AI could help to forecast extreme weather

[GordMay]

AI could help to forecast extreme weather

Two algorithms [1 & 2] have demonstrated the potential for artificial intelligence (AI) to make faster, and more accurate, weather forecasts.

Pangu-Weather [1], trained on 39 years of weather data, can retrospectively predict global temperature, pressure, and wind speed, a week in advance. It’s 10,000 times faster, and no less accurate, than making predictions on the basis of an understanding of physics.

NowcastNet [2] combines deep-learning methods, with physics equations, to give local predictions of heavy rain, up to three hours in advance. The meteorologists who tested NowcastNet judged it to be better, in most cases, than four leading ‘nowcasting’ systems.

AI could help make better weather prediction cheaper, and more widely available, and could one day attempt other complex predictions, such as the spread of wildfire smoke.
Human oversight will be key, note atmospheric scientists Imme Ebert-Uphoff and Kyle Hilburn, in the Nature News & Views article: the systems can struggle with extreme events, which are more likely to occur in a changing climate.

“The Outlook for AI Weather Prediction” ~ by Imme Ebert-Uphoff & Kyle Hilburn
➥ https://www.nature.com/articles/d41586-023-02084-9

“Why AI Could Save Us From the Next Deadly Hurricane” ~ by Tony Ho Tran
➥ https://www.thedailybeast.com/can-ai...nado-and-flood

[1] “Accurate medium-range global weather forecasting with 3D neural networks” ~ by Kaifeng Bi et al
➥ https://www.nature.com/articles/s415...ec6LjK3w3ug%3D

[2] “Skilful nowcasting of extreme precipitation with NowcastNet” ~ by Yuchen Zhang et al
➥ https://www.nature.com/articles/s415...yefWVlfj120%3D

[Jamme]

Very interesting, thanks Gord.

[GordMay]

AI Could Forecast Death:

Researchers say “Chartwatch” an AI early warning system, launched in 2020, at St. Michael's Hospital, is saving lives.

A year-and-a-half-long study [1], on ‘Chartwatch’, published in the Canadian Medical Association Journal, found that: use of the artificial intelligence [AI] system, led to a 26 per cent drop, in the number of unexpected deaths, among hospitalized patients.

‘Chartwatch’ measures about 100 inputs, from a patient's medical record, that are currently routinely gathered, in the process of delivering care. So, a patient's vital signs, their heart rate, their blood pressure … all of the lab test results that are done every day.
Working in the background, alongside clinical teams, the tool monitors any changes in someone's medical record, and makes a dynamic prediction, every hour, about whether that patient is likely to deteriorate in the future.

That could mean identifying someone getting sicker, or requiring intensive care, or even being on the brink of death, giving doctors and nurses a chance to intervene.
In some cases, those interventions involve escalating someone's level of treatment, to save their life, or providing early palliative care, in situations where patients can't be rescued.

In either case, ‘Chartwatch’ appears to complement clinicians' own judgment, and leads to better outcomes, for fragile patients, helping to avoid more sudden, and potentially preventable, deaths.

[1] “Clinical evaluation of a machine learning–based early warning system for patient deterioration” ~ by Amol A. Verma, et al
➥ https://www.cmaj.ca/content/196/30/E1027

“... Interpretation: Implementing a machine learning–based early warning system in the GIM unit was associated with lower risk of non-palliative death than in the pre-intervention period. Machine learning–based early warning systems are promising technologies for improving clinical outcomes. ...”

See also: “GEMINI” - Harnessing hospital data to improve healthcare.
GEMINI is Canada’s largest hospital data sharing network, for research and analytics.
➥ https://geminimedicine.ca/

[Kettlewell]

The key word here is "potential." AI shows potential to do a lot of things well, but so far you still need real human intelligence for quality control and judgment. How many of us would trust an AI system with our lives? Which is what we will be doing soon, and in some cases already are doing.

[Johan Leopard51]

A professor once explained to me that in simple terms what weather “forecasting” does is look at what happened the last time a similar set of weather conditions was present, and present that. Different models look at and weigh differently, different weather data sets and can focus differently. It is why the further into future, the less reliable.

Most everything nowadays is labeled “AI” because it is so hyped. My CCTV system had movement detection and alarms that brought that screen up in guard room, now it boasts Artificial Intelligence Surveillance. What many new entrants do have nowadays is extraordinary computing power for cheap-cheap from (loss-leading) disruptors chasing unicorn valuations.

[GordMay]

Originally Posted by Johan Leopard51

A professor once explained to me that in simple terms what weather “forecasting” does is look at what happened the last time a similar set of weather conditions was present, and present that. Different models look at and weigh differently, different weather data sets and can focus differently. It is why the further into future, the less reliable...

I suspect that you may have slightly misunderstood your professor's simplified explanation.

Modern [numerical] weather models use systems of differential equations, based on the laws of physics, including fluid motion, thermodynamics, radiative transfer, chemistry, and etc.
Ensemble forecasts use multiple simulations, each with a slight variation of its initial conditions, and with slightly perturbed weather models.

Older weather forecasting methods usually relied on [observed] pattern recognition. For example, it was observed that, if the sunset was particularly red, the following day often brought fair weather. This experience accumulated over the generations to produce weather lore. However, not all of these predictions prove reliable, and many of them have since been found not to stand up to rigorous statistical testing.

The weather is a complex, and chaotic, system, with lots of variables. Small errors, in the initial conditions of a forecast, grow rapidly, and affect predictability ["butterfly effect"*].
Furthermore, predictability is limited by model errors, due to the approximate simulation of atmospheric processes, of the numerical models.
The weather does obey the laws of physics, and every change in the weather has a cause. The problem is that, since there are so many possible causes, we can’t know about them all.

* If we don’t know what’s going on, in the atmosphere, and on the Earth’s surface, down to the detail of a butterfly flapping its wings, now, we can’t hope to know, how that’ll affect the weather, in a week’s time. The possible range of consequences grows with time, and consequently, the ability to accurately forecast the weather will decrease with time.