Seeing Airport Delays Before They Happen

Why Airports Slow Down (Even When It’s Not a Blizzard)

We’ve all seen it: your flight is “on time”… until suddenly it isn’t. While it’s easy to blame major storms, most delays come from the everyday stuff—low visibility, strong winds, or even just a bit of rain that slows down how many planes can safely land.

In the U.S., the FAA manages these situations with Ground Delay Programs (GDPs) and Ground Stops. Instead of letting planes circle endlessly, flights are held at the origin until conditions improve. It keeps the skies safe, but it can ripple across the whole network.

Here’s the exciting part: if we could predict these airport constraints earlier, airlines and passengers could act proactively—rerouting aircraft, rescheduling crews, or helping travelers rebook before things snowball.

Weather + Math = A Hard Problem

You might think: “If weather causes delays, can’t we just forecast it?” Turns out, it’s not that simple.

Lumo has been analyzing years of weather forecasts, FAA advisories, and airport conditions to predict things like:

• The chance of a GDP at a specific hour
• How many arrivals per hour an airport can handle
• The likely cause of a slowdown

But here’s the catch: probabilities don’t tell the full story. Knowing there’s a 40% chance of rain at 6 pm and 50% at 7 pm and 30% at 8pm is fine—but what if you want to know the likelihood it rains continuously from 6–8 pm? Those hours aren’t independent, and the math gets messy. Traditional methods break down fast.

Our Big Breakthrough

• What happens if the GDP lasts two hours?
• How do things change if arrival rates drop further?

For example, here is a screenshot of probabilities from our Ops Manager tool: It shows the predictions for Newark on Oct 1st (note this blog post is being published on Sep 29th so we hope this post ages well). The model is predicting 5 potential scenarios - the first with probability 0.275 expects a rate of 36/36/30/30/30 starting at 2300, the second predicts a rate of 30/30/30/22 with probability of 0.26 and so on. 

This shift from numbers to scenarios is a game-changer for planning. How did we do it? It's our secret sauce we can't share, but to give you a sense for what the models are doing, rather than predict a single hour's data, we predict a moving window of hours, which we then stitch together to come up with the scenario.

Beyond U.S. Borders 🌍

So far, we’ve talked about U.S. airports where GDPs are published. But what about the rest of the world?

For most international airports, we do not have detailed public ATC programs to read. So Lumo took another approach: infer constraints directly from flight data. By analyzing historical arrivals and delays, we can estimate an airport’s “hidden” capacity patterns.

That means we can now generate predicted arrival rates for airports globally—even without official data. It’s not perfect (since we can’t see the exact ATC decisions), but it’s powerful enough to guide proactive planning worldwide.

Predicted capacity is just the start. Lumo feeds these scenarios into our flight-level simulator, combining them with deep learning delay models. The result? You don’t just see that JFK is slowing down—you see whether your specific flight is likely to be delayed, and by how much.

That’s how we connect the dots from airport-wide constraints to the seat you’re sitting in.

Wrapping It Up

We’ve gone from raw probabilities → scenarios → flight-level insights. The result: a whole new way to see disruption coming, not just react when it hits.

🚀 Ready to learn more?