One challenge was how to show the data in an easy-to-view visual manner to allow comparison of the data held by Environment Canada and GHCN/GISS for each station – in terms of which years had data. The annual mean data for all stations was in columns in a spreadsheet and using the COUNT function in Excel it was possible to report the presence or absense of data, by station by year, into another spreadsheet. Then, by assigning a number to each station where data was present and a blank cell (“ “) where absent using the function formula:
IF(COUNTA(‘Sheet number’!Cell number)>0, Station number, “ “)
Two things to note here. Firstly the paucity of data prior to the 1940s and secondly that the GHCN/GISS set is reduced to seven stations after 1990. I was considering trying to combine the stations in each set with an attempt at a proper offset and weighting by distance, but I realised that would be futile as these stations are only a small subset of what is available from each data set – I have chosen them only because they are available in both sets. However, just averaging anomaly data from these stations, regardless of offsets, does make some interesting points. So, here is a comparison:
What does this show? First off, GHCN/GISS data does run slightly warmer, by about 0.25⁰C. Then, looking at the period 1890-1940 first, there is a lot of variation in the (GHCN/GISS) data for the two stations Dawson and Hay River. This ‘volatility’ is hardly dampened when Fort Smith and Mayo are added in the 1920s. I guess there were a lot of temperature swings between years back then. Also note that in the Env. Canada data set Mayo essentially ‘reports alone’ for the first 12 years; there is a reasonably good agreement between this and the average of the four stations including Mayo in GHCN/GISS. In the 1940s we reach a station count of eight for both data sets and, suddenly the ‘volatility’ is dampened. Actually, in the this data there is a period of relative stability of temperatures across many of the stations at this time from 1950-1970. This is visible in the ‘spaghetti graph’ below that shows the data for all the stations (GHCN/GISS), although generally having more stations results in a dampening of that volatility and a less spiky graph of averages. There is then a spike in 1998, which shows up most strongly in Fort Smith and Norman Wells, followed by a period where the anomaly hovers around the +1⁰C value.
A few things strike me in looking at this data:
- The Arctic stations which are currently reporting a warm anomaly are not active prior to the 1940s in either data set, so a comparison isn’t possible.
- Cool anomalies in the Nunavut (Eureka, Coral Harbour) are are not necessarily cool in Yukon and NWT.
- In the data from 1945-current, there is that dip then rise that is common across the world.
- Current temperatures are not unprecedented within this set if we go back before 1945.
- Comparing this and the previous graph with E.M.Smith’s dT/dt graph (my version here), the warming in Canada seems to be less about the warming in the high latitude long-lived stations, and more about the loss of stations generally, after 1989/90.
Regarding that older data, the trouble is that Dawson stopped reporting in the GHCN/GISS set in 1990, and has only recently started again, (with lower temperatures). Hay River also does not report in GHCN/GISS after 1989, and we have to go to the Env. Canada data set to get comparative data; Env. Canada however does not have data from prior 1944, so we need both sets for a comparison. Fort Smith has a relatively long run of data in both sets and is decidedly warming!
So what about those anomaly graphs to look at the differences between the two data sets? Although there was not necessarily good agreement between the actual temperatures for stations in the two data sets (here), the anomaly data for Eureka was in reasonable agreement. It was also a similar story for other stations. In some cases there was an offset between the data (see below for examples for Dawson) however these could be almost completely accounted for by the fact that the data periods were different and therefore the value of the monthly anomalies and derived yearly anomalies were different.
So in the case of Hay River, if you calculate the monthly anomalies from each full data setthe there is an offset of 0.83 between the two anomalies. However, if you only use the overlapping period to calculate monthly anomaliesthere is no difference (as below, right).
So it seems that the relationship between the two sets was somehow locked in in 1989/1990 through anomalies. Now that leaves me with one puzzle – if the anomaly values for the data sets are very closely aligned or identical, why are there such differing absolute temperatures between the two datasets? (and… does it matter?).