Quality Control Procedures

I have listed below some of the procedures I use for quality control (non-exhaustive). This page was last updated in 2020; some methods may have been updated since.
How is spurious precipitation identified?
  1. Determine whether or not precipitation fell on each day in the dataset, using various sources (ECCC Hourly data in Archives, and metar/airway reports retrieved from Ogimet and NOAA). 
  2. If no precipitation was reported to have fallen but a precip amount is shown in the Archives (usually 0.2 to 2.0 mm), verify the individual case using radar if available (2008-present) and logic (sunny day? CB/TCU/distant SHRA reported? Did nearby stations report precip? etc). 
  3. If clearly no precipitation fell on the date, mark the precipitation amount as spurious and either: move it to another date (such as in the case that it is suspected to be a remnant amount from the previous day... or if the amount is extremely small (0.2 mm)), or eliminate it (such as in the case that it is a significant spurious amount of 0.5 mm or more, or if it is far from any day with precipitation therefore cannot be leftover precipitation from a previous date).
  • ​Spurious precipitation is mostly a problem with automated weather stations (such as XWG for Winnipeg Airport) because they undergo limited quality control. Spurious precipitation can be the result of leftover precipitation from a previous date, dew from fog or condensation, blowing/drifting snow, or wind. 
Other methods
  • For daily maximum and minimum temperature, compare the value with the hourly data in ECCC Climate Archives. If there is a significant discrepancy between the two, flag the daily value and verify each case manually. This method has found several data entry errors in the Archives, such as missing a minus sign on a value that is supposed to be negative, or adding a ''1'' by accident in front of a single digit number, or typing a 4 instead of a 5 for instance. 
  • Determine whether or not precipitation fell on each day in the dataset, using various sources (ECCC Hourly data in Archives, and metar/airway reports retrieved from Ogimet and NOAA.) Verify manually each day that reported a snowfall amount but no snow was reported. Do the same for rainfall. This method has found several occurrences of reporting the wrong type of precipitation accumulation (such as reporting a rainfall amount in the snowfall column by mistake). This could be due to manual data entry error or a weakness in the automated quality control procedures. 
  • Compare snow depth values with synoptic reports. If there is a discrepancy, verify each case manually. This method has found several data entry errors, such as entering the snow depth value on the incorrect date. 
  • In general, fill in missing data using synoptic reports, metar, backup stations or through inference (such as filling in rainfall from known precipitation amounts). 
  • ​By comparing with nearby stations, determine whether the Airport has an undercatch bias for snow-water equivalent (due to its exposed location around farm fields). Replace with The Forks if undercatch is an issue. If The Forks is unavailable or erroneous (sometimes it has excessive spurious readings), use Charleswood or keep the Airport amounts. This method is meant to reduce a dry bias caused by undercatch of snowfall at the Airport and is only used during the period of automated measurements (2013 to present). 
  • Compare precipitation amounts with nearby stations to find errors or unrealistic values. This method catches unrealistically high precipitation amounts in the winter amounts due to the nipher shield problem, which results in precipitation around two times higher than reality when it is raining. Replace with a nearby station if it is a major issue.
  • If the minimum dewpoint for the day from the hourly data is higher than the daily low temperature, use the daily low temperature as an estimate of minimum daily dewpoint. If no hourly dewpoint values are available (such as before 1953), use the minimum daily temperature as an estimate, but only use these for record low minimum dewpoint records (not for record high minimum dewpoint records). 
  • When no daily wind gust value is available in the Archives or synoptic reports, use metar to retrieve a maximum wind gust for the day, if available. This value might come from the remarks section if ''PK GST ##'' or ''G##'' is reported. Missing daily maximum wind gusts mostly occurred from 1980 and later. Some could not be filled due to the absence of wind gust reports or gusts not being reported at the time of maximum sustained wind. 
  • ECCC errors in rounding are corrected. This is for monthly average high/low/mean temperature. When the average ends up being something like 5.4495, ECCC rounding procedure records this as 5.5 (rounding twice, first as 5.45 then 5.5), but I change this to the mathematically correct version of 5.4. For Winnipeg, this was mostly an issue in the 21st century. As well, in the monthly data in ECCC archives, there are years where ECCC calculated the average monthly mean temperature as the average between the average monthly maximum and minimum temperatures. However, I have changed these to be the average of the daily mean temperatures instead. For Winnipeg, this was mostly an issue in the 1977 to 1989 period. 
  • In hourly data, flag any occurrence of freezing rain or freezing drizzle above freezing, and liquid precipitation (rain, drizzle) below freezing, and correct if an obvious error is detected. 
  • Particularly in 20th century years, fog was reported in moderate or heavy snow due to visibility falling to 1/2 SM or less, but the reduction in visibility was due to the heavy snow and not fog. In these cases, the fog day is eliminated. 
  • Use monthly meteorological summaries from ECCC office from 1977 to 2004 to fill in missing gusts that could not be retrieved from synoptic reports, and to confirm or detect errors such as data entry errors.