The length of the time series is not synonymous with short or long term period as we can say for a forecast for example.

Results of **VGEST** are the probabilities calculated from a time series which is supposed to represent a given stationary hydrological regime. We are interested in events up to a return period of 100 years, and looking at such an event in **VGEST** results is as good as studying one particular event arises from 109 years of data. These rare drought and flood events emerge as the result of specific sequences of events, unlikely to be repeated.

Since the results of **VGEST** are intended to guide the decision-maker for future events, is it wise to base our strategies on past circumstances that will probably never be repeated and will not be exceeded? Certainly not. And the reason is given by https://waterprogramming.wordpress.com/2017/02/07/synthetic-streamflow-generation :

(...) the use of synthetic streamflow can improve the efficacy of planning and management strategies by exposing them to larger and more diverse flood and drought than those in the record (Loucks et al. 1981; Vogel and Stedinger, 1988; Loucks et al. 2005).

Consequently, the longer the time series is, the more diverse are the events in it and the results related to events with a return period up to 100 years should converge to a stable solution.

If you can prove by any relevant indicators that 100 years return period events are stable from a 2000 years generated streamflows and there is no significant change for longer time series, so let's use 2000 years generated streamflows!

I don't understand your reasoning:

- You said playing with more than 2000 years is not relevant. Why not. But after that you propose to play with time series until 5000 years. Where is the coherence?
- I said that having several length is not interesting. You don't react on this and you continue to propose several times series without any argument proving that it's interesting to do so anyway...

Here is the question, what are durations do you want to present? My preference is

`seq(100,2000, by = 100)`

. I felt that we don't need to present synthetic data longer than 2000 years.

What is the goal of using generated streamflows? One answer is given here https://waterprogramming.wordpress.com/2017/02/07/synthetic-streamflow-generation/:

Even under stationarity and even with long hydrologic records, the use of synthetic streamflow can improve the efficacy of planning and management strategies by exposing them to larger and more diverse flood and drought than those in the record (Loucks et al. 1981; Vogel and Stedinger, 1988; Loucks et al. 2005).

So why do you want to provide results on times series of 100, 200, 300..., 1900, 2000 years? As I said in in-wop/article-irmara#6, the main idea is to work with long time series in order to obtain more robust probabilities than with the short historical time series.

Thus, the longer the time series generated, the better, unless there is no significant difference in the results of VGEST when working with a longer time series. I don't see the benefit of working with a large sample of time series length.