From NOT A LOT OF PEOPLE KNOW THAT
By Paul Homewood
Great video by Peter Ridd:
AI formatted transcript below.
Here is the transcript with grammar and spelling corrected, content unchanged:
Computer models predicting weather and climate are actually fairly similar to each other. They use basic physics — the laws of motion, Newton’s laws of motion, thermodynamics, radiative transfer — and a very big computer.
And the weatherman often gets a very bad rap for predictions that don’t always work out. And it’s very unfair, in my opinion, where the models are actually fabulous, and there’s no doubt they’ve improved hugely over the last decades.
But climate models? Not so much.
Now, I live in an area where we have to dodge a few cyclones each year, and I reckon it’s incredible just how good a job, for example, the Weather Bureau does at predicting the future paths of those cyclones. They did a magnificent job for, for example, Tropical Cyclone Narelle. They said it would head across Cape York Peninsula and into the Northern Territory. And that’s exactly what it did. I remember a time when those predictions were pretty well useless — if there was a cyclone in the Coral Sea, the predictions were basically that it could go anywhere.
So let’s have a look at why the weather predictions have got so much better in the last decades. There are four reasons, and we’re only going to concentrate on one.
The first is that we have a better handle on the physics of the atmosphere. It’s not a lot better, and it’s not actually a big deal. The second thing is we’ve got much bigger computers, and this does make a difference. The third reason, and this starts to get important now, is that we now have huge amounts of data from the recent past, and we can use advanced statistical methods — AI, if you like — to help with the predictions. So just as ChatGPT is trained on everything that’s ever been written in the past on the internet to give an answer to your question that kind of sounds okay, the same can be done with the weather. They use the data from the past to work out what would happen if conditions in the past were similar to what we are seeing today, more or less.
But the fourth reason is very, very important, and it’s what we’re going to focus on today. And it’s something that’s not well known: we can now measure the atmosphere much more accurately, and sometimes in some surprising ways.
It turns out that to make a weather prediction for tomorrow, it helps a lot if you know really well — everywhere — what the weather is today.
Now, to understand this, we need to look closely at how weather and climate models work, how they are the same, and how they’re different.
So the great big physics calculations use the laws of thermodynamics — about how air expands, heats, cools, how water vapor forms — and there are plenty of really nice mathematical equations for that. Some are very accurate. Others, about clouds, are pretty hairy, actually — basically just educated guesses. Then there’s the physics of all the radiation: the sun’s radiation coming in, the infrared radiation coming from the ground, from the clouds, even from the air itself.
These sit on top of a great big model that uses Newton’s laws of motion — good old Newton — to work out how the air moves: the wind, the vertical motions, and what that air carries with it — the humidity, its momentum, its kinetic energy. Newton’s second law can be used to work out how that air will accelerate. That’s the A in F = MA — basically how it changes speed or direction.
But we do need to know the mass of the air, and that’s not too hard. We just divide the whole atmosphere into a whole lot of chunks a few kilometers across and say a kilometer in the vertical. We then need to know the F, which is the force on the air. Lots of things contribute to that: the air pressure, the Earth’s rotation, the air density. We then work out the acceleration, or the change in speed or direction, over a given period of time.
We can compare it with calculations of a car accelerating. You’ve got an engine that produces the force, and let’s say the car accelerates at 20 km/h per second — or at least it goes 20 km/h faster every second. So we can work out how fast it will be going in, say, five seconds’ time. Five times 20 is 100 — we will be going 100 km/h faster than we were at the beginning.
But here’s the crucial thing: to know how fast we will be going in five seconds, you need to know how fast you were going at the beginning. If we started sitting still at the traffic lights, then in five seconds we will be doing 100 km/h. But if we were already doing 60 km/h when we floored the throttle, we will now be doing 60 plus 100 — 160 km/h — in five seconds.
You need to know the initial conditions to work out what you’ll finally be doing. The initial condition is the parlance of the differential equations that this all comes from, and it’s the same for weather. You cannot calculate the wind speed or direction tomorrow unless you know what the wind speed and direction is today. And it’s not just the wind — it’s also the pressure, the humidity.
Now, the weather, like many systems, if you get those initial conditions wrong, the final answer you get can be totally different, even if you get that initial condition wrong by a tiny amount. So for example, these balls rattling around here started in just a very slightly different position, and with a little bit of time they’re in a completely different position. This is a classic example of a compound pendulum — the two pendulums started in very slightly different positions, but if you go out in time, eventually they’re nowhere near the same position. And it’s exactly the same with the weather. Get that initial condition wrong, and in a week or two or three weeks down the line, you cannot possibly have an accurate weather prediction.
So how has our ability to measure the weather improved so much?
The first is that satellites can see clouds, but that’s just the start. They can actually monitor and measure the microwave and infrared radiation coming from the air itself, to get a profile of the air temperature and the humidity from the top of the atmosphere down to the surface. Now they still use the old system of releasing balloons that carry humidity and temperature sensors, but they can only do that in a few locations. So when the Bureau does it, as you can see from this map, it’s not very many places, and they can’t do it in the middle of the ocean.
But here’s one thing that I found very interesting — a measurement we can actually use: the GPS system, the same one you use with your phone to work out where you are. It turns out that it can be used to monitor humidity. With the GPS system, there are signals going from the satellite down to ground stations, and the timing of those signals is used to work out your location. But the time of flight of those signals depends on the humidity. So they can do some pretty snazzy calculations and actually probe the atmosphere with the GPS signal.
So what you’re seeing here now is just a revolution in the way we can measure the weather. If you can know the weather today, you can work out what the weather is going to be tomorrow much more accurately.
Now let’s look at climate models. In many ways they are similar, but climate is sort of an average of the weather conditions, and we often take that average over, say, 30 years. In a weather model, we calculate the changes in the weather and add or subtract that to what the conditions are today. For climate models, rather than calculating changes with time, we try to calculate the average conditions over a long period of time.
And it’s actually quite interesting that even here, the present climate models are often gotten wrong by the great big famous climate models — so they disagree with each other by up to a few degrees. I did a video on that.
Now, maybe it doesn’t matter too much, because what we really want to know is how the climate will change if you, say, double the amount of carbon dioxide, or some other parameter. So, for example, modelers often do interesting simulations where they reduce the sun’s output, or even change the position of the continents and the oceans, to simulate past climates.
So this is an important difference: weather models calculate the change in the weather using today’s weather as a starting point, and calculate the weather tomorrow. Climate models start usually with the rough climate and calculate the changes in the climate if we change some important parameter, like maybe carbon dioxide concentration.
Doubling the climate models — climate models don’t need highly accurate measurements of the climate today to calculate the changes if you double the carbon dioxide, for example. But weather models need very accurate measurements of the weather today to calculate the weather a week or a day in the future.
So this means that the weather models have benefited massively from all the extra measurements, but not so the climate models — that extra data we have for weather today does not help us calculate the weather in 100 years’ time. So climate models don’t benefit from the revolution.
So what’s the moral of the story? Don’t keep dissing the weatherman. He’s actually doing an incredibly good job. And same with all those technical people who developed the satellite monitoring systems, the GPS systems, and the statistical techniques — they’ve made a revolution.
Now, as for the climate models — well, are they actually better at prediction than a simple back-of-the-envelope calculation? You can do a very simple calculation and end up with a number very similar to these huge supercomputer outputs. And do the models really predict catastrophe? And do the models have discrepancies with each other — big discrepancies? And do we have a fundamental problem that we won’t be able to tell if they’re right or wrong until 2050 or 2100, when it could well be all too late because we’ve cooked — or maybe we realize that they were all wrong?
Well, my, look at the time. We’ll have to do that on another day.
