Such a prediction requires a model of the underlying processes or at least distribution of values from observations. The distribution of values in a dataset mi

Such a prediction requires a model of the underlying processes or at least distribution of values from observations. The distribution of values in a dataset might suggest that the process producing these values does so in a manner consistent with e.g. a gaussian distribution with a given expectation value and standard variation. Now if a particular value is observed, this model can be used to estimate the probability of that value occuring. If it is also known how often these events happen per year then it's possible to say how many years of observation would be needed to observe such a value with probability of e.g. 50%, 95%, or whatever threshold you choose.