Daily high and low temperatures are averaged for a composite "daily temperature".
The problem is that cities have grown dramatically all over the world. Skeptics say that this is a significant bias in the reported climate data sets. There's a very good discussion of this subject, showing a major difference in 20th Century climate between urban and rural areas:
Because tree ring temperatures disagree with a sharply rising instrumental average, climate scientists officially dubbed this the “divergence problem.”9 However when studies compared tree ring temperatures with only maximum temperatures (instead of the average temperatures that are typically inflated by urbanized minimum temperatures) they found no disagreement and no divergence.10 Similarly a collaboration of German, Swiss, and Finnish scientists found that where average instrumental temperatures were minimally affected by population growth in remote rural stations of northern Scandinavia, tree ring temperatures agreed with instrumental average temperatures.11 As illustrated in Figure B, the 20th century temperature trend in the wilds of northern Scandinavia is strikingly similar to maximum temperature trends of the Sierra Nevada and the contiguous 48 states. All those regions experienced peak temperatures in the 1940s and the recent rise since the 1990s has never exceed that peak.Remember, the whole "hide the decline" episode was triggered by scientists trying to cover up this "divergence" between tree ring proxy data and temperature readings. And the plot thickens when computers massage the data:
It soon became obvious that the homogenization process was unwittingly blending rising minimum temperatures caused by population growth with temperatures from more natural landscapes. Climate scientists cloistered in their offices have no way of knowing to what degree urbanization or other landscape factors have distorted each weather station’s data. So they developed an armchair statistical method that blended trends amongst several neighboring stations,17 using what I term the “blind majority rules” method. The most commonly shared trend among neighboring stations became the computer’s reference, and temperatures from “deviant stations” were adjusted to create a chimeric climate smoothie. Wherever there was a growth in population, this unintentionally allows urbanization warming effects to alter the adjusted trend.Translation: the warming signal in the data may be mostly (or even entirely) due to urbanization, not to carbon dioxide.
In the good old days, weather stations such as the one in Orland, CA (pictured above) would have been a perfect candidate to serve as a reference station. It was well sited, away from pavement and buildings, and its location and thermometers had not changed throughout its history. Clearly Orland did not warrant an adjustment but the data revealed several “change points.” Although those change points were naturally caused by the Pacific Decadal Oscillation (PDO), it attracted the computer’s attention that an “undocumented change” had occurred.
The data sets are frightfully bad. There's very poor transparency, whether by accident or design. Indeed, the CRU even lost their raw temperature data, only retaining the "value added" (adjusted data). The bottom line is that the numbers are cooked, and so headlines referring to "hottest decade" are meaningless. If you really want to understand what's happening in climate science, you should read this post in its entirety.