How soon is too soon?

One of the issues that has come up more and more recently is: "You cant forecast that 10 days out!"

So when can you forecast what, and when should you do it? After all, the atmosphere is not perfectly predictable, there is uncertainty in knowing the current state of the atmosphere and uncertainty contained within the equations and parameterizations that we use to approximate the behavior of the atmosphere.



Not long ago, there was much debate about Hurricane Sandys' landfall and a few models hinted at a close approach or even landfall on the East Coast. How do you know when to start believing these forecasts?

What about the Atlanta #snowjam? Or even my delightful central OK snow?

What is the underlying strategy and what followup tactics are there to approach situations that have embedded uncertainty? Dos anyone really know?

The communication strategy appears to be:
1. Start getting awareness that a storm is brewing. (bad media launches into outlier forecasts)
2. Pinpoint when that storm might "rear its ugly head".
3. Pinpoint what might happen in your area. Maybe a scenario or two would be good, acknowledging when/if the details matter more or less*1.
4. Update frequently.

When should the models/ensembles begin to converge before an event? How many forecasts do you need to say its becoming more certain?

The answers are less than satisfying since it all depends on what you are forecasting! For my central OK snow the models were flip flopping even 9-15 hours out (much like severe weather). The differences were a factor of 2, meaning either 3 or 6 inches of snow*2. And yet the general features of interest were well forecast out 24 hours, maybe longer on the generics.  yet it was the details that produced the differences.

Thus is the quandry we face. We know the details make all the difference and yet sometimes those details come into focus too late, not at all, or worse - in hindsight. How exactly do we frame our skill on certain events and our uncertainty on others? How do we communicate that we have skill in XX situation and little skill in another (i.e. more uncertainty)?

Not easy questions. These questions beg us to understand our forecasts and how/why we made them, to vigorously try to calibrate ourselves in an ever evolving/changing modeling environment, and a fast paced communication environment. It takes serious work to do this scientific forecasting, and making it harder is the fact that these significant events happen to us rarely. This is why on every TV stations weather broadcast, there is a portion of the broadcast dedicated to an event "nearby" but has nothing to do with "our" local weather.

We may have trouble answering these questions, but unlike the groundhog your local meteorologist is always training.


*1 Give your audience something to calibrate to (relate to and remember).
*2 The difference between happiness and sadness lies somewhere in through "here".