L.H. 1991. ‘Homogeneity testing of monthly temperature series. Application of
multiple-phase regression models with mathematical changepoints’, in CCC Report No. 91 -10, Climate Research Branch,
Downsview, Canada, 47 pp.
Gullett, D.W., Vincent, L. and Sajecki, P.J.F. 1990. ‘Testing for homogeneity in temperature time series at Canadian climate
stations’, in CCC Report No. 90 -4, Climate Research Branch, Downsview, Canada, 43 pp.
Jones, P.D., Raper, S.C.B., Bradley, R.S., Diaz, H.F., Kelly, P.M. and Wigley, T.M.L. 1986. ‘Northern hemisphere surface air
temperature variations: 1851–1984’, J. Clim. Appl. Meteor., 25, 161 – 179.
Karl, T.R. and Williams Jr., C.N. 1987. ‘An approach to adjusting climatological time series for discontinuous inhomogeneities’, J.
Clim. Appl. Meteor., 26, 1744–1763.
Karl, T.R., Williams Jr., C.N., Young, P.J. and Wendland, W.M. 1986. ‘A model to estimate the time of observation bias associated
with monthly mean maximum, minimum and mean temperatures for the United States’, J. Clim. Appl. Meteor., 25, 145 – 160.
Lewis, P.J. 1996. Personal communication, Environment Canada, Atlantic Region.
Mitchell Jr., J.M. 1961. ‘The measurement of secular temperature change in the eastern United States’, in Research Paper No. 43,
Office of Climatology, U.S. Weather Bureau, 80 pp.
Peterson, T.C. and Vose, R.S. 1997. ‘An overview of the global historical climatology network temperature database’, Bull. Am.
Meteorol. Soc., 78(12), 2837–2849.
Phillips, D. 1990. The Climates of Canada, K1A 0S9, Catalogue No. En56-1/1990E ISSN 0-660-13459-4, Supply and Services
Canada, Ottawa, Canada, 176 pp.
Schaal, L.A. and Dale, R.F. 1977. ‘Time of observation temperature bias and ‘climate change’, J. Appl. Meteor., 16, 215 – 222.
Skinner, W.R. and Gullett, D.W. 1993. ‘Trends of daily maximum and minimum temperature in Canada during the past Century’,
Climatol. Bull., 27(2), 63–77.
Thom, H.C.S. 1966. Some Methods of Climatological Analysis, Technical Note No. 81, WMO No. 199, TP, 103, 53 pp.
Vincent, L. 1990. ‘Time series analysis: testing the homogeneity of monthly temperature series’, in Sur6ey Paper No. 90 -05, York
University, North York, Canada, 50 pp.
Vincent, L.A. 1998. ‘A technique for the identification of inhomogeneities in Canadian temperature series’, J. Clim., 11, 1094 – 1104.
Vincent, L.A. and Gullett, D.W. 1997. ‘Identification and correction of inhomogeneities in Canadian temperature datasets: an
example’, in Proceedings of the 10th Conference on Applied Climatology, October 19 – 23, 1997, Reno, Nevada, pp. 436 – 440.
World Meteorological Organization 1986. ‘Guidelines on the selection of reference climatological stations (RCSs) from the existing
climatological station network’, in WCP-116, WMO/TD-No. 130, World Climate Programme, WMO, 12 pp.
Copyright © 1999 Royal Meteorological Society
Int. J. Climatol. 19: 1375 – 1388 (1999)