Hacker News new | past | comments | ask | show | jobs | submit login

I found it a bit clickbaity. The title is correct to the letter but not really to the spirit, because down at the bottom there is:

> Temperatures are known to vary wildly over Antarctica, and massive swings are common. Contrasting with this warm spell over eastern Antarctica, the South Pole observed just observed its coldest April to September period on record last year, with an average temperature of minus-78 degrees (minus-61 Celsius).

But to be sure "The heat dome was exceptionally intense, five standard deviations above normal" is nuts.

We should rename Antarctica to something like "Crazy continent".




When you start calling five sigma events "click-baity" you may find others asking legitimate questions about your ability to assess what is significant.

Encountering a headline that tries to get people to read the article is a roughly zero-sigma event. That's the purpose of a headline, and has been for about 150 years. Retailers also sell things for more than they paid for them and supply chains are hard. If you can understand what a five sigma event is, you can understand these things as well.


I must agree on the "click-baity" title.

If it wasn't for the article being posted on HN, I would never have clicked on the title, as it includes something like "Scientists are flabbergasted".


Uhh last I checked 5 standard deviations was the definition of a statistically significant difference.

To me it sounds like a realization that climate changes will have drastic and possibly chaotic effects.


Do you know the temperature probability distribution in Antartica? Not all distributions are normal and not all even have a standard deviation.


The probability that an observation is beyond 5 standard deviations is < 1/25 = 4%, according to Chebyshev's inequality, provided the variance of the distribution is finite (which, as you say, is not guaranteed, eg. for Cauchy distributed random variables).

https://en.wikipedia.org/wiki/Chebyshev's_inequality

https://en.wikipedia.org/wiki/Cauchy_distribution


An observation bound below by 5 standard deviations is at best a one in a million on a standard normal distribution.


Yes, for a normal, which is known to have extremely thin tails (decaying with a squared exponential!). Chebyshev's bound holds for any possible distribution (with finite variance).


Not squared exponential (that's just an exponential), but exponential of a square, even. Anyway.


If the observed variable is “the mean temperature in a given day”, 4% can be expected to occur relatively frequently.

Anyway until we know the probability distribution, and my intuition says that it’s definitely not Normal for wide periods of time, we can’t say if it’s significant or not.


Five standard deviations is not four percent, it is in the 1 in 3 Million range.


Yes, for a normal, which is known to have extremely thin tails (decaying with a squared exponential!). Chebyshev's bound holds for any possible distribution (with finite variance).


You’re the only one introducing the normal distribution here.


When discussing standard deviations, what meaning is there for non-normal distribution?

https://stats.stackexchange.com/questions/108578/what-does-s...


4% would imply that we should expect about two weeks of such weather every year, in total.

Which would make this article somewhat clickbait — nothing more than “wowza, July 15th is another hot one in California!”

Since it’s currently summer, in the Southern Hemisphere.


That's a bound not an expectation.


Standard deviation is defined by the distribution under question.


I think you're hearing "massive swings" and jumping to the conclusion that 5 std dev is somehow normal-ish behavior for the region. It's not. At all. By definition.


Temperature is bounded to a positive values and it is sufficiently reasonable to at least try to look at temperature distribution as a log-normal distribution, where logarithms of temperature are distributed normally.

[1] https://en.wikipedia.org/wiki/Log-normal_distribution

From this vantage point, bigger positive swings are more likely than bigger negative ones - because positive swings are in the range exp(logmean)..infinity and negative ones are in range 0..exp(logmean).

Take a look at second picture in Wiipedia article. The value with CDF(x)=0.75 is about four times bigger for sigma=1 curve than value for CDF(x)=0.25.

And I guess that distribution of temperature in Antarctica is more close to a log-normal distribution than distribution of temperature in, say, Houston or Cancun.


That’s absolutely right.

But as the article itself says, we don’t know if this is the start of some trend or if this is a never previously observed fluke.

Climate change is about regression to a mean that is itself higher than the previous mean. But with or without human contribution, weather events are still going to regress to a mean.


> But as the article itself says, we don’t know if this is the start of some trend or if this is a never previously observed fluke.

...That's not how statistics works.

A 2 sigma deviation can be a never previously observed fluke. A 5 sigma variation ... that means that the model is wrong. Observing a 5 sigma variation means that you need to stop, and consider why your model is incorrect.

> Climate change is about regression to a mean that is itself higher than the previous mean.

That's what we've been assuming, but this result upends that assumption.

We talk about a new mean that's 5F higher than the old mean. But an event that's 70F higher than the current mean is also 65F above the the 'new' mean. And that throws out the possibility what we're talking about is a new mean that's higher than the old mean with the same standard deviation.

The brave new world we're walking into doesn't just have a higher average temperature, it also has a significantly higher standard deviation.

And this is potentially catastrophic. If Antarctica is, on average, 5F warmer, but is not prone to significant deviations, then Antarctica stays frozen. This is our old model; maybe the new extremes are a little bit more extreme, but Antarctica is still so cold and the rare high temperature events are mild enough that it's still gonna be super cold, and all the ice that's held up in the Antarctic ice sheets stay there. But if we have wild swings like this one, then our old models are quite simply wrong. We now need to start to consider models where Antarctica melts, and now we need to consider a 190 foot increase in sea level over the next century.

AFAIK that means that a majority of Earth's population will be displaced. NYC, London, Tokyo, Florida, Rio, all gone.


I agree that if you don't like getting 5 sigma events against your model's forecast, sure, change your model.

But I'm curious whether the model in question is even looking at 5-day events, or more at monthly/annual/decadal averages. This may not be 5 sigma at all at those timescales.

And, it sure looks like it's not headed to a new mean 5 degrees Fahrenheit higher: https://www.timeanddate.com/weather/antarctica/vostok-statio...

Please don't get me wrong. Human-generated greenhouse gas emission increase and carbonization of the planet's atmosphere are big problems, that will contribute to overall global temperature average rises. It's just that this is not that.

This is more like when the CPU spikes momentarily by +50% on your production observability monitors. Panicky teams might react to it. Mature teams might see it as transient cloud (heh) behavior and know not to even bother alerting. Immature teams will tune it out as alert fatigue.

We don't want to give the whole planet alert fatigue about extreme weather events. We want to focus on reducing the long run change to global average temperature.


We at least have day-to-day written history for thousands of years, in human population zones. We have oral beyond that.

We barely have 30 or 40 for a place like antartica.

We can detect gross climate, and not any real weather in fossils.

For example, no fossil record would capture a chinook, a common occurance in Alberta.

https://cwf.ca/research/publications/five-facts-about-chinoo...

In 1962, Pincher Creek saw record temperature rise of 41 °C from -19 to 22 °C in just an hour.

One hour!

And of course, we have only been monitoring chinooks with detail for maybe a 150 years.


It is however fairly predicable that folks will comment on unusual (to them) weather behaviour[0]. First thought - "I guess Leo has never experienced a chinook". Go to school when it's -25°C; go home for lunch at +5°C [1].

[0] https://calgaryherald.com/entertainment/celebrity/that-awkwa...

[1] circa mid 60s




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: