The link above from War on the Rocks has some key thoughts about military forecasting which are worth noting.
First, the author doesn't believe forecasting can go out more than 20 years. Some commentators in the article argue as little as two years.
Second, the author recommends focusing on only developed systems or emergent technologies and not science fiction. Probably a practical suggestion, however, was the impact of the smart phone or the web in the last 20 years a forecastable event?
Third, the author suggests using a bounding approach where course corrections are made every few years. This is good logic I think especially with a long term program, but in real life it is as likely that the plan will be tossed entirely and replaced.
Last a commentator on the article suggests that namely it is the application of new technology to traditional methods and goals. I think this has some justification. The author uses the reference to the blitzkrieg as an example.
If I had a real fault on military related forecasting is its utter inability to forecast the location and timing of the conflict itself. Was the second Iraq war forecastable? 911? The Syrian civil war?
My personal opinion is that forecasting beyond two years is very difficult as the scenarios change so quickly and people are adapting real-time to changing situations. Long term computer programming always pushes mathematical trends toward extremes, but in real life people and complex systems adapt along the way. This kind of forecasting overshoot can be seen in environmental forecasting where variables have to be generalized and simulated. Also my observation is that large scale government programs seem to over promise and under deliver while small widespread technical innovations have huge global impacts - cell phones, smart phones, twitters, etc.