All the automation in the world with useless without a human guide to either transform production into a useful product, or useful knowledge to heed. That's why this act of trying to remove human labor is asinine. Even skilled human can't always get the right readings, so expecting a robot to do it all at this stage is just selling snake oil.
The current human-in-the-loop model exists largely because our technology hasn't been good enough yet, not because there's something inherently special about human judgment in this context. Weather prediction is fundamentally a pattern recognition problem. Pattern analysis at scale is exactly what computers do better than us.
Perhaps someone could apply to YC with this idea. There is one YC startup doing this already: https://www.ycombinator.com/companies/atmo
I'm not too worried about the human factor being replaced as a whole here. Even with AI, someone needs to interpret the output and make sure the the prediction models actually work.
It's worth being specific: the National Weather Service operates some of the most robust automation and radar ingest pipelines on Earth, but the final go/no-go warning call is almost always human—often a single overnight forecaster on a console, monitoring a swath of counties. Automation (e.g., Warn-on-Forecast guidance) can surface threats, but the NWS intentionally doesn't have an 'auto-warn' button for tornadoes, because of the asymmetry of false positives (blow credibility, cost lives in the long run).
Budget cuts reduce redundancy and experience in those overnight shifts. When you have only one person monitoring instead of a team of two or three, you get decision fatigue and coverage holes, especially during clustered, multi-cell outbreaks. We've seen near-misses in the past, and every pro-meteorologist I know says they're playing defense against process errors, not just technology failures.
Before we point fingers or blame 'technology/automation' shortfalls, let's quantify the concrete bottleneck: skilled human decision-makers are the limiting reagent; machine learning warning aids are still years away from majority trust.