I’ve not yet had time to delve into the GISTEMP code changes to use GHCN v3 in place of GHCN v2. This post simply takes a look at the treatment of two stations, Dublin Airport and Reykjavik, which have been mentioned on other blogs, and is intended in part for use to provide graphs to illustrate comments I may make or questions I may ask on those other blogs. I have some further comments to add, but I will keep these for further posts, and for now just put these images and brief comments in this post.
The graphs below show four comparisons:
- The GISTEMP adjusted temperature series using GHCN v3 vs the previous GISTEMP adjusted series using GHCN v2
- The two corresponding “raw” series, i.e. the GHCN v3 adjusted series and the GISTEMP STEP1 combination of records from GHCN v2.
- The GISTEMP adjusted series using GHCN v3 vs the “raw” data, i.e. the GHCN adjusted series
- The GISTEMP adjusted series using GHCN v3 vs the previous “raw” data, i.e. the GISTEMP STEP1 combination of temperature series from GHCN v2
The images have been prepared by combining the PostScript output graphs from the GISTEMP station data page, without wasting time to remove every redundant part of the combined images. Remember that it is the use of GHCN data in GISTEMP which is being considered here, so all references to “raw” data here, unless specifically stated otherwise, are to “raw” data for GISTEMP, which is of course GHCN adjusted data.
Both GISTEMP and GHCN impose a zero adjustment in the final year, so that in principal the 2011 values plotted might be expected to coincide if the raw data is identical. In practice, any missing values change this. The plots for Dublin Airport which were combined here were downloaded from GISS earlier this month, at which time GISS returned v3 data only up to August 2011 (data to February 2012 is returned today). As a result GISTEMP, missing data for September to November, calculates an annual temperature from the earlier nine months (8.75 for Dublin Airport) which is biased low. The temperature returned today, with complete data, is 9.40, but it is the lower value, 8.75, which is used in these plots.
On the other hand, a new error appears to have been introduced on the GISS station data for v2 page. The same September, October and November 2011 data values, 14.0, 11.8 and missing, are returned both now and earlier in the month, but whereas earlier in the month the S-O-N mean was correctly calculated from the two available values as 12.9 – a quarterly mean may be calculated from the available months as long as no more than one monthly value is missing – giving an annual mean of 9.66, now the mean of 14.0 and 11.8 has become 11.4 and the annual mean 9.29! I have been unable to guess how this value of 11.4 was obtained. It would require a November mean temperature lower than that given by the CLIMAT report, 9.8, or by Met Eireann, 9.7. Just to make matters a little more complicated, the actual PostScript file I download for the v2 adjusted data was downloaded after the change, and so plots as 9.287 in 2011, while the text data I downloaded for examination was downloaded before the change.
To summarise, the 2011 values plotted for Dublin Airport will not be identical because of differing missing values, even though all available data values for each individual month, GISS or GHCN, adjusted or unadjusted, are identical.
The difference in plotted values for Reykjavik for 2011 is smaller. The only missing values are the GHCN v2 values for November, as this series ends in October 2011. There is still a small difference in addition to that caused by that missing value, as the GHCN v3 values for 2011 are all 0.1 degree higher than the corresponding GHCN v2 values.
lapogus at Bishop Hill has commented
Sorry Richard, nothing personal, just need to rant now and again to keep sane. By the way, you are in Dublin yes? Have you seen how the historical temperature data for the station at Dublin Airport has been adjusted by GCHNv3 and GISS (compared with GCHNv2)? Truly shocking. Ask BH for my email and I will send you the gifs.
lapogus: I will post a comment at Bishop Hill to ask if the images below correspond to your gifs, and to invite you to comment further here on how you see the adjustment of the historical data for this particular station, Dublin Airport. “Truly shocking” calls for further explanation. Please feel free to add your gifs to your comment. It will get held for moderation, but I will clear it as soon as I see it.
In this case, unlike Reykjavik shown further below and discussed at other blogs, the effect here of the change is to warm temperature data, by very little at either end of the time range (but see next paragraph), but more substantially in the “middle” years, particularly here from 1985 to 2007. The temperature rise from 1881-2011 is virtually unchanged, but the temperature rise from intermediate years to the present is reduced. It is probably worth noting here also that the GHCN v3 adjusted data, which is then used as input to GISTEMP, successfully detects a station relocation in May 1994, which I understand to be the correct year and month. More on this however in a separate post to come.
What is the caveat which started to appear in the last paragraph? The GHCN v3 adjusted data, input to GISTEMP, now, later in March, shows slightly higher values in 1994 and before, rising by 0.04 degrees in 1994 and 0.10 degrees in earlier years when compared to the data downloaded earlier in the month. The adjusted output from GISTEMP however shows a much greater change. From 1881 to 1993 all values are cooled by 0.2 degrees, other than 1902, where the cooling is only 0.1 degrees, and 1925, 1948 to 1950 and 1971 to 1974, where the cooling is 0.3 degrees. In 1994 (where GHCN v3 correctly detected the station relocation) the cooling increases to 0.358 degrees, then to 0.4 degrees for 1995 to 1998, declines to 0.3 degrees for 1999 to 2004, 0.2 degrees for 2005 and 2006, 0.1 degrees for 2007 to 2009, and zero for 2010 (the 2011 difference is not relevant here). This of course adds a further 0.2 degrees temperature rise between 1881 and 2011, and changes the temperature changes found from other years to the present. Rather than illustrating this here I will now finish this post and start another to compare the early March download with the current download.
Reykjavik has featured at a number of blogs
As commented elsewhere, in this case the effect of the change is a cooling of the past (with a very small warming of very recent years), leading to an increased temperature rise over the period shown.