Given a lack of stations, ships, buoys, etc., it is useful to extrapolate, if extrapolation can be performed skillfully. Fortunately, it can. The estimated error, largely due to lack of spatial coverage, is 0.05°C (95% confidence limit).
http://demeterdesign.net/gistemp2010_draft0319.pdf
Regarding Arctic temperatures, here's what the GISS website states:
There are several reasons for the small discrepancies that exist between the three records. Most important, subtleties in the way the scientists from each institution handle regions of the world where temperature-monitoring stations are scarce produce differences.
While developed areas have a dense network of weather stations, temperature monitoring equipment is sparse in some parts of the Amazon, Africa, Antarctica, and Arctic. In the Arctic, particularly, the absence of solid land means there are large areas without weather stations.
The Met Office and the NCDC leave areas of the Arctic Ocean without stations out of their analyses, while GISS approaches the problem by filling in the gaps with data from the nearest land stations, up to a distance of 1200 kilometers (746 miles) away. In this way, the GISS analysis achieves near total coverage in the Arctic.
Both approaches pose problems. By not inferring data, the Met Office assumes that areas without stations have a warming equal to that experienced by the entire Northern Hemisphere, a value that satellite and field measurements suggest is too low given the rate of Arctic sea ice loss.
On the other hand, GISS's approach may either overestimate or underestimate Arctic warming. "There's no doubt that estimates of Arctic warming are uncertain, and should be regarded with caution," Hansen said. "Still, the rapid pace of Arctic ice retreat leaves little question that temperatures in the region are rising fast, perhaps faster than we assume in our analysis."
http://www.giss.nasa.gov/research/news/20110113/