Trenberth: "...climate models do not perform well for blocking anticyclones, monsoons, tropical storms, or most intense rainfall events...These models have been successfully used to clearly demonstrate that human-induced climate change is real and substantial"

How To Relate Climate Extremes to Climate Change | ThinkProgress

[Trenberth] Attribution of climate change requires not only good data but also a good model to take apart any simulated signal. This means that the model must be capable of simulating the relevant phenomena with high integrity. Unfortunately, this is often not the case for individual events owing to model errors. For instance climate models do not perform well for blocking anticyclones, monsoons, tropical storms, or most intense rainfall events. In part this is resolution related, but in part it is because of model biases, so that basic features of the simulated climate are not quite in the right place. Fundamentally the model climate is not identical to the observed.

Uncritical use of a model, without testing its ability on the phenomena thought to be relevant has led to some studies to conclude that “it must be natural variability” owing to the way the testing is done. This is because the main test is to see whether it is due to human influences, rather than the reverse: to test whether it isn’t. Of course the real conclusion should be that the tools are inadequate. Or that the wrong questions are being posed! These days, climate change is pervasive, as the basic environment in which all weather forms is simply different than it was more than 30 years ago. That reflects the memory of all the human influences on climate. In addition there is the changed atmospheric composition and atmospheric heating occurring on a daily basis. How can there not be a human influence on all climate events?

As an aside, having an imperfect model does not preclude it from being used to explore changes in a general sense. These models have been successfully used to clearly demonstrate that human-induced climate change is real and substantial. However, event attribution requires considerable care and better models are essential. Testing models on such events, though, is one way to help improve them. It is important research.