• Alexandersson, H., 1986: A homogeneity test applied to precipitation data. J. Climatol., 6 , 661675.

  • Begert, M., , T. Schlegel, , and W. Kirchhofer, 2005: Homogeneous temperature and precipitation series of Switzerland from 1864 to 2000. Int. J. Climatol., 25 , 6580.

    • Search Google Scholar
    • Export Citation
  • Brown, R. D., 2000: Northern Hemisphere snow cover variability and change, 1915–97. J. Climate, 13 , 23392355.

  • Devine, K. A., , and É Mekis, 2008: Field accuracy of Canadian rain measurements. Atmos.–Ocean, 46 , 213227.

  • González-Rouco, J., , J. L. Jiménez, , V. Quesada, , and F. Valero, 2001: Quality control and homogeneity of precipitation data in the southwest of Europe. J. Climate, 14 , 964978.

    • Search Google Scholar
    • Export Citation
  • Goodison, B. E., 1978: Accuracy of Canadian snow gauge measurements. J. Appl. Meteor., 17 , 15421548.

  • Goodison, B. E., 1981: Compatibility of Canadian snowfall and snow cover data. Water Resour. Res., 17 , 893900.

  • Hanssen-Bauer, I., , and E. J. Førland, 1994: Homogenizing long Norwegian precipitation series. J. Climate, 7 , 10011013.

  • Mekis, É, , and W. D. Hogg, 1999: Rehabilitation and analysis of Canadian daily precipitation time series. Atmos.–Ocean, 37 , 5385.

  • Mekis, É, , and R. Hopkinson, 2004: Derivation of an improved snow water equivalent adjustment factor map for application on snowfall ruler measurements in Canada. Preprints, 14th Conf. on Applied Climatology, Seattle, WA. Amer. Meteor. Soc., 7.12. [Available online at http://ams.confex.com/ams/pdfpapers/68724.pdf.].

    • Search Google Scholar
    • Export Citation
  • Metcalfe, J. R., , B. Routledge, , and K. Devine, 1997: Rainfall measurement in Canada: Changing observational methods and archive adjustment procedures. J. Climate, 10 , 92101.

    • Search Google Scholar
    • Export Citation
  • New, M., , M. Todd, , M. Hulme, , and P. Jones, 2001: Precipitation measurements and trends in the twentieth century. Int. J. Climatol., 21 , 18991922.

    • Search Google Scholar
    • Export Citation
  • Peterson, T. C., , R. Vose, , R. Schmoyer, , and V. Razuvaëv, 1998: Global historical climatology network (GHCN) quality control of monthly temperature data. Int. J. Climatol., 18 , 11691179.

    • Search Google Scholar
    • Export Citation
  • Thom, H. C. S., 1966: Some methods of climatological analysis. WMO Tech. Note 81, 53 pp.

  • Tuomenvirta, H., 2001: Homogeneity adjustments of temperature and precipitation series – Finnish and Nordic data. Int. J. Climatol., 21 , 495506.

    • Search Google Scholar
    • Export Citation
  • Wijngaard, J. B., , A. M. G. Klein Tank, , and G. P. Können, 2003: Homogeneity of the 20th century European daily temperature and precipitation series. Int. J. Climatol., 23 , 679692.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Standardized ratios for annual total (a) rain and (b) snow (base over neighbor 5) for the stations Digby/Bear River, joined in 1957.

  • View in gallery

    Monthly, annual, and long series mean adjustments obtained using the neighbors (black) and overlapping data (gray) for total (a) rain and (b) snow for the stations Digby/Bear River, joined in 1957.

  • View in gallery

    (a) Summer total rain and (b) winter total snow for the joined stations Digby/Bear River before (gray) and after (black) adjustments.

  • View in gallery

    Annual total (a) rain and (b) snow for Digby (gray) and Bear River (black) for the overlapping period 1953–65.

  • View in gallery

    Box plots of the differences between the adjustments from neighbors and adjustments from overlapping data for (a) rain and (b) snow obtained from the 60 stations. The circle denotes the median, the box indicates the 10th and 90th percentiles, and the whiskers specify the minimum and maximum values.

  • View in gallery

    Frequency distribution of the adjustments given in percent in classes of 0.1.

  • View in gallery

    Trends in summer rain (a) before and (b) after adjustments for 1930–2007, and trends in winter snow (c) before and (d) after adjustments for same period. Upward- and downward-pointing triangles indicate positive and negative trends, respectively. The size of the triangle is proportional to the magnitude of the trend.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 1 1 1
PDF Downloads 1 1 1

Discontinuities due to Joining Precipitation Station Observations in Canada

View More View Less
  • 1 Climate Research Division, Science and Technology Branch, Environment Canada, Toronto, Ontario, Canada
© Get Permissions
Full access

Abstract

When a climatological station is relocated or is closing, it is often possible to join the climate observations of a nearby site to create a longer time series. However, joining climate observations can sometimes introduce artificial discontinuity that affects the trend. A procedure to detect discontinuities at the joining dates for precipitation station observations is described. It is based on standardized ratios between a tested station and a neighbor, and the t test is used to determine whether the means before and after the joining dates are statistically significantly different. The procedure is applied to 234 climatological stations across Canada to identify steps in rainfall and snowfall. The results indicate that joining precipitation station observations creates steps of different magnitude for rain and snow. It is concluded that about 35% of the stations need adjustment for rain whereas 58% of the stations need adjustment for snow. The magnitude of the adjustments varies from 0.75 to 1.25 for rain and from 0.65 to 1.60 for snow. The annual and seasonal trends before and after adjustments are also examined for 1930–2007. The results show that the trends computed from the adjusted data present a more consistent regional pattern than do trends computed from the unadjusted observations.

Corresponding author address: Lucie A. Vincent, Environment Canada, 4905 Dufferin St., Toronto, ON M3H 5T4, Canada. Email: lucie.vincent@ec.gc.ca

Abstract

When a climatological station is relocated or is closing, it is often possible to join the climate observations of a nearby site to create a longer time series. However, joining climate observations can sometimes introduce artificial discontinuity that affects the trend. A procedure to detect discontinuities at the joining dates for precipitation station observations is described. It is based on standardized ratios between a tested station and a neighbor, and the t test is used to determine whether the means before and after the joining dates are statistically significantly different. The procedure is applied to 234 climatological stations across Canada to identify steps in rainfall and snowfall. The results indicate that joining precipitation station observations creates steps of different magnitude for rain and snow. It is concluded that about 35% of the stations need adjustment for rain whereas 58% of the stations need adjustment for snow. The magnitude of the adjustments varies from 0.75 to 1.25 for rain and from 0.65 to 1.60 for snow. The annual and seasonal trends before and after adjustments are also examined for 1930–2007. The results show that the trends computed from the adjusted data present a more consistent regional pattern than do trends computed from the unadjusted observations.

Corresponding author address: Lucie A. Vincent, Environment Canada, 4905 Dufferin St., Toronto, ON M3H 5T4, Canada. Email: lucie.vincent@ec.gc.ca

1. Introduction

Reliable long-term climate datasets are essential for the analysis of climate change. High-quality station data with good temporal and spatial resolution are necessary for accurate trend and variability analysis, verification of regional and global models, validation of remotely sensed data from space platforms, and proper detection and attribution of climate change. However, long-term climate records are often not available because of frequent changes in observing networks that sometimes lead to the relocation or closure of climatological stations.

In Canada in the late 1940s and early 1950s, rural stations were often closed and then reopened as airport stations in nearby cities a few kilometers away. In the more recent 1980s and 1990s, stations were closed because of the downsizing of the traditional climate network and increasing use of automated instruments. When a station is relocated, even within close vicinity, a new identification number is assigned to the new location. To achieve a longer time series, the two stations can be combined. In addition, when a station is closing because of the end of an observing program, it is sometimes possible to find a neighbor with similar climate characteristics and to join the observations to produce a long time series. Because the new site can have different elevation and exposure, it becomes essential to determine whether a discontinuity is created at the joining date.

Adjusted precipitation datasets were prepared for the analysis of climate trends in Canada (Mekis and Hogg 1999). Daily rainfall was adjusted for known changes in instruments, and snow density adjustments were applied to ruler measurements to convert all snowfall amounts to snow water equivalent amounts (Mekis and Hopkinson 2004). At times, the observations of two stations were merged into one, and the data prior to the joining date were adjusted using the simple ratio method (Thom 1966) when concurrent observations were available from both locations. The daily values were multiplied by the ratio of the two stations to adjust for a potential step at the joining date. However, this procedure was applied only when overlapping observations were available and several joined stations consequently remained without adjustments.

In Europe, precipitation observations are often archived under the same station number when the instruments are relocated. Therefore, it is necessary to apply a procedure to search for and detect steps due to station relocation within the same city. The standard normal homogeneity test (Alexandersson 1986) was often used along with other methods to address this problem (Hanssen-Bauer and Førland 1994; Tuomenvirta 2001; González-Rouco et al. 2001; Wijngaard et al. 2003; Begert et al. 2005). The test was developed to identify a shift in a sequence of ratios between a tested site series and a reference series produced from surrounding stations. It was mainly applied to annual total precipitation to develop adjustments for annual series and subsequently to adjust monthly series. In some cases, the adjustments were calculated for each month separately but individual monthly corrections seemed to introduce some uncertainties to the time series (Alexandersson 1986; González-Rouco et al. 2001; Begert et al. 2005). The trends resulting from adjusted precipitation have generally presented a more coherent regional pattern than do those from the unadjusted data (González-Rouco et al. 2001; Begert et al. 2005).

Rainfall and snowfall measurements in Canada have been archived since 1840 for a number of stations. Efforts have focused on addressing the inherent difficulties of measuring solid precipitation (Goodison 1978, 1981) and adjusting rainfall measurements for changing observational methods and archive procedures (Metcalfe et al. 1997). Snow cover is also an important climate element in Canada that shows significant spatial variability, exhibits a close relationship with air temperature, and can exert a strong influence on regional albedo (Brown 2000). Therefore, it is important to examine the impending problems in rain and snow data separately, as opposed to examining total precipitation. In the study presented here, the conventional snow ruler measurements are used throughout the entire station record and the Canadian Nipher gauge observations were used only indirectly in the snow water equivalent computation.

The objective of this paper is to determine whether joining precipitation station observations in Canada has created any artificial discontinuities that affect the trends. Annual and monthly adjustments were obtained for rainfall and snowfall separately using a number of neighboring stations with no overlapping data. Sections 2 and 3 describe the data and method, respectively, and the procedure is illustrated in section 4 through the provision of an example. Section 5 presents some validation results by comparing adjustments using neighboring stations with adjustments produced from overlapping observations at 60 climatological stations. In section 6, the procedure is applied to 234 stations and the impact of the adjustments on annual and seasonal trends is examined. A discussion on the difficulties in adjusting monthly rain and snow is presented in section 7, and the conclusions follow in section 8.

2. The data

Daily total rainfall and snowfall were retrieved directly from the National Climate Data Archive of Environment Canada. Daily rain was adjusted for known changes in instruments (Mekis and Hogg 1999): different rain gauges have different wind undercatch, wetting losses, and evaporation (Metcalfe et al. 1997). The main changes were the modification of the Meteorological Service of Canada (MSC) gauge from a copper to a plastic receiver in 1965 and the replacement of the MSC gauge by a type-B gauge in the 1970s. The adjustments were based on long-term detailed rain gauge experiments (Devine and Mekis 2008). For snowfall, ruler measurements were used throughout the entire period. The snow amounts were converted to an equivalent amount of water using a snow water equivalent adjustment map based on concurrent snow ruler and Nipher observations (Mekis and Hopkinson 2004); at any given location the same adjustment was applied throughout the entire station record.

Records were joined together either because a station was moved to a new location or because two closely located shorter-period stations could be combined. On other rare occasions on which a large gap was found within a single station record, a neighbor was used to fill the gap. Before joining station observations, considerations were given to the distance and elevation difference between the stations to be merged. Each station needed at least 5 yr of data with very few missing daily values. In most cases only two stations were joined together, but sometimes it was necessary to join three or four stations. Efforts were made to cover the longest period (e.g., 1900 to present) as much as possible.

In the current study, a monthly value is considered to be missing when more than three consecutive days or more than five random days are missing. The annual value is the sum of the 12 monthly values, and it is missing if at least 1 month is missing. The seasons are defined as follows: winter (from December of the previous year to February of the current year), spring (March–May), summer (June–August), and autumn (September–November).

3. Method

The method used in this study is similar to the approach developed by Alexandersson (1986); however, it does not require a search for the most probable date of change because the joining date is already known. A sequence of standardized ratios {zi} between the tested station and a neighbor is produced. The t test is applied to determine whether the means of the standardized ratios before and after the joining date are statistically significantly different at the 5% level. The test is repeated for each individual neighbor, and it is performed for each monthly and annual series separately. If most neighbors indicate a significant step at the joining date, then it is concluded that the tested station should be adjusted.

a. Ratios

Because rain and snow have high spatial variability and precipitation events do not necessarily occur in the same time at different locations, a rule is needed to establish when to compute the ratios {qi}. Let Ti and Ni be the monthly total rain (or snow) in millimeters at the tested site and at the neighbor, respectively, for year i:
i1558-8432-48-1-156-e1

A monthly series is tested only if the monthly total rain (or snow) is greater than zero at least 50% of the time during the entire station record. For example, snowfall can be observed at southern locations in May but not necessarily every year; therefore, the May snow series is tested only if the percentage of years with snow in May is above 50%.

b. Outliers

Because the neighbors can be located several kilometers away from the tested site, it is possible that a great amount of precipitation occurs at one of the stations during a month but not at the other. This difference in the precipitation amounts can generate outliers in the sequence of ratios {qi}. Because outliers can influence the results of the t test, it is necessary to identify them. Outliers are defined as
i1558-8432-48-1-156-e2
where q0.25 and q0.75 are the 25th and 75th percentiles, respectively. The interquartile range (difference between the 75th and 25th percentiles) has been used in the quality control of climate data because it is resistant to outliers (Peterson et al. 1998). The 25th and 75th percentiles are obtained from the entire period of record, and outliers are replaced by a missing value.

c. The t test

Similar to Alexandersson (1986), a sequence of standardized ratios {zi} is obtained as follows:
i1558-8432-48-1-156-e3
where q is the mean of the sequence {qi} and sq is its standard deviation. This new series {zi} has a mean of 0 and a standard deviation equal to 1. The t test is applied directly to the standardized ratios to determine whether the means before and after the joining date are statistically different at the 5% level. To reduce the influence of other temporal inhomogeneities occurring at either the tested series or neighbors, a window of 60 yr centered on the joining date is applied: 30 yr before and after the joining date. Sometimes the interval before (or after) is shorter if the joining date is near to the beginning (or end) of the record or is near to another joining date. In these cases, the test is applied only if a minimum of 5 yr on each side of the joining date is available.

d. Neighbors

The selection of the neighbors was mainly based on data coverage, correlation, distance, and elevation. If the joining date was in the early 1920s, it was often more difficult to find suitable neighbors with sufficient data because fewer stations have climate observations in Canada in the beginning of the century. Correlation was another important factor, and only neighbors with a correlation above 0.5 for the annual values were retained. In the southern regions, where the station density is better, the distance between the tested site and neighbor was often within 50–100 km; in more isolated areas, such as in the north, the neighbor was frequently as far as 200 km.

In previous studies over Europe (Hanssen-Bauer and Førland 1994; González-Rouco et al. 2001; Begert et al. 2005), a single reference series was calculated from the selected neighbors. In the current study, the test is applied using each individual neighbor. This approach was chosen for the following reasons: the monthly total rain and snow are more spatially variable relative to the annual series and the neighbors can be far away, the starting and ending dates of each neighbor can be different, and the number of missing values in the ratio sequence {qi} can be important because of the number of dry months. It was easier to interpret the results when the test was applied to an individual neighbor. If one neighbor’s result was very different from the other neighbor’s, then it was discarded (e.g., if it indicated a significant step of the opposite sign). Further work is needed to address properly the creation of a reasonable reference series when monthly rain and snow are used.

e. Long series

The 12 monthly series and 1 annual series are tested in the same manner. A main disadvantage of this approach is the small sample size (two averages based on 30 values are compared). To overcome this disadvantage, a long series is created by placing together all months consecutively in a sequence of 12n values, where n is the number of years. Furthermore, the long series is modified to accommodate the months with precipitation events above 50%. For example, if most of the rain occurs in April–October (7 months) and the data cover 1941–2000 (60 yr), then the long series for rain is composed of the April–October values of each year, for a total of 420 values (210 values before and after the joining date of 1971). The results provided by the long series are useful to verify those provided by the annual time series.

f. Adjustments

The difference between the means before and after the joining date is tested for each monthly, annual, and long series. If a step is significant in many months, in the annual values, and in the long series, it is likely that there is an artificial step at the joining date. This becomes more evident when a significant step is identified by many neighbors. The tested site is then adjusted by multiplying the values before the joining date with an adjustment factor A calculated as
i1558-8432-48-1-156-e4
where qa and qb are the means of {qi} after and before the joining date. The final adjustment is the average of the adjustments given by each neighbor. The information provided by the neighbors is helpful to explain the precipitation variations over the years.

4. Example

The precipitation datasets of the station Digby (Nova Scotia) were joined with those of Bear River in 1957 to create a long series from 1915 to 2006. The stations are located 11 km apart and have an elevation difference of 4 m. Eight neighbors were identified with sufficient data to cover the interval 1927–86. The distances, elevation differences, and correlations between the joined station and the eight neighbors are given in Table 1. The series of the ratios and standardized ratios were obtained for each monthly, annual, and long series, and the t test was applied to determine whether the difference between the means before and after 1957 was statistically significant. The adjustment factors are provided in Table 2; the boldface numbers correspond to significant steps.

For rain, the annual and long series results indicate a significant step with only one neighbor; therefore it seems that there is no substantial change in the mean before and after 1957. The standardized ratios for the annual total rain confirm this result (Fig. 1a). However, on a monthly basis, a significant step is detected with at least three stations in January, July, and December (Table 2). A closer look at the results reveals that the adjustments are often above 1.0 during winter and below 1.0 during summer. The monthly mean adjustments (Fig. 2a in black) show a clear seasonal pattern that indicates that the joining of the observations has produced a positive step in the winter and a negative one in the summer. In this particular situation, even if the spring and autumn do not show strong evidence of an artificial discontinuity, the monthly adjustments are applied at Digby to be in agreement with the values of Bear River. Even if the adjusted annual series does not show much change, the adjusted summer series shows less precipitation than the original data over 1915–56 (Fig. 3a).

Snowfall rarely occurs at these locations during the summer months; therefore, the time series of May–October are not tested for snow (Table 2). In the remaining months, the results show a significant step with most neighbors (except for November). The step can be easily identified visually in the standardized ratios of the annual total snow (Fig. 1b). The monthly mean adjustments are considerable for all months except November (Fig. 2b in black). Because there is no distinct seasonal pattern in the monthly adjustments, the long series adjustment is applied to all months of Digby over 1915–56. The adjusted annual series (not shown) and the winter series (Fig. 3b) show more precipitation than the original data over 1915–56. It is important to point out that this does not mean that the snowfall precipitation was incorrect at Digby, but the values had to be increased by about 40% to be in agreement with the snowfall precipitation at Bear River.

The history files were closely examined to explain further the cause of this discontinuity. Overall, the results show more precipitation at Bear River than at Digby. Because the same type of rain gauge and snow ruler were used at both locations, it is likely that the increase in precipitation is due to the site environment change. At Digby, the instruments were located on a rolling terrain sloping down with tall grass, but the exposure of the instruments was generally good, with no buildings or trees in close vicinity. At Bear River, the instruments are located in a valley surrounded by trees and hills. The land rises quickly to an elevation of 150 m to the west and the tall trees along the river bank on the opposite side can have an influence on precipitation catch. Therefore, the second location is more sheltered, which can result in higher precipitation catching rate depending on the wind direction. In addition, the rain gauge is located near a paved parking area and the readings on warm summer days can be influenced by that proximity, which can consequently lead to lower summer rainfall observations.

5. Validation

To validate the procedure described in section 3, the adjustments obtained by neighboring stations are compared with those produced by overlapping data. Daily precipitation observations are sometimes available at both locations for a period of time, and therefore it is possible to compute the adjustments using the simple ratio method (Thom 1966). When a first station is merged with a second station, the ratios can be derived as follows:
i1558-8432-48-1-156-e5
where P1i and P2i are the monthly or annual total rain (or snow) of the first station and second station, respectively, for year i. The ratios are then averaged for each month, the annual values, and the long series for the available overlapping period.

As an example, the parallel observations of Digby and Bear River were examined over 1953–65 (Fig. 4). The monthly and annual ratios (Bear River over Digby) were averaged over the 13 yr (Fig. 2). For rain, the adjustments computed from the overlapping data (in gray) are similar to those derived from the neighbors (in black). Both indicate a positive adjustment in the winter and a negative adjustment in the summer. The overlapping data also show only a slight difference in the annual values (Fig. 4a). For snow, the monthly and annual adjustments from overlapping data (in gray, Fig. 2b) indicate a positive adjustment for all months (except November); however, their magnitudes are smaller than those obtained from the neighbors (in black). In this case, because of the uncertainty corresponding to the monthly adjustments, it is preferable to apply the annual or long series adjustments to each month. The comparison of the overlapping annual snowfall precipitation measurements (Fig. 4b) confirms that there is more snow at Bear River than at Digby for the common 1953–65 period even if there is some year-to-year variability in the annual total snowfall difference between both locations.

The adjustments factors derived from the neighbors are compared with those from overlapping data at 60 stations located across Canada. These 60 stations have at least 10 yr of concurrent daily rain and snow observations at both locations. The difference between the adjustment factors derived from the neighbors and those from overlapping data are examined for each month, the annual, and the long series. The median, the 10th and 90th percentiles, and the minimum and maximum values are given in Fig. 5.

For both rain and snow, the medians are near 0, and they suggest that the adjustment factors derived from the neighbors are generally in agreement with those from overlapping data. For rain, the 10th and 90th percentiles are located near −0.2 and 0.2, which indicates a possible difference between both adjustments that is as large as 20%. The box is smaller from May to October, and therefore there is less uncertainty in the adjustments during these months. For snow, the 10th and 90th percentiles are located further from 0, suggesting a possible difference between the adjustments that is as large as 20%–25%. The boxes are larger for snow than for rain because snow amounts are more variable. For both elements, the annual and long series boxes are definitively smaller than the monthly boxes, suggesting less uncertainty in the annual and long series adjustments. Last, it seems that the long series box is not very much different from the annual box, indicating that the long series adjustments are not substantially better than the annual adjustments.

6. Results

The procedure described in section 3 was applied to 234 stations across the country. Overall, there were 164 stations composed of two segments (or two stations), 58 stations with three segments, and 12 stations with four segments or more. Altogether, a total of 324 joining dates were tested. In cases of three segments and more, the test was applied sequentially, starting from the most recent joining date and moving backward. The results show that not all stations required adjustments and that snow was more often affected by joining station observations than rain was. In the end, rain measurements were adjusted at 35% of the stations and snow measurements were adjusted at 58% of the stations (Table 3). The magnitude of the adjustments was also different: it varied from 0.75 to 1.25 for rain and from 0.65 to 1.60 for snow (Fig. 6). For rain, the annual and long series adjustments were often similar, and monthly adjustments were applied at only 12 stations for which a seasonal pattern was visually evident. For snow, the annual and long series adjustments were at times different and the long series adjustment was applied more often because it was computed from the months with frequent snow events. No monthly adjustments were used for snow because no seasonal pattern was observed.

Annual and seasonal trends were examined before and after adjustments for 1930–2007. The best-fit linear trend was used to estimate the magnitude of the trend. The adjustments for joining site observations depend on the site conditions, and the impact on the trends greatly varies with the location. Overall, it seems that the adjusted data present a more coherent spatial pattern of the trends than do the unadjusted data. Some adjustments have helped to resolve large trends caused by joining station observations. For example, a few large positive trends observed in the summer rain before adjustment (Fig. 7a) have become smaller with the adjustment and more in agreement with the surrounding area (Fig. 7b). The large positive trends observed in winter snow have also diminished after the adjustments (Figs. 7c,d), and the regions with increasing and decreasing snow have become more evident—in particular, in the eastern regions of Canada. There were a few large positive and negative trends observed in the spring and autumn that were not resolved by the adjustments, and these were probably caused by other homogeneity issues, such as changes in observers and relocation of the instruments, that have not been addressed in this work.

7. Discussion

The downsizing of the precipitation network is a problem that is well recognized all around the globe. In the past decades, when precipitation measurements were assembled into datasets for comprehensive global studies of climate change, it was realized that since the 1990s the number of stations reporting precipitation had substantially declined, mainly because of the closure of various networks established for specific projects and increasing restrictions by national agencies (New et al. 2001). In Canada, most stations receive a new identification number when they are relocated, and it becomes necessary to merge the station observations for studying trends. Also, there are a number of stations with long and reliable records that have unfortunately closed in the 1990s. These records sometimes cover 50–70 yr of observations, and they are extremely valuable for climate-change studies. Therefore, joining precipitation observations has become an important task for ensuring longevity and continuity of the observations in Canada.

Several issues should be considered before applying adjustments that are necessitated by joining station observations. First, monthly adjustments versus annual adjustments should be examined. Individual monthly adjustment factors seem to introduce some uncertainties in the time series because the monthly factors are often randomly scattered around the annual mean adjustment. However, in some cases, joining precipitation observations has definitively created a larger discontinuity in specific months, and it would be incorrect to adjust all months by the same amount. The Digby/Bear River rain datasets are a good example of this problem. Therefore, the annual adjustments should be applied to individual months unless there is a strong and visible seasonal pattern in the monthly adjustments.

Consideration should also be given to the record length to be adjusted versus the more recent record length. For example, when a station with observations for 1923–2002 (80 yr) is merged with a second station with observations for 2003–06 (4 yr), the observations of 1923–2002 would be adjusted to be in agreement with the most recent interval. This does not mean that the most recent observations are more reliable or of a higher quality. It is hoped that the same instruments will remain at the same location for a considerable period of time and that it would not be necessary to adjust the data each time that the datasets are updated with new observations. When the most recent segment is too short relative to the first segment, it is more difficult to determine whether a consistent step is well established. Therefore the ratio of 1 to 10 was applied to the Canadian datasets before adjusting the data. In other words, adjustments were applied only if the number of years in the most recent segment was greater than the number of years in the previous segment divided by 10. In the former example, the adjustments would be recalculated and applied only after 8 yr of observations become available, in 2011.

Last, one may consider adjusting monthly total precipitation instead of monthly rain and snow. Adjustments by neighbors and by overlapping data were also compared for monthly and annual total precipitation at the 60 climatological stations with concurrent observations. Overall, the results showed better agreement between both adjustments, and the 10th and 90th percentiles often ranged from −0.12 to 0.12. Therefore, this approach produces better results when it is applied to total precipitation. This is not unexpected for several reasons. Rain and snow have an evident seasonal pattern whereas there is precipitation in every month of the year. Rain and snow can occur in very small amounts during the shoulder seasons, and it can be more difficult to detect steps properly during these months. Snow has very high spatial variability, and, because the procedure is based on neighboring sites, the snow adjustments are more variable. Nonetheless, rain and snow were adjusted separately in this study to ensure continuity and longevity of these two climate observations.

8. Conclusions

A procedure to determine whether joining precipitation station observations has created discontinuities at joining dates is described. First, it was applied to monthly and annual rainfall and snowfall at 60 climatological stations for which overlapping observations existed at both locations to verify that the monthly and annual adjustments were representative. Overall, the adjustments by neighbors and by overlapping data were in agreement, although the difference between the adjustments was sometimes as large as 20%. The procedure was then applied to 234 joined stations. The results indicate that snow was more affected than rain by the joining of station observations, and it was necessary to adjust rain and snow at 35% and 58% of the stations, respectively. The annual and seasonal trends were examined before and after adjustments for 1930–2007. The adjusted data showed a more coherent spatial pattern in trends than did the original data. The impact on the trends greatly varies depending on the site conditions and environment.

More studies can be done to improve this work. An approach for creating a reliable single reference series from surrounding stations could be developed for monthly rain and snow, which have time series often composed of outliers and zero events. Homogeneity testing and adjustments should also be performed to make sure that all other unknown issues such as changes in observers and relocation of the instruments are resolved as much as possible. Several adjustments are needed to make the precipitation datasets free of any discontinuities. Adjustments for well-known issues such as changes in rain gauges, snow water density, and joining of station observations were addressed first because the uncertainties corresponding to these adjustments were more consistent and well documented. Last, more effort should be given in the development of techniques to adjust daily rain and snow to produce reliable daily datasets that can be used for the preparation and analysis of several climate-change indices.

Acknowledgments

The authors thank Robert Whitewood from the Climate Research Division of Environment Canada and three anonymous reviewers for their helpful comments and suggestions. The work done by Delafriya Dorabjee, a student from the University of Waterloo, in retrieving metadata information and organizing datasets was very much appreciated.

REFERENCES

  • Alexandersson, H., 1986: A homogeneity test applied to precipitation data. J. Climatol., 6 , 661675.

  • Begert, M., , T. Schlegel, , and W. Kirchhofer, 2005: Homogeneous temperature and precipitation series of Switzerland from 1864 to 2000. Int. J. Climatol., 25 , 6580.

    • Search Google Scholar
    • Export Citation
  • Brown, R. D., 2000: Northern Hemisphere snow cover variability and change, 1915–97. J. Climate, 13 , 23392355.

  • Devine, K. A., , and É Mekis, 2008: Field accuracy of Canadian rain measurements. Atmos.–Ocean, 46 , 213227.

  • González-Rouco, J., , J. L. Jiménez, , V. Quesada, , and F. Valero, 2001: Quality control and homogeneity of precipitation data in the southwest of Europe. J. Climate, 14 , 964978.

    • Search Google Scholar
    • Export Citation
  • Goodison, B. E., 1978: Accuracy of Canadian snow gauge measurements. J. Appl. Meteor., 17 , 15421548.

  • Goodison, B. E., 1981: Compatibility of Canadian snowfall and snow cover data. Water Resour. Res., 17 , 893900.

  • Hanssen-Bauer, I., , and E. J. Førland, 1994: Homogenizing long Norwegian precipitation series. J. Climate, 7 , 10011013.

  • Mekis, É, , and W. D. Hogg, 1999: Rehabilitation and analysis of Canadian daily precipitation time series. Atmos.–Ocean, 37 , 5385.

  • Mekis, É, , and R. Hopkinson, 2004: Derivation of an improved snow water equivalent adjustment factor map for application on snowfall ruler measurements in Canada. Preprints, 14th Conf. on Applied Climatology, Seattle, WA. Amer. Meteor. Soc., 7.12. [Available online at http://ams.confex.com/ams/pdfpapers/68724.pdf.].

    • Search Google Scholar
    • Export Citation
  • Metcalfe, J. R., , B. Routledge, , and K. Devine, 1997: Rainfall measurement in Canada: Changing observational methods and archive adjustment procedures. J. Climate, 10 , 92101.

    • Search Google Scholar
    • Export Citation
  • New, M., , M. Todd, , M. Hulme, , and P. Jones, 2001: Precipitation measurements and trends in the twentieth century. Int. J. Climatol., 21 , 18991922.

    • Search Google Scholar
    • Export Citation
  • Peterson, T. C., , R. Vose, , R. Schmoyer, , and V. Razuvaëv, 1998: Global historical climatology network (GHCN) quality control of monthly temperature data. Int. J. Climatol., 18 , 11691179.

    • Search Google Scholar
    • Export Citation
  • Thom, H. C. S., 1966: Some methods of climatological analysis. WMO Tech. Note 81, 53 pp.

  • Tuomenvirta, H., 2001: Homogeneity adjustments of temperature and precipitation series – Finnish and Nordic data. Int. J. Climatol., 21 , 495506.

    • Search Google Scholar
    • Export Citation
  • Wijngaard, J. B., , A. M. G. Klein Tank, , and G. P. Können, 2003: Homogeneity of the 20th century European daily temperature and precipitation series. Int. J. Climatol., 23 , 679692.

    • Search Google Scholar
    • Export Citation
Fig. 1.
Fig. 1.

Standardized ratios for annual total (a) rain and (b) snow (base over neighbor 5) for the stations Digby/Bear River, joined in 1957.

Citation: Journal of Applied Meteorology and Climatology 48, 1; 10.1175/2008JAMC2031.1

Fig. 2.
Fig. 2.

Monthly, annual, and long series mean adjustments obtained using the neighbors (black) and overlapping data (gray) for total (a) rain and (b) snow for the stations Digby/Bear River, joined in 1957.

Citation: Journal of Applied Meteorology and Climatology 48, 1; 10.1175/2008JAMC2031.1

Fig. 3.
Fig. 3.

(a) Summer total rain and (b) winter total snow for the joined stations Digby/Bear River before (gray) and after (black) adjustments.

Citation: Journal of Applied Meteorology and Climatology 48, 1; 10.1175/2008JAMC2031.1

Fig. 4.
Fig. 4.

Annual total (a) rain and (b) snow for Digby (gray) and Bear River (black) for the overlapping period 1953–65.

Citation: Journal of Applied Meteorology and Climatology 48, 1; 10.1175/2008JAMC2031.1

Fig. 5.
Fig. 5.

Box plots of the differences between the adjustments from neighbors and adjustments from overlapping data for (a) rain and (b) snow obtained from the 60 stations. The circle denotes the median, the box indicates the 10th and 90th percentiles, and the whiskers specify the minimum and maximum values.

Citation: Journal of Applied Meteorology and Climatology 48, 1; 10.1175/2008JAMC2031.1

Fig. 6.
Fig. 6.

Frequency distribution of the adjustments given in percent in classes of 0.1.

Citation: Journal of Applied Meteorology and Climatology 48, 1; 10.1175/2008JAMC2031.1

Fig. 7.
Fig. 7.

Trends in summer rain (a) before and (b) after adjustments for 1930–2007, and trends in winter snow (c) before and (d) after adjustments for same period. Upward- and downward-pointing triangles indicate positive and negative trends, respectively. The size of the triangle is proportional to the magnitude of the trend.

Citation: Journal of Applied Meteorology and Climatology 48, 1; 10.1175/2008JAMC2031.1

Table 1.

Identification number (in Table 2), distance, elevation difference, and annual correlation between Bear River and each neighbor.

Table 1.
Table 2.

Monthly, annual, and long series (LS) adjustments given by each neighbor for the stations Digby/Bear River, joined in 1957. Boldface type indicates that the adjustment corresponds to a step significant at the 5% level.

Table 2.
Table 3.

Number of stations with zero, one, and two adjustments for rain and snow. In total, 234 stations were tested.

Table 3.
Save