That would be a good test. Probably they could go back only 20 or 30 years when they have better data to start from and find out how well it works. They still probably won't get tomorrow's weather right anyway.
The prediction of tomorrow's weather is not necessary, but the prediction of variation that matches reality is. What is not needed for climate modeling is accurate timing of weather features. What is needed is accurate depiction of weather features and increasing time and space resolution as these people are doing is a step in that direction.