Reasonable length of forecast horizon in a time series

[ssd]

Suppose we have monthly totals of observed data for last 35 years. That data is of inflow of a river in a reservoir and monthly demands from the reservoir. We are interested to check the effect of construction of a dam in the upstream. The effect is, whether the downstream reservoir will have enough water to meet its dependent demands after the upstream construction. We also have 12 estimated monthly demands for the proposed dam which are assumed to be fixed for future years.
For future prediction of inflow and demand, I primarily will use a simple time series model using regression for linear trend, without cyclic component. My query is, how many future years can I predict reasonably. I know that there is no hard and fast rule, but still can I find a logical length of forecast horizon from some sort of thumb rule?

 

[.Scott]

The future inflow should not be a problem. Just compare weather conditions for the past 35 months to the full weather record. That's usually many decades and provides enough information to determine the range on annual inflows you can expect - long term effects (such as global warming) not withstanding.

To predict future demand, investigate your consumer base. Is the area you are serving about to experience a serious build-up. This information should be easily accessible through government agencies, for example those who issue building permits and who are as interested as you are in planning.

 

[FactChecker]

If you want to project into the future, I recommend using Monte Carlo simulations based on historical trends and variance. You can run many simulations with reasonable projected random data and see what the distribution of results is.