The notorious (in the UK at least) failure to predict the storm in 1987 was because they only had the compute power to run the simulation once. These days if they run their programme on the same data many times then they can see the storm in about 40% of the runs.
Biting the hand that feeds IT
