The problem is that it is strongly varying and thus likely to diverge very quickly
That's a common problem, the usual answer is to frequently recompute the basis , with new observations.
If your model doesn't fit the data in the first place it's time to work out why. No deterministic model that comes close to reality will give exponential growth. A stochastic model may but you're supposed to run those a number of times and average, the average will only give exponential growth if there is something strange with the random number generator. Just continuously re-fitting a bad model doesn't strike me as being useful.
Just continuously re-fitting a bad model doesn't strike me as being useful.
you have new observations over time . So of course it's useful, as it tells you how inaccurate your original model is likely to be, since the initial condition is by definition solid. hence, however far into the future your model is reliable is extended out. It's the only way, for example, to make orbit adjustments to satelites for example, to keep them on station.
The problem is that it is strongly varying and thus likely to diverge very quickly
That's a common problem, the usual answer is to frequently recompute the basis , with new observations.
If your model doesn't fit the data in the first place it's time to work out why. No deterministic model that comes close to reality will give exponential growth. A stochastic model may but you're supposed to run those a number of times and average, the average will only give exponential growth if there is something strange with the random number generator. Just continuously re-fitting a bad model doesn't strike me as being useful.
Of course you are right. But the climate change Priests won't even do that. They want their models that replace real world truth to be the only means to support thei claims
The problem is that it is strongly varying and thus likely to diverge very quickly
That's a common problem, the usual answer is to frequently recompute the basis , with new observations.
If your model doesn't fit the data in the first place it's time to work out why.
All models are different from reality. The best we could ever expect is that its behavior is close enough to reality that it is a useful way to represent reality, flixibly, so that the importance of various parameters can be probed.
Just continuously re-fitting a bad model doesn't strike me as being useful.
As a general rule, if a model is useful, stale data often makes it less useful or even useless.