Article: How good is the NASA GISS global temperature dataset?

[And an update with more adjusted data]

Take away - 82.5% of all weather stations have siting issues that increase the recorded temperature. In addition, these were then used as the standard by which pristine sites were adjusted, increasing their temperatures as well.

It is generally accepted that there are two major land temperature record issues: microsite problems, and urban heat island (UHI) effects. Both introduce warming biases. 

The SurfaceStations.org project manually inspected and rated 1007 of 1221 USHCN stations (82.5%) using the 2002 Climate Reference Network (CRN) classification scheme (handbook section 2.2.1). The resulting preliminary paper shows a large temperature trend difference (about 0.1C/decade) between acceptably sited stations (CRN 1 or 2) and those with material microsite problems (CRN 3, 4, 5).

[snip]

In the Surface Stations supplemental materials (available at www.surfacestations.org) only 14 CONUS stations have pristine CRN 1 siting (1.2%). 4 are labeled urban, 3 are suburban, and 7 are rural. Since these 14 have zero microsite issues, they can be used to examine the GISS UHI homogenization.

[snip]

A spurious warming trend was introduced into all three suburban and 6 of 7 rural CRN 1 stations. Only Apalachicola FL emerged from GISS unscathed. 

Automated homogenization algorithms like GISS use some form of a regional expectation, comparing a station to ‘neighbors’ to detect/correct ‘outliers’. BUT 92% of US stations have microsite issues. So most neighbors are artificially warm. So the GISS algorithm makes the hash illustrated above. How could it not? And by extension NCDC, BEST, Australian BOM, …

Additionally, the older data from some of the pristine sites was adjusted cooler which either introduced or exaggerated a warming trend.

Update: HadCRUT4 joins the terrestrial temperature tamperers

Even though the satellites of RSS and UAH are watching, all three of the terrestrial record-keepers have tampered with their datasets to nudge the apparent warming rate upward yet again. There have now been so many adjustments with so little justification – nearly all of them calculated to steepen the apparent rate of warming – that between a third and a fifth of the entire warming of the 20th century arises solely from the adjustments, which ought to have been in the opposite direction because, as McKitrick & Michaels showed in a still-unchallenged 2007 paper, the overland warming in the datasets over recent decades is twice what actually occurred.