Raise the Hammer

I feel I need to jump in here, despite my better judgement, because there seems to be some confusion about basic aspects of the difference between climate and weather models and the difference between predicting averages and fluctuations.

In climate models one is attempting to predict the long term trends of climate normals, i.e. average rainfall, average temperature etc.

Predicting averages is much easier than predicting instantaneous values when the underlying systems is very complicated and prone to large fluctuations. Note that these are averages over time and space (i.e. over many years and over large land areas).

Weather prediction, on the other hand, attempts to predict instantaneous values (i.e. whether it will rain in Hamilton 48 hours from now). This is much more difficult because rainfall fluctuates enormously both in time and space (e.g. it might rain in Dundas, but not in Stoney Creek). This is why rainfall predictions are now accompanied with probabilities.

Both climate models and weather models assimilate data to keep them on track, and there are sophisticated mathematical models that ensure the data is assimilated into the models in the optimal way. Obviously, climate models must rely mostly on historic data, whereas weather models can be continuously updated. However, climate models can be tested using "post-diction", i.e. using them to predict past climate variation using historic data. By varying initial conditions and model parameters, one can estimate the degree of sensitivity of climate models: this is why a range of possible scenarios is always presented.

Finally, although weather fluctuates enormously from day to day and even from year to year, the averages defining the climate evolve much more slowly, which is why it is possible in principle to predict climate changes even when the underlying weather systems vary enormously.

To understand the basics of why the mean (or expected value) can be well defined even when the data has large fluctuations, google "law of large numbers". Even data with infinite variance can have a well-defined expected value, which can be estimated from the sample average.