Airplanes Unexpectedly Modify Weather

reillymj writes "Commercial airliners have a strange ability to create rain and snow when they fly through certain clouds. Scientists have known for some time that planes can make outlandish 'hole-punch' and 'canal' features in clouds. A new study has found that these odd formations are in fact evidence that planes are seeding clouds and changing local weather patterns as they fly through. In one case, researchers noted that a plane triggered several inches of snowfall directly beneath its flight path."

Comments

Surprised?

[confused one]

So, we're surprised when a large metal object that sucks in cold air and spits out water vapor (and CO2) by the ton, affects cloud formation?

Re:Opposite effect

[forceman130]

I'm guessing that has more to do with the heat coming off all that tarmac than it does with the aircraft themselves.

Re:Cloud Seeding

[The_Wilschon]

I'm very glad that you've taken classes, and I'm even more glad that you're taking obviously quite a lot of time to think about this. However, as a working scientist (HEP) myself, I have to assert that you are quite wrong. Take a large number of clouds. Fly a plane through each one of them, half of the time dropping cloud seeding crystals, half of the time not (choose by flipping a coin). Fill two histograms with the integrated radar echo strength, beginning from the moment the pilot reports entering the cloud, and ending with the time the cloud ceases to precipitate, or one hour later, or some such. Obviously, put the data from the seeding runs in one histogram, and the data from the non-seeding runs in the other. At this point, you have obtained approximations to two distributions. You can obtain error bars on each bin of each histogram (poisson statistics), and estimate systematic errors on top (in quadrature) of those. Now, you can do a K-S test, or a Chi-squared test, or an eyeball test, and determine whether the two distributions are commensurate within experimental error or not. Quote Bayesian credibility, or confidence levels, or whatnot. Done. You have a successful experiment and a publication.

The key to the experiment is that the set of all clouds has some (currently unknown, but definitely fixed) distribution of rainfall amounts. As you draw samples from this distribution and fill a histogram, you get an idea (perhaps fairly coarse) of what that distribution is. Then you draw samples from a different distribution (seeded clouds), and get an idea of what that distribution is, too. Do these distributions appear to be different, or are they similar enough that we can't tell? Since what matters is the distribution as a whole, we don't need to worry about matched pairs in control and experimental groups, or what the characteristics of individual clouds are. Trust me, we have exactly the same situation in HEP. No two collisions are ever exactly (or really even close to exactly) alike, so if matched pairs were required, we'd never get anywhere at all.

The kicker is, of course, getting enough samples to populate your histograms sufficiently to get a good enough idea of the distributions. You are asserting that there are too many variables in cloud configuration space (and you're right, there certainly are an awful lot). But we don't care about filling up cloud configuration space. What we care about is filling up integrated radar echo (as an approximation to rainfall amount) space, which is one dimensional, and therefore much, much easier to populate.