Statistical models are all about quantifying uncertainty. As Tim notes, if there wasn’t any uncertainty (or if the uncertainty was not quantified) that would be a cause for suspicion of the results. These models give us predictions AND confidence intervals for those predictions. Those confidence intervals are interesting specifically because they describe the uncertainty in the predictions.
There is more that will not be obvious. You can have a statistically significant model that gives terrible predictions. This might happen because the wrong sort of model is being used. There are methods to test for Goodness-of-Fit, which test for variability in model predictions in excess of (or lacking from) expected random variation.
I haven’t read the report, but the graph Tim posted appears to be a sort of Variance Components analysis (VC):
The difference here is VC analysis specifically focuses on sources of uncertainty, and tells us how much uncertainty is due to each. Here the authors have combined predictions and uncertainty (expressed as confidence intervals) to show sources of uncertainty over time. Uncertainty increases as the predictions go farther into the future, as they should (predicting the future is hard!).