I feel like for the last 20 years I’ve felt like tomorrow’s forecast is very accurate, 2 days from now is an educated guess (if it says it’s raining, it’ll rain, but not sure what time of day) and 3+ days is unknown.
What advancements have we made in meteorology and what are the current limitations for weather forecasting?
I’ve definitely seen my phone be very accurate in forecasts for the current day like “in 16 mins, it’ll rain for 22 mins and then stop for an hour”
In: Planetary Science
“Prediction is very difficult, especially if it’s about the future.” — Niels Bohr
The “in 16 mins, it’ll rain for 22 mins” thing isn’t a forecast. It’s a radar observation that there is a rain cell moving toward your location that’s 16 minutes away and 22 minutes thick.
There are many measurements so that better models can be produced. However, there are places where the forecasts are almost always right (= Hawaii) because the weather is almost always the same as yesterday. Other places are hard to predict because of a combination of wind patterns and terrain, like the Mason-Dixon line area.
Predicting weather is something we’ve been trying to do for a *very* long time. The math isn’t even super hard, it’s just that you have to do a lot of it, and each time point you step forward in the future is iterative. In fact, when we first invented computers, meteorologists were super excited to finally put these models to the test. “Finally!” they said, “We can just plug in all the information and it will predict the weather perfectly!”
What they found is that it actually did a pretty good job predicting the next few hours… but then would go wildly off course. They would adjust their models, add in more data, and still, insanely wrong after a short period of time, but somehow it would be a different type of wrong even when they only changed a tiny thing. This sort of random destabilization is what would eventually be called “chaos theory”. It turns out that even a miniscule change in inputs would eventually generate a diverging system after a while. It’s the same idea as the “butterfly effect”: that a single beat of a butterfly’s wings in Texas can be the inciting incident for a hurricane in Brazil.
Our instruments are vastly better at collecting data than they used to be. Our computers are more accurate at doing the math. We’ve got layers on top of that where meteorologists look at weather as a system with a history and relationship to other weather systems. There’s even some really exciting work with using machine-learning to identify subtle trends. But the fundamental issue is that it is mathematically impossible to accurately describe, *down to the molecule*, the current conditions. All of our forecasts are doomed to eventually destabilize and become meaningless.
Where I live the forecasts are 50/50 about what it is doing now. The only thing a (future) forecast tells me is that it won’t do exactly that.
Meteorologists are like economists, they can’t tell you what’s going to happen, but they can explain in minute detail why it happened.
[Here’s a plot showing how forecast accuracy has improved from the 1980s to today](https://ourworldindata.org/images/published/improved-weather-forecasting_1350.png). So yes, there have been advancements made. And that’s mostly because of the rapid progress in computing technology.
What makes weather prediction difficult is that it is a chaotic system, which means that very small changes in initial conditions can lead to wildly different outcomes over time. This is an issue because no simulation is perfect – we cannot measure exactly how fast the wind is blowing on every square millimeter of the Earth’s surface, and even if we could we don’t have enough computing power to simulate it with that resolution.
But the nature of chaotic systems is that the errors grow exponentially with time, so if you want to predict further into the future you need *exponentially* more accurate initial data, and exponentially more computing power, in order to have a linear increase in how far ahead you can predict. That’s why, despite *massively* more sophisticated models, and computing power that would have been unthinkable 40 years ago, today’s 7-day forecast is still worse than a 3-day forecast from the 80s.
What the exact accuracy is each day out will heavily dependent on the region and what you mean by “accurate”. So it’s probably difficult to answer that question.
IIRC after about a week simulations are less accurate than historical averages. That’s because the air molecules have circulated around the earth like 3 times.
Latest Answers