I wonder why when we seem to get inside a certain time frame we see these major flips...it certainly doesn't happen every time but it always seems like we get inside this magic number and poof...a complete 180 happens.
Is it perhaps that everything is being sampled better? Perhaps its a parameterization function (outside of a certain time range doesn't the parameterization schemes change?)? Or maybe its a type of bias resulting from the ensemble mean smoothing key features just a bit too much?