I have no particular brief for Ferguson and I prefer solid data to forecasts too. Ferguson probably does not generate the data on which the forecasts are based, so I guess what happened is that the data he was provided with justified a pessimistic assumption, but then as more data came ion, it changed. The problem is that data tends to change, although gets better with time. It often starts closer to one extreme on the optimistic-pessimistic spectrum, then shifts to the other end, but gradually centralises around the "true" answer with time. Probably best not to keep changing the model output in the early stages as the data bounces around looking for more accurate answers.
The question of which data set is true/accurate and which is the proper basis for policy decisions is a trickier one. There is a principle in public health called the precautionary principle, which basically says that you should assume the worst as it is easier to course-correct from there than from assuming the best and having to scrabble to increase activities. So it is wise to have a margin for error and if you think there will be 1000 deaths, plan for 5000 instead. The effects of getting these things wrong tend to occur on an exponential scale.
Thanks. Makes a lot of sense.