Vertical temperature profiles over Germany

In an age full of bold claims that AI will solve every problem, it’s worth remembering:

You can’t reliably predict what you don’t measure.
(If no one else has said it, I’ll happily call dibs!)

This is especially true for weather forecasting. Many apps promise hyper-local predictions at resolutions of just a few hundred meters, yet the underlying models are built on sparse observations. For example, if temperatures are measured in two cities 100 km apart, the “forecast” for the midpoint is simply an estimate based on physics and dynamics.

What about the vertical dimension instead? To truly capture the atmosphere aloft, we need radiosondes: balloons that measure temperature, humidity, and winds up to tens of kilometers high. These soundings reveal evolving structures and correlations in the atmosphere, as shown in this animation I made of high-resolution German radiosonde data from January up to now. In the video the points represent the radiosonde location, while the color maps the temperature observed.

Notice how sparse these observations are: launching sondes is far more expensive than operating ground stations. Thankfully, conditions higher up are often more predictable than at the surface, so that less observations are needed to maintain a good forecast accuracy.

We should not forget that, no matter how powerful AI becomes, it cannot replace high-quality measurements. Expanding and improving observations has always been, and must remain, the top priority.