Elephant in the room: There are factors like volcanos and human involvement that create unpredictable variables

As above. And let me lead with this: I’m a smart guy, but i know little about meteorology. My question is grounded in ignorance.

Right now we can predict with relative accuracy storms and events 3-7 days out, depending on the scope of the storm. Temperatures are even ‘easier’.

It seems like we’re close (if not already there) to having global weather data at all times. We live in a world of scientific order, so shouldn’t we be able to create a model that very accurately predicts specific weather on a long timeline?

Exa: I’m ignorant in meteorology, but I would think all of the fronts and movements that will lead to a hurricane in the Atlantic in 2025 are already in motion. Couldn’t we track that *now*? Play out the equation, so to speak?

And then we’d be able to reverse engineer the moments that disrupted the model, to accurately determine what artificially affected the weather pattern.

Thanks! And sorry if I’m dumb.

In: Planetary Science

We can predict the *chance*s of what the weather will be this afternoon.

We can predict the *chances* of what the weather will be tomorrow, but that will change depending on what actually happens this afternoon.

We can predict the *chances* of what the weather will be next week, but it’ll have a lot of possibility for failure because it will depend the actual weather Sunday, which will depend on the actual weather Saturday, which will depend on the actual weather Friday, which will depend…

If you’re trying to predict **years** out then you may as well just graph the past several years, extend the line forwards, and shrug.

The answer is probably “no,” the system is fundamentally probabilistic and not deterministic.

This more generally has been a long-standing question about the universe in general – can you mathematically predict the entire thing from a known starting condition or are certain behaviors inherently random?

Right now things appear to be fundamentally random at the smallest scales – two identical uranium atoms will decay at different times and there is no mathematical model to predict their individual behavior.

Small deviations like this build up in a system and introduce increasingly large levels of variation as you try to predict behavior further in the future.

You can predict tomorrow’s weather based on today’s weather. You can’t predict May 2025’s weather with any confidence at all – the system is extremely complex and chaotic and tiny perturbations will have an entire year to cascade.

The weather, I am told, is (in the mathematical sense) a chaotic system. That means that even the tiniest error in the data we have, or the calculations we apply, can lead over time to a prediction far removed from what actually happens. Hold your next breath for 30 seconds, and the worldwide weather in a year or two may, quite literally, be totally different to what it would have been if you hadn’t. Or the famous phrase, the Butterfly Effect – a butterfly flaps its wings, and next year there’s a storm at the other side of the world that wouldn’t have otherwise happened.

Our actual data is massively incomplete; our computers only work to a given level of accuracy (significant digits, rounding and so on). We can refine that, and increase our accuracy – but, no, it’s never going to be perfect.

(Actual weather forecasting works on the basis of recognising that the data is imperfect. Run thousands of predictions, perturbing the data slightly each time, and see what trends dominate. Sometimes the system is pretty stable, many of the predictions are very similar, and you can say what’s likely to happen with a high degree of confidence. Other times, less so.)

In 1961, a meteorologist called Edward Lorenz began using computer simulations to try to predict the dynamics of the atmosphere. He created a system of interrelated equations that described properties of air motion and temperature. He then iterated the equation to obtain the atmospheric conditions at the next time step. These conditions were then fed back into the equations to obtain the conditions at the next time step.

When he found an interesting pattern in the data for certain initial conditions, he ran the simulation with the same initial conditions and got a completely different result. He expected to reproduce the same result because his equations were deterministic, i.e. not random but reproducible.

The source of the difference turned out to be a rounding error in the last few digits of one of the input numbers. He had entered 0.486 instead of the full 0.486134 which was the original starting value. He had assumed that such a small difference or error would not be significant and would disappear in the calculations. But instead of disappearing, the error grew with each iteration until the result was completely different from the original answer.

This is why weather models are so difficult. Google “butterfly effect” and “chaos theory”.

Roughly, a weather model creates a globe with ~1 cubed mile grid points… Plugs in the observations (satellites, ground stations, etc), and then uses well studied math tools to “fill in the grid” and then start stepping forward in time.

The more powerful computers become, the tighter those grids become, the more accurate the forecast. And presumably tighter grids will require “updates” to the equations used for the weather (for example, assumptions made with a 100 cubic mile grid might be invalid for a 1 cubic Mile.)

All that said, it will always be a “continuum model,” in that it is looking at bulk movements of air, not individual particles crashing around into each other. So being able to predict weather like we can predict Haley’s Comet isn’t ever likely.

Edit: cubic, not square.

Also continuum as a model is less an issue than the data problem of factories, cars, volcanoes etc… the point being it’s inherently averages even at a cubic meter of resolution

Weather is a classic example of a *chaotic* system. In casual conversation, “chaotic” implies randomness, but here I mean it in reference to the field of study called “chaos theory”. Chaos Theory studies deterministic systems that are very sensitive to initial conditions, or “When the present determines the future, but the approximate present does not approximately determine the future”. This means that while a chaotic system is *theoretically* perfectly predictable, in practice, small errors in measurements grow exponentially as you try to predict further and further in the future.

We have a solid understanding of how weather works, but we’re limited by the accuracy of our measurements and the level of detail of our computer models. We can’t measure or model every atom in the atmosphere, every droplet of a cloud, or every tiny bit of turbulence. Due to the chaotic nature of the weather, these limitations very quickly limit how far into the future we can accurately predict. We have **really good** measurements and models that let us predict the weather **really well** over a period of 5 days, but as you push the prediction further into the future, the accuracy rapidly drops off.

You’re right that satellites provide a wealth of data for meteorologists, but they have their limits. The data they provide only has so many significant figures and so much granularity – they can’t constantly measure the temperature of every cubic meter of air to thousandths of a degree. To improve how far we can predict a *tiny* bit would require *massively* better measurements and more detailed models.

Maybe in our lifetimes we’ll have somewhat accurate two-week forecasts, but multiple months in advance is practically impossible.

Weather is part of chaos theory, which is when small perturbations make a large difference. The 3 body problem is another example of this that’s recently gained spotlight because of the show.

The butterfly effect was coined because allegedly the force of a butterfly flapping its wings was enough perturbation to effect the path of a tornado (this concept is not practically true because there are much larger effects such as cars driving, but mathematically it is true).

So until we reach a point where we can calculate the force and location of every butterfly flap on the planet, predicting weather a year away is impossible.

In theory, yes, but we need a time-traveling magic supercomputer.

The reality is the weather depends on so many variables, making a great simulation of it for further than 3-7 days out would take our current computers more than 3-7 days to complete. A forecast isn’t very useful if it takes longer to get it than just waiting for the weather to happen.

So our models are simplified, focus on the near future, and omit things that we know don’t usually cause “big” differences. Just in case they do, we have multiple models and some of them omit different things. Meteorologists look at all of them and, in the end, go with an educated hunch which one will be right.

As we get stronger computers and more worldwide sensors, we can get more accurate. But we’re still a very long way away from a true, very accurate model of weather. We also occasionally discover new things impacting weather we weren’t accounting for. I’ve seen some theories and convincing data that changes in shipping traffic can change weather patterns in areas, including one theory that an industry switch to different emissions standards caused measurable changes. We’re encountering things like “the warmest average ocean temperatures in recorded history”. When we don’t have recorded history that means we’re pretty bad at understanding how those conditions affect the weather. We have to observe it and record it before it can become part of our models.

So right now? We don’t really have evidence that we CAN truly predict the weather. But we can’t mathematically seal the deal that we’ll NEVER be able to do it, either. If there is some major breakthrough in computing, like a new model that increases our computational ability by several thousandfold, we’ll have to redefine a lot of “impossibles”. That sounds like a big deal, and it is. But it was accomplished once within the last 2 or 3 lifetimes.

## Latest Answers