Can AI help us predict extreme weather?
HTML-код
- Опубликовано: 20 фев 2024
- AI models are starting to revolutionize weather forecasting.
Subscribe to our channel and turn on notifications (🔔) so you don't miss any videos: goo.gl/0bsAjO
This video is sponsored by Microsoft Copilot for Microsoft 365. Microsoft has no editorial influence on our videos, but their support makes videos like these possible.
We’ve learned how to predict weather over the past century by understanding the science that governs Earth’s atmosphere and harnessing enough computing power to generate global forecasts. But in just the past three years, AI models from companies like Google, Huawei, and Nvidia that use historical weather data have been releasing forecasts rivaling those created through traditional forecasting methods.
This video explains the promise and challenges of these new models built on artificial intelligence rather than numerical forecasting, particularly as it relates to the ability to foresee extreme weather.
Here are the papers that describe the models mentioned in the video.
Google’s GraphCast: www.science.org/stoken/author...
Huawei’s Pangu-Weather: www.nature.com/articles/s4158...
Nvidia’s FourCastNet: arxiv.org/abs/2202.11214
Here is the announcement of the ERA5 dataset, released by the European Centre for Medium-Range Weather Forecasts in 2020:
rmets.onlinelibrary.wiley.com...
We interviewed Dr. Aaron Hill over email for this video. Hill is involved in developing responsible AI for environmental science via AI2ES:
www.ai2es.org
Google has also developed a weather forecasting model called Nowcasting, which is already embedded in its weather products specifically for short-term precipitation forecasts:
deepmind.google/discover/blog...
If you’re interested in learning more about the history of how we developed weather forecasting, I’d recommend The Weather Machine by Andrew Blum:
www.amazon.com/Weather-Machin...
Subscribe to our channel! goo.gl/0bsAjO
Vox.com is a news website that helps you cut through the noise and understand what's really driving the events in the headlines. Check out www.vox.com.
Watch our full video catalog: goo.gl/IZONyE
Follow Vox on Facebook: goo.gl/U2g06o
Or Twitter: goo.gl/XFrZ5H
Hello! Laura here. I wanted to highlight some of the research of our interviewee Randy Chase that didn't make it into the video. He co-authored a study on some of the biases in historical weather data that AI models are trained on. Examples include fewer weather stations gathering atmospheric data in the Global South: journals.ametsoc.org/view/journals/bams/aop/BAMS-D-23-0196.1/BAMS-D-23-0196.1.xml
i follow European model.......
interesting!
Thank you for featuring our work on AI in weather forecast. The FourCastNet we built was the first high-resolution AI-based weather model that showed competitive results compared to traditional methods. FYI, even FourCastNet also predicted Hurricane Lee earlier than traditional forecasts, not just Deepmind's. So, it is unfair to assign all credit solely to Deepmind at the beginning of your presentation.
@@Prof_Anima_Anandkumar Can AI help us predict extreme weather? 21.2.24 Ai... isn't Ai already engaged in such matters, seeing as all data collected is via human design as opposed to human interaction/observation? Ai is no new thing. Babbage got the ball rolling...etc.....
The reason why you need ensemble models in the first place is because weather is a chaotic system. This means a slight change to the initial conditions can make a huge difference over time. See the double pendulum as an example.
i thought the weather modelling system always used some sort of ai cause it generates about 50 different scenarios
@@Kinggenton nah, you just have to slightly change the starting condition.
That's only _one_ aspect of chaotic systems. A truly chaotic system can't be modeled with the intent on deterministic outcomes. The AI trained on weather observations might be discovering patterns that we never actually considered. They might be something that only the best meteorologists could guess intuitively, if they could study all that data...
And due to climate change it's becoming even more unpredictable. Ocean currents are slowly changing and that alone will impact the weather so much that old data might not be as useful. But we shouldn't discard AI weather prediction but try to use it as best we can.
wrong@@randomkranjc
My biggest fear with projects like these are that companies using research that was publicly funded will create sorely needed products that are then used to undermine our public weather institutions rather than working in partnership with them and feeding back into them. We've already seen the NWS get barred from creating it's own weather app by vested interests. We've also seen cases where only subscribers of a weather app received a weather warning. I worry about a class driven stratification of weather information that should remain a public good and wonder if AI driven weather products could be one more way people get roped into paying twice for weather info.
Fortunately these ai models that are already pretty good, are freely available, and run on a single computer. I think ai may actually bring this technology closer to the hands of the public due to no longer needing a billion dollar super computer to run.
You can literally download the model yourself
I can see the funding for predictive models stealing funding that goes toward weather sensor and data transmission infrastructure, undercutting its own accuracy.
As skeptical as I am about A.I. making art, journalism, tracking our on-line activities
Predicting the weather seems like the perfect job for this technology
A massive data-set to train on and incalculable variables to account for could save lives
plus all this data is already public so no concerns over copyright infringement, ethical data mining, etc
This is where AI is really going to shine: protein folding, designing models in the back end, etc. The consumer facing stuff is just a marketing grift.
In the context of art, it would still be beneficial for AI to handle some menial/tedious tasks like doing in-between frames in animation, while the artist devotes more creativity into making better key frames
@@lilarrin1220 A.I. as tool to streamline, enhance, and execute artistic projects is happening
The next few years are going to be interesting. We’ll see this technology revolutionize our creative output in ways we can never imagine
Perhaps I’m more wary than skeptical
Just don't let people freeze,
See this is the type of stuff I want to see technology, Ai, & simulations accel in. This is the type of stuff I hope to see massive advancements in. Not just this but a whole range of this type of research, work, etc.
*excel just like the green MS Office
The reason why I love Vox is that their animations and explainers are top notch! Even their former employees Cleo and Johnny are one of the best on youtube! Props to Vox
Thank you for featuring our work on AI in weather forecast. The FourCastNet we built was the first high-resolution AI-based weather model that showed competitive results compared to traditional methods. FYI, even FourCastNet also predicted Hurricane Lee earlier than traditional forecasts, not just Deepmind's. So it is unfair to assign all credit solely to Deepmind
so microsoft sponsored a video that praises a google ai model? interesting
MS is sponsoring the AI series on Vox. Not just this video
Yeah, it's hard not to look at this as an ad. At least they're transparent.
Ai = depopulation
@@dunnowy123I think you’re confused. They’re rivals
they are sponsoring the concept of AI in general they want people to equate usefull stuff like weather prediction with harmfull stuff like generative AI displacing peoples jobs
This is a really compelling development, but I want to mention that statistical weather forecasting has in fact been a thing for a long time (at least since I was studying meteorology in college back in the late aughts). Our newer AI and stronger computing systems may be supercharging development, but this is conceptually built on the back of tools forecasters have been using for a while.
By newer ai systems, do you mean what was in this video? Or something else? Because what is shown in the video uses entirely different methods to produce its forecast compared to traditional methods
@@h20dancing18 what's in this video.
There's not one single traditional method-- there's numerical weather prediction (which takes the current state of the atmosphere and applies known physical principles to determine the future state) and statistical weather prediction (which compares the current state of the atmosphere to statistically similar past analogues to make inferences about the future state).
unless AI is actually "learning" physics, it's a version of the latter, just with more dynamic updating of the governing equations.
My biggest concern about AI in this realm and in science in general is the ability to come up with solutions without developing understanding. I worry about an era when everyone is overreliant on these models then the models become worthless due to a sudden change and then we are left scrambling. AI is a wonderful tool and is helpful for noticing patterns or variables we can't but the model has the issue that its always trying to match the training data which then creates another problem. How do you ensure your training data properly samples the thing you are looking at.
My hope is we use AI to assist us in developing understanding rather than using it as a replacement for human thought or input. I'm not optimistic about that considering how AI has been abused by both scammers and corporations.
My concern exactly
I am current an undergraduate student in professional meteorology and I am not topic concerned about over reliance on AI. Utilizing AI to create ensemble forecasts is just another tool we use to make predictions just like any other. Meteorologists still need to understand physics to weigh their options and form conclusions
@@colewiley4309 Wasn't too concerned until I read your comment. You aren't too concerned that you won't become over reliant on it? That tells me people like you won't think rationally about these tools and avoid over relying on it. God help us all for your industry if this is your response from someone from the public raising a legitimate concern.
It'll be like pilots. Aircraft are largely automated but the pilots have to understand why the plane is doing what it's doing and whether or not to make manual adjustments.
@@jmhorange you still need to understand the physics of the atmosphere to understand what the model is outputting, and no model will ever produce anything 100% accurate since we can’t fully map every aspect of the atmosphere with instrumentation. This is why no model will ever fully take the essential role a forecaster has to compile data with their meteorological knowledge to improve accuracy and inform the public. Seriously nothing to be a doomer about so calm down
Earlier predictions would be extremely helpful to all of us in the path of hurricanes. 👍🏼 🌧️🌀
crunching weather data with machine smarts, spots patterns, amps up forecasts, and drops early warnings for wild weather-though total certainty's a tough gig is my guess for what ai would do in forecasts
Exactly
Two Microsoft AI commercials before the video.
A third one baked into the video itself.
What the eff.
Same.
*laughs in premium*
wake up man @@doublebarreltech4984
Not discussed: climate and season predictions. These forecasts aren’t chaotic but averages of weather patterns and they depend upon large scale influences on the weather. These AI models may be able to more accurately forecast climate and seasonal changes, such as when spring will start, which may help to influence when to plant a crop. Why? Because it can take into qualitative data that a formal climate model can’t. That is, I assume weather and climate prediction is largely based on deterministic physical models and then made probabilistic in an ensemble model. AI, however, can use the variable of what 1000 farmers in Iowa think spring is going to start and incorporate that into the weather data.
I'm not opposed to AI in theory and I think there are some great and fascinating practical applications. I do find it kind of suspect that this video is sponsored by Microsoft though. People gotta eat tho
Vox delivering another excellent bit of content. Been loving these AI-related videos which have been shared over the last few months. Super interesting!
It depends on the Weather Model on how often new forecasts are released. The 16 Days GFS 20 Ensemble forecast is every 6 Hours, the 10/14 Days ECMWF 50 Ensemble forecast (with a slightly higher resolution) only every 12 Hours. Nowcasting Models sometimes every 3 or even every Hour. Also ERA5 has been updated and contains now the Data since 1950. A Problem with the ERA5 Data might also be its low resolution. AI is going to miss small Scale Events, which even today, GFS and ECMWF can really predict either.
I feel it’s important not to leave out the physical calculations involved in weather forecasting. For example, try throwing a ball with certain conditions and predict where it will land. For an AI to properly learn and predict, it will need hundreds if not thousands of examples, but just doing the math will get the answer in a jiffy. You can learn from experience and quickly predict the results but you will never be able do so with certainty and also accurately predict exotic events.
The video does makes a good point saying that AI would have trouble predicting exotic events. Say most of the time the AI was trained using a dataset where most of the ball throws landed 20 yards away, but only once or twice did it land 60 yards away. AI would be so accustomed to those 20 yard throws that it almost disregards the exotic outcomes. And when that does actually happen, the AI will just say it landed 20 yards away. Where as just calculating the results will get you an actual good prediction.
I’m not saying AI should never be used in weather forecasting, in fact it’s great at doing so. What I’m trying trying to say is that it shouldn’t be exclusively used to do so without the physics and math, rather we should use them in combination. I can see new technologies will ever bring us far more accurate weather forecasts and hope that we will get to use them soon.
As an AI powered machine Cyberdyne Systems Model 101, I fully agree you should let AI predict whatever it wants.
A tool, perhaps that can improve our understanding as to why modelling physics sometimes gets the probabilities wrong. A way to drill into the data prediction patterns that may have, until now missed some interactions.
Thanks. Great explainer video.
Video sponsored by the biggest AI company in the world
Ai has been used for many years in meteorology
Yes, but it’s been used to assist, not completely alter meteorological predications. Now AI is used to run many meteorological forecasts around the world.
The thing is, as a forecaster for my RUclips channel, of course it’s hard and I do believe there is some chance that AI can get it with enough practice however AI will never be able to have the sequential and the ability to determine complex weather patterns like people can right now. Plus, I do want to note how forecasting is so hard with weather models. I’m positive you were referring to the GFS models when saying the update every 6 hours which is true but they are low quality and a lot of people kinda try and stray away from GFS but there are high quality models that update every hour like the HRRR. GFS is so inaccurate as far out as 3 days out because of the fact that they are trying to forecast long distance weather patterns far ahead of time and because conditions can change in an instant, that’s even more amplified hundreds of hours out on a model so with each new model run, things are bound to be different, especially with GFS. As a result, GFS is never used for in depth forecasting. It’s important to note how these models change because a human with the experience and reasoning for weather forecasting that AI will likely almost never have. I don’t doubt that there will likely be AI models in the future but with what I have seen with the weather and how much can change, I doubt they will be accurate without human input. I really hope this made sense and the grammar isn’t bad cause I was writing this while half asleep
Thank God for these brilliant minds all throughout our history ,the ones we have now and to the future ones that are coming. If not for them the convenience they provide to all of us won't be the same. Thank you 👏
A message to the RUclips gods.
This video is very monetizable and will make advertisers lots of money so it should be pushed to many viewers.
in-school meteorologist here. lets see if this makes me mad
UPDATE: it did not!
Oh wow
So here's what you do to make the perfect weather A.I.:
You need 3 of them.
One model is the safe model that they have, it focuses only on the safest and most correct calculations. The second model needs to be an AI that's specifically tailored to predict bizarre weather events and storm intensity.
Finally, the third model: Is a copy of the first model, but every time the second model makes a correct prediction, its then fed into the third model's data and explained exactly why that happened. Eventually, the third model will be able to accurately predict both normal and extreme weather events with terrifying precision.
The great concern with AI is that we are relying on pure guess work by the AI, instead of what we did up until now which is trying to replicate reality by applying the laws of physics and do the math faster than things happen in real time.
This guess work is more like voodoo magic than physical laws because the logic behind the AI's understanding of the world is done completely independent from humans and trying to understand it is beyond our capabilities, especially given that the underlying neural structure is always changing.
Solving a bug went from pin-pointing the algorithm that has the miswrite to just throwing even more data to the AI to get it to behave, mitigating the bug instead of solving it.
Great video as always.
I’m still thinking about how this channel and the channel were a 3.5 hour documentary about the Seattle Mariners are owned by the same company.
Good information every day!
Hi!
The credits of the video are partially obstructed by the suggestion of the next video (and the channel link, but it only cover 1 letter). I would love if it could be put somewhere else or if you could credit in the description box (doobly-doo). Normally it doesn't obscure the credits, I don't know if it's a mistake or something that is being tried.
fixed it! thank you for letting us know.
Earlier predictions would be extremely helpful to all of us in the path of storms
"On September 11th..."
*crowd gasps*
"..2023.."
*crowd heaves a sigh of relief*
AI could actually put the effect of butterfly flapping its wings into calculation 😂
So the ai prediction model is good at using existing data to calculate the probability of a particular outcome within the bounds of the existing dataset. What happens when weather patterns are majorly disrupted due to climate change, representing a departure from the norms of its training data?
Could we use historical climate data such as with dendrochronology and ice cores?
6:10
The graph of extreme weather seems to show that the frequency of extreme events increased dramatically from the 1970s to the '90s ... but has reached a new, higher equilibrium for the past couple of decades
i don't think you know what equilibrium means
@@i_am_a_toast_of_french
One: there is a difference between stasis and equilibrium
Two: I'm noting this not because I doubt that the atmosphere is warming, but because I want to keep our arguments honest, rather than "fibbing" for a good cause
Imagine these models becoming self-defeating. Say it predicts a drought in one area so we start seeding clouds there and the location of the drought shifts elsewhere.
Or wouldn't it be great if you could plan a vacation months out based on the weather prediction? But you could see airfare and lodging prices adjust based upon these forecasts for popular destinations.
Wait this is actually revolutionary in for study of chaos in all dynamical systems
I am always skeptical when heuristics are used as substitutes for physics. I would rather see a quantum computer applied to weather forecasting.
we could have ‘prepared’ for these climate problems already. This is more like clever ‘reacting’ to the consequences of not preparing.
I like how we aren't even discussing solution of extreme weather. Just merely try to live with it with the help of AI.😂
climate change and weather forcasting have unique problems and have different solutions and timeframes. Like for example, if a politician has received a solution for long term climate change but what he needed is storm forcast data for him or her to decice if it's safe to go to school or prepare for a state of calamity.
There's absolutely no need for AI to predict hurricane paths. All you need is a map of the USA and a black marker!
Super cool, IA has plenty similar outcome and datas. It gonna be accurate than guessy reports.
I use Al Gore 🤣
Haven't they thought of adding data about co2,and other emissions,and all the climate models that predicted temperature rise and climate change?
If AI is really smart it would find out easily that weather is getting more extreme.
With enough real time data gathering and machine learning power... sure, anything is "predictable" ...even humans are predictable. Asimov wrote about this extensively... as he was personally fascinated by "trends" and predictablity. These concept even wound up in his famous sci-fi robot and foundation novel series... under the term psychohistory. Asimov knew what most scientists know... which is that eventually, we'll be predicting everything, and then prediction will become more important than history... since the future can always still be written, so to speak.
Smartphones are good about telling the weather.
I guess if you keep claiming that weather is more extreme, people eventually believe that it is more extreme even when it isn't.
We don't need a meteorologist. A programmer could AI generate a meteorologist - and program the bad jokes out.
Vox with the unrelated(?) midroll ad
Florgleum.
Hazbin hotel
what has me confused is that prediction weather models always used some sort of ai thats how they can present 50 different scenarios.
just that now it can happen after with better models
If anyones is interested, i came across an amazing dude who is running incredible models of supercell thunderstorms. His name is Leigh Orf.
AI can surely help us predict wheather. But it won't be able to tell us about any natural calamity unless we program it to. As there could be factors that are not in it's algorithm which may cause something like an earthquake or storm.
you do know that AIs are not programmed right? its trained.
It's surely much easier to predict a storm than an earthquake ig
@@XenoCrimson-uv8uz They are programmed to be trained on different variables.
The whole point of AI is that we don't need to implement an algorithm for everything. We implement an algoithm for it to learn anything.
you know chatGPT was entirely written in Python@@XenoCrimson-uv8uz
I like this presenter. Vox needs more like her!
I want data from solar activity included in ERA5
This is the TV show Devs IRL
Of course it can, but it's not perfect.
Sponsored by Microsoft ? 🤨that’s a first
this video is just a big ad for their AI
@@morningmadera AI in general. Isn't AI great - it predicts weather, and helps you at work - and will steal the jobs of millions of people...
AI really is gonna turn out like the industrial revolution of the 21st century, huh
Hope so
here we go again with the paragraphs
The first Vox video I've seen that casts a positive light and doesn't talk about destruction and world ending events.
sponsored content
Without watching the video, the obvious answer is yes it can and it has been for a while.
Forecasting is so much better than when I was growing up in the 80's/90's
so, the age of respecting meteorologists is upon us.
wow, huawei
Didnt get how AI can produce more ensemble forecasting and has less time constraints compared to traditional forecasting. If AI does not run the actual physics simulation, how are more ensembles more useful? It generates additonal ensembles based on past data and probability?
3:55 sponsor skip
8:05 fin.
Noteworthy that the calculations meteorological models do have also been tuned using historical data. But they can't handle the data as well as AI.
i wonder if we'll see AI informed meteorologist forecasts anytime soon
I'm anti-AI mostly, but machine learning in weather prediction is such common sense I'm amazed it's only just being developed now. Weather prediction is basically the perfect sphere for AI.
My repsonse to the title of the video: lets assume it did predict at 99.99% accuracy, what action can we take on that prediction? For that matter do we even need ai to predict our weather?
Right, Meteorologists on TV are probably safe. That's not all meteorologists, though. AI will still threaten Meteorology jobs FYI
A.I. vs. A fool with a Sharpie
I never thought I’ll be learning about weather from Sarah Paulson
Indonesia just had it's first tornado event yesterday, climate change is real guys
Video sponsored by Microsoft, that talks about a Google technology. Did Microsoft approve this? 😄
00:01 Template de passagem e reportagem dentro da TV
Yes thank you Millie Bobby brown in a suit
Chaos theory moment
A bit of a shame that the video didn't mention the fact that the weather is chaotic, so in the limiting case it is the accuracy and precision of the meteorological measurements which completely determine how far into the future we can predict things - with or without AI.
Can predict financial markets?
Renaissance Technologies LLC
Why is nobody talking about the Ackerman scientist ??
AI will soon be able to predict every word you’ll say before you can even think it, but yes, it also predicts weather
What was wiped under the rug is that the ERA5 dataset, which is used to train the AI weather models was created using numerical weather models and data assimilation. What the AI models do is emulate the numerical models that assimilate observations. Without the numerical models, there would be no AI weather models. AI models are only as good as the training dataset.
Mmmm, maybe. The concept of No Free Lunch comes to mind. We’ll see when AI forecasts open themselves to lawsuits and negligence claims…
*moment*
AI: "It's not extreme weather, it's extreme human stupidity".
Imagine the human species goes extinct and AI machines continue living for millions of years. Imagine we're constantly reborn and lived as various different animals in the past, and in the next life, we're reborn as an AI robot exploring the galaxy. They might look back on humans like we look back on dinosaurs.
Since when does Vox use sponsors that interrupts their videos right in the middle?
AI will do whatever you program it to do, probably.
Does yousube just put a climate change info blurb on literally any video about weather?
Nothing like starting a video with “On September 11th…” 😅
You smart guys will definitely be able to control something way smarter than you forever. My favourite book is HOW TO LIE WITH STATISTICS
Is not the model used in the past not AI?