OMSZ főoldal |  Szolgálatunkról |  Pályázatok, projektek |  Rendezvények |  Irodalom 
Felbontás: KicsiFelbontás: NormálFelbontás: KözepesFelbontás: NagyobbFelbontás: Nagy Copyright © 
Éva Mekis and Lucie Vincent
Climate Research Branch, Meteorological Service of Canada
4905 Dufferin Street, Downsview, Ontario, M3H 5T4, Canada
Tel: 1-416-739-4374; Fax: 1-416-739-5700; E-mail:
Tel: 1-416-739-4337; Fax: 1-416-739-5700; E-mail:


The identification of local, regional and global climate change has become an important issue in climatology. Temperature and precipitation are the key elements of climate. They are of primary importance for society since changes of climate to warmer or colder, or to wetter or drier conditions could have large social and economic consequences. The availability of long term, continuous and homogeneous time-series for temperature and precipitation provides tremendous advantages for climate researchers. However, most long-term climate datasets contain variations due to non-climatic factors such as relocation of stations, replacement of instruments, changes in observing procedures, growth of surrounding vegetation, urbanization, automation and others. These changes can introduce variations in historical records totally unrelated to actual changes in the regional climate. Therefore, rigorous examination of climate datasets and adjustment for data problems are a necessity before any climate change analysis can be done.
This paper describes the methodologies used to adjust historical temperature and precipitation datasets. Different approaches were applied for homogenization of temperature and precipitation due to the different behavior of the climate variable. For temperature, a statistical technique was first applied to identify all potential inhomogeneities, and the cause of each inhomogeneity was retrieved through the historical reports when possible. For precipitation the Canadian network density is insufficient to allow widespread use of surrounding station data to identify and adjust inhomogeneities. Instead, the primary goal was to correct the measurements for all known inhomogeneities - such as instrument changes, gauge undercatch, snow density differences, trace measurements, etc. - using meta data information.
As the result of several years of research, homogenized temperatures time series (Vincent and Gullett, 1999) and rehabilitated precipitation time series (Mekis and Hogg, 1999) have been prepared for analysis of climate change and variability in Canada. Figure 1 presents the location of 210 homogeneous temperature and 496 rehabilitated precipitation stations across Canada, most with data covering the period 1900-present. Data availability in much of the Canadian Arctic was restricted to 1948-present.

Figure 1:
Rehabilitated Canadian temperature and precipitation stations location.

Inhomogeneities are identified in the annual mean maximum and minimum temperature time series using a technique based on regression models (Vincent, 1998). Annual temperature anomalies from the 1961-1990 reference period are obtained at a candidate station and at a number of surrounding stations, and the reference series is produced by averaging the anomalies of the surrounding stations. To ensure that the inhomogeneities of the surrounding stations do not obscure the homogeneity assessment of the candidate station, graphs of station pairs are examined and large steps in surrounding stations are adjusted.

2.1 Monthly adjustments

The first step consists of determining if the candidate series is homogeneous. To accomplish this, model 1 is applied to the datasets as follows:

yi = a1 +c1xi + ei               i = 1,...,n Model 1

where the dependent variable yi is the anomaly of the candidate series and the independent variable xi is the anomaly of the reference series at time i. The autocorrelations of the residuals ei are computed for several lags k. If the autocorrelations are not significantly different from zero, model 1 is accepted, the candidate series is homogeneous for the tested period of time and the procedure stops. However, if there is significant autocorrelations at several consecutive low lags then model 1 is rejected: the candidate series is not homogeneous and a second model is applied in order to search for the inhomogeneities.

The second step consists of the identification of the position in time and magnitude of a step. Model 2 is applied to the temperature anomalies as follows:

yi = a2 + b2I + c2xi + ei i = 1,...,n Model 2

where I is an indicator variable that simulates a step in the candidate series. The variable I takes the following values:

I = 0 for i = 1,...,p-1
I = 1 for i = p,...,n.

The changepoint p indicates the position in time of a potential step. Since the position of the step is unknown a priori, model 2 is fitted successively for p equal 4 to n-3, and each time the sum of square residuals SSE is calculated. The minimum SSE indicates the best fit of model 2: the corresponding p is the position the step and the estimated parameter b2 is its magnitude. The candidate series is then divided at the changepoint p and each section is re-tested separately starting with model 1.

Figure 2:
a) Annual mean maximum temperature for Harrow CDA, Ontario, 1918-1995: the dashed line shows a trend of -1.6°C over the 78 years;

b) difference between the annual mean maximum temperature of Harrow CDA and the reference series: a step of -1.0°C is identified in 1956;

c) adjusted annual mean maximum temperature for Harrow CDA: the dashed line shows a trend of -0.1°C over the 78 years.

For example, the annual mean maximum temperatures of Harrow CDA, Ontario, are assessed for homogeneity for the period 1918-1995 (Fig. 2a): the series shows a decreasing trend of -1.6°C over the 78 years of observations. When model 1 is fitted to the datasets, the autocorrelations of the residuals are significantly different from zero at many consecutive low lags (Fig. 3): it is concluded that Harrow CDA is not homogeneous for the tested period of time. When model 2 is applied successively for p equal 1921 to 1992, the minimum SSE is reached for 1956 (Fig. 4). The difference between the annual mean maximum temperatures of Harrow CDA and the reference series clearly shows the step in 1956 (Fig. 2b). The magnitude of the step is -1.0°C which is significantly different from zero using the t-test. Base and reference series are divided at 1956. Each segment is re-tested separately and both intervals are founded to be homogeneous.

Figure 3:
Autocorrelation of the residuals at different lags after fitting model 1 to the annual mean maximum temperature of Harrow CDA, 1918-1995. The dashed lines represent the approximate 95% confidence interval.
Figure 4:
Sum of square errors obtained from fitting model 2 with changepoint p.

To obtain the monthly adjustment factors, model 2 is applied for p equals 1956 on the time series of the 12 individual months and the magnitude of the step for each month is the monthly adjustment (Fig. 5). The step identified in 1956 corresponds to a small relocation of the instruments away from the main building for a better exposure. The adjusted time series shows a trend of -0.1°C over 78 years (Fig. 2c) and similar trends are also observed in the surrounding stations.

Figure 5:
Monthly adjustment factors by regression and daily factors by interpolation for the step of -1.0 °C in 1956.
2.2. Daily adjustments.

Daily maximum and minimum temperatures are adjusted to reflect the same long term climate variations as presented by the adjusted monthly time series (Vincent et al. 2000). The procedure is based on an interpolation procedure described by Sheng and Zwiers (1998) which was applied to the monthly mean sea surface temperatures.
The monthly adjustment factors are used to generate the daily factors. The daily factors correspond to the interpolated line between mid-month "target" values chosen so that the average of the daily adjustments over a given month is equal to the monthly adjustment. The target values are related to the monthly adjustment factors by the system of equations:

AT = M,

where A is a tridiagonal 12 x 12 matrix,

T is a 12 x 1 vector of the target values and M is a 12 x 1 vector of the monthly adjustment factors. The target values are then obtained by solving the equations:

T = A-1M.

The main advantages are that the procedure does not produce artificial steps at the join of the calendar months and it also preserves the monthly adjustments. In the example of Harrow CDA, the daily adjustments factors for the step identified in the annual mean maximum temperatures in 1956 are illustrated by the interpolated line between the target values (Fig. 5).

2.3 Impact of the adjustments on the temperature trends

To assess the impact of the adjustments, the best fit linear trend was applied to the annual mean maximum and minimum temperatures before and after adjustment respectively for the period 1950-1998.

Figure 6:
Trends in
a) the annual mean maximum temperatures,
b) the annual mean minimum temperatures, over 1950-1998, before and after adjustment.
Grey and black dots indicate positive and negative trends only for stations experiencing a change in trend cause by the adjustment. Size of the dots is proportional to the magnitude of the trend.

Figure 6a shows the trends in the annual maximum temperatures. The trends are given only for the stations experiencing a change in trends due to the adjustments (about 16% of the stations). The maps show very little impact with slightly less warming in the west and less cooling in the east parts of the country. A greater impact is observed in the annual minimum temperature trends (Figure 6b), over 40% of the stations have a change in trends caused by the adjustments. Generally, there is less cooling in the minimum temperatures in eastern Canada after adjustments. It is mostly due to the adjustments applied to the minimum temperatures as result of a change in observing window in July 1916 at the principal stations (Vincent et Gullett, 1999).


There are specific problems in Canadian precipitation data, which are difficult (if not impossible) to correct on a monthly level (e.g. wetting loss and trace measurements). The methodology for correction of systematic biases can be improved when performed on daily rain gauge and snow ruler data. Part of the adjustment methodology was based on procedures developed for 6 hourly synoptic station data by Metcalfe et al.(1994). Since their methodology was designed for a different time-step and purpose, several modifications had to be implemented.

For each of the ~ 500 locations, station history files were searched thoroughly for: date of any relocation; installation date of all rain gauges; introduction date of 6 hourly measurement program and; introduction date of hourly weather type measurement program. Computerized metadata files were created for each station to aid in the task of correction. For the selected stations, daily total rainfall (gauge) and snowfall (ruler) measurements were extracted from the Canadian National Climate Data Archive for the maximum available interval.

3.1 Adjustments associated with daily rain gauge measurements

Canadian rain measurement methodologies have been modified several times. The official rain gauge in Canada is currently the Type B, which was introduced at most locations during the 1970's (see Fig. 7). Prior to the 70's the Meteorological Service of Canada (MSC) gauge was used. The MSC gauges were originally manufactured totally from copper but the inside container was modified to a soft plastic material with different wetting characteristics around 1965. Metcalfe et al. (1997) give a more complete description of these gauges. The instruments have different measurement efficiencies. Both are mounted low to the ground to reduce the undercatch due to wind but the undercatch is not the same. Ongoing intercomparison confirms that systematic differences between the MSC gauge and a pit gauge (buried gauge with orifice at ground level) is about 4% (Goodison and Louie, 1986) and between the Type B and a pit gauge is only about 2%.

Figure 7:
Canadian MSC copper/plastic and Type-B rain gauges

The wetting loss has two basic components: the water subject to evaporation from the surface of the funnel and the inner walls after an event before emptying and the water retained on the walls of the gauge and on the funnel after emptying. The MSC copper and plastic inserts and the all plastic Type B gauge all have different wetting loss characteristics. The adjustments associated with instrument change for daily rain gauge measurements can be summarized as:

The trace adjustment is very important in Canada especially in the northern part of the country, where precipitation amounts are relatively low and many trace events are recorded. Metcalfe et al. (1994) report that trace amounts can account for 80 % of precipitation occurrences at some locations in northern Canada. Trace amounts are identified by a "flag" in the archive data files but are assigned an absolute value of zero when monthly precipitation totals are calculated. This leads to an underestimation of monthly and annual precipitation totals. The present study makes an attempt to estimate the actual amount of daily rain and snow accumulated in trace events, for each station, in a systematic and consistent way.

Because of the long winters in the Arctic and the high frequency of trace observations during frozen precipitation events, it is important to adjust for traces observed during snowfall. For completeness, even though rainfall amounts are small in the Arctic, adjustments for rainfall trace were also implemented. Based on a few experiments all rain amounts for the interval 0.0-0.2 mm were considered equally probable for events classified as trace and the interval average of 0.1 mm was considered representative of a trace measurement amount. Since rain trace is also exposed to wetting, evaporation and retention losses prior to measurement, use of 0.1 mm for the rain only trace adjustment should be conservatively low (for details see Routledge, 1997). As well, rain trace events in the North have relatively little effect on the final total precipitation accumulation, so approximation procedures were considered adequate.

The practice of measuring precipitation every 6 hours at synoptic stations but archiving daily total precipitation may lead to a further underestimation. There could be as many as 4 observations of trace contributing to a single archived daily trace flag. Since the present adjustment is performed on daily data, it is important to determine the average number of 6-hourly trace measurements included in one daily trace flag. In order to be able to properly adjust for trace at different locations, the original 6-hourly archive information had to be classified based on the type of precipitation occurring when trace was recorded. Three major classes of trace were determined -- rain, ice crystal and snow -- through the comparison of the hourly archive weather-type information and the 6-hourly total precipitation measurements. For relating the daily and 6-hourly trace flag observations the Trace Occurrence Ratio (Tor,6) was computed for each station. The Tor,6 values include all 6 hourly trace counts, even for those cases when measurable precipitation and trace events happened on the same day (if precipitation and trace occur on the same day, the daily observation file will only contain the precipitation amount without the effect of trace). Thus Tor,6 inherently includes an allowance for adjustment of trace on days with measured precipitation in other synoptic periods. Since the computed mean of the long-term Tor,6 for the full period is 3 for each station over the examined period with relatively little variance, this factor was accepted for use at each station. For climate station, where two daily measurements are taken, the resulting trace occurrence ratio was assumed to be proportional to the number of measurements taken. For ordinary climate stations the Tor,12 value was half (i.e. Tor,12 = 1.5) of the factor applied at synoptic stations.

3.2 Adjustments for daily snow ruler measurements

The depth of freshly fallen snow, measured by ruler, has always been the Canadian standard climate measurement of snowfall. For all stations, prior to the 1960's, and for non-synoptic stations over the entire record, precipitation amount (water equivalent) for snowfall events is determined by assuming a density for fresh snow of 100 kg m-3. At synoptic stations, a Nipher shielded snow gauge was introduced in the 1960's to directly measure snow water equivalent for determination of precipitation amount, but snowfall measurements with the ruler were continued.

Since the process of ruler measurements has undergone fewer changes over time, daily snow ruler data were used for the full period of this study. For computation of snow water equivalent, the use of the 100 kg m-3 standard density of fresh snow was rejected. Instead, a new fresh snow density adjustment (rnew) is calculated for each station separately. The adjustment method is based on the ratio of corrected Nipher gauge to ruler depth measurements during the period of overlap (Metcalfe et al. 1994). Using data from 63 stations, a map of the values of rnew was produced and used to obtain estimates for each station. This density was applied to all ruler measurements to generate time series in snow water equivalent at each location.

Accumulation of solid precipitation trace measurements is an even more complex problem than that for rainfall trace. Due to the measurement methodology, snow ruler trace measurements are not exposed to wetting and retention losses, and evaporation losses are minimized by the season. A value less than the rainfall trace allowance (.1 mm) is quite consistent and .07 mm for snowfall trace was deemed appropriate in Southern Canada. However, north of 55 oN, this value is less appropriate. In Canada, observation of ice crystals are considered precipitation. Ice crystals are recorded at very low temperatures and contribute extremely little water to precipitation totals. In addition, the number of ice crystal events increases as latitude increases while annual precipitation decreases. Thus, the appropriate value for the trace adjustment varies with climate and location. If the observed increase in number of trace observations in the Arctic is due to an increase in the number of ice crystal events, then it is very important to properly adjust for the trace observations due to ice crystals to avoid introducing an artificial trend in arctic precipitation. The amount assigned to trace observations was reduced in proportion to the average number of ice crystal events observed at the site.

The adjustments associated with trace observations and snow ruler measurements can be summarized as:

3.3 Effect of adjustments

To evaluate the effect of adjustments for trace and other changes, several sensitivity studies were performed. Results of the selected Resolute station (located in the Arctic Islands 75 oN) are plotted in Figure 8. Here the amount of adjustment is relatively high compared to the total due to the frequent occurrence of ice crystal trace events, high snowfall density adjustment and small annual precipitation.

Figure 8:
Sensitivity study of annual total precipitation adjustment at Resolute:

curve (1): archive rain + snow, trend (1) = 1.04 mm/year;

curve (2): rain + snow ( new density (32 % change),
trend (2) = 1.61 mm/year;

curve (3): (2) + gauge correction (36 % change),
trend (3) = 1.52 mm/year;

curve (4): (3) + rain trace correction (42 % change),
trend (4) = 1.64 mm/year;

curve (5): (4) + snow trace correction (56 % change),
trend (5) = 1.81 mm/year.

Step-by-step adjustments were performed and the percentage change along with trends of total annual precipitation amounts were computed. In the first step the actual snow values were multiplied by the new density adjustment. Since Pnew is a multiplicative factor, it has a direct and significant effect on total precipitation amounts but changes in the trend are dependent only upon changes in the fraction of total precipitation falling as snow. Adjustment for undercatch and wetting loss due to rain gauge change accounts for 3 % increases in total precipitation. Since the new Type-B gauge has smaller loss and approximates "true" rainfall amount better, the trend is decreased by this step. Adjustment for rain trace days adds to the accumulation and increases the trends slightly. In the last step the solid trace adjustment values were added to the daily snow ruler measurements. In spite of the relatively low solid trace intensity rates (0.04 mm), the effect of this adjustment was quite high on this Northern station because of the large percentage of ice crystal events at Resolute. The trend is also increased slightly because of the recorded increase in number of measurements of trace. After application of all adjustments the trend for Resolute almost doubled.

Figure 9:
Magnitude of adjustments of total precipitation, snow and rain expressed in % change [adjusted / measured] over the 1961-90 normal period.

The magnitude of adjustments were computed for each station where the normal was available and mapped for Canada (Figure 9). As expected, the highest values are located in the North, decreasing towards the South. The spatial variance of snow is much stronger compared to rain and dominated by the variability in snow density.

It was also studied and proved, that trace adjustments did not modify the direction of trends at individual stations. Trace correction application effected only 4 station's trend out of the total of 425 computed for the 1950-1998 period, the originally significant trend become insignificant.


Analysis of the temperature and precipitation trends can be performed at individual stations or using gridded datasets. Monthly grids of temperature and precipitation anomalies (departures from the 1961-1990 reference period for temperature and normalized departures for precipitation) were produced using an optimal interpolation technique. The gridded anomalies were then combined with the gridded normals obtained from a multiple regression procedure (Seglenieks and Soulis, 2000) to generate grids of monthly temperature and precipitation for the period 1900-1998 (Zhang et al., 2000/b). The grid covers Canada with a 50 km resolution on a polar stereographic projection.

A recent analysis using these datasets shows that annual mean temperature has increased between 0.5 to 1.5°C in southern Canada (south of 60°N) during the 20th

Several further analyses were performed using the above datasets. The adjusted precipitation and temperature data have already been used to improve the skill of seasonal climate forecasts (Shabbar and Barnston, 1996), as input data for the water balance computation on the MacKenzie Basin (Hogg et al., 1996), as input data for climate variability computation over the instrumental record (Skinner, 1996) and for Canadian Prairie growing season variability computation (Bonsal et al., 1999). Extreme value analysis of spatial and temporal characteristics of heavy precipitation events (Zhang et al., 2000/a), extremes of multi-day precipitation accumulation (Mekis and Zhang, 2000) and changes in daily temperature characteristics and extremes (Bonsal, et al., 2000) were also computed over Canada. The adjusted dataset will provide a valuable source of information about Canada's changing climate for years to come.


The development of reliable homogenized temperature and precipitation datasets is an ongoing task. Monthly maximum, minimum and mean temperatures and monthly rainfall, snowfall and total precipitation are available at
The gridded dataset is also available on CD-ROM from the authors. The datasets in the present form are annually updated but further developments are also required.

Automation of climate observations is inevitable in Canada. Almost one third of the stations selected for this project have already been automated. New methods need to be developed to determine if human and automated observations can be joined together in order to produce reliable and homogenized time series for climate change analysis in Canada.


Bonsal, B.R., X. Zhang and W.D. Hogg, 1999: Canadian Prairie growing season precipitation variability and associated atmospheric circulation. Climate Research, Vol. 11: 191-208.
Bonsal, B.R., X. Zhang, L.A. Vincent and W.D. Hogg, 2000: Characteristics of daily and extreme temperatures over Canada. J. Climate, In press.
Goodison, B.E. and P.Y.T. Louie, 1986: Canadian methods for precipitation measurement and correction. International Workshop on Correction of Precipitation Measurements, Zurich, Switzerland, 141-145.
Hogg, W.D., P.Y.T. Louie, A. Niitsoo, E. Milewska and B. Routledge, 1996: Gridded water balance for the Mackenzie Basin GEWEX study area. Second International Scientific Conference on GEWEX, June 17-21, Washington DC, USA.
Mekis, E. and W.D. Hogg, 1999: Rehabilitation and analysis of Canadian daily precipitation time series. Atmosphere-Ocean, 37, 53-85.
Mekis, E. and X. Zhang, 2000: Extremes of Canadian multi-day precipitation accumulation. 12th Conference on Applied Climatology, Asheville NC, American Meteorology Society, 107-110.
Metcalfe, J.R., S. Ishida and B.E. Goodison, 1994: A corrected precipitation archive for the Northwest Territories. EC - Mackenzie Basin Impact Study, Interim Report #2, 110-117.
Metcalfe, J.R., B. Routledge and K. Devine, 1997: Rainfall measurement in Canada: changing observational methods and archive adjustment procedures. J. Climate, 10, 92-101.
Routledge, B., 1997: Corrections for Canadian standard rain gauges. Internal Report, Environment Canada, Atmospheric Environment Service, Downsview, 8 pp.
Seglenieks, F. and R. Soulis, 2000: Generation of square grid normals for Canada - Phase 1. U. Waterloo Contractor's Report to Climate Research Branch, Meteorological Service of Canada, Downsview.
Shabbar, A. and A.G. Barnston, 1996: Skill of seasonal Climate Forecasts in Canada Using Canonical Correlation Analysis. Monthly Weather Review, Vol. 124, 2370-2385.
Sheng, J. and F. Zwiers, 1998: An improved scheme for time-dependent boundary conditions in atmospheric general circulation models. Climate Dynamics, 14, 609-613.
Skinner, W., 1996: Climate Variability over the Instrumental Record with special emphasis on Western Canada. Inter American Institute for Global Change Workshop, October 6-11, Canada.
Vincent, L.A., 1998: A technique for the identification of inhomogeneities in Canadian temperature series. Journal of Climate, 11, 1094-1104.
Vincent, L.A., and D. W. Gullett, 1999: Canadian historical and homogeneous temperature datasets for climate change analyses. International Journal of Climatology, 19, 1375-1388.
Vincent, L.A., X. Zhang, B.R. Bonsal, W.D. Hogg, 2000: Homogenized daily temperatures for trend analyses in extremes over Canada. 12th Conf. on Applied Clim., Asheville, AMS, 86-89.
Zhang, X., W.D. Hogg and E. Mekis, 2000/a: Spatial and temporal characteristics of heavy precipitation events over Canada. Journal of Climate, In press.
Zhang, X., L.A. Vincent, W.D. Hogg and A. Niitsoo, 2000/b: Temperature and precipitation trends in Canada during the 20th Century. Atmosphere and Ocean. In press.

COST Action 0601 
Városklíma 2011 
Climate variability and climate change 
RCM Workshop 2008 
HIRLAM / AAA Workshop 2007 
The preparation of climate atlas 
17th EGOWS Meeting 
ALADIN 2002 
6th Seminar for homogenization 
5th Seminar for homogenization 
Természeti katasztrófák megelőzése, hatásainak csökkentése

2005 Időjárás, éghajlat, víz és
fenntartható fejlődés
2004 Időjárás, klíma és víz az
információs társadalom korában