: With the incredible technology that we have today, why is it still impossible to have 100% accuracy on predicting the weather?
In: 551
Serious question. Not trying to start a fight here. How much credibility should be put in long range weather projections when we can’t effectively predict short term changes? This is not climate change denial. Just a question
The accuracy of the prediction goes up exponentially the closer you get to the time you predict. So (and bearing in mind this is a made up example) the prediction for 7 days out might only have say a 20% certainty, the prediction for an hour out will be close to 100% certainty. Given where we started (weather as completely uncertain, unpredictable, act of the gods) to where we are now (prediction capabilities that get close to 100% certainty as the event horizon is looming), I’d say we’re doing stupendously well. Don’t let the perfect be the enemy of the good.
“But I don’t predict it. Nobody does, ‘cause i-it’s just wind. It’s wind. It blows all over the place!”
Dave Spritz – The Weatherman
Perfect predictive accuracy would take several things:
Having 100% accuracy on knowing the current state of all parts of the atmosphere. (including any changes caused by measuring the current state of the atmosphere).
Having 100% accurate predictions on everything that influences the atmosphere(the sun has weather too, which can slightly change how much light/heat is hitting Earth). Also need to predict things like the impact of people, animals, plants, volcanoes, etc.
Having a mathematical model that perfectly captures the interactions of everything in the atmosphere, and everything that influences the atmosphere.
Having a computer that can run that mathematical model with enough resolution to simulate every atom in the atmosphere, and fast enough to outpace the changes in the atmosphere.
A way to send a personalized forecast to every person, based on where they plan on being.
At some point you’d reach a level of accuracy where sending people the forecast changes the outcome, and to improve from there you’d basically be engineering a self fulfilling prophecy.
A better prediction is always possible, but a perfect prediction is not.
A lot of it is that we don’t have complete data to input into our computer models. There’s typically a couple of weather stations per city but they miss all the stuff that’s in between the stations. And then once you get out into rural areas, weather stations are even further apart which means you’re missing even more data.
The further out you try to forecast, the more this missing data affects you.
Here’s how I would ELI5:
Take two cups of water. One hot, one cold. Put a couple drops of food coloring into the hot water and mix it.
Now, pour both cups into a bowl. There will be swirls, vortexes, things will happen and eventually, they’ll mix.
The thing is, every time you do this, it will look different. The swirls will be in different locations, some will be bigger, some will be smaller. You can do this 1,000 times and no two swirls will be exactly the same. You can generally predict that the hot water will rise and the cold water will fall, but the exact locations are different every time.
Let’s say that there’s an ant at the bottom of the bowl. This ant really doesn’t like the food coloring. If he gets coloring on him, it’ll ruin his day. Can you accurately predict where and when the food coloring will reach him?
The bowl in this example can represent an entire state, or country, or even continent. There are a bunch of areas of varying pressures, densities, humidities, and temperatures, all interacting with each other. There are extremely powerful supercomputers who are constantly fed data from weather centers, airports, and weather balloons from all around the world, but because of the complexity of the system, even the computers can’t agree (think of the “spaghetti plots” you see in hurricane predictions)
We’re all like tiny ants, in a bowl with 50 different kinds of food coloring, at 50 different temperatures, that’s spinning, and we want to know whether or not the conditions are perfect to make water droplets form in mid-air and fall from the sky.
Along with everything else that’s been commented, there’s a misconception about what “chance of rain” means. A 30% chance doesn’t mean that you in your location have a 30% chance to see rain, it means that 30% of the area covered in the conversation will 100% see rain. So they are actually very accurate on where rain will show up but describe it in a way that’s easily misunderstood.
Knowing a dice has a 1/6 chance of rolling a 6 , how come I can’t tell you exactly how many 6s I will roll when I knock over this tub of 100 dice?
It is much the same. We can know the general liklihood but there are just far too many variables to be precise.
Reminds me of that physicist that said if god was real, when he died he would only have 2 questions for him. “Why relativity?” And “Why turbulence?”
Imagine you have a series of swimming pools filled with water to that sum to 1.3 billion km^2 in surface area – 2/3rds the size of the globe. They are covered by a series of air pockets the size of the globe. The swimming pools are constantly emptying humidity into the air pockets, and then the air pockets are dumping the water out of the pool or into a different pool. This is happening in billions of square km all over the globe, more than 2/3rds of which is water. Every time it happens, the heat content and humidity of each square km changes, and it’s changing non-stop.
So if you want a resolution of 1 square km, you have to track ~2 billion data points *just on the surface*. And the pockets of air are many kilometers deep, so really, you have 10s of billions of data points – one for each cubic km. And they’re changing non-stop. And they’re interacting with one another non-stop. Let’s say we want to have predictions that are accurate to the hour interval, and assume that only 10 km of atmosphere are relevant to the weather. We’re now on the order of 250 billion data points, on an hourly, cubic km resolution.
So a weather model at that resolution is required to have 250 billion sensors all over the globe, all networked and feeding information back to a central data repository. But that’s too costly. Plus people in other countries would feel kinda weird about being observed that closely, so we have to make due with a much smaller number of sensors, which are trying to interpolate (figure out in-between points), often from high altitudes/space. So we use that smaller number of data points to create models about what we’d expect to be there in between, and use that incomplete data model to make predictions about what will happen, in terms of weather. And that means there’s a lot more error than there might be with perfect data.
Latest Answers