More than numbers

Weather forecasts are much more reliable than they used to be, mostly thanks to more powerful computers.  Input from people is still important, though.

Here in Alexandria we are looking at a fall of snow the week after shifting to Daylight Savings Time.  It was no surprise, being forecast some days ago, and appears to have followed the forecast closely, even to the amount of snowfall.  This must be gratifying to the many people who have worked hard to improve the weather services.

Weather is a hard problem.  There are many variables and the basic structure is mathematically chaotic, meaning (among other things) that small differences in input lead to big differences in output after a certain time.  Even the biggest and best computers cannot see beyond something like eight or ten days; adding another day might require ten times the computing power.  Even then, the fact that we do not have weather instruments everywhere on the globe means the input data are incomplete.

So there are still weatherpeople who are not (primarily) computer programmers.  If you look at the National Weather Service’s web page and put in your location for a forecast, scrolling down a bit below the map you’ll find a “forecast discussion.”  It will refer to computer models but also to other arcane bits of scientific weather lore, upper-level lows and 750mb maps, the effect of daytime warming on humidity and whatnot.  It is obviously written by a human.  Moreover, the human often mentions judgements made by humans.

The point is not that humans are superior to machines, as our Romantic inclinations would like it to be (consider “the Force” vs. the Death Star, people vs. the Terminator).  Nor is it that human judgement is by nature superior to calculation; it isn’t.  (People are particularly bad at ranking risks.)  It’s that human minds work differently than number-crunching computers, and can in some cases see useful features in ambiguous data.

With weather this can work at several different levels.  Our navigator remembers a number of rules for weather prediction at sea, good for a few days into the future even in the absence of any numerical forecast.  At the highest level is the meteorologist with all the data and computation at hand, choosing among ambiguities.

Maybe quantum computing will change this.  Maybe not.  For now, you might take comfort in the fact that this hard problem requires both the best computers and the finest human judgement.