“Only he that has travelled the road knows where the holes are deep” (Chinese Proverb)
And so it is with climate data for China, for which there is only word – ‘interesting’…but I leave that to you to judge.
China is of similar size to the US – 9,600,000 sq. kilometers (3,700,000 sq. miles) and has 417 stations reporting climate data, of which (as of January 2009) 414 made it through into the GIStemp set of stations contributing to their climate analysis. The longest serving stations are Bejing and Shanghai. As we know from the “station drop-out problem“, China is one of those countries that for which there was a large addition of thermometers in the 1950s and a massive loss of them from the GHCN record in 1990. Most of the discussions of this problem, for example at Lucia’s site The Blackboard here and here, insist there is no effect on a global scale of this loss of thermometers. Nor should there be any effect from dropping thermometers from specifically cold regions (here). Nonetheless I am interested to see how this change affects the raw data.
I’ve been mainly concerned with trend data – i.e. using tables of the trend produced by a linear regression of individual station data. I’ve analysed data from the GISS combined/adjusted data set – i.e. that which has already gone though the GIStemp adjustment process that corrects for UHI effects etc. and therefore will actually contribute to the GIStemp output data; here I’ve only divided it by whether individual stations have an overall warming or cooling trend across the station data history.
providing data in the GHCN v2.mean file.
I’ve used this data to plot (Figure 2) counts of years when stations were dropped from the database and clearly the data shows a small drop-off in thermometer numbers in 1988 and a huge loss in 1990. This is where, if there is one, I would expect to see an immediate effect on temperature differences. In 2009 we are left with twenty eight stations with an overall warming trend, of which eight are rural stations, and just three with a cooling trend (only one rural – 1939-present).
I’ve also examined in the same way the year of addition of stations. In this case an immediate effect would perhaps not be apparent as I would expect it might take several years for the trends of the new stations to assert themselves over the ‘noise’ of yearly weather fluctuations.
- A change from the addition of stations from 1908 (and perhaps from 1933)
- Change from the addition of stations in the 1950s
- Change from the loss of stations from 1988 and/or 1990
Linear regressions for periods are as follows:
1908-1950: y = 0.0118x – 24.2451, R^2 = 0.03141
1951-1988: y = 0.0013x – 4.3841, R^2 = 0.0032
1988-2010: y = 0.0668x – 134.24, R^2 = 0.7828