A voxel-based forecasting model years before AI and supercomputers caught up
Here’s one of those ideas I had that, at the time, felt so obvious I assumed everyone else had already built it. Spoiler; they hadn’t.
Back in the early 90s, same development-fueled enthusiasm, I had this notion that the world’s weather could be modeled elegantly if you simply divided the entire planet into a grid of 10km × 10km cells. Then you add vertical slices every 1km up, all the way to where weather stops mattering.
Each cell would store its own mini weather dataset; temperature, humidity, pressure, wind vectors, all the usual suspects. Think of it as a 3D matrix of weather-voxels, ticking over hour by hour.
Update the array, feed it into a forecasting algorithm, and off you go.
In my mind it made perfect sense.
You aren’t trying to simulate chaos; you’re structuring it.
And, naturally, I assumed some giant organisation like NASA or the Met Office must’ve built this already.
Turns out… they hadn’t.
Not back then.
What shocked me years later is discovering that modern weather prediction, the seriously heavy stuff like ECMWF, NOAA’s GFS models, Germany’s ICON system, actually do exactly what I imagined. The Earth is now divided into grid cells, each with its own atmospheric state, and supercomputers crank through these cells layer by layer.
Even more surreal?
Google DeepMind’s GraphCast AI model, one of the most advanced forecasting systems in the world, is built on the same conceptual foundation; a global grid, time-based updates, AI smoothing out the physics and generating predictions.
I had the concept. The world just didn’t have the computer power it has today.
That’s the thing about ideas, hey?
Sometimes the only difference between a crazy thought and a global standard is the technology finally catching up.
And honestly, I still look back at that weather-grid idea and smile. Not because I “missed an opportunity,” but because I somehow saw a system that would only become possible 30 years later. And the elegance still appeals to me.
Simple rules. Structured space. Predictable outcomes.
That’s always been my design language, even before I had the words for it.