Couldn’t tell you. However, it’s been the most pleasant summer I can ever remember in my lifetime and we’re below average for tornadoes.
Because we only have weather records for the last century or so, a |“1000 year event” is of necessity an extrapolation from the data we do have.
If the data, when plotted, looks like a Gaussian distribution (“Bell curve”) , then it is reasonable to assume that is what a larger data set would look like, and you can get a good idea of what the standard deviation is, and from there you can work out the likelihood of being X far out on the tail of the curve.
But extrapolation is always risky. The real distribution may not be Gaussian, and the climate is changing, so the old data set may not reflect the current reality.
So, while these claims do reflect the best scientific thinking, the scientists are limited by very incomplete data.
Well, yes, you could ring in reporting bias. However, I don’t think things like the California/Southwestern drought (1200 year), Carolina flooding (500 year?), Katrina, Sandy, the extreme temps seen by much of the US this summer, etc. are things that would have gone unreported pre-Twitter.
Nor does the lowest rainfall ever recorded in East Jesusville, KY make much difference in the overall picture, whether reported only in the East Jesusville Monitor or as a Fox News headline.
It’s not a lot of minor things being reported. It’s an increasing number of increasingly severe things.
But we had an exceptionally mild hurricane season, so I guess it’s all in the interpretation.
[QUOTE=]
If the data, when plotted, looks like a Gaussian distribution (“Bell curve”) , then it is reasonable to assume that is what a larger data set would look like, and you can get a good idea of what the standard deviation is, and from there you can work out the likelihood of being X far out on the tail of the curve.
[/QUOTE]
In what formulation, pray tell, might plotting storm intensity look Gaussian?
I called attention to Hurst’s discovery of power-law statistics for such events, and have been soundly ignored. I guess I forgot to include the buzz word “fractal.”
Even if they did look Gaussian, you still really don’t want to extrapolate out into the tails. There are a lot of distributions, including a lot encountered in the real world, that look like Gaussians but differ significantly out in the tails.
Ever hear of tree rings?
yes, they tell the age of the tree. They do not however act as thermometers. At best you can get an idea of how much water and sunlight the tree consumed.
The Central Pacific is having an all-time record year, with eight major (Cat 3+) hurricanes with six of them being Cat 4’s. The Western Pacific had a really bad year as well, including four Cat 5 Super Typhoons.
Setting aside the debate about climate change …
The definitions of 100- or 1000- year events includes (at least implicitly) a location or area.
There is nothing surprising or confusing or invalid about the USA collectively experiencing, say, ten 100-year floods in a single year. One in northern Florida, one on a different date in southern Florida, another in Oregon, a different one in South Carolina, etc.
The whole point is “what are the odds that this particular location will experience this particular phenomenon in any one-year period?”
To be sure, if we see that collectively these events are recurring at much greater frequency or at many more locations than our statistics expect, we should conclude that our statistics are not valid for predicting the future because something has changed since the historical data was gathered.
But that’s a separate analysis.
Additionally to everything mentioned above, we have more people … lots more people.
I was comparing Hurricane Betsy (1965) to Hurricane Katrina (2005). Although both storms were roughly the same, the death toll from Katrina was better than ten times that of Betsy, and both flooded New Orleans. Certainly the dynamics were different but also there’s just a hell of a lot more people living on the Gulf Coast.
Charleston, South Carolina, is subject to Category 5 Hurricanes … this recent flooding is to be expected there, and it’s to be expected anywhere along the Atlantic and Gulf Coasts … it’s a non-zero probability.
I deal with this questions a lot in my work. I now work in Property Insurance. We get argumentative clients which think we propose excessive risk from nat haz events to them. These events are typically are grouped in these <X year bands.
The key point raised by several above is that this:
“what are the odds that this particular location will experience this particular phenomenon in any one-year period?”
But people very often misapply the basic mathematics attached to the probability, depending on how the local jurisdiction has applied them. A good summary that i copied from our publicly available data sheets is below. I think its all-but-mirrored on wikipedia as well.
The mean return period of an event (e.g., damaging ground motion) is the average number of years between successive events. A mean return period of 500 years does not imply that successive events will be exactly 500 years apart. Nor does it imply that there is 100% probability of its occurrence in a 500-year period.
(Compare rolling a 6-sided die. There is a one-in-six chance of rolling a “3” (i.e., a “return period” of 6)
However, in six rolls of the die, it is possible that a “3” will not be rolled and it is also possible that a “3” will be rolled more than once.) The following relationship gives the probability of an event in a given period:
p= 1-exp(-t/T)
where, P is the probability that an event of mean return period T will occur at least once in a time period t.
The probability of a 500-year event occurring at least once in 50 years is:
P= 1-exp(-50/500)= 0.0952 (9.5%).
The probability of a 500-year event occurring at least once in 500 years is:
P= 1-exp(-500/500)= 0.632 (63.2%).
Put another way, there is about a 37% chance that a 500-year event will not occur in a given 500-year period. Assuming independence of seismic events, the probability of a 500-year event occurring at least twice in
500 years is: 0.6320 x 0.632 = 0.4 (40%).
Maps provide the mean return period of damaging ground motions at a site. They do not provide the mean return period of damaging earthquakes below the site. To distinguish between the two, refer to the sketch below (confidential).
anyway my stock standard response, if it within probabilistic range of a major loss - "yep, you’re right, it’s pretty unlikely to occur in area x. Just like all insurance events are pretty unlikely. So why do you have insurance then? " which is a good logical argument.
Nitpick: the probability isn’t addative. Assuming that the probabilities are normally distributed on a Guassian distribution function, for a single tailed three sigma (99.87%) threshold of occurance the probability per annum of exceeding a once in one thousand year event is ~0.66%. In terms of the correlation between stations, they cannot be condidered independent, e.g. the storm condition that will cause an exceedance at one is probably not so localized that it will not cause an exceedance at another. In general, when measuring a large scale system like weather, you would look at a weighted aggregate of a cluster using some kind of kernel density estimate rather than a localized single event exceedance.
As Chronos and others have noted, assuming that weather events are distributed on a Gaussian distribution is just a model, and we know what George E.P. Box says about models. Many real world distirubtions are approximately Gaussian, or logaritmic Gaussian in the main body, but the distributions at the extremes (tails) require shape parameters or even more complex spectral models to precisely represent them, and this requires a large body of data. The fact that complex systems like Earth’s climate are never in stable equilibria complicates the application of any predictive modeling, e.e. you have to be able to discern trends that change your time-specific predictions. The trend projections themselves are proabilitistic in nature, hence why there are such wildly divergent estimates of the significance of global climate change (but there is no debate that an observable trend shift in climate behavior has occurred over the last century)… The “1,000 year weather event” is just a cursory estimate of likely we would expect an event of that threshold would occur no more than once over a 1000 year period for which the aggregate measurements represent a typical distribution.
Stranger
Sure, but Hurricane Katrina isn’t an example of anything extreme, except for incompetence at the local and state levels in Louisiana, and within the Corps of Engineers. Katrina itself was nothing special- it made landfall as a Category 3 storm, and was down to a 1-2 in terms of wind by the time it hit New Orleans. What fucked the city was the failure of the levee system, not the intensity of the storm itself; other storms like Ike have had higher storm surges, and others have had higher winds, yet none of them had the degree of loss of life that Katrina did.
I think a lot of people tend to equate financial damage with actual intensity of the weather pattern- this isn’t so at all. The two most expensive hurricanes recently were Sandy and Katrina, and neither of them were anything particularly special at all. Katrina’s main damage was due to a wretched levee system that failed, and Sandy’s was for the most part, due to it hitting an utterly unprepared part of the country. I suspect had it hit Florida or the Gulf Coast, the story would have been quite different.
A “storm” and a “flood” are not comparably measured events, and should not be thrown into the same basket of nomenclature. The concept of a xxx-year event began with floods, for the purposes of the security of land use in a given locality. A flood is very easy to quantify according to a single measurement – the height of the water, without which, there is no flood. A “storm” can include so many meteorological variables (wind speed, rainfall, temperature, duration,time of year, etc.) that an extreme of any one of them doesn’t necessarily define a “storm” in totality.
Agree with almost all you said.
Quibble: Katrina damaged a lot more than just New Orleans. Because although it wasn’t extra windy or extra rainy or particularly extra storm-surgy, it was friggin’ extra-specially huge. And as such it devastated a couple hundred miles of Gulf Coast, not 20 or 40 miles as is typical.
NO got all the headlines, but the total seriously damaged area was larger than all of England.
This flood event in South Carolina has been (arbitrarily) set at having a 0.1% chance of occurring in any given year. This translates into lay-speak as a “1,000 year flood event”. I’m extending the definition of a “100-year flood event”, which is a flood event having a 1% chance of occurring in a given year. We’ve only a couple hundreds years of hard data, these probabilities are more guesswork than anything.
I’ll admit to not having knowledge about how these event estimates are formulated, but the notion of taking the period over which an event is expected to occur not more than once and dividing it into 100% to get a per annum probability of occurance, or vice versa, is completely wrong in even a naive sense. Applying a statistial distribution to even a modest population of historical data, however, can give a credible estimate and is the the typical way of estimating a probability of occurring over a specified time interval, assuming that the phenomenon is normally distributed and understanding that events at the extremes will certainly deviate from the normal.
Stranger
When you get into the 1 in 100,000 year event territory, you have to start including things like 800 foot tsunami. That begins to stretch the definition of “weather”.
Here’s a 1720 foot wave back just in 1958.
Yes. The Boulder floods of 2013 were only a 100 year flood, due to it being at the end of a dry summer, very dry soil, and the creeks being at a trickle with the snowmelt long since over. It was, however, a 1000 year rain event, with 19" of rain falling in five days in a month that averages about 2" of rain (all month.)