Despite what some folks will try to tell you, global warming IS happening. This is now observable fact, beyond all reasonable doubt. Most scientists have long been pretty confident that the release of massive quantities of carbon dioxide into the atmosphere, and the consequent rise in the amount of carbon dioxide in the atmosphere [link], would have this effect. However, measuring the Earth’s temperature is not straightforward. It varies from place to place and from time to time in a chaotic* manner. We can’t measure it everywhere continuously to get an average – and even if we could, that average over the Earth’s surface would vary randomly in time around an average value.
This is a general problem with most measurements – for some things, like a child’s height for example, the random variations (usually called noise by scientists) are quite small compared with the changes of the average over time that we’re interested in (the signal). In the case of global climate, the noise is significant compared to the signal, which makes it difficult to measure the signal. Until recently, this was sufficient to make it uncertain whether the global temperature really was rising.
Noise isn’t the only issue. There’s also a problem of consistency in the method of measurement from year to year, and a problem of whether the places that temperatures are measured are really representative of the world as a whole.
We have more measurements in some parts of the world than others, and more measurements from some kinds of places (particularly towns and cities) than others (particularly deserts, mountains and tundra). It’s possible that the places we measure may experience temperature changes different in some consistent way from those in the places we don’t measure – for example, we might have more measurements in or near towns, and towns might be warming while the countryside is not. How should we weight the different measurements when we work out an average?
The methods of measurement have changed over the decades since measurements began. Can we be sure they’re actually consistent? For example, for several years satellite measurements of global temperatures appeared to show a falling trend, while surface measurements showed a rising trend; the satellite measurements turned out to be mistaken because they weren’t allowing for gradual changes in the orbits of the satellites. When the orbital changes were taken into account, the two sets of measurements came into line [link].
How can we estimate temperatures from times before people were taking measurements at all? Right back into the ice ages – and beyond? There are methods, but they’re obviously indirect, based on things like measurements of isotope ratios in ice cores from Antarctica or Greenland, tree rings, or layers of sediment in lakes. The interpretation of these proxies for temperature is open to some debate [link], but generally the pattern is consistent and reasonably clear.
All these complications mean that there is always going to be some room for argument about whether we’re really seeing a rise in global temperatures or not. It isn’t a case of “one day we’re uncertain, the next day we know”. As the signal size gets larger, we can be more and more confident that what we’re seeing is a real signal, not just a random bit of noise that happens to be going one way a bit more than usual. Confidence levels in the reality of the global warming signal are now very high indeed.
*Chaos is an interesting branch of mathematics, see Butterflies, Chaos & Feedback.