Predicting Stratospheric Winds

Last week the USA shot down a suspected Chinese balloon with an air-to-air missile. The balloon itself was obviously just following the stratospheric jet stream and was not completely controllable in terms of navigation [1]. This brings up an interesting issue, as the state-of-the-art in atmospheric science has never been able to predict which direction the jet streams will follow at any future time. Nowhere is this more evident than with the stratospheric winds that encircle the Earth along the equator, known as the QBO. Scientists have been able to heuristically gauge when this wind will reverse it’s direction (hence the name Quasi-Biennial Oscillation) but have never been able to explain why it reverses. The first time they were able to systematically measure it was via the launching of instrumented weather balloons (radiosondes) in the 1950’s.

The QBO is visualized by the following animation:

Read More

Explaining Natural Climate Variations

Get this out of the way first. Making a prediction and then waiting 20-30 years to gather sufficient data to validate the prediction is not the way to do science. Yet this is the expectation in geoscience and earth science, where quick-turn-around of controlled experiments is not possible. The result of this lack of experimental control is that progress in climate science is glacially slow — since most models will fail to some degree, a geophysicist or climate scientist working in a specific domain may only get one chance in their career to test a long lead-time hypothesis. Feedback on results occurs over decades and breakthroughs are rare (see plate tectonics). So what I want to get out of the way first is: don’t expect any predictions from me. And I guarantee someone will ask for one.

However there are ways around having to do a prediction. One option is to consider the application of cross-validation of a model against existing data. As I was trained in the lab world of controlled experiments, this seems quite reasonable. One gets a fast turnaround on evaluating a model so you can try something else, or move on to another problem.

Read More