The same way any programme that has risk in in it is developed- you take a best guess. Link from task 1 to task 2 to determine the earliest start date, then put a duration for task 2. you can, if you have time and motivation, do a set of programmes showing pessimistic, optimistic and realistic views, but I've never seen a point to it.
Start dates will vary as you update a programme to reflect actual progress, so the risk will be lessened as time goes by.
Generally, risk bars can be inserted in the programme- either one bar to represent realistic risk at the end of the programme, a succession of bars at the end of each section, or a bar for every task where there is a high risk. In this case you can filter to show risk bars, then set to max or 0 to get best and worst cases.
For me, a reasonable estimate with risk bars will give you as good as you are going to get. Experience is essential to gauging risk, and is the only real way of getting something that will work.
Generally, using Time Risk Allocation within tasks will give you something that will work. In cases where tasks are subject to delays beyond your control, you will never be able to give accurate forecasts, but can give fairly accurate overall completion dates.