AI Can Predict If It Will Rain In Two Hours' Time

Artificial intelligence can tell whether it is going to rain in the next two hours, research suggests. The BBC reports: Scientists at Google-owned London AI lab DeepMind and the University of Exeter partnered with the Met Office to build the so-called nowcasting system. Traditional methods use complex equations and often forecast for only between six hours and two weeks' time. The AI system can make more accurate short-term predictions, including for critical storms and floods. The system learned how to identify common patterns of rainfall, using UK radar maps from 2016 to 2018, was tested on maps from 2019 and found, by 50 Met Office meteorologists, to be accurate in 89% of cases. The research, published in the journal Nature, found: "Meteorologists significantly preferred the [AI] approach to competing methods."

Not very impressive

[nagora]

Give me a high building to stand on and I can do that. Radar maps just make it trivial.

Re:

[Joce640k]

Classic: https://www.youtube.com/watch?... [youtube.com]

Re:

[Joce640k]

I guess that might need some explanation over the other side of the pond:

https://www.youtube.com/watch?... [youtube.com]

Re:

[BAReFO0t]

That actually needed explanation over here too, unless you're from GB I guess.

But yeah, that's the whole problem with chaotic systems. There will always be a remainder that's unpredictable. Because there are too many factors and you can't look at it with a coarser resolution and averaging out between the points without missing some high frequency parts of the picture.
Same problem as standing on a mountain and to save resources, avoid walking the entire valley to check for local minima and maxima, and just t

Re:

[tlhIngan]

Yeah, a 2 hour prediction isn't actually very impressive.

It wasn't that long ago when a 5 day prediction was about all you could stretch it - with the real forecast being about 3 days and the last 2 a wishful thought.

The late 90s then extended it to 7 days with about 5 days being truthful. These days we can do it to about 10 days with 7-8 days of that being pretty spot on.

Given this, a 2 hour forecast isn't that hard. Even our general models can compute the hour-by-hour probability of rain.

Re:

[gosso920]

My mother's arthritis lets her know it's going to rain three hours ahead of time.

Re:

[sabbede]

Arthritis is the best barometer.

Re:

[arglebargle_xiv]

Was just going to add something similar, my ninety-year-old neighbour can do this without any AI, he'll go out onto the deck, look at the sky, feel the air a bit, and then say "yep, looks like we're gunna git a bit of rain right soon now". Doesn't need any AI or computing to do it.

Re:

[CubicleZombie]

Give me a high building to stand on and I can do that. Radar maps just make it trivial.

Or automate that process with a weather rock [wikipedia.org].

Re:

[dogsbreath]

My dog crawls behind the toilet hours before a storm hits.

Re:

[holophrastic]

I'd like to know exactly when it became acceptable to have the words "Traditional methods" refer to "complex equations".

I don't need a high building. High buildings are expensive to build, and I don't want them blocking my view.

Instead, I flock-source my weather report to a higher power, for whom the weather means life and death.

I have a few bird feeders in the backyard. Acrobatic squirrels, gravity-defying chipmunks, ballpark-peanut-loving bluejays, rotisserie morning doves and the coopers hawk farming t

Re:

[dsgrntlxmply]

I used to live across from a school playground and town ball fields, around 35 km from the coast. Winter storms were often preceded by large numbers of seagulls arriving from the coast and hunkering down on the fields.

Re:

[holophrastic]

I need to apologize profusely in-advance here. I am a man who sleeps with a professional lawyer and amateur ornithologist; my training is very strict: there is no bird classified as a "seagulls". They are simply "gulls".

Yeah, I know, I don't care either. She's in my head!

LMFTFY

[sTERNKERN]

AI Can Predict If It Will Rain In Two Hours' Time With around 89% probability

How impressive?

[dsvilko]

Well, if it actually rains only 11% of the time and the 'AI' never predicts that it will rain, it's still 89% accurate so I am finding it difficult to judge how impressive that result is.

Re:

[Anonymous Coward]

Worse - this is the UK, where it never stops raining.

Yeah, my weather radar can too.

[BAReFO0t]

My weather app already does that.

But go on peddling your S(hitty)ANN scam. I will never call it "AI" no matter how much you repeat it. Because it isn't AI. And this is clearly not a news site anymore.

Re:

[mu22le]

My weather app already does that.

Honest curiosity, what app is that?

Re:

[thegarbz]

It's called the "BAReFO0t" app. BAReFO0t being German for "I only ever speak out of my arse and comment before reading TFA".

UK?

[Bert64]

But this is in the UK... You can always say with certainly that it's going to rain.

Not any more

[Viol8]

Climate change has affected our weather just like the rest of the world. Long hot dry spells seem to be the norm during the summer now.

Re: Not any more

[OneSmartFellow]

By long hot dry spells, I presume you mean all afternoon it doesn't rain and the temperature is almost comfortable.

Re:

[Viol8]

No, I mean like in 2019 when it hit almost 39C in Kent and it didn't rain for a month. The cliched view of england being a soggy place are over - its climate has been fucked just like rest of the world. Now its extreme (for northern europe) heat following by torrential rain. Rinse and repeat.

Accuracy should not be used in this way

[Flownez]

Recall and Precision would be better metrics to gauge performance of the classification algorithm than with accuracy, or even better - an F-Score [wikipedia.org].

Weather rainfall prediction is an imbalanced classification problem for which Accuracy is a poor performance metric.

Re:

[Viol8]

"Weather rainfall prediction is an imbalanced classification problem for which Accuracy is a poor performance metric."

What a load of statistical buzzword bullshit. Accuracy with weather forecasts is the ONLY measure that matters you moron. If you don't believe that go out onto the ocean in a small boat with an approaching storm that may or may not come your way and get back to us.

Re:

[Flownez]

Maybe you can help my puny, moron mind with this issue then:
If the system categorises 100% of all time as a potential rain event, then we would accurately predict 100% of all rain events, meanwhile we would have incorrectly classified a whole bunch of non rain events.

FTA: "The system learned how to identify common patterns of rainfall, using UK radar maps from 2016 to 2018, was tested on maps from 2019 and found, by 50 Met Office meteorologists, to be accurate in 89% of cases."

So if it was raining 89

Re:

[nevermindme]

There is no temporal precision term when measuring a binary such as "OBSERVERD RAINING when PREDICTED RAINING". The temporal precision is the prediction of wind speed at the storm cloud level, with radar data, if rain starting at 5:05 vs 5:35 at two hours out was not spot on with modern doppler radar I would be shocked.

So predicting if a set of clouds contiue to drop rain is a processed reflectivity of the cloud vs processed reflectivity of the lost of water, the rainm, below it. Quite honestly on a two h

Re:

[Viol8]

"Out of interest, and for the sake of us peons, how does one with a superior mind such as yours discern a descriptive concept from a 'bullshit buzzword' anyway"

When someone like you spouts a load of words you stole from a thesaurus in order to sound like you have a coherent argument when in fact you're talking out your arse.

Re:

[Flownez]

"Out of interest, and for the sake of us peons, how does one with a superior mind such as yours discern a descriptive concept from a 'bullshit buzzword' anyway"

When someone like you spouts a load of words you stole from a thesaurus in order to sound like you have a coherent argument when in fact you're talking out your arse.

Oooh, this is like Christmas! I cannot wait for you to read my other response to your post!

I'm giving it a slim chance of you responding, maybe 3%?
...Lets see how accurate my prediction is :D

Re:

[Viol8]

Aww diddums, did someones chain get yanked so you have to fire off indignant replies to lots of my posts? And probably mod some of them down using another account. Poor ickle baby waby , why not run along to your safe space poppet and cuddle the therapy teddy :)

Re:

[Flownez]

Seriously, for your sake: you know there's no delete function here right?

Re:

[Viol8]

I think that should be more of a worry for you pal.

Re:

[Flownez]

That tracks ;)

The BBC stated 89% Accuracy is bogus

[Flownez]

Sometimes in life, on a cloudy day when everything is shitting you the clouds clear, and the universe precipitates validation unto you. It would seem that for me today is just such a day...

So having just read the Nature article [nature.com] referenced by the BBC article it just so happens to be the actual case that the main measure for success in meteorological forecasting is a measure called Critical Success Index (CSI) which, wait for it, uses RECALL and PRECISION

According to the literature referenced in

hey.....

[Vomitgod]

basement kids know better scientists....
fuck trying anything different... they know better

This is child's play

[Miles_O'Toole]

Of course I can predict whether it will rain in two hours. I simply wait to see how many of you left your umbrellas at home, then order my cloud-seeding drones to act accordingly.

You needn't bother to plead. It's far too late for that.

Something is off...

[Junta]

Based on my interactions with meteorologists, their weather models are precise and they seem good at providing enough predictions to power alerts like "It's going to rain in 25 minutes" or "rain stops in 15 minutes".

In fact, this meeting was specifically about AI, and the meteorologist consensus was that they could have seen AI as providing value if the scientific community hadn't already worked out the math/physics to do predictions, but since we had the more precise body of scientific meteorology at hand and instrumentation that already distilled the key metrics into computationally friendly form, the AI demonstrations were always underwhelming. The AI models were unable to augment the models nor compete with them, so no one was able to see how AI could apply to this specific field.

AI continues to excel at characterizing unstructured inputs (human text, imagery and audio) into orderly data. That sort of problem is common, but generally not applicable to modern meteorology, last members of that community told me.

Re:

[lazy genes]

In the winter of 2011, I used a newly discovered variable to make 30-day temperature predictions. I was able to beat the historical average by an average of almost 2 degrees per day. I brought my data and explained my experiment to the national weather service in Duluth MN. I'm still waiting for an answer. It is getting harder by the day to believe we are an intelligent species.

Whoopie

[Snotnose]

My knees have been doing that for years.

Yawn

[Ol Olsoc]

So can my knees and right ankle.

So can I

[OneHundredAndTen]

I don't always get it right though. But neither does this AI.

My Shelty Izzy can do that too.

[Buffalo Al]

Wow so an AI can look at the radar with wind flow direction and velocity. do some algebra and predict when the rain will arrive? Stunning technology! My dog Izzy just goes outside and then quickly returns to be inside because she doesn't like rain. Can that AI predict when my wife is five minutes away from the house? Izzy sits at the front door about five minutes before she gets home from work. Now when an AI can do that I'll be impressed. Hey Google when is Evie going to be home? "I'm afraid I don't know

Any updates? What did the AI say?

[mark-t]

This headline is more than two hours old at the time I am posting this... what did the AI eventually say?

Useless

[sentiblue]

What good does it do to predict rain 2hrs ahead? Most outdoors activities are planned much farther ahead. I guess this prediction makes allows you to shutoff your automated irrigation to conserve water. But if you're not home that probably doesn't matter anyway.

Big deal.

[Oligonicella]

My right knee has been doing that for almost forty years.

So Do Meteoroligists

[lsllll]

And I bet both AI and meteorologists make mistakes, perhaps at the same rate.

"AI can predict..." I can also predict

[ayesnymous]

But that doesn't mean my prediction will be correct.

89% accurate predictions aren't necessarily

[aRTeeNLCH]

If it rains less than half the time, getting a 50% accurate prediction is trivial. A quick search taught that in the Netherlands, it actually rains (precipitation, really) 7% of the time. Predicting that it won't rain in the next 2 hours will then yield a 93% accuracy over time. Which is also totally useless.